| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 | |
| ; RUN: llc -mtriple=amdgpu9.00-amd-amdhsa < %s | FileCheck -check-prefixes=GFX9,GFX900 %s | |
| ; RUN: llc -mtriple=amdgpu9.0a-amd-amdhsa < %s | FileCheck -check-prefixes=GFX9,GFX90APLUS,GFX90A %s | |
| ; RUN: llc -mtriple=amdgpu9.42-amd-amdhsa < %s | FileCheck -check-prefixes=GFX9,GFX90APLUS,GFX942 %s | |
| define void @v_shuffle_v2bf16_v8bf16__u_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX9-LABEL: v_shuffle_v2bf16_v8bf16__u_u: | |
| ; GFX9: ; %bb.0: | |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX9-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> poison | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v1, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v2, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v3, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX9-LABEL: v_shuffle_v2bf16_v8bf16__8_u: | |
| ; GFX9: ; %bb.0: | |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX9-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v1, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v2, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v3, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_u(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 poison> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v4, 16 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v5, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v5, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v4, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v5, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v5, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v5, 16 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v5, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v5, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v5, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v5, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v5, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v6, 16 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v7, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v7, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v6, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v7, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v7, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v7, 16 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v7, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v7, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v7, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v7, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v7, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__15_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__15_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__15_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__15_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v0 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v0 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v0 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> zeroinitializer | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v0 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v0 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v0 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v4, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v4, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v4, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v4, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_0(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v4, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v5, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v5, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 0> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v0 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v0 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v0 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v0 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v2, v0 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v2, v0 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v0 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v0 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v0 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v0 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v0 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v0 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v0 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v4, v0 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v4, v0 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v4, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v4, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_1(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v4, v0 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v5, v0 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v5, v0 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 1> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v1 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v1 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v1 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v1 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v1 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v1 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v4, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v4, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v4, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v4, 16 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v4, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v4, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_2(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v5, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v5, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v5, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 2> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v1, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v1 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v1 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v1 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v1, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v1 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v2, v1 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v2, v1 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v1 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v1 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v1 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v1, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v1 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v1 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v1 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v4, v1 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v4, v1 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v4, v1 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v4, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v4, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v4, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_3(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v5, v1 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v5, v1 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v5, v1 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 3> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v2 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v2 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v2 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v2 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v4, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v4, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v4, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v5, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v5, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v4, 16 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v5, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v5, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v5, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v6, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v6, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v5, 16 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v6, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v6, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_4(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v6, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v7, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v7, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 4> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v2, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v2 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v2 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v2 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v2 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v2, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v2 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v2 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v2, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v4, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v4, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v4, v2 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v5, v2 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v5, v2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v4, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v5, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v5, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v5, v2 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v6, v2 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v6, v2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v5, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v6, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v6, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_5(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v6, v2 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v7, v2 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v7, v2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 5> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v3 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v3 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v3 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v3 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v3 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v3 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v4, 16 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v4, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v4, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v5, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v5, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v5, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v5, 16 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v5, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v5, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v6, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v6, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v6, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v6, 16 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v6, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v6, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_6(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v7, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v7, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v7, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 6> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v3, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v3 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v3 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v3 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v3 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v3 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v3 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v3 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v2, v3 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v2, v3 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v3, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v3, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v4, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v4, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v4, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v5, v3 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v5, v3 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v5, v3 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v5, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v5, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v5, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v6, v3 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v6, v3 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v6, v3 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v6, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v6, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v6, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_7(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v7, v3 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v7, v3 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v7, v3 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 7> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX9-LABEL: v_shuffle_v2bf16_v8bf16__u_8: | |
| ; GFX9: ; %bb.0: | |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX9-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v1, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v2, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v3, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, s4, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, s4, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, s0, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX9-LABEL: v_shuffle_v2bf16_v8bf16__8_8: | |
| ; GFX9: ; %bb.0: | |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX9-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v0, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v0, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v0, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_8(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 8> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v1 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v2 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v2 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v2 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v3 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v2, v4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v2, v4 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v4, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v4, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v4 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v4, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v4, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v4, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v0 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v0 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v0 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v0 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v2, v0 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v2, v0 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v0, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v0, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_9(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v0 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v0 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v0 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 9> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v1 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v1 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v1 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v4, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v5, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v5, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v4, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v5, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v5, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v5, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v5, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v5, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v5, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v5, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v5, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v1, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v1, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v1, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_10(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 10> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v1, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v2 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v3 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v3 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v3 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v3 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v3 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v2, v5 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v2, v5 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v4, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v5, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v5, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v5 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v5 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v5 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v5, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v5, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v5, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v1 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v1 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v1 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v1, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v1, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v1 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v2, v1 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v2, v1 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v1, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v1, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v1, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_11(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v1 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v1 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v1 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 11> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v2 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v2 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v4, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v4, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v4, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v4, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v4, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v4, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v4, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v4, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v4, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v4, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v5, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v6, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v6, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v5, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v6, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v6, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v6, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v6, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v6, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v6, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v6, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v6, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v2, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v2, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v2, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_12(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 12> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v2, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v3 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v4 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v4, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v4, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v4 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v4, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v4, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v4, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v5 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v2, v6 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v2, v6 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v5, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v6, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v6, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v6 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v6 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v6 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v6, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v6, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v6, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v2 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v2 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v2 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v2 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v2, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v2, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v2, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v2, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v2, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_13(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v2 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v2 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 13> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_lshlrev_b32_e32 v0, 16, v3 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_lshlrev_b32_e32 v0, 16, v3 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_lshlrev_b32_e32 v0, 16, v3 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v4, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v5, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v5, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v4, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v5, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v5, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v5, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v5, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v5, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v5, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v5, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v5, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v6, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v7, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v7, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v6, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v7, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v7, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v7, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v7, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v7, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v7, v3, 16 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v7, v3, 16 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v7, v3, 16 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v0, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v0, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v0, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v1, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v1, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v1, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_alignbit_b32 v0, v3, v2, 16 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_alignbit_b32 v0, v3, v2, 16 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_alignbit_b32 v0, v3, v2, 16 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_14(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x5040100 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x5040100 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 14> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__u_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__u_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__u_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__u_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v3, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__0_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__0_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__0_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v5 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__0_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v5 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__1_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__1_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v5, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[1:4] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v4, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v5, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__1_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v5, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__1_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v5, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__2_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__2_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v5 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__2_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v5 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__2_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v5 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__3_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__3_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[2:5] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v5, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__3_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[2:5] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v5, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v6, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__3_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v6, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[2:5] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v5, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v6, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__4_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__4_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v6 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__4_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v2, v7 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__4_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v2, v7 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__5_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__5_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v7, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[3:6] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v6, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v7, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__5_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v7, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__5_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v7, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__6_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__6_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v3, v7 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__6_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v3, v7 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__6_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v3, v7 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__7_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__7_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: v_perm_b32 v0, v7, v3, s4 | |
| ; GFX900-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__7_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: v_perm_b32 v0, v7, v3, s4 | |
| ; GFX90A-NEXT: global_store_dword v8, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__7_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v8, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v7, v3, s2 | |
| ; GFX942-NEXT: global_store_dword v8, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__8_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__8_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v0, v3 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__8_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v0, v3 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__8_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v0, v3 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__9_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__9_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__9_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v0, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__9_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v0, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__10_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__10_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v1, v3 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__10_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v1, v3 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__10_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v1, v3 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__11_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__11_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__11_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v1, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__11_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v1, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__12_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__12_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_bfi_b32 v0, s4, v2, v3 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__12_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0xffff | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_bfi_b32 v0, s4, v2, v3 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__12_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0xffff | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_bfi_b32 v0, s2, v2, v3 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__13_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__13_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: v_perm_b32 v0, v3, v2, s4 | |
| ; GFX900-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__13_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s4, 0x7060302 | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: v_perm_b32 v0, v3, v2, s4 | |
| ; GFX90A-NEXT: global_store_dword v4, v0, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__13_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s2, 0x7060302 | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: v_perm_b32 v0, v3, v2, s2 | |
| ; GFX942-NEXT: global_store_dword v4, v0, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @v_shuffle_v2bf16_v8bf16__14_15(ptr addrspace(1) inreg %ptr) { | |
| ; GFX900-LABEL: v_shuffle_v2bf16_v8bf16__14_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def v[0:3] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: v_shuffle_v2bf16_v8bf16__14_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def v[0:3] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: global_store_dword v4, v3, s[16:17] | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: v_shuffle_v2bf16_v8bf16__14_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: v_mov_b32_e32 v4, 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def v[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: global_store_dword v4, v3, s[0:1] | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=v"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 15> | |
| store <2 x bfloat> %shuf, ptr addrspace(1) %ptr, align 4 | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_u() { | |
| ; GFX9-LABEL: s_shuffle_v2bf16_v8bf16__u_u: | |
| ; GFX9: ; %bb.0: | |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX9-NEXT: ;;#ASMSTART | |
| ; GFX9-NEXT: ; use s8 | |
| ; GFX9-NEXT: ;;#ASMEND | |
| ; GFX9-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> poison | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[8:11] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s4, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s4, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s5, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s5, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s6, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s6, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s7, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s7, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_u() { | |
| ; GFX9-LABEL: s_shuffle_v2bf16_v8bf16__8_u: | |
| ; GFX9: ; %bb.0: | |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX9-NEXT: ;;#ASMSTART | |
| ; GFX9-NEXT: ; use s8 | |
| ; GFX9-NEXT: ;;#ASMEND | |
| ; GFX9-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s4, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s4, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s5, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s5, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s6, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s6, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_u() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_u: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s7, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_u: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s7, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_u: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 poison> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__15_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__15_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__15_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__15_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 15, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s4, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s4, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> zeroinitializer | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s4, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s4, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s9, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s9, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s5, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s10, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s10, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s6, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_0() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_0: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s11, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_0: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s11, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_0: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s7, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 0> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[8:11] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[8:11] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[8:11] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s9, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s9, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s5, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s10, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s10, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s6, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_1() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_1: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s11, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_1: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s11, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_1: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s7, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 1> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s5, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s5, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s5, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s5, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s9, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s9, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s5, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s10, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s10, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s6, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_2() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_2: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s11, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_2: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s11, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_2: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s7, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 2> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s9, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s9, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s5, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s10, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s10, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s6, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_3() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_3: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s11, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_3: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s11, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_3: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s7, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 3> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s6, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s6, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s6, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s6, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s9, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s9, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s5, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s10, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s10, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s6, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_4() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_4: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s11, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_4: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s11, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_4: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s7, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 4> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s9, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s9, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s5, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s10, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s10, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s6, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_5() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_5: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s11, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_5: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s11, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_5: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s7, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 5> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s7, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s7, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s7, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s7, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s9, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s9, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s5, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s10, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s10, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s6, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_6() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_6: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s11, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_6: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s11, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_6: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s7, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 6> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s9, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s9, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s5, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s10, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s10, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s6, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_7() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_7: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s11, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_7: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s11, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_7: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s7, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 7> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_8() { | |
| ; GFX9-LABEL: s_shuffle_v2bf16_v8bf16__u_8: | |
| ; GFX9: ; %bb.0: | |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX9-NEXT: ;;#ASMSTART | |
| ; GFX9-NEXT: ; use s8 | |
| ; GFX9-NEXT: ;;#ASMEND | |
| ; GFX9-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 poison, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[8:11] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 0, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s4, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s4, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 1, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 2, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s5, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s5, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 3, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 4, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s6, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s6, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 5, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 6, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s8, s7, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s8, s7, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s8, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 7, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_8() { | |
| ; GFX9-LABEL: s_shuffle_v2bf16_v8bf16__8_8: | |
| ; GFX9: ; %bb.0: | |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX9-NEXT: ;;#ASMSTART | |
| ; GFX9-NEXT: ; use s8 | |
| ; GFX9-NEXT: ;;#ASMEND | |
| ; GFX9-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> poison, <2 x i32> <i32 8, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_8() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_8: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_8: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_8: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 8> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[8:11] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s8, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s4, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s4, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s4, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s4, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s8, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s4, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[8:11] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_nop 0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_9() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_9: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_9: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_9: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 9> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s5, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s5, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s9 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s9 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s5 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s9 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s9 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s5 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s9 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s9 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s5 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s9 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s9 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s5 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s9 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s9 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s5 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s9 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s9 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s5 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s9 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s9 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s5 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s9 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s9 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s5 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_10() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_10: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_10: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_10: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 10> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s9, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s5, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s5, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s5, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s5, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s9, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s5, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_11() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_11: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_11: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_11: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 11> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s6, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s6, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s10 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s10 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s6 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s10 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s10 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s6 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s10 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s10 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s6 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s10 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s10 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s6 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s10 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s10 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s6 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s10 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s10 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s6 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s10 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s10 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s6 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s10 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s10 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s6 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_12() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_12: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_12: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_12: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 12> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s10, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s6, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s6, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s6, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s6, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s10, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s6, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s6 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s6 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s2 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_13() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_13: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_13: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_13: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 13> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshl_b32 s8, s7, 16 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshl_b32 s8, s7, 16 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshl_b32 s8, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s11 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s11 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s7 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s11 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s11 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s7 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s11 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s11 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s7 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s11 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s11 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s7 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s11 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s11 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s7 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s11 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s11 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s7 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s11 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s11 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s7 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s11 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s11 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s7 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_14() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_14: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_14: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_14: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 14> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__u_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__u_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__u_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__u_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 poison, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__0_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__0_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__0_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__0_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 0, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__1_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__1_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__1_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s11, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__1_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s7, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 1, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__2_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__2_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__2_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__2_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 2, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__3_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__3_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__3_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__3_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s7, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 3, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__4_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__4_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__4_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__4_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 4, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__5_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__5_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__5_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__5_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s7, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 5, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__6_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__6_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__6_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s7, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__6_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s7, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s3, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 6, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__7_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__7_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[8:11] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__7_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[8:11] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s11, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__7_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[4:7] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s7, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 7, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__8_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__8_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__8_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__8_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 8, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__9_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__9_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__9_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s7, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s4, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s4, s5 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__9_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s1, s3, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s0, s0, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s0, s1 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 9, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__10_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__10_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__10_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__10_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 10, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__11_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__11_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__11_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s5, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__11_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s1, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 11, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__12_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__12_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__12_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s6, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__12_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s2, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 12, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__13_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__13_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX900-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX900-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__13_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_lshr_b32 s4, s7, 16 | |
| ; GFX90A-NEXT: s_lshr_b32 s5, s6, 16 | |
| ; GFX90A-NEXT: s_pack_ll_b32_b16 s8, s5, s4 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__13_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_lshr_b32 s0, s3, 16 | |
| ; GFX942-NEXT: s_lshr_b32 s1, s2, 16 | |
| ; GFX942-NEXT: s_pack_ll_b32_b16 s8, s1, s0 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 13, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| define void @s_shuffle_v2bf16_v8bf16__14_15() { | |
| ; GFX900-LABEL: s_shuffle_v2bf16_v8bf16__14_15: | |
| ; GFX900: ; %bb.0: | |
| ; GFX900-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; def s[4:7] | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_mov_b32 s8, s7 | |
| ; GFX900-NEXT: ;;#ASMSTART | |
| ; GFX900-NEXT: ; use s8 | |
| ; GFX900-NEXT: ;;#ASMEND | |
| ; GFX900-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX90A-LABEL: s_shuffle_v2bf16_v8bf16__14_15: | |
| ; GFX90A: ; %bb.0: | |
| ; GFX90A-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; def s[4:7] | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_mov_b32 s8, s7 | |
| ; GFX90A-NEXT: ;;#ASMSTART | |
| ; GFX90A-NEXT: ; use s8 | |
| ; GFX90A-NEXT: ;;#ASMEND | |
| ; GFX90A-NEXT: s_setpc_b64 s[30:31] | |
| ; | |
| ; GFX942-LABEL: s_shuffle_v2bf16_v8bf16__14_15: | |
| ; GFX942: ; %bb.0: | |
| ; GFX942-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; def s[0:3] | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_mov_b32 s8, s3 | |
| ; GFX942-NEXT: ;;#ASMSTART | |
| ; GFX942-NEXT: ; use s8 | |
| ; GFX942-NEXT: ;;#ASMEND | |
| ; GFX942-NEXT: s_setpc_b64 s[30:31] | |
| %vec0 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %vec1 = call <8 x bfloat> asm "; def $0", "=s"() | |
| %shuf = shufflevector <8 x bfloat> %vec0, <8 x bfloat> %vec1, <2 x i32> <i32 14, i32 15> | |
| call void asm sideeffect "; use $0", "{s8}"(<2 x bfloat> %shuf) | |
| ret void | |
| } | |
| ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: | |
| ; GFX90APLUS: {{.*}} |