| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py | |
| ; RUN: llc -global-isel=0 -mtriple=amdgpu9.00 < %s | FileCheck %s --check-prefixes=GFX9,GFX9-SDAG | |
| ; RUN: llc -global-isel=1 -mtriple=amdgpu9.00 < %s | FileCheck %s --check-prefixes=GFX9,GFX9-GISEL | |
| ; RUN: llc -global-isel=0 -mtriple=amdgpu10.10 < %s | FileCheck %s --check-prefixes=GFX10,GFX10-SDAG | |
| ; RUN: llc -global-isel=1 -mtriple=amdgpu10.10 < %s | FileCheck %s --check-prefixes=GFX10,GFX10-GISEL | |
| ; RUN: llc -global-isel=0 -mtriple=amdgpu11.00 < %s | FileCheck %s --check-prefixes=GFX11,GFX11-SDAG | |
| ; RUN: llc -global-isel=1 -mtriple=amdgpu11.00 < %s | FileCheck %s --check-prefixes=GFX11,GFX11-GISEL | |
| define amdgpu_kernel void @test0() { | |
| ; GFX9-LABEL: test0: | |
| ; GFX9: ; %bb.0: | |
| ; GFX9-NEXT: s_endpgm | |
| ; | |
| ; GFX10-LABEL: test0: | |
| ; GFX10: ; %bb.0: | |
| ; GFX10-NEXT: s_endpgm | |
| ; | |
| ; GFX11-LABEL: test0: | |
| ; GFX11: ; %bb.0: | |
| ; GFX11-NEXT: s_endpgm | |
| tail call void @llvm.amdgcn.endpgm() | |
| unreachable | |
| } | |
| define void @test1() { | |
| ; GFX9-LABEL: test1: | |
| ; GFX9: ; %bb.0: | |
| ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX9-NEXT: s_endpgm | |
| ; | |
| ; GFX10-LABEL: test1: | |
| ; GFX10: ; %bb.0: | |
| ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX10-NEXT: s_endpgm | |
| ; | |
| ; GFX11-LABEL: test1: | |
| ; GFX11: ; %bb.0: | |
| ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) | |
| ; GFX11-NEXT: s_endpgm | |
| tail call void @llvm.amdgcn.endpgm() | |
| unreachable | |
| } | |
| define amdgpu_kernel void @test2(ptr %p, i32 %x) { | |
| ; GFX9-SDAG-LABEL: test2: | |
| ; GFX9-SDAG: ; %bb.0: | |
| ; GFX9-SDAG-NEXT: s_load_dword s0, s[4:5], 0x2c | |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX9-SDAG-NEXT: s_cmp_lt_i32 s0, 1 | |
| ; GFX9-SDAG-NEXT: s_cbranch_scc0 .LBB2_2 | |
| ; GFX9-SDAG-NEXT: ; %bb.1: ; %else | |
| ; GFX9-SDAG-NEXT: s_load_dwordx2 s[2:3], s[4:5], 0x24 | |
| ; GFX9-SDAG-NEXT: v_mov_b32_e32 v2, s0 | |
| ; GFX9-SDAG-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX9-SDAG-NEXT: v_mov_b32_e32 v0, s2 | |
| ; GFX9-SDAG-NEXT: v_mov_b32_e32 v1, s3 | |
| ; GFX9-SDAG-NEXT: flat_store_dword v[0:1], v2 | |
| ; GFX9-SDAG-NEXT: s_endpgm | |
| ; GFX9-SDAG-NEXT: .LBB2_2: ; %then | |
| ; GFX9-SDAG-NEXT: s_endpgm | |
| ; | |
| ; GFX9-GISEL-LABEL: test2: | |
| ; GFX9-GISEL: ; %bb.0: | |
| ; GFX9-GISEL-NEXT: s_load_dword s0, s[4:5], 0x2c | |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX9-GISEL-NEXT: s_cmp_le_i32 s0, 0 | |
| ; GFX9-GISEL-NEXT: s_cbranch_scc0 .LBB2_2 | |
| ; GFX9-GISEL-NEXT: ; %bb.1: ; %else | |
| ; GFX9-GISEL-NEXT: s_load_dwordx2 s[2:3], s[4:5], 0x24 | |
| ; GFX9-GISEL-NEXT: v_mov_b32_e32 v2, s0 | |
| ; GFX9-GISEL-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX9-GISEL-NEXT: v_mov_b32_e32 v0, s2 | |
| ; GFX9-GISEL-NEXT: v_mov_b32_e32 v1, s3 | |
| ; GFX9-GISEL-NEXT: flat_store_dword v[0:1], v2 | |
| ; GFX9-GISEL-NEXT: s_endpgm | |
| ; GFX9-GISEL-NEXT: .LBB2_2: ; %then | |
| ; GFX9-GISEL-NEXT: s_endpgm | |
| ; | |
| ; GFX10-SDAG-LABEL: test2: | |
| ; GFX10-SDAG: ; %bb.0: | |
| ; GFX10-SDAG-NEXT: s_load_dword s0, s[4:5], 0x2c | |
| ; GFX10-SDAG-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX10-SDAG-NEXT: s_cmp_lt_i32 s0, 1 | |
| ; GFX10-SDAG-NEXT: s_cbranch_scc0 .LBB2_2 | |
| ; GFX10-SDAG-NEXT: ; %bb.1: ; %else | |
| ; GFX10-SDAG-NEXT: s_load_dwordx2 s[2:3], s[4:5], 0x24 | |
| ; GFX10-SDAG-NEXT: v_mov_b32_e32 v2, s0 | |
| ; GFX10-SDAG-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX10-SDAG-NEXT: v_mov_b32_e32 v0, s2 | |
| ; GFX10-SDAG-NEXT: v_mov_b32_e32 v1, s3 | |
| ; GFX10-SDAG-NEXT: flat_store_dword v[0:1], v2 | |
| ; GFX10-SDAG-NEXT: s_endpgm | |
| ; GFX10-SDAG-NEXT: .LBB2_2: ; %then | |
| ; GFX10-SDAG-NEXT: s_endpgm | |
| ; | |
| ; GFX10-GISEL-LABEL: test2: | |
| ; GFX10-GISEL: ; %bb.0: | |
| ; GFX10-GISEL-NEXT: s_load_dword s0, s[4:5], 0x2c | |
| ; GFX10-GISEL-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX10-GISEL-NEXT: s_cmp_le_i32 s0, 0 | |
| ; GFX10-GISEL-NEXT: s_cbranch_scc0 .LBB2_2 | |
| ; GFX10-GISEL-NEXT: ; %bb.1: ; %else | |
| ; GFX10-GISEL-NEXT: s_load_dwordx2 s[2:3], s[4:5], 0x24 | |
| ; GFX10-GISEL-NEXT: v_mov_b32_e32 v2, s0 | |
| ; GFX10-GISEL-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX10-GISEL-NEXT: v_mov_b32_e32 v0, s2 | |
| ; GFX10-GISEL-NEXT: v_mov_b32_e32 v1, s3 | |
| ; GFX10-GISEL-NEXT: flat_store_dword v[0:1], v2 | |
| ; GFX10-GISEL-NEXT: s_endpgm | |
| ; GFX10-GISEL-NEXT: .LBB2_2: ; %then | |
| ; GFX10-GISEL-NEXT: s_endpgm | |
| ; | |
| ; GFX11-SDAG-LABEL: test2: | |
| ; GFX11-SDAG: ; %bb.0: | |
| ; GFX11-SDAG-NEXT: s_load_b32 s0, s[4:5], 0x2c | |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX11-SDAG-NEXT: s_cmp_lt_i32 s0, 1 | |
| ; GFX11-SDAG-NEXT: s_cbranch_scc0 .LBB2_2 | |
| ; GFX11-SDAG-NEXT: ; %bb.1: ; %else | |
| ; GFX11-SDAG-NEXT: s_load_b64 s[2:3], s[4:5], 0x24 | |
| ; GFX11-SDAG-NEXT: v_mov_b32_e32 v2, s0 | |
| ; GFX11-SDAG-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX11-SDAG-NEXT: v_dual_mov_b32 v0, s2 :: v_dual_mov_b32 v1, s3 | |
| ; GFX11-SDAG-NEXT: flat_store_b32 v[0:1], v2 | |
| ; GFX11-SDAG-NEXT: s_endpgm | |
| ; GFX11-SDAG-NEXT: .LBB2_2: ; %then | |
| ; GFX11-SDAG-NEXT: s_endpgm | |
| ; | |
| ; GFX11-GISEL-LABEL: test2: | |
| ; GFX11-GISEL: ; %bb.0: | |
| ; GFX11-GISEL-NEXT: s_load_b32 s0, s[4:5], 0x2c | |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX11-GISEL-NEXT: s_cmp_le_i32 s0, 0 | |
| ; GFX11-GISEL-NEXT: s_cbranch_scc0 .LBB2_2 | |
| ; GFX11-GISEL-NEXT: ; %bb.1: ; %else | |
| ; GFX11-GISEL-NEXT: s_load_b64 s[2:3], s[4:5], 0x24 | |
| ; GFX11-GISEL-NEXT: s_waitcnt lgkmcnt(0) | |
| ; GFX11-GISEL-NEXT: v_dual_mov_b32 v2, s0 :: v_dual_mov_b32 v1, s3 | |
| ; GFX11-GISEL-NEXT: v_mov_b32_e32 v0, s2 | |
| ; GFX11-GISEL-NEXT: flat_store_b32 v[0:1], v2 | |
| ; GFX11-GISEL-NEXT: s_endpgm | |
| ; GFX11-GISEL-NEXT: .LBB2_2: ; %then | |
| ; GFX11-GISEL-NEXT: s_endpgm | |
| %cond = icmp sgt i32 %x, 0 | |
| br i1 %cond, label %then, label %else | |
| then: | |
| tail call void @llvm.amdgcn.endpgm() | |
| unreachable | |
| else: | |
| store i32 %x, ptr %p | |
| ret void | |
| } | |
| declare void @llvm.amdgcn.endpgm() |