Google Git
Sign in
llvm / llvm-project / llvm / cd44f13d389ec38f03f9e668045ef78fdf8b594f / . / test / CodeGen / AMDGPU / atomic_optimizations_struct_buffer.ll
blob: 0d37f0e814dd529db3e0f83c9d5ffc54da2f40c5 [file] [log] [blame]
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -march=amdgcn -amdgpu-atomic-optimizations=true -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX6 %s
; RUN: llc -march=amdgcn -mcpu=tonga -amdgpu-atomic-optimizations=true -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX8 %s
; RUN: llc -march=amdgcn -mcpu=gfx900 -amdgpu-atomic-optimizations=true -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX9 %s
; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=-wavefrontsize32,+wavefrontsize64 -amdgpu-atomic-optimizations=true -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX10,GFX10W64 %s
; RUN: llc -march=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-atomic-optimizations=true -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX10,GFX10W32 %s
; RUN: llc -march=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -amdgpu-atomic-optimizations=true -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11W64 %s
; RUN: llc -march=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32,-wavefrontsize64 -amdgpu-atomic-optimizations=true -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11W32 %s
declare i32 @llvm.amdgcn.workitem.id.x()
declare i32 @llvm.amdgcn.struct.buffer.atomic.add(i32, <4 x i32>, i32, i32, i32, i32)
declare i32 @llvm.amdgcn.struct.buffer.atomic.sub(i32, <4 x i32>, i32, i32, i32, i32)
; Show what the atomic optimization pass will do for struct buffers.
define amdgpu_kernel void @add_i32_constant(ptr addrspace(1) %out, <4 x i32> %inout) {
; GFX6-LABEL: add_i32_constant:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_mov_b64 s[4:5], exec
; GFX6-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s4, 0
; GFX6-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s5, v0
; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX6-NEXT: ; implicit-def: $vgpr1
; GFX6-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX6-NEXT: s_cbranch_execz .LBB0_2
; GFX6-NEXT: ; %bb.1:
; GFX6-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xd
; GFX6-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX6-NEXT: s_mul_i32 s4, s4, 5
; GFX6-NEXT: v_mov_b32_e32 v1, s4
; GFX6-NEXT: v_mov_b32_e32 v2, 0
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: buffer_atomic_add v1, v2, s[8:11], 0 idxen glc
; GFX6-NEXT: .LBB0_2:
; GFX6-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: v_readfirstlane_b32 s4, v1
; GFX6-NEXT: v_mad_u32_u24 v0, v0, 5, s4
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: add_i32_constant:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_mov_b64 s[4:5], exec
; GFX8-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX8-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT: ; implicit-def: $vgpr1
; GFX8-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT: s_cbranch_execz .LBB0_2
; GFX8-NEXT: ; %bb.1:
; GFX8-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX8-NEXT: s_mul_i32 s4, s4, 5
; GFX8-NEXT: v_mov_b32_e32 v1, s4
; GFX8-NEXT: v_mov_b32_e32 v2, 0
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: buffer_atomic_add v1, v2, s[8:11], 0 idxen glc
; GFX8-NEXT: .LBB0_2:
; GFX8-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: v_readfirstlane_b32 s2, v1
; GFX8-NEXT: v_mad_u32_u24 v2, v0, 5, s2
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX9-LABEL: add_i32_constant:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: s_mov_b64 s[4:5], exec
; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT: ; implicit-def: $vgpr1
; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT: s_cbranch_execz .LBB0_2
; GFX9-NEXT: ; %bb.1:
; GFX9-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX9-NEXT: s_mul_i32 s4, s4, 5
; GFX9-NEXT: v_mov_b32_e32 v1, s4
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: buffer_atomic_add v1, v2, s[8:11], 0 idxen glc
; GFX9-NEXT: .LBB0_2:
; GFX9-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_readfirstlane_b32 s2, v1
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: v_mad_u32_u24 v0, v0, 5, s2
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v2, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; GFX10W64-LABEL: add_i32_constant:
; GFX10W64: ; %bb.0: ; %entry
; GFX10W64-NEXT: s_mov_b64 s[4:5], exec
; GFX10W64-NEXT: ; implicit-def: $vgpr1
; GFX10W64-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W64-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX10W64-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX10W64-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT: s_cbranch_execz .LBB0_2
; GFX10W64-NEXT: ; %bb.1:
; GFX10W64-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX10W64-NEXT: v_mov_b32_e32 v2, 0
; GFX10W64-NEXT: s_mul_i32 s4, s4, 5
; GFX10W64-NEXT: v_mov_b32_e32 v1, s4
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: buffer_atomic_add v1, v2, s[8:11], 0 idxen glc
; GFX10W64-NEXT: .LBB0_2:
; GFX10W64-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT: s_waitcnt vmcnt(0)
; GFX10W64-NEXT: v_readfirstlane_b32 s2, v1
; GFX10W64-NEXT: v_mov_b32_e32 v1, 0
; GFX10W64-NEXT: v_mad_u32_u24 v0, v0, 5, s2
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: global_store_dword v1, v0, s[0:1]
; GFX10W64-NEXT: s_endpgm
;
; GFX10W32-LABEL: add_i32_constant:
; GFX10W32: ; %bb.0: ; %entry
; GFX10W32-NEXT: s_mov_b32 s3, exec_lo
; GFX10W32-NEXT: ; implicit-def: $vgpr1
; GFX10W32-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
; GFX10W32-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10W32-NEXT: s_and_saveexec_b32 s2, vcc_lo
; GFX10W32-NEXT: s_cbranch_execz .LBB0_2
; GFX10W32-NEXT: ; %bb.1:
; GFX10W32-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10W32-NEXT: s_bcnt1_i32_b32 s3, s3
; GFX10W32-NEXT: v_mov_b32_e32 v2, 0
; GFX10W32-NEXT: s_mul_i32 s3, s3, 5
; GFX10W32-NEXT: v_mov_b32_e32 v1, s3
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: buffer_atomic_add v1, v2, s[4:7], 0 idxen glc
; GFX10W32-NEXT: .LBB0_2:
; GFX10W32-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT: s_or_b32 exec_lo, exec_lo, s2
; GFX10W32-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT: s_waitcnt vmcnt(0)
; GFX10W32-NEXT: v_readfirstlane_b32 s2, v1
; GFX10W32-NEXT: v_mov_b32_e32 v1, 0
; GFX10W32-NEXT: v_mad_u32_u24 v0, v0, 5, s2
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: global_store_dword v1, v0, s[0:1]
; GFX10W32-NEXT: s_endpgm
;
; GFX11W64-LABEL: add_i32_constant:
; GFX11W64: ; %bb.0: ; %entry
; GFX11W64-NEXT: s_mov_b64 s[4:5], exec
; GFX11W64-NEXT: s_mov_b64 s[2:3], exec
; GFX11W64-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W64-NEXT: ; implicit-def: $vgpr1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX11W64-NEXT: v_cmpx_eq_u32_e32 0, v0
; GFX11W64-NEXT: s_cbranch_execz .LBB0_2
; GFX11W64-NEXT: ; %bb.1:
; GFX11W64-NEXT: s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX11W64-NEXT: v_mov_b32_e32 v2, 0
; GFX11W64-NEXT: s_mul_i32 s4, s4, 5
; GFX11W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT: v_mov_b32_e32 v1, s4
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: buffer_atomic_add_u32 v1, v2, s[8:11], 0 idxen glc
; GFX11W64-NEXT: .LBB0_2:
; GFX11W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT: s_waitcnt vmcnt(0)
; GFX11W64-NEXT: v_readfirstlane_b32 s2, v1
; GFX11W64-NEXT: v_mov_b32_e32 v1, 0
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT: v_mad_u32_u24 v0, v0, 5, s2
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: global_store_b32 v1, v0, s[0:1]
; GFX11W64-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT: s_endpgm
;
; GFX11W32-LABEL: add_i32_constant:
; GFX11W32: ; %bb.0: ; %entry
; GFX11W32-NEXT: s_mov_b32 s3, exec_lo
; GFX11W32-NEXT: s_mov_b32 s2, exec_lo
; GFX11W32-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
; GFX11W32-NEXT: ; implicit-def: $vgpr1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT: v_cmpx_eq_u32_e32 0, v0
; GFX11W32-NEXT: s_cbranch_execz .LBB0_2
; GFX11W32-NEXT: ; %bb.1:
; GFX11W32-NEXT: s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W32-NEXT: s_bcnt1_i32_b32 s3, s3
; GFX11W32-NEXT: v_mov_b32_e32 v2, 0
; GFX11W32-NEXT: s_mul_i32 s3, s3, 5
; GFX11W32-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W32-NEXT: v_mov_b32_e32 v1, s3
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: buffer_atomic_add_u32 v1, v2, s[4:7], 0 idxen glc
; GFX11W32-NEXT: .LBB0_2:
; GFX11W32-NEXT: s_or_b32 exec_lo, exec_lo, s2
; GFX11W32-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT: s_waitcnt vmcnt(0)
; GFX11W32-NEXT: v_readfirstlane_b32 s2, v1
; GFX11W32-NEXT: v_mov_b32_e32 v1, 0
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W32-NEXT: v_mad_u32_u24 v0, v0, 5, s2
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: global_store_b32 v1, v0, s[0:1]
; GFX11W32-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT: s_endpgm
entry:
%old = call i32 @llvm.amdgcn.struct.buffer.atomic.add(i32 5, <4 x i32> %inout, i32 0, i32 0, i32 0, i32 0)
store i32 %old, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @add_i32_uniform(ptr addrspace(1) %out, <4 x i32> %inout, i32 %additive) {
; GFX6-LABEL: add_i32_uniform:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_mov_b64 s[4:5], exec
; GFX6-NEXT: s_load_dword s6, s[0:1], 0x11
; GFX6-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s4, 0
; GFX6-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s5, v0
; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX6-NEXT: ; implicit-def: $vgpr1
; GFX6-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX6-NEXT: s_cbranch_execz .LBB1_2
; GFX6-NEXT: ; %bb.1:
; GFX6-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xd
; GFX6-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_mul_i32 s4, s6, s4
; GFX6-NEXT: v_mov_b32_e32 v1, s4
; GFX6-NEXT: v_mov_b32_e32 v2, 0
; GFX6-NEXT: buffer_atomic_add v1, v2, s[8:11], 0 idxen glc
; GFX6-NEXT: .LBB1_2:
; GFX6-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: v_readfirstlane_b32 s4, v1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: v_mul_lo_u32 v0, s6, v0
; GFX6-NEXT: v_add_i32_e32 v0, vcc, s4, v0
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: add_i32_uniform:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_load_dword s6, s[0:1], 0x44
; GFX8-NEXT: s_mov_b64 s[4:5], exec
; GFX8-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX8-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT: ; implicit-def: $vgpr1
; GFX8-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT: s_cbranch_execz .LBB1_2
; GFX8-NEXT: ; %bb.1:
; GFX8-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: s_mul_i32 s4, s6, s4
; GFX8-NEXT: v_mov_b32_e32 v1, s4
; GFX8-NEXT: v_mov_b32_e32 v2, 0
; GFX8-NEXT: buffer_atomic_add v1, v2, s[8:11], 0 idxen glc
; GFX8-NEXT: .LBB1_2:
; GFX8-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mul_lo_u32 v0, s6, v0
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: v_readfirstlane_b32 s2, v1
; GFX8-NEXT: v_add_u32_e32 v2, vcc, s2, v0
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX9-LABEL: add_i32_uniform:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: s_load_dword s6, s[0:1], 0x44
; GFX9-NEXT: s_mov_b64 s[4:5], exec
; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT: ; implicit-def: $vgpr1
; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT: s_cbranch_execz .LBB1_2
; GFX9-NEXT: ; %bb.1:
; GFX9-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_mul_i32 s4, s6, s4
; GFX9-NEXT: v_mov_b32_e32 v1, s4
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: buffer_atomic_add v1, v2, s[8:11], 0 idxen glc
; GFX9-NEXT: .LBB1_2:
; GFX9-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mul_lo_u32 v0, s6, v0
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_readfirstlane_b32 s2, v1
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: v_add_u32_e32 v0, s2, v0
; GFX9-NEXT: global_store_dword v2, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; GFX10W64-LABEL: add_i32_uniform:
; GFX10W64: ; %bb.0: ; %entry
; GFX10W64-NEXT: s_load_dword s6, s[0:1], 0x44
; GFX10W64-NEXT: s_mov_b64 s[4:5], exec
; GFX10W64-NEXT: ; implicit-def: $vgpr1
; GFX10W64-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W64-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX10W64-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX10W64-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT: s_cbranch_execz .LBB1_2
; GFX10W64-NEXT: ; %bb.1:
; GFX10W64-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX10W64-NEXT: v_mov_b32_e32 v2, 0
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: s_mul_i32 s4, s6, s4
; GFX10W64-NEXT: v_mov_b32_e32 v1, s4
; GFX10W64-NEXT: buffer_atomic_add v1, v2, s[8:11], 0 idxen glc
; GFX10W64-NEXT: .LBB1_2:
; GFX10W64-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT: s_waitcnt vmcnt(0)
; GFX10W64-NEXT: v_readfirstlane_b32 s2, v1
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: v_mad_u64_u32 v[0:1], s[2:3], s6, v0, s[2:3]
; GFX10W64-NEXT: v_mov_b32_e32 v1, 0
; GFX10W64-NEXT: global_store_dword v1, v0, s[0:1]
; GFX10W64-NEXT: s_endpgm
;
; GFX10W32-LABEL: add_i32_uniform:
; GFX10W32: ; %bb.0: ; %entry
; GFX10W32-NEXT: s_load_dword s2, s[0:1], 0x44
; GFX10W32-NEXT: s_mov_b32 s4, exec_lo
; GFX10W32-NEXT: ; implicit-def: $vgpr1
; GFX10W32-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W32-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10W32-NEXT: s_and_saveexec_b32 s3, vcc_lo
; GFX10W32-NEXT: s_cbranch_execz .LBB1_2
; GFX10W32-NEXT: ; %bb.1:
; GFX10W32-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W32-NEXT: s_bcnt1_i32_b32 s4, s4
; GFX10W32-NEXT: v_mov_b32_e32 v2, 0
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: s_mul_i32 s4, s2, s4
; GFX10W32-NEXT: v_mov_b32_e32 v1, s4
; GFX10W32-NEXT: buffer_atomic_add v1, v2, s[8:11], 0 idxen glc
; GFX10W32-NEXT: .LBB1_2:
; GFX10W32-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT: s_or_b32 exec_lo, exec_lo, s3
; GFX10W32-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT: s_waitcnt vmcnt(0)
; GFX10W32-NEXT: v_readfirstlane_b32 s4, v1
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: v_mad_u64_u32 v[0:1], s2, s2, v0, s[4:5]
; GFX10W32-NEXT: v_mov_b32_e32 v1, 0
; GFX10W32-NEXT: global_store_dword v1, v0, s[0:1]
; GFX10W32-NEXT: s_endpgm
;
; GFX11W64-LABEL: add_i32_uniform:
; GFX11W64: ; %bb.0: ; %entry
; GFX11W64-NEXT: s_load_b32 s6, s[0:1], 0x44
; GFX11W64-NEXT: s_mov_b64 s[4:5], exec
; GFX11W64-NEXT: s_mov_b64 s[2:3], exec
; GFX11W64-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W64-NEXT: ; implicit-def: $vgpr1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX11W64-NEXT: v_cmpx_eq_u32_e32 0, v0
; GFX11W64-NEXT: s_cbranch_execz .LBB1_2
; GFX11W64-NEXT: ; %bb.1:
; GFX11W64-NEXT: s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX11W64-NEXT: v_mov_b32_e32 v2, 0
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: s_mul_i32 s4, s6, s4
; GFX11W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT: v_mov_b32_e32 v1, s4
; GFX11W64-NEXT: buffer_atomic_add_u32 v1, v2, s[8:11], 0 idxen glc
; GFX11W64-NEXT: .LBB1_2:
; GFX11W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT: s_waitcnt vmcnt(0)
; GFX11W64-NEXT: v_readfirstlane_b32 s2, v1
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11W64-NEXT: v_mad_u64_u32 v[1:2], null, s6, v0, s[2:3]
; GFX11W64-NEXT: v_mov_b32_e32 v0, 0
; GFX11W64-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX11W64-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT: s_endpgm
;
; GFX11W32-LABEL: add_i32_uniform:
; GFX11W32: ; %bb.0: ; %entry
; GFX11W32-NEXT: s_load_b32 s2, s[0:1], 0x44
; GFX11W32-NEXT: s_mov_b32 s4, exec_lo
; GFX11W32-NEXT: s_mov_b32 s3, exec_lo
; GFX11W32-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W32-NEXT: ; implicit-def: $vgpr1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT: v_cmpx_eq_u32_e32 0, v0
; GFX11W32-NEXT: s_cbranch_execz .LBB1_2
; GFX11W32-NEXT: ; %bb.1:
; GFX11W32-NEXT: s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W32-NEXT: s_bcnt1_i32_b32 s4, s4
; GFX11W32-NEXT: v_mov_b32_e32 v2, 0
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: s_mul_i32 s4, s2, s4
; GFX11W32-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W32-NEXT: v_mov_b32_e32 v1, s4
; GFX11W32-NEXT: buffer_atomic_add_u32 v1, v2, s[8:11], 0 idxen glc
; GFX11W32-NEXT: .LBB1_2:
; GFX11W32-NEXT: s_or_b32 exec_lo, exec_lo, s3
; GFX11W32-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT: s_waitcnt vmcnt(0)
; GFX11W32-NEXT: v_readfirstlane_b32 s4, v1
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT: v_mad_u64_u32 v[1:2], null, s2, v0, s[4:5]
; GFX11W32-NEXT: v_mov_b32_e32 v0, 0
; GFX11W32-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX11W32-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT: s_endpgm
entry:
%old = call i32 @llvm.amdgcn.struct.buffer.atomic.add(i32 %additive, <4 x i32> %inout, i32 0, i32 0, i32 0, i32 0)
store i32 %old, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @add_i32_varying_vdata(ptr addrspace(1) %out, <4 x i32> %inout) {
; GFX6-LABEL: add_i32_varying_vdata:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xd
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT: v_mov_b32_e32 v1, 0
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: buffer_atomic_add v0, v1, s[4:7], 0 idxen glc
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: add_i32_varying_vdata:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: v_mov_b32_e32 v3, 0
; GFX8-NEXT: v_mbcnt_lo_u32_b32 v4, exec_lo, 0
; GFX8-NEXT: v_mbcnt_hi_u32_b32 v4, exec_hi, v4
; GFX8-NEXT: v_mov_b32_e32 v1, v0
; GFX8-NEXT: s_not_b64 exec, exec
; GFX8-NEXT: v_mov_b32_e32 v1, 0
; GFX8-NEXT: s_not_b64 exec, exec
; GFX8-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX8-NEXT: v_mov_b32_e32 v2, 0
; GFX8-NEXT: s_nop 0
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX8-NEXT: s_nop 1
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX8-NEXT: s_nop 1
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX8-NEXT: s_nop 1
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
; GFX8-NEXT: s_nop 1
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
; GFX8-NEXT: v_readlane_b32 s4, v1, 63
; GFX8-NEXT: s_nop 0
; GFX8-NEXT: v_mov_b32_dpp v2, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
; GFX8-NEXT: s_mov_b64 exec, s[2:3]
; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
; GFX8-NEXT: ; implicit-def: $vgpr0
; GFX8-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT: s_cbranch_execz .LBB2_2
; GFX8-NEXT: ; %bb.1:
; GFX8-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT: v_mov_b32_e32 v0, s4
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: buffer_atomic_add v0, v3, s[8:11], 0 idxen glc
; GFX8-NEXT: .LBB2_2:
; GFX8-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: v_readfirstlane_b32 s2, v0
; GFX8-NEXT: v_mov_b32_e32 v0, v2
; GFX8-NEXT: v_add_u32_e32 v0, vcc, s2, v0
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v4, s1
; GFX8-NEXT: v_mov_b32_e32 v3, s0
; GFX8-NEXT: flat_store_dword v[3:4], v0
; GFX8-NEXT: s_endpgm
;
; GFX9-LABEL: add_i32_varying_vdata:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: v_mov_b32_e32 v3, 0
; GFX9-NEXT: v_mbcnt_lo_u32_b32 v4, exec_lo, 0
; GFX9-NEXT: v_mbcnt_hi_u32_b32 v4, exec_hi, v4
; GFX9-NEXT: v_mov_b32_e32 v1, v0
; GFX9-NEXT: s_not_b64 exec, exec
; GFX9-NEXT: v_mov_b32_e32 v1, 0
; GFX9-NEXT: s_not_b64 exec, exec
; GFX9-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: s_nop 0
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX9-NEXT: s_nop 1
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX9-NEXT: s_nop 1
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX9-NEXT: s_nop 1
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
; GFX9-NEXT: s_nop 1
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
; GFX9-NEXT: v_readlane_b32 s4, v1, 63
; GFX9-NEXT: s_nop 0
; GFX9-NEXT: v_mov_b32_dpp v2, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
; GFX9-NEXT: s_mov_b64 exec, s[2:3]
; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
; GFX9-NEXT: ; implicit-def: $vgpr0
; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT: s_cbranch_execz .LBB2_2
; GFX9-NEXT: ; %bb.1:
; GFX9-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT: v_mov_b32_e32 v0, s4
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: buffer_atomic_add v0, v3, s[8:11], 0 idxen glc
; GFX9-NEXT: .LBB2_2:
; GFX9-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_readfirstlane_b32 s2, v0
; GFX9-NEXT: v_mov_b32_e32 v0, v2
; GFX9-NEXT: v_add_u32_e32 v0, s2, v0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v3, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; GFX10W64-LABEL: add_i32_varying_vdata:
; GFX10W64: ; %bb.0: ; %entry
; GFX10W64-NEXT: v_mov_b32_e32 v1, v0
; GFX10W64-NEXT: s_not_b64 exec, exec
; GFX10W64-NEXT: v_mov_b32_e32 v1, 0
; GFX10W64-NEXT: s_not_b64 exec, exec
; GFX10W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W64-NEXT: v_mov_b32_e32 v3, 0
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W64-NEXT: v_mov_b32_e32 v2, v1
; GFX10W64-NEXT: v_permlanex16_b32 v2, v2, -1, -1
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
; GFX10W64-NEXT: v_readlane_b32 s4, v1, 31
; GFX10W64-NEXT: v_mov_b32_e32 v2, s4
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
; GFX10W64-NEXT: v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
; GFX10W64-NEXT: v_readlane_b32 s4, v1, 15
; GFX10W64-NEXT: v_readlane_b32 s5, v1, 31
; GFX10W64-NEXT: v_writelane_b32 v3, s4, 16
; GFX10W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX10W64-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX10W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX10W64-NEXT: v_readlane_b32 s4, v1, 63
; GFX10W64-NEXT: v_readlane_b32 s6, v1, 47
; GFX10W64-NEXT: v_writelane_b32 v3, s5, 32
; GFX10W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX10W64-NEXT: v_mbcnt_hi_u32_b32 v4, exec_hi, v0
; GFX10W64-NEXT: v_mov_b32_e32 v0, 0
; GFX10W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX10W64-NEXT: v_writelane_b32 v3, s6, 48
; GFX10W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX10W64-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
; GFX10W64-NEXT: ; implicit-def: $vgpr4
; GFX10W64-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT: s_cbranch_execz .LBB2_2
; GFX10W64-NEXT: ; %bb.1:
; GFX10W64-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT: v_mov_b32_e32 v4, s4
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: buffer_atomic_add v4, v0, s[8:11], 0 idxen glc
; GFX10W64-NEXT: .LBB2_2:
; GFX10W64-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT: s_waitcnt vmcnt(0)
; GFX10W64-NEXT: v_readfirstlane_b32 s2, v4
; GFX10W64-NEXT: v_mov_b32_e32 v4, v3
; GFX10W64-NEXT: v_add_nc_u32_e32 v4, s2, v4
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: global_store_dword v0, v4, s[0:1]
; GFX10W64-NEXT: s_endpgm
;
; GFX10W32-LABEL: add_i32_varying_vdata:
; GFX10W32: ; %bb.0: ; %entry
; GFX10W32-NEXT: v_mov_b32_e32 v1, v0
; GFX10W32-NEXT: s_not_b32 exec_lo, exec_lo
; GFX10W32-NEXT: v_mov_b32_e32 v1, 0
; GFX10W32-NEXT: s_not_b32 exec_lo, exec_lo
; GFX10W32-NEXT: s_or_saveexec_b32 s2, -1
; GFX10W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W32-NEXT: v_mov_b32_e32 v3, 0
; GFX10W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W32-NEXT: v_mov_b32_e32 v2, v1
; GFX10W32-NEXT: v_permlanex16_b32 v2, v2, -1, -1
; GFX10W32-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
; GFX10W32-NEXT: v_readlane_b32 s4, v1, 31
; GFX10W32-NEXT: v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
; GFX10W32-NEXT: v_readlane_b32 s3, v1, 15
; GFX10W32-NEXT: s_mov_b32 exec_lo, s2
; GFX10W32-NEXT: v_mbcnt_lo_u32_b32 v4, exec_lo, 0
; GFX10W32-NEXT: v_mov_b32_e32 v0, 0
; GFX10W32-NEXT: s_or_saveexec_b32 s2, -1
; GFX10W32-NEXT: v_writelane_b32 v3, s3, 16
; GFX10W32-NEXT: s_mov_b32 exec_lo, s2
; GFX10W32-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v4
; GFX10W32-NEXT: ; implicit-def: $vgpr4
; GFX10W32-NEXT: s_and_saveexec_b32 s2, vcc_lo
; GFX10W32-NEXT: s_cbranch_execz .LBB2_2
; GFX10W32-NEXT: ; %bb.1:
; GFX10W32-NEXT: s_mov_b32 s3, s4
; GFX10W32-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10W32-NEXT: v_mov_b32_e32 v4, s3
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: buffer_atomic_add v4, v0, s[4:7], 0 idxen glc
; GFX10W32-NEXT: .LBB2_2:
; GFX10W32-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT: s_or_b32 exec_lo, exec_lo, s2
; GFX10W32-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT: s_waitcnt vmcnt(0)
; GFX10W32-NEXT: v_readfirstlane_b32 s2, v4
; GFX10W32-NEXT: v_mov_b32_e32 v4, v3
; GFX10W32-NEXT: v_add_nc_u32_e32 v4, s2, v4
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: global_store_dword v0, v4, s[0:1]
; GFX10W32-NEXT: s_endpgm
;
; GFX11W64-LABEL: add_i32_varying_vdata:
; GFX11W64: ; %bb.0: ; %entry
; GFX11W64-NEXT: v_mov_b32_e32 v1, v0
; GFX11W64-NEXT: s_not_b64 exec, exec
; GFX11W64-NEXT: v_mov_b32_e32 v1, 0
; GFX11W64-NEXT: s_not_b64 exec, exec
; GFX11W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W64-NEXT: v_mov_b32_e32 v3, 0
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_mov_b32_e32 v2, v1
; GFX11W64-NEXT: v_permlanex16_b32 v2, v2, -1, -1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
; GFX11W64-NEXT: v_readlane_b32 s4, v1, 31
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_mov_b32_e32 v2, s4
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX11W64-NEXT: v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
; GFX11W64-NEXT: v_readlane_b32 s4, v1, 15
; GFX11W64-NEXT: v_readlane_b32 s5, v1, 31
; GFX11W64-NEXT: v_writelane_b32 v3, s4, 16
; GFX11W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX11W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX11W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX11W64-NEXT: v_readlane_b32 s4, v1, 63
; GFX11W64-NEXT: v_readlane_b32 s6, v1, 47
; GFX11W64-NEXT: v_writelane_b32 v3, s5, 32
; GFX11W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX11W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT: v_mbcnt_hi_u32_b32 v4, exec_hi, v0
; GFX11W64-NEXT: v_mov_b32_e32 v0, 0
; GFX11W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX11W64-NEXT: v_writelane_b32 v3, s6, 48
; GFX11W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX11W64-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
; GFX11W64-NEXT: ; implicit-def: $vgpr4
; GFX11W64-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX11W64-NEXT: s_cbranch_execz .LBB2_2
; GFX11W64-NEXT: ; %bb.1:
; GFX11W64-NEXT: s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT: v_mov_b32_e32 v4, s4
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: buffer_atomic_add_u32 v4, v0, s[8:11], 0 idxen glc
; GFX11W64-NEXT: .LBB2_2:
; GFX11W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT: s_waitcnt vmcnt(0)
; GFX11W64-NEXT: v_readfirstlane_b32 s2, v4
; GFX11W64-NEXT: v_mov_b32_e32 v4, v3
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11W64-NEXT: v_add_nc_u32_e32 v4, s2, v4
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: global_store_b32 v0, v4, s[0:1]
; GFX11W64-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT: s_endpgm
;
; GFX11W32-LABEL: add_i32_varying_vdata:
; GFX11W32: ; %bb.0: ; %entry
; GFX11W32-NEXT: v_mov_b32_e32 v1, v0
; GFX11W32-NEXT: s_not_b32 exec_lo, exec_lo
; GFX11W32-NEXT: v_mov_b32_e32 v1, 0
; GFX11W32-NEXT: s_not_b32 exec_lo, exec_lo
; GFX11W32-NEXT: s_or_saveexec_b32 s2, -1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W32-NEXT: v_mov_b32_e32 v3, 0
; GFX11W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W32-NEXT: v_mov_b32_e32 v2, v1
; GFX11W32-NEXT: v_permlanex16_b32 v2, v2, -1, -1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W32-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
; GFX11W32-NEXT: v_readlane_b32 s4, v1, 31
; GFX11W32-NEXT: v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
; GFX11W32-NEXT: v_readlane_b32 s3, v1, 15
; GFX11W32-NEXT: s_mov_b32 exec_lo, s2
; GFX11W32-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W32-NEXT: v_mbcnt_lo_u32_b32 v4, exec_lo, 0
; GFX11W32-NEXT: v_mov_b32_e32 v0, 0
; GFX11W32-NEXT: s_or_saveexec_b32 s2, -1
; GFX11W32-NEXT: v_writelane_b32 v3, s3, 16
; GFX11W32-NEXT: s_mov_b32 exec_lo, s2
; GFX11W32-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v4
; GFX11W32-NEXT: ; implicit-def: $vgpr4
; GFX11W32-NEXT: s_and_saveexec_b32 s2, vcc_lo
; GFX11W32-NEXT: s_cbranch_execz .LBB2_2
; GFX11W32-NEXT: ; %bb.1:
; GFX11W32-NEXT: s_mov_b32 s3, s4
; GFX11W32-NEXT: s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W32-NEXT: v_mov_b32_e32 v4, s3
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: buffer_atomic_add_u32 v4, v0, s[4:7], 0 idxen glc
; GFX11W32-NEXT: .LBB2_2:
; GFX11W32-NEXT: s_or_b32 exec_lo, exec_lo, s2
; GFX11W32-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT: s_waitcnt vmcnt(0)
; GFX11W32-NEXT: v_readfirstlane_b32 s2, v4
; GFX11W32-NEXT: v_mov_b32_e32 v4, v3
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT: v_add_nc_u32_e32 v4, s2, v4
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: global_store_b32 v0, v4, s[0:1]
; GFX11W32-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT: s_endpgm
entry:
%lane = call i32 @llvm.amdgcn.workitem.id.x()
%old = call i32 @llvm.amdgcn.struct.buffer.atomic.add(i32 %lane, <4 x i32> %inout, i32 0, i32 0, i32 0, i32 0)
store i32 %old, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @add_i32_varying_vindex(ptr addrspace(1) %out, <4 x i32> %inout) {
; GFX6-LABEL: add_i32_varying_vindex:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xd
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT: v_mov_b32_e32 v1, 1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: buffer_atomic_add v1, v0, s[4:7], 0 idxen glc
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: buffer_store_dword v1, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: add_i32_varying_vindex:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX8-NEXT: v_mov_b32_e32 v2, 1
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: buffer_atomic_add v2, v0, s[4:7], 0 idxen glc
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX9-LABEL: add_i32_varying_vindex:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX9-NEXT: v_mov_b32_e32 v1, 1
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: buffer_atomic_add v1, v0, s[4:7], 0 idxen glc
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; GFX10-LABEL: add_i32_varying_vindex:
; GFX10: ; %bb.0: ; %entry
; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10-NEXT: v_mov_b32_e32 v1, 1
; GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
; GFX10-NEXT: buffer_atomic_add v1, v0, s[4:7], 0 idxen glc
; GFX10-NEXT: v_mov_b32_e32 v0, 0
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: global_store_dword v0, v1, s[0:1]
; GFX10-NEXT: s_endpgm
;
; GFX11-LABEL: add_i32_varying_vindex:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_load_b128 s[4:7], s[0:1], 0x34
; GFX11-NEXT: v_mov_b32_e32 v1, 1
; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: buffer_atomic_add_u32 v1, v0, s[4:7], 0 idxen glc
; GFX11-NEXT: v_mov_b32_e32 v0, 0
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
entry:
%lane = call i32 @llvm.amdgcn.workitem.id.x()
%old = call i32 @llvm.amdgcn.struct.buffer.atomic.add(i32 1, <4 x i32> %inout, i32 %lane, i32 0, i32 0, i32 0)
store i32 %old, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @add_i32_varying_offset(ptr addrspace(1) %out, <4 x i32> %inout) {
; GFX6-LABEL: add_i32_varying_offset:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: v_mov_b32_e32 v1, v0
; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xd
; GFX6-NEXT: s_mov_b32 s2, 0
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT: v_mov_b32_e32 v0, s2
; GFX6-NEXT: v_mov_b32_e32 v2, 1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: buffer_atomic_add v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: buffer_store_dword v2, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: add_i32_varying_offset:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX8-NEXT: s_mov_b32 s2, 0
; GFX8-NEXT: v_mov_b32_e32 v1, v0
; GFX8-NEXT: v_mov_b32_e32 v0, s2
; GFX8-NEXT: v_mov_b32_e32 v2, 1
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: buffer_atomic_add v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX9-LABEL: add_i32_varying_offset:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX9-NEXT: s_mov_b32 s2, 0
; GFX9-NEXT: v_mov_b32_e32 v1, v0
; GFX9-NEXT: v_mov_b32_e32 v0, s2
; GFX9-NEXT: v_mov_b32_e32 v2, 1
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: buffer_atomic_add v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v2, s[0:1]
; GFX9-NEXT: s_endpgm
;
; GFX10-LABEL: add_i32_varying_offset:
; GFX10: ; %bb.0: ; %entry
; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10-NEXT: s_mov_b32 s2, 0
; GFX10-NEXT: v_mov_b32_e32 v1, v0
; GFX10-NEXT: v_mov_b32_e32 v0, s2
; GFX10-NEXT: v_mov_b32_e32 v2, 1
; GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
; GFX10-NEXT: buffer_atomic_add v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX10-NEXT: v_mov_b32_e32 v0, 0
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: global_store_dword v0, v2, s[0:1]
; GFX10-NEXT: s_endpgm
;
; GFX11W64-LABEL: add_i32_varying_offset:
; GFX11W64: ; %bb.0: ; %entry
; GFX11W64-NEXT: s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W64-NEXT: s_mov_b32 s2, 0
; GFX11W64-NEXT: v_mov_b32_e32 v1, v0
; GFX11W64-NEXT: v_mov_b32_e32 v0, s2
; GFX11W64-NEXT: v_mov_b32_e32 v2, 1
; GFX11W64-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: buffer_atomic_add_u32 v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX11W64-NEXT: v_mov_b32_e32 v0, 0
; GFX11W64-NEXT: s_waitcnt vmcnt(0)
; GFX11W64-NEXT: global_store_b32 v0, v2, s[0:1]
; GFX11W64-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT: s_endpgm
;
; GFX11W32-LABEL: add_i32_varying_offset:
; GFX11W32: ; %bb.0: ; %entry
; GFX11W32-NEXT: s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W32-NEXT: s_mov_b32 s2, 0
; GFX11W32-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W32-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s2
; GFX11W32-NEXT: v_mov_b32_e32 v2, 1
; GFX11W32-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: buffer_atomic_add_u32 v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX11W32-NEXT: v_mov_b32_e32 v0, 0
; GFX11W32-NEXT: s_waitcnt vmcnt(0)
; GFX11W32-NEXT: global_store_b32 v0, v2, s[0:1]
; GFX11W32-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT: s_endpgm
entry:
%lane = call i32 @llvm.amdgcn.workitem.id.x()
%old = call i32 @llvm.amdgcn.struct.buffer.atomic.add(i32 1, <4 x i32> %inout, i32 0, i32 %lane, i32 0, i32 0)
store i32 %old, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @sub_i32_constant(ptr addrspace(1) %out, <4 x i32> %inout) {
; GFX6-LABEL: sub_i32_constant:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_mov_b64 s[4:5], exec
; GFX6-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s4, 0
; GFX6-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s5, v0
; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX6-NEXT: ; implicit-def: $vgpr1
; GFX6-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX6-NEXT: s_cbranch_execz .LBB5_2
; GFX6-NEXT: ; %bb.1:
; GFX6-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xd
; GFX6-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX6-NEXT: s_mul_i32 s4, s4, 5
; GFX6-NEXT: v_mov_b32_e32 v1, s4
; GFX6-NEXT: v_mov_b32_e32 v2, 0
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: buffer_atomic_sub v1, v2, s[8:11], 0 idxen glc
; GFX6-NEXT: .LBB5_2:
; GFX6-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: v_readfirstlane_b32 s4, v1
; GFX6-NEXT: v_mul_u32_u24_e32 v0, 5, v0
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, s4, v0
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: sub_i32_constant:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_mov_b64 s[4:5], exec
; GFX8-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX8-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT: ; implicit-def: $vgpr1
; GFX8-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT: s_cbranch_execz .LBB5_2
; GFX8-NEXT: ; %bb.1:
; GFX8-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX8-NEXT: s_mul_i32 s4, s4, 5
; GFX8-NEXT: v_mov_b32_e32 v1, s4
; GFX8-NEXT: v_mov_b32_e32 v2, 0
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: buffer_atomic_sub v1, v2, s[8:11], 0 idxen glc
; GFX8-NEXT: .LBB5_2:
; GFX8-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: v_readfirstlane_b32 s2, v1
; GFX8-NEXT: v_mul_u32_u24_e32 v0, 5, v0
; GFX8-NEXT: v_sub_u32_e32 v2, vcc, s2, v0
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX9-LABEL: sub_i32_constant:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: s_mov_b64 s[4:5], exec
; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT: ; implicit-def: $vgpr1
; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT: s_cbranch_execz .LBB5_2
; GFX9-NEXT: ; %bb.1:
; GFX9-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX9-NEXT: s_mul_i32 s4, s4, 5
; GFX9-NEXT: v_mov_b32_e32 v1, s4
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: buffer_atomic_sub v1, v2, s[8:11], 0 idxen glc
; GFX9-NEXT: .LBB5_2:
; GFX9-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_readfirstlane_b32 s2, v1
; GFX9-NEXT: v_mul_u32_u24_e32 v0, 5, v0
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: v_sub_u32_e32 v0, s2, v0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v2, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; GFX10W64-LABEL: sub_i32_constant:
; GFX10W64: ; %bb.0: ; %entry
; GFX10W64-NEXT: s_mov_b64 s[4:5], exec
; GFX10W64-NEXT: ; implicit-def: $vgpr1
; GFX10W64-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W64-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX10W64-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX10W64-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT: s_cbranch_execz .LBB5_2
; GFX10W64-NEXT: ; %bb.1:
; GFX10W64-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX10W64-NEXT: v_mov_b32_e32 v2, 0
; GFX10W64-NEXT: s_mul_i32 s4, s4, 5
; GFX10W64-NEXT: v_mov_b32_e32 v1, s4
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: buffer_atomic_sub v1, v2, s[8:11], 0 idxen glc
; GFX10W64-NEXT: .LBB5_2:
; GFX10W64-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT: s_waitcnt vmcnt(0)
; GFX10W64-NEXT: v_readfirstlane_b32 s2, v1
; GFX10W64-NEXT: v_mul_u32_u24_e32 v0, 5, v0
; GFX10W64-NEXT: v_mov_b32_e32 v1, 0
; GFX10W64-NEXT: v_sub_nc_u32_e32 v0, s2, v0
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: global_store_dword v1, v0, s[0:1]
; GFX10W64-NEXT: s_endpgm
;
; GFX10W32-LABEL: sub_i32_constant:
; GFX10W32: ; %bb.0: ; %entry
; GFX10W32-NEXT: s_mov_b32 s3, exec_lo
; GFX10W32-NEXT: ; implicit-def: $vgpr1
; GFX10W32-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
; GFX10W32-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10W32-NEXT: s_and_saveexec_b32 s2, vcc_lo
; GFX10W32-NEXT: s_cbranch_execz .LBB5_2
; GFX10W32-NEXT: ; %bb.1:
; GFX10W32-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10W32-NEXT: s_bcnt1_i32_b32 s3, s3
; GFX10W32-NEXT: v_mov_b32_e32 v2, 0
; GFX10W32-NEXT: s_mul_i32 s3, s3, 5
; GFX10W32-NEXT: v_mov_b32_e32 v1, s3
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: buffer_atomic_sub v1, v2, s[4:7], 0 idxen glc
; GFX10W32-NEXT: .LBB5_2:
; GFX10W32-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT: s_or_b32 exec_lo, exec_lo, s2
; GFX10W32-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT: s_waitcnt vmcnt(0)
; GFX10W32-NEXT: v_readfirstlane_b32 s2, v1
; GFX10W32-NEXT: v_mul_u32_u24_e32 v0, 5, v0
; GFX10W32-NEXT: v_mov_b32_e32 v1, 0
; GFX10W32-NEXT: v_sub_nc_u32_e32 v0, s2, v0
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: global_store_dword v1, v0, s[0:1]
; GFX10W32-NEXT: s_endpgm
;
; GFX11W64-LABEL: sub_i32_constant:
; GFX11W64: ; %bb.0: ; %entry
; GFX11W64-NEXT: s_mov_b64 s[4:5], exec
; GFX11W64-NEXT: s_mov_b64 s[2:3], exec
; GFX11W64-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W64-NEXT: ; implicit-def: $vgpr1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX11W64-NEXT: v_cmpx_eq_u32_e32 0, v0
; GFX11W64-NEXT: s_cbranch_execz .LBB5_2
; GFX11W64-NEXT: ; %bb.1:
; GFX11W64-NEXT: s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX11W64-NEXT: v_mov_b32_e32 v2, 0
; GFX11W64-NEXT: s_mul_i32 s4, s4, 5
; GFX11W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT: v_mov_b32_e32 v1, s4
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: buffer_atomic_sub_u32 v1, v2, s[8:11], 0 idxen glc
; GFX11W64-NEXT: .LBB5_2:
; GFX11W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT: s_waitcnt vmcnt(0)
; GFX11W64-NEXT: v_readfirstlane_b32 s2, v1
; GFX11W64-NEXT: v_mul_u32_u24_e32 v0, 5, v0
; GFX11W64-NEXT: v_mov_b32_e32 v1, 0
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT: v_sub_nc_u32_e32 v0, s2, v0
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: global_store_b32 v1, v0, s[0:1]
; GFX11W64-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT: s_endpgm
;
; GFX11W32-LABEL: sub_i32_constant:
; GFX11W32: ; %bb.0: ; %entry
; GFX11W32-NEXT: s_mov_b32 s3, exec_lo
; GFX11W32-NEXT: s_mov_b32 s2, exec_lo
; GFX11W32-NEXT: v_mbcnt_lo_u32_b32 v0, s3, 0
; GFX11W32-NEXT: ; implicit-def: $vgpr1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT: v_cmpx_eq_u32_e32 0, v0
; GFX11W32-NEXT: s_cbranch_execz .LBB5_2
; GFX11W32-NEXT: ; %bb.1:
; GFX11W32-NEXT: s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W32-NEXT: s_bcnt1_i32_b32 s3, s3
; GFX11W32-NEXT: v_mov_b32_e32 v2, 0
; GFX11W32-NEXT: s_mul_i32 s3, s3, 5
; GFX11W32-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W32-NEXT: v_mov_b32_e32 v1, s3
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: buffer_atomic_sub_u32 v1, v2, s[4:7], 0 idxen glc
; GFX11W32-NEXT: .LBB5_2:
; GFX11W32-NEXT: s_or_b32 exec_lo, exec_lo, s2
; GFX11W32-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT: s_waitcnt vmcnt(0)
; GFX11W32-NEXT: v_readfirstlane_b32 s2, v1
; GFX11W32-NEXT: v_mul_u32_u24_e32 v0, 5, v0
; GFX11W32-NEXT: v_mov_b32_e32 v1, 0
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W32-NEXT: v_sub_nc_u32_e32 v0, s2, v0
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: global_store_b32 v1, v0, s[0:1]
; GFX11W32-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT: s_endpgm
entry:
%old = call i32 @llvm.amdgcn.struct.buffer.atomic.sub(i32 5, <4 x i32> %inout, i32 0, i32 0, i32 0, i32 0)
store i32 %old, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @sub_i32_uniform(ptr addrspace(1) %out, <4 x i32> %inout, i32 %subitive) {
; GFX6-LABEL: sub_i32_uniform:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_mov_b64 s[4:5], exec
; GFX6-NEXT: s_load_dword s6, s[0:1], 0x11
; GFX6-NEXT: v_mbcnt_lo_u32_b32_e64 v0, s4, 0
; GFX6-NEXT: v_mbcnt_hi_u32_b32_e32 v0, s5, v0
; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX6-NEXT: ; implicit-def: $vgpr1
; GFX6-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX6-NEXT: s_cbranch_execz .LBB6_2
; GFX6-NEXT: ; %bb.1:
; GFX6-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0xd
; GFX6-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_mul_i32 s4, s6, s4
; GFX6-NEXT: v_mov_b32_e32 v1, s4
; GFX6-NEXT: v_mov_b32_e32 v2, 0
; GFX6-NEXT: buffer_atomic_sub v1, v2, s[8:11], 0 idxen glc
; GFX6-NEXT: .LBB6_2:
; GFX6-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: v_readfirstlane_b32 s4, v1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: v_mul_lo_u32 v0, s6, v0
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, s4, v0
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: sub_i32_uniform:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_load_dword s6, s[0:1], 0x44
; GFX8-NEXT: s_mov_b64 s[4:5], exec
; GFX8-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX8-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX8-NEXT: ; implicit-def: $vgpr1
; GFX8-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT: s_cbranch_execz .LBB6_2
; GFX8-NEXT: ; %bb.1:
; GFX8-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: s_mul_i32 s4, s6, s4
; GFX8-NEXT: v_mov_b32_e32 v1, s4
; GFX8-NEXT: v_mov_b32_e32 v2, 0
; GFX8-NEXT: buffer_atomic_sub v1, v2, s[8:11], 0 idxen glc
; GFX8-NEXT: .LBB6_2:
; GFX8-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mul_lo_u32 v0, s6, v0
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: v_readfirstlane_b32 s2, v1
; GFX8-NEXT: v_sub_u32_e32 v2, vcc, s2, v0
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX9-LABEL: sub_i32_uniform:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: s_load_dword s6, s[0:1], 0x44
; GFX9-NEXT: s_mov_b64 s[4:5], exec
; GFX9-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX9-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX9-NEXT: ; implicit-def: $vgpr1
; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT: s_cbranch_execz .LBB6_2
; GFX9-NEXT: ; %bb.1:
; GFX9-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_mul_i32 s4, s6, s4
; GFX9-NEXT: v_mov_b32_e32 v1, s4
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: buffer_atomic_sub v1, v2, s[8:11], 0 idxen glc
; GFX9-NEXT: .LBB6_2:
; GFX9-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mul_lo_u32 v0, s6, v0
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_readfirstlane_b32 s2, v1
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: v_sub_u32_e32 v0, s2, v0
; GFX9-NEXT: global_store_dword v2, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; GFX10W64-LABEL: sub_i32_uniform:
; GFX10W64: ; %bb.0: ; %entry
; GFX10W64-NEXT: s_load_dword s6, s[0:1], 0x44
; GFX10W64-NEXT: s_mov_b64 s[4:5], exec
; GFX10W64-NEXT: ; implicit-def: $vgpr1
; GFX10W64-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W64-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX10W64-NEXT: v_cmp_eq_u32_e32 vcc, 0, v0
; GFX10W64-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT: s_cbranch_execz .LBB6_2
; GFX10W64-NEXT: ; %bb.1:
; GFX10W64-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX10W64-NEXT: v_mov_b32_e32 v2, 0
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: s_mul_i32 s4, s6, s4
; GFX10W64-NEXT: v_mov_b32_e32 v1, s4
; GFX10W64-NEXT: buffer_atomic_sub v1, v2, s[8:11], 0 idxen glc
; GFX10W64-NEXT: .LBB6_2:
; GFX10W64-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: v_mul_lo_u32 v0, s6, v0
; GFX10W64-NEXT: s_waitcnt vmcnt(0)
; GFX10W64-NEXT: v_readfirstlane_b32 s2, v1
; GFX10W64-NEXT: v_mov_b32_e32 v1, 0
; GFX10W64-NEXT: v_sub_nc_u32_e32 v0, s2, v0
; GFX10W64-NEXT: global_store_dword v1, v0, s[0:1]
; GFX10W64-NEXT: s_endpgm
;
; GFX10W32-LABEL: sub_i32_uniform:
; GFX10W32: ; %bb.0: ; %entry
; GFX10W32-NEXT: s_load_dword s2, s[0:1], 0x44
; GFX10W32-NEXT: s_mov_b32 s4, exec_lo
; GFX10W32-NEXT: ; implicit-def: $vgpr1
; GFX10W32-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX10W32-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v0
; GFX10W32-NEXT: s_and_saveexec_b32 s3, vcc_lo
; GFX10W32-NEXT: s_cbranch_execz .LBB6_2
; GFX10W32-NEXT: ; %bb.1:
; GFX10W32-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W32-NEXT: s_bcnt1_i32_b32 s4, s4
; GFX10W32-NEXT: v_mov_b32_e32 v2, 0
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: s_mul_i32 s4, s2, s4
; GFX10W32-NEXT: v_mov_b32_e32 v1, s4
; GFX10W32-NEXT: buffer_atomic_sub v1, v2, s[8:11], 0 idxen glc
; GFX10W32-NEXT: .LBB6_2:
; GFX10W32-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT: s_or_b32 exec_lo, exec_lo, s3
; GFX10W32-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: v_mul_lo_u32 v0, s2, v0
; GFX10W32-NEXT: s_waitcnt vmcnt(0)
; GFX10W32-NEXT: v_readfirstlane_b32 s2, v1
; GFX10W32-NEXT: v_mov_b32_e32 v1, 0
; GFX10W32-NEXT: v_sub_nc_u32_e32 v0, s2, v0
; GFX10W32-NEXT: global_store_dword v1, v0, s[0:1]
; GFX10W32-NEXT: s_endpgm
;
; GFX11W64-LABEL: sub_i32_uniform:
; GFX11W64: ; %bb.0: ; %entry
; GFX11W64-NEXT: s_load_b32 s6, s[0:1], 0x44
; GFX11W64-NEXT: s_mov_b64 s[4:5], exec
; GFX11W64-NEXT: s_mov_b64 s[2:3], exec
; GFX11W64-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W64-NEXT: ; implicit-def: $vgpr1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_mbcnt_hi_u32_b32 v0, s5, v0
; GFX11W64-NEXT: v_cmpx_eq_u32_e32 0, v0
; GFX11W64-NEXT: s_cbranch_execz .LBB6_2
; GFX11W64-NEXT: ; %bb.1:
; GFX11W64-NEXT: s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT: s_bcnt1_i32_b64 s4, s[4:5]
; GFX11W64-NEXT: v_mov_b32_e32 v2, 0
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: s_mul_i32 s4, s6, s4
; GFX11W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT: v_mov_b32_e32 v1, s4
; GFX11W64-NEXT: buffer_atomic_sub_u32 v1, v2, s[8:11], 0 idxen glc
; GFX11W64-NEXT: .LBB6_2:
; GFX11W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: v_mul_lo_u32 v0, s6, v0
; GFX11W64-NEXT: s_waitcnt vmcnt(0)
; GFX11W64-NEXT: v_readfirstlane_b32 s2, v1
; GFX11W64-NEXT: v_mov_b32_e32 v1, 0
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W64-NEXT: v_sub_nc_u32_e32 v0, s2, v0
; GFX11W64-NEXT: global_store_b32 v1, v0, s[0:1]
; GFX11W64-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT: s_endpgm
;
; GFX11W32-LABEL: sub_i32_uniform:
; GFX11W32: ; %bb.0: ; %entry
; GFX11W32-NEXT: s_load_b32 s2, s[0:1], 0x44
; GFX11W32-NEXT: s_mov_b32 s4, exec_lo
; GFX11W32-NEXT: s_mov_b32 s3, exec_lo
; GFX11W32-NEXT: v_mbcnt_lo_u32_b32 v0, s4, 0
; GFX11W32-NEXT: ; implicit-def: $vgpr1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT: v_cmpx_eq_u32_e32 0, v0
; GFX11W32-NEXT: s_cbranch_execz .LBB6_2
; GFX11W32-NEXT: ; %bb.1:
; GFX11W32-NEXT: s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W32-NEXT: s_bcnt1_i32_b32 s4, s4
; GFX11W32-NEXT: v_mov_b32_e32 v2, 0
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: s_mul_i32 s4, s2, s4
; GFX11W32-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W32-NEXT: v_mov_b32_e32 v1, s4
; GFX11W32-NEXT: buffer_atomic_sub_u32 v1, v2, s[8:11], 0 idxen glc
; GFX11W32-NEXT: .LBB6_2:
; GFX11W32-NEXT: s_or_b32 exec_lo, exec_lo, s3
; GFX11W32-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: v_mul_lo_u32 v0, s2, v0
; GFX11W32-NEXT: s_waitcnt vmcnt(0)
; GFX11W32-NEXT: v_readfirstlane_b32 s2, v1
; GFX11W32-NEXT: v_mov_b32_e32 v1, 0
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX11W32-NEXT: v_sub_nc_u32_e32 v0, s2, v0
; GFX11W32-NEXT: global_store_b32 v1, v0, s[0:1]
; GFX11W32-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT: s_endpgm
entry:
%old = call i32 @llvm.amdgcn.struct.buffer.atomic.sub(i32 %subitive, <4 x i32> %inout, i32 0, i32 0, i32 0, i32 0)
store i32 %old, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @sub_i32_varying_vdata(ptr addrspace(1) %out, <4 x i32> %inout) {
; GFX6-LABEL: sub_i32_varying_vdata:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xd
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT: v_mov_b32_e32 v1, 0
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: buffer_atomic_sub v0, v1, s[4:7], 0 idxen glc
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: sub_i32_varying_vdata:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: v_mov_b32_e32 v3, 0
; GFX8-NEXT: v_mbcnt_lo_u32_b32 v4, exec_lo, 0
; GFX8-NEXT: v_mbcnt_hi_u32_b32 v4, exec_hi, v4
; GFX8-NEXT: v_mov_b32_e32 v1, v0
; GFX8-NEXT: s_not_b64 exec, exec
; GFX8-NEXT: v_mov_b32_e32 v1, 0
; GFX8-NEXT: s_not_b64 exec, exec
; GFX8-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX8-NEXT: v_mov_b32_e32 v2, 0
; GFX8-NEXT: s_nop 0
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX8-NEXT: s_nop 1
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX8-NEXT: s_nop 1
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX8-NEXT: s_nop 1
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
; GFX8-NEXT: s_nop 1
; GFX8-NEXT: v_add_u32_dpp v1, vcc, v1, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
; GFX8-NEXT: v_readlane_b32 s4, v1, 63
; GFX8-NEXT: s_nop 0
; GFX8-NEXT: v_mov_b32_dpp v2, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
; GFX8-NEXT: s_mov_b64 exec, s[2:3]
; GFX8-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
; GFX8-NEXT: ; implicit-def: $vgpr0
; GFX8-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX8-NEXT: s_cbranch_execz .LBB7_2
; GFX8-NEXT: ; %bb.1:
; GFX8-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX8-NEXT: v_mov_b32_e32 v0, s4
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: buffer_atomic_sub v0, v3, s[8:11], 0 idxen glc
; GFX8-NEXT: .LBB7_2:
; GFX8-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: v_readfirstlane_b32 s2, v0
; GFX8-NEXT: v_mov_b32_e32 v0, v2
; GFX8-NEXT: v_sub_u32_e32 v0, vcc, s2, v0
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v4, s1
; GFX8-NEXT: v_mov_b32_e32 v3, s0
; GFX8-NEXT: flat_store_dword v[3:4], v0
; GFX8-NEXT: s_endpgm
;
; GFX9-LABEL: sub_i32_varying_vdata:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: v_mov_b32_e32 v3, 0
; GFX9-NEXT: v_mbcnt_lo_u32_b32 v4, exec_lo, 0
; GFX9-NEXT: v_mbcnt_hi_u32_b32 v4, exec_hi, v4
; GFX9-NEXT: v_mov_b32_e32 v1, v0
; GFX9-NEXT: s_not_b64 exec, exec
; GFX9-NEXT: v_mov_b32_e32 v1, 0
; GFX9-NEXT: s_not_b64 exec, exec
; GFX9-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX9-NEXT: v_mov_b32_e32 v2, 0
; GFX9-NEXT: s_nop 0
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX9-NEXT: s_nop 1
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX9-NEXT: s_nop 1
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX9-NEXT: s_nop 1
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_bcast:15 row_mask:0xa bank_mask:0xf
; GFX9-NEXT: s_nop 1
; GFX9-NEXT: v_add_u32_dpp v1, v1, v1 row_bcast:31 row_mask:0xc bank_mask:0xf
; GFX9-NEXT: v_readlane_b32 s4, v1, 63
; GFX9-NEXT: s_nop 0
; GFX9-NEXT: v_mov_b32_dpp v2, v1 wave_shr:1 row_mask:0xf bank_mask:0xf
; GFX9-NEXT: s_mov_b64 exec, s[2:3]
; GFX9-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
; GFX9-NEXT: ; implicit-def: $vgpr0
; GFX9-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX9-NEXT: s_cbranch_execz .LBB7_2
; GFX9-NEXT: ; %bb.1:
; GFX9-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX9-NEXT: v_mov_b32_e32 v0, s4
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: buffer_atomic_sub v0, v3, s[8:11], 0 idxen glc
; GFX9-NEXT: .LBB7_2:
; GFX9-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_readfirstlane_b32 s2, v0
; GFX9-NEXT: v_mov_b32_e32 v0, v2
; GFX9-NEXT: v_sub_u32_e32 v0, s2, v0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: global_store_dword v3, v0, s[0:1]
; GFX9-NEXT: s_endpgm
;
; GFX10W64-LABEL: sub_i32_varying_vdata:
; GFX10W64: ; %bb.0: ; %entry
; GFX10W64-NEXT: v_mov_b32_e32 v1, v0
; GFX10W64-NEXT: s_not_b64 exec, exec
; GFX10W64-NEXT: v_mov_b32_e32 v1, 0
; GFX10W64-NEXT: s_not_b64 exec, exec
; GFX10W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W64-NEXT: v_mov_b32_e32 v3, 0
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W64-NEXT: v_mov_b32_e32 v2, v1
; GFX10W64-NEXT: v_permlanex16_b32 v2, v2, -1, -1
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
; GFX10W64-NEXT: v_readlane_b32 s4, v1, 31
; GFX10W64-NEXT: v_mov_b32_e32 v2, s4
; GFX10W64-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
; GFX10W64-NEXT: v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
; GFX10W64-NEXT: v_readlane_b32 s4, v1, 15
; GFX10W64-NEXT: v_readlane_b32 s5, v1, 31
; GFX10W64-NEXT: v_writelane_b32 v3, s4, 16
; GFX10W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX10W64-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX10W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX10W64-NEXT: v_readlane_b32 s4, v1, 63
; GFX10W64-NEXT: v_readlane_b32 s6, v1, 47
; GFX10W64-NEXT: v_writelane_b32 v3, s5, 32
; GFX10W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX10W64-NEXT: v_mbcnt_hi_u32_b32 v4, exec_hi, v0
; GFX10W64-NEXT: v_mov_b32_e32 v0, 0
; GFX10W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX10W64-NEXT: v_writelane_b32 v3, s6, 48
; GFX10W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX10W64-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
; GFX10W64-NEXT: ; implicit-def: $vgpr4
; GFX10W64-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX10W64-NEXT: s_cbranch_execz .LBB7_2
; GFX10W64-NEXT: ; %bb.1:
; GFX10W64-NEXT: s_load_dwordx4 s[8:11], s[0:1], 0x34
; GFX10W64-NEXT: v_mov_b32_e32 v4, s4
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: buffer_atomic_sub v4, v0, s[8:11], 0 idxen glc
; GFX10W64-NEXT: .LBB7_2:
; GFX10W64-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX10W64-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W64-NEXT: s_waitcnt vmcnt(0)
; GFX10W64-NEXT: v_readfirstlane_b32 s2, v4
; GFX10W64-NEXT: v_mov_b32_e32 v4, v3
; GFX10W64-NEXT: v_sub_nc_u32_e32 v4, s2, v4
; GFX10W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W64-NEXT: global_store_dword v0, v4, s[0:1]
; GFX10W64-NEXT: s_endpgm
;
; GFX10W32-LABEL: sub_i32_varying_vdata:
; GFX10W32: ; %bb.0: ; %entry
; GFX10W32-NEXT: v_mov_b32_e32 v1, v0
; GFX10W32-NEXT: s_not_b32 exec_lo, exec_lo
; GFX10W32-NEXT: v_mov_b32_e32 v1, 0
; GFX10W32-NEXT: s_not_b32 exec_lo, exec_lo
; GFX10W32-NEXT: s_or_saveexec_b32 s2, -1
; GFX10W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W32-NEXT: v_mov_b32_e32 v3, 0
; GFX10W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX10W32-NEXT: v_mov_b32_e32 v2, v1
; GFX10W32-NEXT: v_permlanex16_b32 v2, v2, -1, -1
; GFX10W32-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
; GFX10W32-NEXT: v_readlane_b32 s4, v1, 31
; GFX10W32-NEXT: v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
; GFX10W32-NEXT: v_readlane_b32 s3, v1, 15
; GFX10W32-NEXT: s_mov_b32 exec_lo, s2
; GFX10W32-NEXT: v_mbcnt_lo_u32_b32 v4, exec_lo, 0
; GFX10W32-NEXT: v_mov_b32_e32 v0, 0
; GFX10W32-NEXT: s_or_saveexec_b32 s2, -1
; GFX10W32-NEXT: v_writelane_b32 v3, s3, 16
; GFX10W32-NEXT: s_mov_b32 exec_lo, s2
; GFX10W32-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v4
; GFX10W32-NEXT: ; implicit-def: $vgpr4
; GFX10W32-NEXT: s_and_saveexec_b32 s2, vcc_lo
; GFX10W32-NEXT: s_cbranch_execz .LBB7_2
; GFX10W32-NEXT: ; %bb.1:
; GFX10W32-NEXT: s_mov_b32 s3, s4
; GFX10W32-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10W32-NEXT: v_mov_b32_e32 v4, s3
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: buffer_atomic_sub v4, v0, s[4:7], 0 idxen glc
; GFX10W32-NEXT: .LBB7_2:
; GFX10W32-NEXT: s_waitcnt_depctr 0xffe3
; GFX10W32-NEXT: s_or_b32 exec_lo, exec_lo, s2
; GFX10W32-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10W32-NEXT: s_waitcnt vmcnt(0)
; GFX10W32-NEXT: v_readfirstlane_b32 s2, v4
; GFX10W32-NEXT: v_mov_b32_e32 v4, v3
; GFX10W32-NEXT: v_sub_nc_u32_e32 v4, s2, v4
; GFX10W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX10W32-NEXT: global_store_dword v0, v4, s[0:1]
; GFX10W32-NEXT: s_endpgm
;
; GFX11W64-LABEL: sub_i32_varying_vdata:
; GFX11W64: ; %bb.0: ; %entry
; GFX11W64-NEXT: v_mov_b32_e32 v1, v0
; GFX11W64-NEXT: s_not_b64 exec, exec
; GFX11W64-NEXT: v_mov_b32_e32 v1, 0
; GFX11W64-NEXT: s_not_b64 exec, exec
; GFX11W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W64-NEXT: v_mov_b32_e32 v3, 0
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_mov_b32_e32 v2, v1
; GFX11W64-NEXT: v_permlanex16_b32 v2, v2, -1, -1
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
; GFX11W64-NEXT: v_readlane_b32 s4, v1, 31
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W64-NEXT: v_mov_b32_e32 v2, s4
; GFX11W64-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xc bank_mask:0xf
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_2) | instid1(VALU_DEP_2)
; GFX11W64-NEXT: v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
; GFX11W64-NEXT: v_readlane_b32 s4, v1, 15
; GFX11W64-NEXT: v_readlane_b32 s5, v1, 31
; GFX11W64-NEXT: v_writelane_b32 v3, s4, 16
; GFX11W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX11W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT: v_mbcnt_lo_u32_b32 v0, exec_lo, 0
; GFX11W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX11W64-NEXT: v_readlane_b32 s4, v1, 63
; GFX11W64-NEXT: v_readlane_b32 s6, v1, 47
; GFX11W64-NEXT: v_writelane_b32 v3, s5, 32
; GFX11W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX11W64-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W64-NEXT: v_mbcnt_hi_u32_b32 v4, exec_hi, v0
; GFX11W64-NEXT: v_mov_b32_e32 v0, 0
; GFX11W64-NEXT: s_or_saveexec_b64 s[2:3], -1
; GFX11W64-NEXT: v_writelane_b32 v3, s6, 48
; GFX11W64-NEXT: s_mov_b64 exec, s[2:3]
; GFX11W64-NEXT: v_cmp_eq_u32_e32 vcc, 0, v4
; GFX11W64-NEXT: ; implicit-def: $vgpr4
; GFX11W64-NEXT: s_and_saveexec_b64 s[2:3], vcc
; GFX11W64-NEXT: s_cbranch_execz .LBB7_2
; GFX11W64-NEXT: ; %bb.1:
; GFX11W64-NEXT: s_load_b128 s[8:11], s[0:1], 0x34
; GFX11W64-NEXT: v_mov_b32_e32 v4, s4
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: buffer_atomic_sub_u32 v4, v0, s[8:11], 0 idxen glc
; GFX11W64-NEXT: .LBB7_2:
; GFX11W64-NEXT: s_or_b64 exec, exec, s[2:3]
; GFX11W64-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT: s_waitcnt vmcnt(0)
; GFX11W64-NEXT: v_readfirstlane_b32 s2, v4
; GFX11W64-NEXT: v_mov_b32_e32 v4, v3
; GFX11W64-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11W64-NEXT: v_sub_nc_u32_e32 v4, s2, v4
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: global_store_b32 v0, v4, s[0:1]
; GFX11W64-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT: s_endpgm
;
; GFX11W32-LABEL: sub_i32_varying_vdata:
; GFX11W32: ; %bb.0: ; %entry
; GFX11W32-NEXT: v_mov_b32_e32 v1, v0
; GFX11W32-NEXT: s_not_b32 exec_lo, exec_lo
; GFX11W32-NEXT: v_mov_b32_e32 v1, 0
; GFX11W32-NEXT: s_not_b32 exec_lo, exec_lo
; GFX11W32-NEXT: s_or_saveexec_b32 s2, -1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:1 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W32-NEXT: v_mov_b32_e32 v3, 0
; GFX11W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:2 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:4 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W32-NEXT: v_add_nc_u32_dpp v1, v1, v1 row_shr:8 row_mask:0xf bank_mask:0xf bound_ctrl:1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W32-NEXT: v_mov_b32_e32 v2, v1
; GFX11W32-NEXT: v_permlanex16_b32 v2, v2, -1, -1
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11W32-NEXT: v_add_nc_u32_dpp v1, v2, v1 quad_perm:[0,1,2,3] row_mask:0xa bank_mask:0xf
; GFX11W32-NEXT: v_readlane_b32 s4, v1, 31
; GFX11W32-NEXT: v_mov_b32_dpp v3, v1 row_shr:1 row_mask:0xf bank_mask:0xf
; GFX11W32-NEXT: v_readlane_b32 s3, v1, 15
; GFX11W32-NEXT: s_mov_b32 exec_lo, s2
; GFX11W32-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W32-NEXT: v_mbcnt_lo_u32_b32 v4, exec_lo, 0
; GFX11W32-NEXT: v_mov_b32_e32 v0, 0
; GFX11W32-NEXT: s_or_saveexec_b32 s2, -1
; GFX11W32-NEXT: v_writelane_b32 v3, s3, 16
; GFX11W32-NEXT: s_mov_b32 exec_lo, s2
; GFX11W32-NEXT: v_cmp_eq_u32_e32 vcc_lo, 0, v4
; GFX11W32-NEXT: ; implicit-def: $vgpr4
; GFX11W32-NEXT: s_and_saveexec_b32 s2, vcc_lo
; GFX11W32-NEXT: s_cbranch_execz .LBB7_2
; GFX11W32-NEXT: ; %bb.1:
; GFX11W32-NEXT: s_mov_b32 s3, s4
; GFX11W32-NEXT: s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W32-NEXT: v_mov_b32_e32 v4, s3
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: buffer_atomic_sub_u32 v4, v0, s[4:7], 0 idxen glc
; GFX11W32-NEXT: .LBB7_2:
; GFX11W32-NEXT: s_or_b32 exec_lo, exec_lo, s2
; GFX11W32-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT: s_waitcnt vmcnt(0)
; GFX11W32-NEXT: v_readfirstlane_b32 s2, v4
; GFX11W32-NEXT: v_mov_b32_e32 v4, v3
; GFX11W32-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11W32-NEXT: v_sub_nc_u32_e32 v4, s2, v4
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: global_store_b32 v0, v4, s[0:1]
; GFX11W32-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT: s_endpgm
entry:
%lane = call i32 @llvm.amdgcn.workitem.id.x()
%old = call i32 @llvm.amdgcn.struct.buffer.atomic.sub(i32 %lane, <4 x i32> %inout, i32 0, i32 0, i32 0, i32 0)
store i32 %old, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @sub_i32_varying_vindex(ptr addrspace(1) %out, <4 x i32> %inout) {
; GFX6-LABEL: sub_i32_varying_vindex:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xd
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT: v_mov_b32_e32 v1, 1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: buffer_atomic_sub v1, v0, s[4:7], 0 idxen glc
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: buffer_store_dword v1, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: sub_i32_varying_vindex:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX8-NEXT: v_mov_b32_e32 v2, 1
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: buffer_atomic_sub v2, v0, s[4:7], 0 idxen glc
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX9-LABEL: sub_i32_varying_vindex:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX9-NEXT: v_mov_b32_e32 v1, 1
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: buffer_atomic_sub v1, v0, s[4:7], 0 idxen glc
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: global_store_dword v0, v1, s[0:1]
; GFX9-NEXT: s_endpgm
;
; GFX10-LABEL: sub_i32_varying_vindex:
; GFX10: ; %bb.0: ; %entry
; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10-NEXT: v_mov_b32_e32 v1, 1
; GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
; GFX10-NEXT: buffer_atomic_sub v1, v0, s[4:7], 0 idxen glc
; GFX10-NEXT: v_mov_b32_e32 v0, 0
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: global_store_dword v0, v1, s[0:1]
; GFX10-NEXT: s_endpgm
;
; GFX11-LABEL: sub_i32_varying_vindex:
; GFX11: ; %bb.0: ; %entry
; GFX11-NEXT: s_load_b128 s[4:7], s[0:1], 0x34
; GFX11-NEXT: v_mov_b32_e32 v1, 1
; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11-NEXT: s_waitcnt lgkmcnt(0)
; GFX11-NEXT: buffer_atomic_sub_u32 v1, v0, s[4:7], 0 idxen glc
; GFX11-NEXT: v_mov_b32_e32 v0, 0
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
entry:
%lane = call i32 @llvm.amdgcn.workitem.id.x()
%old = call i32 @llvm.amdgcn.struct.buffer.atomic.sub(i32 1, <4 x i32> %inout, i32 %lane, i32 0, i32 0, i32 0)
store i32 %old, ptr addrspace(1) %out
ret void
}
define amdgpu_kernel void @sub_i32_varying_offset(ptr addrspace(1) %out, <4 x i32> %inout) {
; GFX6-LABEL: sub_i32_varying_offset:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: v_mov_b32_e32 v1, v0
; GFX6-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0xd
; GFX6-NEXT: s_mov_b32 s2, 0
; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
; GFX6-NEXT: v_mov_b32_e32 v0, s2
; GFX6-NEXT: v_mov_b32_e32 v2, 1
; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: buffer_atomic_sub v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX6-NEXT: s_mov_b32 s3, 0xf000
; GFX6-NEXT: s_mov_b32 s2, -1
; GFX6-NEXT: s_waitcnt vmcnt(0)
; GFX6-NEXT: buffer_store_dword v2, off, s[0:3], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: sub_i32_varying_offset:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX8-NEXT: s_mov_b32 s2, 0
; GFX8-NEXT: v_mov_b32_e32 v1, v0
; GFX8-NEXT: v_mov_b32_e32 v0, s2
; GFX8-NEXT: v_mov_b32_e32 v2, 1
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: buffer_atomic_sub v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
; GFX8-NEXT: v_mov_b32_e32 v0, s0
; GFX8-NEXT: v_mov_b32_e32 v1, s1
; GFX8-NEXT: s_waitcnt vmcnt(0)
; GFX8-NEXT: flat_store_dword v[0:1], v2
; GFX8-NEXT: s_endpgm
;
; GFX9-LABEL: sub_i32_varying_offset:
; GFX9: ; %bb.0: ; %entry
; GFX9-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX9-NEXT: s_mov_b32 s2, 0
; GFX9-NEXT: v_mov_b32_e32 v1, v0
; GFX9-NEXT: v_mov_b32_e32 v0, s2
; GFX9-NEXT: v_mov_b32_e32 v2, 1
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: buffer_atomic_sub v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT: global_store_dword v0, v2, s[0:1]
; GFX9-NEXT: s_endpgm
;
; GFX10-LABEL: sub_i32_varying_offset:
; GFX10: ; %bb.0: ; %entry
; GFX10-NEXT: s_load_dwordx4 s[4:7], s[0:1], 0x34
; GFX10-NEXT: s_mov_b32 s2, 0
; GFX10-NEXT: v_mov_b32_e32 v1, v0
; GFX10-NEXT: v_mov_b32_e32 v0, s2
; GFX10-NEXT: v_mov_b32_e32 v2, 1
; GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
; GFX10-NEXT: buffer_atomic_sub v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX10-NEXT: v_mov_b32_e32 v0, 0
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: global_store_dword v0, v2, s[0:1]
; GFX10-NEXT: s_endpgm
;
; GFX11W64-LABEL: sub_i32_varying_offset:
; GFX11W64: ; %bb.0: ; %entry
; GFX11W64-NEXT: s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W64-NEXT: s_mov_b32 s2, 0
; GFX11W64-NEXT: v_mov_b32_e32 v1, v0
; GFX11W64-NEXT: v_mov_b32_e32 v0, s2
; GFX11W64-NEXT: v_mov_b32_e32 v2, 1
; GFX11W64-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W64-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W64-NEXT: buffer_atomic_sub_u32 v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX11W64-NEXT: v_mov_b32_e32 v0, 0
; GFX11W64-NEXT: s_waitcnt vmcnt(0)
; GFX11W64-NEXT: global_store_b32 v0, v2, s[0:1]
; GFX11W64-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W64-NEXT: s_endpgm
;
; GFX11W32-LABEL: sub_i32_varying_offset:
; GFX11W32: ; %bb.0: ; %entry
; GFX11W32-NEXT: s_load_b128 s[4:7], s[0:1], 0x34
; GFX11W32-NEXT: s_mov_b32 s2, 0
; GFX11W32-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11W32-NEXT: v_dual_mov_b32 v1, v0 :: v_dual_mov_b32 v0, s2
; GFX11W32-NEXT: v_mov_b32_e32 v2, 1
; GFX11W32-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
; GFX11W32-NEXT: s_waitcnt lgkmcnt(0)
; GFX11W32-NEXT: buffer_atomic_sub_u32 v2, v[0:1], s[4:7], 0 idxen offen glc
; GFX11W32-NEXT: v_mov_b32_e32 v0, 0
; GFX11W32-NEXT: s_waitcnt vmcnt(0)
; GFX11W32-NEXT: global_store_b32 v0, v2, s[0:1]
; GFX11W32-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11W32-NEXT: s_endpgm
entry:
%lane = call i32 @llvm.amdgcn.workitem.id.x()
%old = call i32 @llvm.amdgcn.struct.buffer.atomic.sub(i32 1, <4 x i32> %inout, i32 0, i32 %lane, i32 0, i32 0)
store i32 %old, ptr addrspace(1) %out
ret void
}