maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-amdgpu-atomic-c...

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

827 lines
59 KiB
Plaintext
Raw Normal View History

AMDGPU/GlobalISel: Select MUBUF path for global atomic cmpxchg I'm not sure why this isn't a pattern, but the DAG manually selects this.
2020-02-10 11:32:44 +08:00
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX6 %s
# RUN: llc -march=amdgcn -mcpu=hawaii -mattr=-flat-for-global -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX7 %s
# RUN: llc -march=amdgcn -mcpu=hawaii -mattr=+flat-for-global -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX7-FLAT %s
# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX8 %s
# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9 %s
# RUN: llc -march=amdgcn -mcpu=gfx1010 -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX10 %s
---
name: amdgpu_atomic_cmpxchg_s32_global
legalized: true
regBankSelected: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
; GFX6: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX6: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX6: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX6: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX6: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX6: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX6: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX6: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX6: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX6: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX6: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
; GFX6: $vgpr0 = COPY [[COPY3]]
; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX7: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX7: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX7: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX7: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
; GFX7: $vgpr0 = COPY [[COPY3]]
; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX7-FLAT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7-FLAT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7-FLAT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7-FLAT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7-FLAT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX7-FLAT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX8: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX8: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX8: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX9: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX9: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX10: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX10: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s32) = COPY $vgpr2
%2:vgpr(s32) = COPY $vgpr3
%3:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
%4:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %0, %3 :: (load store seq_cst 4, addrspace 1)
$vgpr0 = COPY %4
...
---
name: amdgpu_atomic_cmpxchg_s32_global_gep4
legalized: true
regBankSelected: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
; GFX6: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX6: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX6: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX6: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX6: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX6: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX6: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX6: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX6: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX6: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 4, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX6: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
; GFX6: $vgpr0 = COPY [[COPY3]]
; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX7: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX7: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX7: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX7: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 4, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
; GFX7: $vgpr0 = COPY [[COPY3]]
; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX7-FLAT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7-FLAT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7-FLAT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7-FLAT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7-FLAT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4, implicit $exec
; GFX7-FLAT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX7-FLAT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
; GFX7-FLAT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
; GFX7-FLAT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
; GFX7-FLAT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
; GFX7-FLAT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
AMDGPU: Rename add/sub with carry out instructions The hardware has created a real mess in the naming for add/sub, which have been renamed basically every generation. Switch the carry out pseudos to have the gfx9/gfx10 names. We were using the original SI/CI v_add_i32/v_sub_i32 names. Later targets reintroduced these names as carryless instructions with a saturating clamp bit, which we do not define. Do this rename so we can unambiguously add these missing instructions. The carry-in versions should also be renamed, but at least those had a consistent _u32 name to begin with. The 16-bit instructions were also renamed, but aren't ambiguous. This does regress assembler error message quality in some cases. In mismatched wave32/wave64 situations, this will switch from "unsupported instruction" to "invalid operand", with the error pointing at the wrong position. I couldn't quite follow how the assembler selects these, but the previous behavior seemed accidental to me. It looked like there was a partial attempt to handle this which was never completed (i.e. there is an AMDGPUOperand::isBoolReg but it isn't used for anything).
2020-07-14 21:18:36 +08:00
; GFX7-FLAT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
; GFX7-FLAT: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
; GFX7-FLAT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
AMDGPU/GlobalISel: Select MUBUF path for global atomic cmpxchg I'm not sure why this isn't a pattern, but the DAG manually selects this.
2020-02-10 11:32:44 +08:00
; GFX7-FLAT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX7-FLAT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX8: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX8: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4, implicit $exec
; GFX8: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX8: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
; GFX8: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
; GFX8: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
; GFX8: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
; GFX8: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
AMDGPU: Rename add/sub with carry out instructions The hardware has created a real mess in the naming for add/sub, which have been renamed basically every generation. Switch the carry out pseudos to have the gfx9/gfx10 names. We were using the original SI/CI v_add_i32/v_sub_i32 names. Later targets reintroduced these names as carryless instructions with a saturating clamp bit, which we do not define. Do this rename so we can unambiguously add these missing instructions. The carry-in versions should also be renamed, but at least those had a consistent _u32 name to begin with. The 16-bit instructions were also renamed, but aren't ambiguous. This does regress assembler error message quality in some cases. In mismatched wave32/wave64 situations, this will switch from "unsupported instruction" to "invalid operand", with the error pointing at the wrong position. I couldn't quite follow how the assembler selects these, but the previous behavior seemed accidental to me. It looked like there was a partial attempt to handle this which was never completed (i.e. there is an AMDGPUOperand::isBoolReg but it isn't used for anything).
2020-07-14 21:18:36 +08:00
; GFX8: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
; GFX8: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
; GFX8: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
AMDGPU/GlobalISel: Select MUBUF path for global atomic cmpxchg I'm not sure why this isn't a pattern, but the DAG manually selects this.
2020-02-10 11:32:44 +08:00
; GFX8: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX8: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX9: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 4, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX9: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gep4
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX10: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 4, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX10: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s32) = COPY $vgpr2
%2:vgpr(s32) = COPY $vgpr3
%3:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
%4:vgpr(s64) = G_CONSTANT i64 4
%5:vgpr(p1) = G_PTR_ADD %0, %4
%6:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %5, %3 :: (load store seq_cst 4, addrspace 1)
$vgpr0 = COPY %6
...
---
name: amdgpu_atomic_cmpxchg_s64_global
legalized: true
regBankSelected: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
; GFX6: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX6: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX6: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX6: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX6: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX6: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX6: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX6: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX6: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX6: [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN:%[0-9]+]]:vreg_128 = BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX6: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN]].sub0_sub1
; GFX6: $vgpr0_vgpr1 = COPY [[COPY3]]
; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX7: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX7: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX7: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX7: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX7: [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN:%[0-9]+]]:vreg_128 = BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX7: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN]].sub0_sub1
; GFX7: $vgpr0_vgpr1 = COPY [[COPY3]]
; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX7-FLAT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7-FLAT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7-FLAT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX7-FLAT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX7-FLAT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8, addrspace 1)
; GFX7-FLAT: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_CMPSWAP_X2_RTN]]
; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX8: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX8: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8, addrspace 1)
; GFX8: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_CMPSWAP_X2_RTN]]
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX9: [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX9: $vgpr0_vgpr1 = COPY [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN]]
; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s64_global
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX10: [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX10: $vgpr0_vgpr1 = COPY [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = COPY $vgpr2_vgpr3
%2:vgpr(s64) = COPY $vgpr4_vgpr5
%3:vgpr(<2 x s64>) = G_BUILD_VECTOR %1, %2
%4:vgpr(s64) = G_AMDGPU_ATOMIC_CMPXCHG %0, %3 :: (load store seq_cst 8, addrspace 1)
$vgpr0_vgpr1 = COPY %4
...
---
name: amdgpu_atomic_cmpxchg_s64_global_gep4
legalized: true
regBankSelected: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
; GFX6: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX6: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX6: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX6: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX6: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX6: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX6: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX6: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX6: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX6: [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN:%[0-9]+]]:vreg_128 = BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 4, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX6: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN]].sub0_sub1
; GFX6: $vgpr0_vgpr1 = COPY [[COPY3]]
; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX7: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX7: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX7: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX7: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX7: [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN:%[0-9]+]]:vreg_128 = BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 4, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX7: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN]].sub0_sub1
; GFX7: $vgpr0_vgpr1 = COPY [[COPY3]]
; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX7-FLAT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7-FLAT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7-FLAT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX7-FLAT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX7-FLAT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4, implicit $exec
; GFX7-FLAT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX7-FLAT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
; GFX7-FLAT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
; GFX7-FLAT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
; GFX7-FLAT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
; GFX7-FLAT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
AMDGPU: Rename add/sub with carry out instructions The hardware has created a real mess in the naming for add/sub, which have been renamed basically every generation. Switch the carry out pseudos to have the gfx9/gfx10 names. We were using the original SI/CI v_add_i32/v_sub_i32 names. Later targets reintroduced these names as carryless instructions with a saturating clamp bit, which we do not define. Do this rename so we can unambiguously add these missing instructions. The carry-in versions should also be renamed, but at least those had a consistent _u32 name to begin with. The 16-bit instructions were also renamed, but aren't ambiguous. This does regress assembler error message quality in some cases. In mismatched wave32/wave64 situations, this will switch from "unsupported instruction" to "invalid operand", with the error pointing at the wrong position. I couldn't quite follow how the assembler selects these, but the previous behavior seemed accidental to me. It looked like there was a partial attempt to handle this which was never completed (i.e. there is an AMDGPUOperand::isBoolReg but it isn't used for anything).
2020-07-14 21:18:36 +08:00
; GFX7-FLAT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
; GFX7-FLAT: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
; GFX7-FLAT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
AMDGPU/GlobalISel: Select MUBUF path for global atomic cmpxchg I'm not sure why this isn't a pattern, but the DAG manually selects this.
2020-02-10 11:32:44 +08:00
; GFX7-FLAT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8, addrspace 1)
; GFX7-FLAT: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_CMPSWAP_X2_RTN]]
; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX8: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX8: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4, implicit $exec
; GFX8: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX8: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
; GFX8: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
; GFX8: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
; GFX8: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
; GFX8: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
AMDGPU: Rename add/sub with carry out instructions The hardware has created a real mess in the naming for add/sub, which have been renamed basically every generation. Switch the carry out pseudos to have the gfx9/gfx10 names. We were using the original SI/CI v_add_i32/v_sub_i32 names. Later targets reintroduced these names as carryless instructions with a saturating clamp bit, which we do not define. Do this rename so we can unambiguously add these missing instructions. The carry-in versions should also be renamed, but at least those had a consistent _u32 name to begin with. The 16-bit instructions were also renamed, but aren't ambiguous. This does regress assembler error message quality in some cases. In mismatched wave32/wave64 situations, this will switch from "unsupported instruction" to "invalid operand", with the error pointing at the wrong position. I couldn't quite follow how the assembler selects these, but the previous behavior seemed accidental to me. It looked like there was a partial attempt to handle this which was never completed (i.e. there is an AMDGPUOperand::isBoolReg but it isn't used for anything).
2020-07-14 21:18:36 +08:00
; GFX8: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
; GFX8: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
; GFX8: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
AMDGPU/GlobalISel: Select MUBUF path for global atomic cmpxchg I'm not sure why this isn't a pattern, but the DAG manually selects this.
2020-02-10 11:32:44 +08:00
; GFX8: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8, addrspace 1)
; GFX8: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_CMPSWAP_X2_RTN]]
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX9: [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 4, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX9: $vgpr0_vgpr1 = COPY [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN]]
; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_gep4
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX10: [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 4, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX10: $vgpr0_vgpr1 = COPY [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = COPY $vgpr2_vgpr3
%2:vgpr(s64) = COPY $vgpr4_vgpr5
%3:vgpr(<2 x s64>) = G_BUILD_VECTOR %1, %2
%4:vgpr(s64) = G_CONSTANT i64 4
%5:vgpr(p1) = G_PTR_ADD %0, %4
%6:vgpr(s64) = G_AMDGPU_ATOMIC_CMPXCHG %5, %3 :: (load store seq_cst 8, addrspace 1)
$vgpr0_vgpr1 = COPY %6
...
---
name: amdgpu_atomic_cmpxchg_s32_global_gepm4
legalized: true
regBankSelected: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
; GFX6: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX6: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX6: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX6: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX6: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294967292, implicit $exec
; GFX6: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
; GFX6: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
; GFX6: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
; GFX6: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
; GFX6: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
; GFX6: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
AMDGPU: Rename add/sub with carry out instructions The hardware has created a real mess in the naming for add/sub, which have been renamed basically every generation. Switch the carry out pseudos to have the gfx9/gfx10 names. We were using the original SI/CI v_add_i32/v_sub_i32 names. Later targets reintroduced these names as carryless instructions with a saturating clamp bit, which we do not define. Do this rename so we can unambiguously add these missing instructions. The carry-in versions should also be renamed, but at least those had a consistent _u32 name to begin with. The 16-bit instructions were also renamed, but aren't ambiguous. This does regress assembler error message quality in some cases. In mismatched wave32/wave64 situations, this will switch from "unsupported instruction" to "invalid operand", with the error pointing at the wrong position. I couldn't quite follow how the assembler selects these, but the previous behavior seemed accidental to me. It looked like there was a partial attempt to handle this which was never completed (i.e. there is an AMDGPUOperand::isBoolReg but it isn't used for anything).
2020-07-14 21:18:36 +08:00
; GFX6: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
; GFX6: %18:vgpr_32, dead %20:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
; GFX6: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %18, %subreg.sub1
AMDGPU/GlobalISel: Select MUBUF path for global atomic cmpxchg I'm not sure why this isn't a pattern, but the DAG manually selects this.
2020-02-10 11:32:44 +08:00
; GFX6: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX6: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX6: [[REG_SEQUENCE3:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX6: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX6: [[REG_SEQUENCE4:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE3]], %subreg.sub2_sub3
; GFX6: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], [[REG_SEQUENCE4]], 0, 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX6: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
; GFX6: $vgpr0 = COPY [[COPY7]]
; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294967292, implicit $exec
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
; GFX7: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
AMDGPU: Rename add/sub with carry out instructions The hardware has created a real mess in the naming for add/sub, which have been renamed basically every generation. Switch the carry out pseudos to have the gfx9/gfx10 names. We were using the original SI/CI v_add_i32/v_sub_i32 names. Later targets reintroduced these names as carryless instructions with a saturating clamp bit, which we do not define. Do this rename so we can unambiguously add these missing instructions. The carry-in versions should also be renamed, but at least those had a consistent _u32 name to begin with. The 16-bit instructions were also renamed, but aren't ambiguous. This does regress assembler error message quality in some cases. In mismatched wave32/wave64 situations, this will switch from "unsupported instruction" to "invalid operand", with the error pointing at the wrong position. I couldn't quite follow how the assembler selects these, but the previous behavior seemed accidental to me. It looked like there was a partial attempt to handle this which was never completed (i.e. there is an AMDGPUOperand::isBoolReg but it isn't used for anything).
2020-07-14 21:18:36 +08:00
; GFX7: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
; GFX7: %18:vgpr_32, dead %20:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
; GFX7: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %18, %subreg.sub1
AMDGPU/GlobalISel: Select MUBUF path for global atomic cmpxchg I'm not sure why this isn't a pattern, but the DAG manually selects this.
2020-02-10 11:32:44 +08:00
; GFX7: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX7: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX7: [[REG_SEQUENCE3:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX7: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX7: [[REG_SEQUENCE4:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE3]], %subreg.sub2_sub3
; GFX7: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], [[REG_SEQUENCE4]], 0, 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX7: [[COPY7:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
; GFX7: $vgpr0 = COPY [[COPY7]]
; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX7-FLAT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7-FLAT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7-FLAT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7-FLAT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7-FLAT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294967292, implicit $exec
; GFX7-FLAT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
; GFX7-FLAT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
; GFX7-FLAT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
; GFX7-FLAT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
; GFX7-FLAT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
; GFX7-FLAT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
AMDGPU: Rename add/sub with carry out instructions The hardware has created a real mess in the naming for add/sub, which have been renamed basically every generation. Switch the carry out pseudos to have the gfx9/gfx10 names. We were using the original SI/CI v_add_i32/v_sub_i32 names. Later targets reintroduced these names as carryless instructions with a saturating clamp bit, which we do not define. Do this rename so we can unambiguously add these missing instructions. The carry-in versions should also be renamed, but at least those had a consistent _u32 name to begin with. The 16-bit instructions were also renamed, but aren't ambiguous. This does regress assembler error message quality in some cases. In mismatched wave32/wave64 situations, this will switch from "unsupported instruction" to "invalid operand", with the error pointing at the wrong position. I couldn't quite follow how the assembler selects these, but the previous behavior seemed accidental to me. It looked like there was a partial attempt to handle this which was never completed (i.e. there is an AMDGPUOperand::isBoolReg but it isn't used for anything).
2020-07-14 21:18:36 +08:00
; GFX7-FLAT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
; GFX7-FLAT: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
; GFX7-FLAT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
AMDGPU/GlobalISel: Select MUBUF path for global atomic cmpxchg I'm not sure why this isn't a pattern, but the DAG manually selects this.
2020-02-10 11:32:44 +08:00
; GFX7-FLAT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX7-FLAT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX8: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX8: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294967292, implicit $exec
; GFX8: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
; GFX8: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
; GFX8: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
; GFX8: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
; GFX8: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
; GFX8: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
AMDGPU: Rename add/sub with carry out instructions The hardware has created a real mess in the naming for add/sub, which have been renamed basically every generation. Switch the carry out pseudos to have the gfx9/gfx10 names. We were using the original SI/CI v_add_i32/v_sub_i32 names. Later targets reintroduced these names as carryless instructions with a saturating clamp bit, which we do not define. Do this rename so we can unambiguously add these missing instructions. The carry-in versions should also be renamed, but at least those had a consistent _u32 name to begin with. The 16-bit instructions were also renamed, but aren't ambiguous. This does regress assembler error message quality in some cases. In mismatched wave32/wave64 situations, this will switch from "unsupported instruction" to "invalid operand", with the error pointing at the wrong position. I couldn't quite follow how the assembler selects these, but the previous behavior seemed accidental to me. It looked like there was a partial attempt to handle this which was never completed (i.e. there is an AMDGPUOperand::isBoolReg but it isn't used for anything).
2020-07-14 21:18:36 +08:00
; GFX8: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
; GFX8: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
; GFX8: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
AMDGPU/GlobalISel: Select MUBUF path for global atomic cmpxchg I'm not sure why this isn't a pattern, but the DAG manually selects this.
2020-02-10 11:32:44 +08:00
; GFX8: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[REG_SEQUENCE2]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX8: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX9: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], -4, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX9: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_gepm4
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX10: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], -4, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX10: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s32) = COPY $vgpr2
%2:vgpr(s32) = COPY $vgpr3
%3:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
%4:vgpr(s64) = G_CONSTANT i64 -4
%5:vgpr(p1) = G_PTR_ADD %0, %4
%6:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %5, %3 :: (load store seq_cst 4, addrspace 1)
$vgpr0 = COPY %6
...
---
name: amdgpu_atomic_cmpxchg_s32_global_nortn
legalized: true
regBankSelected: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
; GFX6: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX6: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX6: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX6: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX6: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX6: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX6: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX6: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX6: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX6: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX6: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX7: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX7: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX7: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX7: [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_ADDR64_RTN]].sub0
; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX7-FLAT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7-FLAT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7-FLAT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7-FLAT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7-FLAT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX8: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX8: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX9: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_nortn
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2, $vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX10: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s32) = COPY $vgpr2
%2:vgpr(s32) = COPY $vgpr3
%3:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
%4:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %0, %3 :: (load store seq_cst 4, addrspace 1)
...
---
name: amdgpu_atomic_cmpxchg_s64_global_nortn
legalized: true
regBankSelected: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
; GFX6: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX6: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX6: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX6: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX6: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX6: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX6: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX6: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX6: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX6: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX6: [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN:%[0-9]+]]:vreg_128 = BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX6: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN]].sub0_sub1
; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX7: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX7: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX7: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 0
; GFX7: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[S_MOV_B64_]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX7: [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN:%[0-9]+]]:vreg_128 = BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN [[REG_SEQUENCE]], [[COPY]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX7: [[COPY3:%[0-9]+]]:vreg_64 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_X2_ADDR64_RTN]].sub0_sub1
; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
; GFX7-FLAT: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX7-FLAT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX7-FLAT: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX7-FLAT: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX7-FLAT: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX7-FLAT: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8, addrspace 1)
; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
; GFX8: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX8: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX8: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX8: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX8: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX8: [[FLAT_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8, addrspace 1)
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX9: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX9: [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s64_global_nortn
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $vgpr4_vgpr5
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
; GFX10: [[COPY2:%[0-9]+]]:vreg_64 = COPY $vgpr4_vgpr5
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY1]], %subreg.sub0_sub1, [[COPY2]], %subreg.sub2_sub3
; GFX10: [[GLOBAL_ATOMIC_CMPSWAP_X2_RTN:%[0-9]+]]:vreg_64 = GLOBAL_ATOMIC_CMPSWAP_X2_RTN [[COPY]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst 8, addrspace 1)
%0:vgpr(p1) = COPY $vgpr0_vgpr1
%1:vgpr(s64) = COPY $vgpr2_vgpr3
%2:vgpr(s64) = COPY $vgpr4_vgpr5
%3:vgpr(<2 x s64>) = G_BUILD_VECTOR %1, %2
%4:vgpr(s64) = G_AMDGPU_ATOMIC_CMPXCHG %0, %3 :: (load store seq_cst 8, addrspace 1)
...
---
name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
legalized: true
regBankSelected: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
; GFX6: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX6: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX6: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX6: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX6: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX6: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294967295
; GFX6: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX6: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX6: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX6: [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX6: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN]].sub0
; GFX6: $vgpr0 = COPY [[COPY3]]
; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
; GFX7: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX7: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294967295
; GFX7: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX7: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX7: [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], 0, 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN]].sub0
; GFX7: $vgpr0 = COPY [[COPY3]]
; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
; GFX7-FLAT: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX7-FLAT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX7-FLAT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7-FLAT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7-FLAT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7-FLAT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[COPY]]
; GFX7-FLAT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY3]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX7-FLAT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
; GFX8: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX8: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX8: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX8: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[COPY]]
; GFX8: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY3]], [[REG_SEQUENCE]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX8: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
; GFX9: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX9: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[COPY]]
; GFX9: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY3]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX9: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr
; GFX10: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX10: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[COPY]]
; GFX10: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY3]], [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX10: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
%0:sgpr(p1) = COPY $sgpr0_sgpr1
%1:vgpr(s32) = COPY $vgpr2
%2:vgpr(s32) = COPY $vgpr3
%3:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
%4:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %0, %3 :: (load store seq_cst 4, addrspace 1)
$vgpr0 = COPY %4
...
---
name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX6-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
; GFX6: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX6: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX6: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX6: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX6: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX6: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294967295
; GFX6: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX6: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX6: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX6: [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], 0, 4095, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX6: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN]].sub0
; GFX6: $vgpr0 = COPY [[COPY3]]
; GFX7-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
; GFX7: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX7: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294967295
; GFX7: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 61440
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX7: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0_sub1, [[REG_SEQUENCE1]], %subreg.sub2_sub3
; GFX7: [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN:%[0-9]+]]:vreg_64 = BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN [[REG_SEQUENCE]], [[REG_SEQUENCE2]], 0, 4095, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY killed [[BUFFER_ATOMIC_CMPSWAP_OFFSET_RTN]].sub0
; GFX7: $vgpr0 = COPY [[COPY3]]
; GFX7-FLAT-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
; GFX7-FLAT: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX7-FLAT: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX7-FLAT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX7-FLAT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX7-FLAT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
; GFX7-FLAT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX7-FLAT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX7-FLAT: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
; GFX7-FLAT: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
; GFX7-FLAT: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
; GFX7-FLAT: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
; GFX7-FLAT: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY3]], [[COPY4]], implicit-def $scc
; GFX7-FLAT: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY5]], [[COPY6]], implicit-def $scc, implicit $scc
; GFX7-FLAT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
; GFX7-FLAT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX7-FLAT: [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
; GFX7-FLAT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY7]], [[REG_SEQUENCE2]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX7-FLAT: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
; GFX8-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
; GFX8: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX8: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX8: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX8: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX8: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
; GFX8: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX8: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX8: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
; GFX8: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
; GFX8: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
; GFX8: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
; GFX8: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY3]], [[COPY4]], implicit-def $scc
; GFX8: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY5]], [[COPY6]], implicit-def $scc, implicit $scc
; GFX8: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
; GFX8: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX8: [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
; GFX8: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY7]], [[REG_SEQUENCE2]], 0, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4, addrspace 1)
; GFX8: $vgpr0 = COPY [[FLAT_ATOMIC_CMPSWAP_RTN]]
; GFX9-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
; GFX9: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX9: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX9: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[COPY]]
; GFX9: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY3]], [[REG_SEQUENCE]], 4095, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX9: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
; GFX10-LABEL: name: amdgpu_atomic_cmpxchg_s32_global_sgpr_ptr_offset_4095
; GFX10: liveins: $sgpr0_sgpr1, $vgpr2, $vgpr3
; GFX10: $vcc_hi = IMPLICIT_DEF
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX10: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
; GFX10: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
; GFX10: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
; GFX10: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
; GFX10: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
; GFX10: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
; GFX10: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY3]], [[COPY4]], implicit-def $scc
; GFX10: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY5]], [[COPY6]], implicit-def $scc, implicit $scc
; GFX10: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
; GFX10: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX10: [[COPY7:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
; GFX10: [[GLOBAL_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_CMPSWAP_RTN [[COPY7]], [[REG_SEQUENCE2]], 0, 0, implicit $exec :: (load store seq_cst 4, addrspace 1)
; GFX10: $vgpr0 = COPY [[GLOBAL_ATOMIC_CMPSWAP_RTN]]
%0:sgpr(p1) = COPY $sgpr0_sgpr1
%1:vgpr(s32) = COPY $vgpr2
%2:vgpr(s32) = COPY $vgpr3
%3:sgpr(s64) = G_CONSTANT i64 4095
%4:sgpr(p1) = G_PTR_ADD %0, %3
%5:vgpr(<2 x s32>) = G_BUILD_VECTOR %1, %2
%6:vgpr(s32) = G_AMDGPU_ATOMIC_CMPXCHG %4, %5 :: (load store seq_cst 4, addrspace 1)
$vgpr0 = COPY %6
...