llvm-for-llvmta/test/CodeGen/AMDGPU/copy-overlap-vgpr-kill.mir

88 lines
4.4 KiB
YAML

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx906 -verify-machineinstrs -run-pass=postrapseudos -o - %s | FileCheck %s
# Don't set a kill of the super register on the last instruction with
# an overlapping copy. This would kill part of the values in the
# result copies.
---
name: overlapping_copy_kill_undef_reg_after_copy
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr30_sgpr31, $vgpr1_vgpr2_vgpr3
; CHECK-LABEL: name: overlapping_copy_kill_undef_reg_after_copy
; CHECK: liveins: $sgpr30_sgpr31, $vgpr1_vgpr2_vgpr3
; CHECK: $vgpr0 = V_MOV_B32_e32 $vgpr1, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2, implicit $vgpr1_vgpr2_vgpr3
; CHECK: $vgpr1 = V_MOV_B32_e32 $vgpr2, implicit $exec, implicit $vgpr1_vgpr2_vgpr3
; CHECK: $vgpr2 = V_MOV_B32_e32 $vgpr3, implicit $exec, implicit $vgpr1_vgpr2_vgpr3
; CHECK: renamable $vgpr1 = nofpexcept V_MUL_F32_e32 0, $vgpr1, implicit $mode, implicit $exec
; CHECK: S_SETPC_B64 $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
renamable $vgpr0_vgpr1_vgpr2 = COPY killed renamable $vgpr1_vgpr2_vgpr3
renamable $vgpr1 = nofpexcept V_MUL_F32_e32 0, $vgpr1, implicit $mode, implicit $exec
S_SETPC_B64 $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
...
---
name: overlapping_copy_kill_undef_reg_after_copy_1
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr30_sgpr31, $vgpr2_vgpr3_vgpr4
; CHECK-LABEL: name: overlapping_copy_kill_undef_reg_after_copy_1
; CHECK: liveins: $sgpr30_sgpr31, $vgpr2_vgpr3_vgpr4
; CHECK: $vgpr0 = V_MOV_B32_e32 $vgpr2, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2, implicit $vgpr2_vgpr3_vgpr4
; CHECK: $vgpr1 = V_MOV_B32_e32 $vgpr3, implicit $exec, implicit $vgpr2_vgpr3_vgpr4
; CHECK: $vgpr2 = V_MOV_B32_e32 $vgpr4, implicit $exec, implicit $vgpr2_vgpr3_vgpr4
; CHECK: renamable $vgpr1 = nofpexcept V_MUL_F32_e32 0, $vgpr1, implicit $mode, implicit $exec
; CHECK: S_SETPC_B64 $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
renamable $vgpr0_vgpr1_vgpr2 = COPY killed renamable $vgpr2_vgpr3_vgpr4
renamable $vgpr1 = nofpexcept V_MUL_F32_e32 0, $vgpr1, implicit $mode, implicit $exec
S_SETPC_B64 $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
...
---
name: nonoverlapping_copy_kill
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr30_sgpr31, $vgpr3_vgpr4_vgpr5
; CHECK-LABEL: name: nonoverlapping_copy_kill
; CHECK: liveins: $sgpr30_sgpr31, $vgpr3_vgpr4_vgpr5
; CHECK: $vgpr0 = V_MOV_B32_e32 $vgpr3, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2, implicit $vgpr3_vgpr4_vgpr5
; CHECK: $vgpr1 = V_MOV_B32_e32 $vgpr4, implicit $exec, implicit $vgpr3_vgpr4_vgpr5
; CHECK: $vgpr2 = V_MOV_B32_e32 $vgpr5, implicit $exec, implicit killed $vgpr3_vgpr4_vgpr5
; CHECK: renamable $vgpr1 = nofpexcept V_MUL_F32_e32 0, $vgpr1, implicit $mode, implicit $exec
; CHECK: S_SETPC_B64 $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
renamable $vgpr0_vgpr1_vgpr2 = COPY killed renamable $vgpr3_vgpr4_vgpr5
renamable $vgpr1 = nofpexcept V_MUL_F32_e32 0, $vgpr1, implicit $mode, implicit $exec
S_SETPC_B64 $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2
...
---
name: overlapping_copy_kill_half_s128
tracksRegLiveness: true
body: |
bb.0:
liveins: $sgpr30_sgpr31, $vgpr2_vgpr3_vgpr4_vgpr5
; CHECK-LABEL: name: overlapping_copy_kill_half_s128
; CHECK: liveins: $sgpr30_sgpr31, $vgpr2_vgpr3_vgpr4_vgpr5
; CHECK: $vgpr0 = V_MOV_B32_e32 $vgpr2, implicit $exec, implicit-def $vgpr0_vgpr1_vgpr2_vgpr3, implicit $vgpr2_vgpr3_vgpr4_vgpr5
; CHECK: $vgpr1 = V_MOV_B32_e32 $vgpr3, implicit $exec, implicit $vgpr2_vgpr3_vgpr4_vgpr5
; CHECK: $vgpr2 = V_MOV_B32_e32 $vgpr4, implicit $exec, implicit $vgpr2_vgpr3_vgpr4_vgpr5
; CHECK: $vgpr3 = V_MOV_B32_e32 $vgpr5, implicit $exec, implicit $vgpr2_vgpr3_vgpr4_vgpr5
; CHECK: renamable $vgpr1 = V_OR_B32_e32 1, $vgpr1, implicit $exec
; CHECK: S_SETPC_B64 $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
renamable $vgpr0_vgpr1_vgpr2_vgpr3 = COPY killed renamable $vgpr2_vgpr3_vgpr4_vgpr5
renamable $vgpr1 = V_OR_B32_e32 1, $vgpr1, implicit $exec
S_SETPC_B64 $sgpr30_sgpr31, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3
...