diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/accvgpr-copy.mir')
-rw-r--r-- | llvm/test/CodeGen/AMDGPU/accvgpr-copy.mir | 83 |
1 files changed, 83 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/accvgpr-copy.mir b/llvm/test/CodeGen/AMDGPU/accvgpr-copy.mir index 2bd1b8b..d22a4b9 100644 --- a/llvm/test/CodeGen/AMDGPU/accvgpr-copy.mir +++ b/llvm/test/CodeGen/AMDGPU/accvgpr-copy.mir @@ -45,6 +45,9 @@ define amdgpu_kernel void @copy_agpr_to_agpr_tuple() #0 { ret void } define amdgpu_kernel void @copy_agpr_to_agpr_tuple_kill() #0 { ret void } + define amdgpu_kernel void @look_for_vgpr_killed() #0 { ret void } + define amdgpu_kernel void @look_for_vgpr_killed_tuple() #0 { ret void } + attributes #0 = { "amdgpu-flat-work-group-size"="1,256" } ... @@ -1517,3 +1520,83 @@ body: | renamable $agpr4_agpr5_agpr6_agpr7 = COPY renamable killed $agpr0_agpr1_agpr2_agpr3, implicit $exec S_ENDPGM 0, implicit $agpr4_agpr5_agpr6_agpr7 ... + +# Make sure the expansion of the a-to-a copy doesn't introduce a use +# after kill of the source vgpr +--- +name: look_for_vgpr_killed +tracksRegLiveness: true +body: | + bb.0: + liveins: $agpr0 + + ; GFX908-LABEL: name: look_for_vgpr_killed + ; GFX908: liveins: $agpr0 + ; GFX908-NEXT: {{ $}} + ; GFX908-NEXT: $vgpr0 = V_MOV_B32_e32 0, implicit $exec + ; GFX908-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 $vgpr0, implicit $exec + ; GFX908-NEXT: S_NOP 0, implicit $vgpr0 + ; GFX908-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 $vgpr0, implicit $exec + ; + ; GFX90A-LABEL: name: look_for_vgpr_killed + ; GFX90A: liveins: $agpr0 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: $vgpr0 = V_MOV_B32_e32 0, implicit $exec + ; GFX90A-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 $vgpr0, implicit $exec + ; GFX90A-NEXT: S_NOP 0, implicit killed $vgpr0 + ; GFX90A-NEXT: $agpr1 = V_ACCVGPR_MOV_B32 $agpr0, implicit $exec + ; + ; GFX942-LABEL: name: look_for_vgpr_killed + ; GFX942: liveins: $agpr0 + ; GFX942-NEXT: {{ $}} + ; GFX942-NEXT: $vgpr0 = V_MOV_B32_e32 0, implicit $exec + ; GFX942-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 $vgpr0, implicit $exec + ; GFX942-NEXT: S_NOP 0, implicit killed $vgpr0 + ; GFX942-NEXT: $agpr1 = V_ACCVGPR_MOV_B32 $agpr0, implicit $exec + $vgpr0 = V_MOV_B32_e32 0, implicit $exec + $agpr0 = COPY $vgpr0 + S_NOP 0, implicit killed $vgpr0 + $agpr1 = COPY $agpr0 + +... + +--- +name: look_for_vgpr_killed_tuple +tracksRegLiveness: true +body: | + bb.0: + liveins: $agpr0 + + ; GFX908-LABEL: name: look_for_vgpr_killed_tuple + ; GFX908: liveins: $agpr0 + ; GFX908-NEXT: {{ $}} + ; GFX908-NEXT: $vgpr0 = V_MOV_B32_e32 0, implicit $exec, implicit-def $vgpr0_vgpr1 + ; GFX908-NEXT: $vgpr1 = V_MOV_B32_e32 1, implicit $exec + ; GFX908-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 $vgpr0, implicit $exec + ; GFX908-NEXT: S_NOP 0, implicit $vgpr0_vgpr1 + ; GFX908-NEXT: $agpr1 = V_ACCVGPR_WRITE_B32_e64 $vgpr0, implicit $exec + ; + ; GFX90A-LABEL: name: look_for_vgpr_killed_tuple + ; GFX90A: liveins: $agpr0 + ; GFX90A-NEXT: {{ $}} + ; GFX90A-NEXT: $vgpr0 = V_MOV_B32_e32 0, implicit $exec, implicit-def $vgpr0_vgpr1 + ; GFX90A-NEXT: $vgpr1 = V_MOV_B32_e32 1, implicit $exec + ; GFX90A-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 $vgpr0, implicit $exec + ; GFX90A-NEXT: S_NOP 0, implicit killed $vgpr0_vgpr1 + ; GFX90A-NEXT: $agpr1 = V_ACCVGPR_MOV_B32 $agpr0, implicit $exec + ; + ; GFX942-LABEL: name: look_for_vgpr_killed_tuple + ; GFX942: liveins: $agpr0 + ; GFX942-NEXT: {{ $}} + ; GFX942-NEXT: $vgpr0 = V_MOV_B32_e32 0, implicit $exec, implicit-def $vgpr0_vgpr1 + ; GFX942-NEXT: $vgpr1 = V_MOV_B32_e32 1, implicit $exec + ; GFX942-NEXT: $agpr0 = V_ACCVGPR_WRITE_B32_e64 $vgpr0, implicit $exec + ; GFX942-NEXT: S_NOP 0, implicit killed $vgpr0_vgpr1 + ; GFX942-NEXT: $agpr1 = V_ACCVGPR_MOV_B32 $agpr0, implicit $exec + $vgpr0 = V_MOV_B32_e32 0, implicit $exec, implicit-def $vgpr0_vgpr1 + $vgpr1 = V_MOV_B32_e32 1, implicit $exec + $agpr0 = COPY $vgpr0 + S_NOP 0, implicit killed $vgpr0_vgpr1 + $agpr1 = COPY $agpr0 + +... |