# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=amdgcn -mcpu=tahiti -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX6 %s
# RUN: llc -O0 -mtriple=amdgcn -mcpu=tahiti -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX6 %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX9 %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx1010 -mattr=+wavefrontsize64 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX9 %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX11 %s
# RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX12 %s
---
name: function_store_private_s32_to_4
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: function_store_private_s32_to_4
; GFX6: liveins: $vgpr0, $vgpr1
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_s32_to_4
; GFX9: liveins: $vgpr0, $vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_s32_to_4
; GFX11: liveins: $vgpr0, $vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_s32_to_4
; GFX12: liveins: $vgpr0, $vgpr1
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (s32), align 4, addrspace 5)
...
---
name: function_store_private_s32_to_2
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: function_store_private_s32_to_2
; GFX6: liveins: $vgpr0, $vgpr1
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_SHORT_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s16), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_s32_to_2
; GFX9: liveins: $vgpr0, $vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_SHORT_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s16), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_s32_to_2
; GFX11: liveins: $vgpr0, $vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_SHORT [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s16), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_s32_to_2
; GFX12: liveins: $vgpr0, $vgpr1
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_SHORT [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s16), addrspace 5)
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (s16), align 2, addrspace 5)
...
---
name: function_store_private_s32_to_1
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: function_store_private_s32_to_1
; GFX6: liveins: $vgpr0, $vgpr1
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_BYTE_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_s32_to_1
; GFX9: liveins: $vgpr0, $vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_BYTE_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_s32_to_1
; GFX11: liveins: $vgpr0, $vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_BYTE [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_s32_to_1
; GFX12: liveins: $vgpr0, $vgpr1
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_BYTE [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (s8), align 1, addrspace 5)
...
---
name: function_store_private_v2s16
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: function_store_private_v2s16
; GFX6: liveins: $vgpr0, $vgpr1
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (<2 x s16>), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_v2s16
; GFX9: liveins: $vgpr0, $vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (<2 x s16>), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_v2s16
; GFX11: liveins: $vgpr0, $vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (<2 x s16>), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_v2s16
; GFX12: liveins: $vgpr0, $vgpr1
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (<2 x s16>), addrspace 5)
%0:vgpr(<2 x s16>) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (<2 x s16>), align 4, addrspace 5)
...
---
name: function_store_private_p3
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: function_store_private_p3
; GFX6: liveins: $vgpr0, $vgpr1
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (p3), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_p3
; GFX9: liveins: $vgpr0, $vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (p3), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_p3
; GFX11: liveins: $vgpr0, $vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (p3), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_p3
; GFX12: liveins: $vgpr0, $vgpr1
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (p3), addrspace 5)
%0:vgpr(p3) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (p3), align 4, addrspace 5)
...
---
name: function_store_private_p5
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: function_store_private_p5
; GFX6: liveins: $vgpr0, $vgpr1
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (p5), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_p5
; GFX9: liveins: $vgpr0, $vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (p5), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_p5
; GFX11: liveins: $vgpr0, $vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (p5), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_p5
; GFX12: liveins: $vgpr0, $vgpr1
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (p5), addrspace 5)
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (p5), align 4, addrspace 5)
...
---
name: function_store_private_s32_to_1_fi_offset_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
stack:
- { id: 0, size: 4096, alignment: 4 }
body: |
bb.0:
; GFX6-LABEL: name: function_store_private_s32_to_1_fi_offset_4095
; GFX6: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX6-NEXT: BUFFER_STORE_BYTE_OFFEN [[V_MOV_B32_e32_]], %stack.0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_s32_to_1_fi_offset_4095
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX9-NEXT: BUFFER_STORE_BYTE_OFFEN [[V_MOV_B32_e32_]], %stack.0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_s32_to_1_fi_offset_4095
; GFX11: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX11-NEXT: SCRATCH_STORE_BYTE_SADDR [[V_MOV_B32_e32_]], %stack.0, 4095, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_s32_to_1_fi_offset_4095
; GFX12: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX12-NEXT: SCRATCH_STORE_BYTE_SADDR [[V_MOV_B32_e32_]], %stack.0, 4095, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
%0:vgpr(p5) = G_FRAME_INDEX %stack.0
%1:vgpr(s32) = G_CONSTANT i32 4095
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_CONSTANT i32 0
G_STORE %3, %2 :: (store (s8), align 1, addrspace 5)
...
---
name: function_store_private_s32_to_1_constant_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
stack:
- { id: 0, size: 4096, alignment: 4 }
body: |
bb.0:
; GFX6-LABEL: name: function_store_private_s32_to_1_constant_4095
; GFX6: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX6-NEXT: BUFFER_STORE_BYTE_OFFSET [[V_MOV_B32_e32_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_s32_to_1_constant_4095
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX9-NEXT: BUFFER_STORE_BYTE_OFFSET [[V_MOV_B32_e32_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_s32_to_1_constant_4095
; GFX11: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
; GFX11-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX11-NEXT: SCRATCH_STORE_BYTE [[V_MOV_B32_e32_1]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_s32_to_1_constant_4095
; GFX12: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
; GFX12-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX12-NEXT: SCRATCH_STORE_BYTE [[V_MOV_B32_e32_1]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
%0:vgpr(p5) = G_CONSTANT i32 4095
%1:vgpr(s32) = G_CONSTANT i32 0
G_STORE %1, %0 :: (store (s8), align 1, addrspace 5)
...
---
name: function_store_private_s32_to_1_constant_4096
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
stack:
- { id: 0, size: 4096, alignment: 4 }
body: |
bb.0:
; GFX6-LABEL: name: function_store_private_s32_to_1_constant_4096
; GFX6: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX6-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX6-NEXT: BUFFER_STORE_BYTE_OFFEN [[V_MOV_B32_e32_]], [[V_MOV_B32_e32_1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_s32_to_1_constant_4096
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX9-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX9-NEXT: BUFFER_STORE_BYTE_OFFEN [[V_MOV_B32_e32_]], [[V_MOV_B32_e32_1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_s32_to_1_constant_4096
; GFX11: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX11-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX11-NEXT: SCRATCH_STORE_BYTE [[V_MOV_B32_e32_1]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_s32_to_1_constant_4096
; GFX12: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX12-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX12-NEXT: SCRATCH_STORE_BYTE [[V_MOV_B32_e32_1]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
%0:vgpr(p5) = G_CONSTANT i32 4096
%1:vgpr(s32) = G_CONSTANT i32 0
G_STORE %1, %0 :: (store (s8), align 1, addrspace 5)
...
---
name: kernel_store_private_s32_to_4
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-LABEL: name: kernel_store_private_s32_to_4
; GFX6: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX9-LABEL: name: kernel_store_private_s32_to_4
; GFX9: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX11-LABEL: name: kernel_store_private_s32_to_4
; GFX11: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
;
; GFX12-LABEL: name: kernel_store_private_s32_to_4
; GFX12: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (s32), align 4, addrspace 5)
...
---
name: kernel_store_private_s32_to_2
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-LABEL: name: kernel_store_private_s32_to_2
; GFX6: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_SHORT_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s16), addrspace 5)
;
; GFX9-LABEL: name: kernel_store_private_s32_to_2
; GFX9: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_SHORT_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s16), addrspace 5)
;
; GFX11-LABEL: name: kernel_store_private_s32_to_2
; GFX11: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_SHORT [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s16), addrspace 5)
;
; GFX12-LABEL: name: kernel_store_private_s32_to_2
; GFX12: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_SHORT [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s16), addrspace 5)
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (s16), align 2, addrspace 5)
...
---
name: kernel_store_private_s32_to_1
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-LABEL: name: kernel_store_private_s32_to_1
; GFX6: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_BYTE_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX9-LABEL: name: kernel_store_private_s32_to_1
; GFX9: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_BYTE_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX11-LABEL: name: kernel_store_private_s32_to_1
; GFX11: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_BYTE [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
;
; GFX12-LABEL: name: kernel_store_private_s32_to_1
; GFX12: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_BYTE [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (s8), align 1, addrspace 5)
...
---
name: kernel_store_private_v2s16
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-LABEL: name: kernel_store_private_v2s16
; GFX6: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (<2 x s16>), addrspace 5)
;
; GFX9-LABEL: name: kernel_store_private_v2s16
; GFX9: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (<2 x s16>), addrspace 5)
;
; GFX11-LABEL: name: kernel_store_private_v2s16
; GFX11: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (<2 x s16>), addrspace 5)
;
; GFX12-LABEL: name: kernel_store_private_v2s16
; GFX12: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (<2 x s16>), addrspace 5)
%0:vgpr(<2 x s16>) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (<2 x s16>), align 4, addrspace 5)
...
---
name: kernel_store_private_p3
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-LABEL: name: kernel_store_private_p3
; GFX6: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (p3), addrspace 5)
;
; GFX9-LABEL: name: kernel_store_private_p3
; GFX9: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (p3), addrspace 5)
;
; GFX11-LABEL: name: kernel_store_private_p3
; GFX11: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (p3), addrspace 5)
;
; GFX12-LABEL: name: kernel_store_private_p3
; GFX12: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (p3), addrspace 5)
%0:vgpr(p3) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (p3), align 4, addrspace 5)
...
---
name: kernel_store_private_p5
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
body: |
bb.0:
liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-LABEL: name: kernel_store_private_p5
; GFX6: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (p5), addrspace 5)
;
; GFX9-LABEL: name: kernel_store_private_p5
; GFX9: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[COPY1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (p5), addrspace 5)
;
; GFX11-LABEL: name: kernel_store_private_p5
; GFX11: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (p5), addrspace 5)
;
; GFX12-LABEL: name: kernel_store_private_p5
; GFX12: liveins: $vgpr0, $vgpr1, $sgpr0_sgpr1_sgpr2_sgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[COPY1]], 0, 0, implicit $exec, implicit $flat_scr :: (store (p5), addrspace 5)
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(p5) = COPY $vgpr1
G_STORE %0, %1 :: (store (p5), align 4, addrspace 5)
...
---
name: kernel_store_private_s32_to_1_fi_offset_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stack:
- { id: 0, size: 4096, alignment: 4 }
body: |
bb.0:
liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-LABEL: name: kernel_store_private_s32_to_1_fi_offset_4095
; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX6-NEXT: BUFFER_STORE_BYTE_OFFEN [[V_MOV_B32_e32_]], %stack.0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX9-LABEL: name: kernel_store_private_s32_to_1_fi_offset_4095
; GFX9: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX9-NEXT: BUFFER_STORE_BYTE_OFFEN [[V_MOV_B32_e32_]], %stack.0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX11-LABEL: name: kernel_store_private_s32_to_1_fi_offset_4095
; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX11-NEXT: SCRATCH_STORE_BYTE_SADDR [[V_MOV_B32_e32_]], %stack.0, 4095, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
;
; GFX12-LABEL: name: kernel_store_private_s32_to_1_fi_offset_4095
; GFX12: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX12-NEXT: SCRATCH_STORE_BYTE_SADDR [[V_MOV_B32_e32_]], %stack.0, 4095, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
%0:vgpr(p5) = G_FRAME_INDEX %stack.0
%1:vgpr(s32) = G_CONSTANT i32 4095
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_CONSTANT i32 0
G_STORE %3, %2 :: (store (s8), align 1, addrspace 5)
...
---
name: kernel_store_private_s32_to_1_constant_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stack:
- { id: 0, size: 4096, alignment: 4 }
body: |
bb.0:
liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-LABEL: name: kernel_store_private_s32_to_1_constant_4095
; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX6-NEXT: BUFFER_STORE_BYTE_OFFSET [[V_MOV_B32_e32_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX9-LABEL: name: kernel_store_private_s32_to_1_constant_4095
; GFX9: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX9-NEXT: BUFFER_STORE_BYTE_OFFSET [[V_MOV_B32_e32_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX11-LABEL: name: kernel_store_private_s32_to_1_constant_4095
; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
; GFX11-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX11-NEXT: SCRATCH_STORE_BYTE [[V_MOV_B32_e32_1]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
;
; GFX12-LABEL: name: kernel_store_private_s32_to_1_constant_4095
; GFX12: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
; GFX12-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX12-NEXT: SCRATCH_STORE_BYTE [[V_MOV_B32_e32_1]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
%0:vgpr(p5) = G_CONSTANT i32 4095
%1:vgpr(s32) = G_CONSTANT i32 0
G_STORE %1, %0 :: (store (s8), align 1, addrspace 5)
...
---
name: kernel_store_private_s32_to_1_constant_4096
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: true
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stack:
- { id: 0, size: 4096, alignment: 4 }
body: |
bb.0:
liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-LABEL: name: kernel_store_private_s32_to_1_constant_4096
; GFX6: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX6-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX6-NEXT: BUFFER_STORE_BYTE_OFFEN [[V_MOV_B32_e32_]], [[V_MOV_B32_e32_1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX9-LABEL: name: kernel_store_private_s32_to_1_constant_4096
; GFX9: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX9-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX9-NEXT: BUFFER_STORE_BYTE_OFFEN [[V_MOV_B32_e32_]], [[V_MOV_B32_e32_1]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s8), addrspace 5)
;
; GFX11-LABEL: name: kernel_store_private_s32_to_1_constant_4096
; GFX11: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX11-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX11-NEXT: SCRATCH_STORE_BYTE [[V_MOV_B32_e32_1]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
;
; GFX12-LABEL: name: kernel_store_private_s32_to_1_constant_4096
; GFX12: liveins: $sgpr0_sgpr1_sgpr2_sgpr3
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX12-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX12-NEXT: SCRATCH_STORE_BYTE [[V_MOV_B32_e32_1]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s8), addrspace 5)
%0:vgpr(p5) = G_CONSTANT i32 4096
%1:vgpr(s32) = G_CONSTANT i32 0
G_STORE %1, %0 :: (store (s8), align 1, addrspace 5)
...
---
name: function_store_private_s32_to_4_wave_address
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: function_store_private_s32_to_4_wave_address
; GFX6: liveins: $vgpr0, $vgpr1
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFSET [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_s32_to_4_wave_address
; GFX9: liveins: $vgpr0, $vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 0, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_s32_to_4_wave_address
; GFX11: liveins: $vgpr0, $vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[V_LSHRREV_B32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_s32_to_4_wave_address
; GFX12: liveins: $vgpr0, $vgpr1
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[V_LSHRREV_B32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(p5) = G_AMDGPU_WAVE_ADDRESS $sgpr32
G_STORE %0, %1 :: (store (s32), align 4, addrspace 5)
...
# Has regbank copy of constant
---
name: function_store_private_s32_to_4_wave_address_offset_copy_constant_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: function_store_private_s32_to_4_wave_address_offset_copy_constant_4095
; GFX6: liveins: $vgpr0, $vgpr1
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFSET [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4095, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_s32_to_4_wave_address_offset_copy_constant_4095
; GFX9: liveins: $vgpr0, $vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4095, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_s32_to_4_wave_address_offset_copy_constant_4095
; GFX11: liveins: $vgpr0, $vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec
; GFX11-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4095
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
; GFX11-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[V_LSHRREV_B32_e64_]], [[COPY1]], 0, implicit $exec
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[V_ADD_U32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_s32_to_4_wave_address_offset_copy_constant_4095
; GFX12: liveins: $vgpr0, $vgpr1
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[V_LSHRREV_B32_e64_]], 4095, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(p5) = G_AMDGPU_WAVE_ADDRESS $sgpr32
%2:sgpr(s32) = G_CONSTANT i32 4095
%3:vgpr(s32) = COPY %2
%4:vgpr(p5) = G_PTR_ADD %1, %3
G_STORE %0, %4 :: (store (s32), align 4, addrspace 5)
...
---
name: function_store_private_s32_to_4_wave_address_offset_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: function_store_private_s32_to_4_wave_address_offset_4095
; GFX6: liveins: $vgpr0, $vgpr1
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFSET [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4095, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_s32_to_4_wave_address_offset_4095
; GFX9: liveins: $vgpr0, $vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFSET [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, $sgpr32, 4095, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_s32_to_4_wave_address_offset_4095
; GFX11: liveins: $vgpr0, $vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
; GFX11-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[V_LSHRREV_B32_e64_]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[V_ADD_U32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_s32_to_4_wave_address_offset_4095
; GFX12: liveins: $vgpr0, $vgpr1
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[V_LSHRREV_B32_e64_]], 4095, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(p5) = G_AMDGPU_WAVE_ADDRESS $sgpr32
%2:vgpr(s32) = G_CONSTANT i32 4095
%3:vgpr(p5) = G_PTR_ADD %1, %2
G_STORE %0, %3 :: (store (s32), align 4, addrspace 5)
...
---
name: function_store_private_s32_to_4_wave_address_offset_copy_constant_4096
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; GFX6-LABEL: name: function_store_private_s32_to_4_wave_address_offset_copy_constant_4096
; GFX6: liveins: $vgpr0, $vgpr1
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 6, $sgpr32, implicit $exec
; GFX6-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
; GFX6-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
; GFX6-NEXT: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, dead [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[V_LSHRREV_B32_e64_]], [[COPY1]], 0, implicit $exec
; GFX6-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[V_ADD_CO_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX9-LABEL: name: function_store_private_s32_to_4_wave_address_offset_copy_constant_4096
; GFX9: liveins: $vgpr0, $vgpr1
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 6, $sgpr32, implicit $exec
; GFX9-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
; GFX9-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[V_LSHRREV_B32_e64_]], [[COPY1]], 0, implicit $exec
; GFX9-NEXT: BUFFER_STORE_DWORD_OFFEN [[COPY]], [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, implicit $exec :: (store (s32), addrspace 5)
;
; GFX11-LABEL: name: function_store_private_s32_to_4_wave_address_offset_copy_constant_4096
; GFX11: liveins: $vgpr0, $vgpr1
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec
; GFX11-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4096
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]]
; GFX11-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[V_LSHRREV_B32_e64_]], [[COPY1]], 0, implicit $exec
; GFX11-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[V_ADD_U32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
;
; GFX12-LABEL: name: function_store_private_s32_to_4_wave_address_offset_copy_constant_4096
; GFX12: liveins: $vgpr0, $vgpr1
; GFX12-NEXT: {{ $}}
; GFX12-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX12-NEXT: [[V_LSHRREV_B32_e64_:%[0-9]+]]:vgpr_32 = V_LSHRREV_B32_e64 5, $sgpr32, implicit $exec
; GFX12-NEXT: SCRATCH_STORE_DWORD [[COPY]], [[V_LSHRREV_B32_e64_]], 4096, 0, implicit $exec, implicit $flat_scr :: (store (s32), addrspace 5)
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(p5) = G_AMDGPU_WAVE_ADDRESS $sgpr32
%2:sgpr(s32) = G_CONSTANT i32 4096
%3:vgpr(s32) = COPY %2
%4:vgpr(p5) = G_PTR_ADD %1, %3
G_STORE %0, %4 :: (store (s32), align 4, addrspace 5)
...
Codebase Browser
llvm