caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
12ccdd67aaf56bd05ab3375d81c58dfd5a2db02c
clang-p2996/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-private.mir
Joe Nash 18ed279a3a [AMDGPU] gfx11 subtarget features & early tests 
Tablegen definitions for subtarget features and cpp predicate functions to
access the features.
New Sub-TargetProcessors and common latencies.
Simple changes to MIR codegen tests which pass on gfx11 because they have the
same output as previous subtargets or operate on pseudo instructions which
are reused from previous subtargets.

Contributors:
Jay Foad <jay.foad@amd.com>
Petar Avramovic <Petar.Avramovic@amd.com>

Patch 4/N for upstreaming of AMDGPU gfx11 architecture

Depends on D124538

Reviewed By: Petar.Avramovic, foad

Differential Revision: https://reviews.llvm.org/D125261
2022-05-11 10:31:49 -04:00

1104 lines
50 KiB
YAML
Raw Blame History

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX6 %s
# RUN: llc -O0 -march=amdgcn -mcpu=tahiti -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX6 %s
# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX9 %s
# RUN: llc -march=amdgcn -mcpu=gfx1010 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX9 %s
# RUN: llc -march=amdgcn -mcpu=gfx1100 -run-pass=instruction-select -verify-machineinstrs -global-isel-abort=0 -o - %s | FileCheck -check-prefix=GFX11 %s
---
name: load_private_s32_from_4
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_4
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s32), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_4
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s32), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_4
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_DWORD:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_DWORD [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_DWORD]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_LOAD %0 :: (load (s32), align 4, addrspace 5)
$vgpr0 = COPY %1
...
---
name: load_private_s32_from_2
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_2
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[BUFFER_LOAD_USHORT_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_USHORT_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s16), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_USHORT_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_2
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[BUFFER_LOAD_USHORT_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_USHORT_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s16), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_USHORT_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_2
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_USHORT:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_USHORT [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s16), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_USHORT]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_LOAD %0 :: (load (s16), align 2, addrspace 5)
$vgpr0 = COPY %1
...
---
name: load_private_s32_from_1
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_LOAD %0 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %1
...
---
name: load_private_p3_from_4
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_p3_from_4
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (p3), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX9-LABEL: name: load_private_p3_from_4
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (p3), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX11-LABEL: name: load_private_p3_from_4
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_DWORD:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_DWORD [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (p3), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_DWORD]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(p3) = G_LOAD %0 :: (load (p3), align 4, addrspace 5)
$vgpr0 = COPY %1
...
---
name: load_private_p5_from_4
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_p5_from_4
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (p5), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX9-LABEL: name: load_private_p5_from_4
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (p5), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX11-LABEL: name: load_private_p5_from_4
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_DWORD:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_DWORD [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (p5), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_DWORD]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(p5) = G_LOAD %0 :: (load (p5), align 4, addrspace 5)
$vgpr0 = COPY %1
...
---
name: load_private_v2s16
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
isEntryFunction: false
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_v2s16
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (<2 x s16>), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX9-LABEL: name: load_private_v2s16
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (<2 x s16>), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX11-LABEL: name: load_private_v2s16
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_DWORD:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_DWORD [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load (<2 x s16>), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_DWORD]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(<2 x s16>) = G_LOAD %0 :: (load (<2 x s16>), align 4, addrspace 5)
$vgpr0 = COPY %1
...
################################################################################
### Stress addressing modes
################################################################################
---
name: load_private_s32_from_1_gep_2047
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_2047
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2047, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_2047
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 2047, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_2047
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[COPY]], 2047, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 2047
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_2047_known_bits
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_2047_known_bits
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2147483647, implicit $exec
; GFX6-NEXT: [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY]], [[V_MOV_B32_e32_]], implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_AND_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 2047, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_2047_known_bits
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2147483647, implicit $exec
; GFX9-NEXT: [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY]], [[V_MOV_B32_e32_]], implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_AND_B32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 2047, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_2047_known_bits
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2147483647, implicit $exec
; GFX11-NEXT: [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[COPY]], [[V_MOV_B32_e32_]], implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[V_AND_B32_e64_]], 2047, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(s32) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 2147483647
%2:vgpr(s32) = G_AND %0, %1
%3:vgpr(p5) = G_INTTOPTR %2
%4:vgpr(s32) = G_CONSTANT i32 2047
%5:vgpr(p5) = G_PTR_ADD %3, %4
%6:vgpr(s32) = G_LOAD %5 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %6
...
---
name: load_private_s32_from_1_gep_2048
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_2048
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2048, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_2048
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 2048, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_2048
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[COPY]], 2048, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 2048
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_m2047
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_m2047
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -2047, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_m2047
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -2047, implicit $exec
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_m2047
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[COPY]], -2047, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 -2047
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_m2048
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_m2048
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -2048, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_m2048
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -2048, implicit $exec
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_m2048
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[COPY]], -2048, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 -2048
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_4095
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_4095
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[COPY]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_4095
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[COPY]], 4095, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 4095
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_4096
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_4096
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_4096
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_4096
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX11-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[V_ADD_U32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 4096
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_m4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_m4095
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -4095, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_m4095
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -4095, implicit $exec
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_m4095
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[COPY]], -4095, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 -4095
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_m4096
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_m4096
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -4096, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_m4096
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -4096, implicit $exec
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_m4096
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[COPY]], -4096, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 -4096
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_8191
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_8191
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 8191, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_8191
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 8191, implicit $exec
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_8191
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 8191, implicit $exec
; GFX11-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[V_ADD_U32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 8191
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_8192
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_8192
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 8192, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_8192
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 8192, implicit $exec
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_8192
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 8192, implicit $exec
; GFX11-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[V_ADD_U32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 8192
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_m8191
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_m8191
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -8191, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_m8191
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -8191, implicit $exec
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_m8191
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -8191, implicit $exec
; GFX11-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[V_ADD_U32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 -8191
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_1_gep_m8192
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
liveins: $vgpr0
; GFX6-LABEL: name: load_private_s32_from_1_gep_m8192
; GFX6: liveins: $vgpr0
; GFX6-NEXT: {{ $}}
; GFX6-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX6-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -8192, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_gep_m8192
; GFX9: liveins: $vgpr0
; GFX9-NEXT: {{ $}}
; GFX9-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX9-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -8192, implicit $exec
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_gep_m8192
; GFX11: liveins: $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -8192, implicit $exec
; GFX11-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[COPY]], [[V_MOV_B32_e32_]], 0, implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[V_ADD_U32_e64_]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = COPY $vgpr0
%1:vgpr(s32) = G_CONSTANT i32 -8192
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
---
name: load_private_s32_from_4_constant_0
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
; GFX6-LABEL: name: load_private_s32_from_4_constant_0
; GFX6: [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s32), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFSET]]
; GFX9-LABEL: name: load_private_s32_from_4_constant_0
; GFX9: [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s32), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFSET]]
; GFX11-LABEL: name: load_private_s32_from_4_constant_0
; GFX11: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_DWORD:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_DWORD [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_DWORD]]
%0:vgpr(p5) = G_CONSTANT i32 0
%1:vgpr(s32) = G_LOAD %0 :: (load (s32), align 4, addrspace 5)
$vgpr0 = COPY %1
...
---
name: load_private_s32_from_4_constant_sgpr_16
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
; GFX6-LABEL: name: load_private_s32_from_4_constant_sgpr_16
; GFX6: [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, 0, 16, 0, 0, 0, implicit $exec :: (load (s32), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFSET]]
; GFX9-LABEL: name: load_private_s32_from_4_constant_sgpr_16
; GFX9: [[BUFFER_LOAD_DWORD_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, 0, 16, 0, 0, 0, implicit $exec :: (load (s32), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFSET]]
; GFX11-LABEL: name: load_private_s32_from_4_constant_sgpr_16
; GFX11: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xexec_hi = S_MOV_B32 16
; GFX11-NEXT: [[SCRATCH_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_DWORD_SADDR [[S_MOV_B32_]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_DWORD_SADDR]]
%0:sgpr(p5) = G_CONSTANT i32 16
%1:vgpr(s32) = G_LOAD %0 :: (load (s32), align 4, addrspace 5)
$vgpr0 = COPY %1
...
---
name: load_private_s32_from_1_constant_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
; GFX6-LABEL: name: load_private_s32_from_1_constant_4095
; GFX6: [[BUFFER_LOAD_UBYTE_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFSET]]
; GFX9-LABEL: name: load_private_s32_from_1_constant_4095
; GFX9: [[BUFFER_LOAD_UBYTE_OFFSET:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFSET $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFSET]]
; GFX11-LABEL: name: load_private_s32_from_1_constant_4095
; GFX11: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = G_CONSTANT i32 4095
%1:vgpr(s32) = G_LOAD %0 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %1
...
---
name: load_private_s32_from_1_constant_4096
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
; GFX6-LABEL: name: load_private_s32_from_1_constant_4096
; GFX6: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_MOV_B32_e32_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_constant_4096
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_MOV_B32_e32_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_constant_4096
; GFX11: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE]]
%0:vgpr(p5) = G_CONSTANT i32 4096
%1:vgpr(s32) = G_LOAD %0 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %1
...
---
name: load_private_s32_from_fi
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
stack:
- { id: 0, size: 4, alignment: 4 }
body: |
bb.0:
; GFX6-LABEL: name: load_private_s32_from_fi
; GFX6: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN %stack.0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s32), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_fi
; GFX9: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN %stack.0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s32), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_fi
; GFX11: [[SCRATCH_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_DWORD_SADDR %stack.0, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_DWORD_SADDR]]
%0:vgpr(p5) = G_FRAME_INDEX %stack.0
%1:vgpr(s32) = G_LOAD %0 :: (load (s32), align 4, addrspace 5)
$vgpr0 = COPY %1
...
---
name: load_private_s32_from_1_fi_offset_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
stack:
- { id: 0, size: 4096, alignment: 4 }
body: |
bb.0:
; GFX6-LABEL: name: load_private_s32_from_1_fi_offset_4095
; GFX6: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %stack.0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_fi_offset_4095
; GFX9: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %stack.0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_fi_offset_4095
; GFX11: [[SCRATCH_LOAD_UBYTE_SADDR:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE_SADDR %stack.0, 4095, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE_SADDR]]
%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_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
# Have to hack around the copy of the constant to VGPR
---
name: load_private_s32_from_1_fi_offset_sgpr_4095
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
stack:
- { id: 0, size: 4096, alignment: 4 }
body: |
bb.0:
; GFX6-LABEL: name: load_private_s32_from_1_fi_offset_sgpr_4095
; GFX6: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %stack.0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_fi_offset_sgpr_4095
; GFX9: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %stack.0, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 4095, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_fi_offset_sgpr_4095
; GFX11: [[SCRATCH_LOAD_UBYTE_SADDR:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE_SADDR %stack.0, 4095, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE_SADDR]]
%0:vgpr(p5) = G_FRAME_INDEX %stack.0
%1:sgpr(s32) = G_CONSTANT i32 4095
%2:vgpr(s32) = COPY %1
%3:vgpr(p5) = G_PTR_ADD %0, %2
%4:vgpr(s32) = G_LOAD %3 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %4
...
---
name: load_private_s32_from_1_fi_offset_4096
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
stack:
- { id: 0, size: 8192, alignment: 4 }
body: |
bb.0:
; GFX6-LABEL: name: load_private_s32_from_1_fi_offset_4096
; GFX6: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 %stack.0, implicit $exec
; GFX6-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX6-NEXT: %2:vgpr_32, dead %4:sreg_64_xexec = V_ADD_CO_U32_e64 [[V_MOV_B32_e32_]], [[V_MOV_B32_e32_1]], 0, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN %2, $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_1_fi_offset_4096
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 %stack.0, implicit $exec
; GFX9-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX9-NEXT: [[V_ADD_U32_e64_:%[0-9]+]]:vgpr_32 = V_ADD_U32_e64 [[V_MOV_B32_e32_]], [[V_MOV_B32_e32_1]], 0, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_UBYTE_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_OFFEN [[V_ADD_U32_e64_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0, 0, 0, 0, implicit $exec :: (load (s8), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_1_fi_offset_4096
; GFX11: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_UBYTE_SVS:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_UBYTE_SVS [[V_MOV_B32_e32_]], %stack.0, 0, 0, implicit $exec, implicit $flat_scr :: (load (s8), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_UBYTE_SVS]]
%0:vgpr(p5) = G_FRAME_INDEX %stack.0
%1:vgpr(s32) = G_CONSTANT i32 4096
%2:vgpr(p5) = G_PTR_ADD %0, %1
%3:vgpr(s32) = G_LOAD %2 :: (load (s8), align 1, addrspace 5)
$vgpr0 = COPY %3
...
# Should not fold offset if this is a null dereference.
---
name: load_private_s32_from_neg1
legalized: true
regBankSelected: true
tracksRegLiveness: true
machineFunctionInfo:
scratchRSrcReg: $sgpr0_sgpr1_sgpr2_sgpr3
stackPtrOffsetReg: $sgpr32
body: |
bb.0:
; GFX6-LABEL: name: load_private_s32_from_neg1
; GFX6: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
; GFX6-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_MOV_B32_e32_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, -1, 0, 0, 0, implicit $exec :: (load (s32), addrspace 5)
; GFX6-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX9-LABEL: name: load_private_s32_from_neg1
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
; GFX9-NEXT: [[BUFFER_LOAD_DWORD_OFFEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_OFFEN [[V_MOV_B32_e32_]], $sgpr0_sgpr1_sgpr2_sgpr3, 0, -1, 0, 0, 0, implicit $exec :: (load (s32), addrspace 5)
; GFX9-NEXT: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_OFFEN]]
; GFX11-LABEL: name: load_private_s32_from_neg1
; GFX11: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
; GFX11-NEXT: [[SCRATCH_LOAD_DWORD:%[0-9]+]]:vgpr_32 = SCRATCH_LOAD_DWORD [[V_MOV_B32_e32_]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32), addrspace 5)
; GFX11-NEXT: $vgpr0 = COPY [[SCRATCH_LOAD_DWORD]]
%0:vgpr(p5) = G_CONSTANT i32 -1
%1:vgpr(s32) = G_LOAD %0 :: (load (s32), align 4, addrspace 5)
$vgpr0 = COPY %1
...
Reference in New Issue View Git Blame Copy Permalink