731 lines
37 KiB
YAML
731 lines
37 KiB
YAML
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
|
|
# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9 %s
|
|
# RUN: llc -march=amdgcn -mcpu=gfx1010 -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX10 %s
|
|
|
|
---
|
|
name: load_global_s32_from_sgpr
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX9: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX10: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:vgpr(p1) = COPY %0
|
|
%2:vgpr(s32) = G_LOAD %1 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %2
|
|
|
|
...
|
|
|
|
# FIXME: This zext wouldn't select on its own.
|
|
---
|
|
|
|
name: load_global_s32_from_sgpr_zext_vgpr
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1, $vgpr0
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_zext_vgpr
|
|
; GFX9: liveins: $sgpr0_sgpr1, $vgpr0
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_zext_vgpr
|
|
; GFX10: liveins: $sgpr0_sgpr1, $vgpr0
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX10: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:vgpr(s32) = COPY $vgpr0
|
|
%2:vgpr(p1) = COPY %0
|
|
%3:vgpr(s64) = G_ZEXT %1
|
|
%4:vgpr(p1) = G_PTR_ADD %2, %3
|
|
%5:vgpr(s32) = G_LOAD %4 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %5
|
|
|
|
...
|
|
|
|
# Test with zext lowered to G_MERGE_VALUES
|
|
---
|
|
|
|
name: load_global_s32_from_sgpr_merge_zext_vgpr
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1, $vgpr0
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_merge_zext_vgpr
|
|
; GFX9: liveins: $sgpr0_sgpr1, $vgpr0
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_merge_zext_vgpr
|
|
; GFX10: liveins: $sgpr0_sgpr1, $vgpr0
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX10: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[COPY1]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:vgpr(s32) = COPY $vgpr0
|
|
%2:vgpr(p1) = COPY %0
|
|
%zero:vgpr(s32) = G_CONSTANT i32 0
|
|
%3:vgpr(s64) = G_MERGE_VALUES %1, %zero
|
|
%4:vgpr(p1) = G_PTR_ADD %2, %3
|
|
%5:vgpr(s32) = G_LOAD %4 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %5
|
|
|
|
...
|
|
|
|
---
|
|
|
|
name: load_global_s32_from_sgpr_merge_not_0_vgpr
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1, $vgpr0
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_merge_not_0_vgpr
|
|
; GFX9: liveins: $sgpr0_sgpr1, $vgpr0
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: [[COPY2:%[0-9]+]]:vreg_64 = COPY [[COPY]]
|
|
; GFX9: %notzero:vgpr_32 = V_MOV_B32_e32 1, implicit $exec
|
|
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, %notzero, %subreg.sub1
|
|
; GFX9: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY2]].sub0
|
|
; GFX9: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX9: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY2]].sub1
|
|
; GFX9: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX9: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_64_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
|
|
; GFX9: %12:vgpr_32, dead %14:sreg_64_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX9: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
|
|
; GFX9: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_merge_not_0_vgpr
|
|
; GFX10: liveins: $sgpr0_sgpr1, $vgpr0
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX10: [[COPY2:%[0-9]+]]:vreg_64 = COPY [[COPY]]
|
|
; GFX10: %notzero:vgpr_32 = V_MOV_B32_e32 1, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, %notzero, %subreg.sub1
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY2]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY2]].sub1
|
|
; GFX10: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX10: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
|
|
; GFX10: %12:vgpr_32, dead %14:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %12, %subreg.sub1
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[REG_SEQUENCE1]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:vgpr(s32) = COPY $vgpr0
|
|
%2:vgpr(p1) = COPY %0
|
|
%notzero:vgpr(s32) = G_CONSTANT i32 1
|
|
%3:vgpr(s64) = G_MERGE_VALUES %1, %notzero
|
|
%4:vgpr(p1) = G_PTR_ADD %2, %3
|
|
%5:vgpr(s32) = G_LOAD %4 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %5
|
|
|
|
...
|
|
|
|
---
|
|
|
|
name: load_global_s32_from_sgpr_zext_vgpr_offset4095
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1, $vgpr0
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_zext_vgpr_offset4095
|
|
; GFX9: liveins: $sgpr0_sgpr1, $vgpr0
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[COPY1]], 4095, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_zext_vgpr_offset4095
|
|
; GFX10: liveins: $sgpr0_sgpr1, $vgpr0
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX10: [[COPY2:%[0-9]+]]:vreg_64 = COPY [[COPY]]
|
|
; GFX10: %zero:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX10: %zext:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, %zero, %subreg.sub1
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY2]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY %zext.sub0
|
|
; GFX10: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY2]].sub1
|
|
; GFX10: [[COPY6:%[0-9]+]]:vgpr_32 = COPY %zext.sub1
|
|
; GFX10: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
|
|
; GFX10: %24:vgpr_32, dead %26:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %24, %subreg.sub1
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
|
|
; GFX10: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
|
|
; GFX10: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX10: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
|
|
; GFX10: [[V_ADD_CO_U32_e64_2:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_3:%[0-9]+]]:sreg_32_xm0_xexec = V_ADD_CO_U32_e64 [[COPY7]], [[COPY8]], 0, implicit $exec
|
|
; GFX10: %14:vgpr_32, dead %16:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY9]], [[COPY10]], killed [[V_ADD_CO_U32_e64_3]], 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_2]], %subreg.sub0, %14, %subreg.sub1
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[REG_SEQUENCE2]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:vgpr(s32) = COPY $vgpr0
|
|
%2:vgpr(p1) = COPY %0
|
|
%zero:vgpr(s32) = G_CONSTANT i32 0
|
|
%zext:vgpr(s64) = G_MERGE_VALUES %1, %zero
|
|
%4:vgpr(p1) = G_PTR_ADD %2, %zext
|
|
%5:vgpr(s64) = G_CONSTANT i64 4095
|
|
%6:vgpr(p1) = G_PTR_ADD %4, %5
|
|
%7:vgpr(s32) = G_LOAD %6 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %7
|
|
|
|
...
|
|
|
|
---
|
|
|
|
name: load_global_s32_from_sgpr_zext_vgpr_offset_neg4096
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1, $vgpr0
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_zext_vgpr_offset_neg4096
|
|
; GFX9: liveins: $sgpr0_sgpr1, $vgpr0
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX9: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[COPY1]], -4096, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_zext_vgpr_offset_neg4096
|
|
; GFX10: liveins: $sgpr0_sgpr1, $vgpr0
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX10: [[COPY2:%[0-9]+]]:vreg_64 = COPY [[COPY]]
|
|
; GFX10: %zero:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX10: %zext:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, %zero, %subreg.sub1
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[COPY2]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY %zext.sub0
|
|
; GFX10: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[COPY2]].sub1
|
|
; GFX10: [[COPY6:%[0-9]+]]:vgpr_32 = COPY %zext.sub1
|
|
; GFX10: [[V_ADD_CO_U32_e64_:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_1:%[0-9]+]]:sreg_32_xm0_xexec = V_ADD_CO_U32_e64 [[COPY3]], [[COPY4]], 0, implicit $exec
|
|
; GFX10: %24:vgpr_32, dead %26:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY5]], [[COPY6]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %24, %subreg.sub1
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294963200, implicit $exec
|
|
; GFX10: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 -1, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub0
|
|
; GFX10: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX10: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE1]].sub1
|
|
; GFX10: [[V_ADD_CO_U32_e64_2:%[0-9]+]]:vgpr_32, [[V_ADD_CO_U32_e64_3:%[0-9]+]]:sreg_32_xm0_xexec = V_ADD_CO_U32_e64 [[COPY7]], [[COPY8]], 0, implicit $exec
|
|
; GFX10: %14:vgpr_32, dead %16:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY9]], [[COPY10]], killed [[V_ADD_CO_U32_e64_3]], 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_2]], %subreg.sub0, %14, %subreg.sub1
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[REG_SEQUENCE2]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:vgpr(s32) = COPY $vgpr0
|
|
%2:vgpr(p1) = COPY %0
|
|
%zero:vgpr(s32) = G_CONSTANT i32 0
|
|
%zext:vgpr(s64) = G_MERGE_VALUES %1, %zero
|
|
%4:vgpr(p1) = G_PTR_ADD %2, %zext
|
|
%5:vgpr(s64) = G_CONSTANT i64 -4096
|
|
%6:vgpr(p1) = G_PTR_ADD %4, %5
|
|
%7:vgpr(s32) = G_LOAD %6 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %7
|
|
|
|
...
|
|
---
|
|
name: load_global_s32_from_sgpr_base_offset_4096
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_base_offset_4096
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
|
|
; GFX9: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_base_offset_4096
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
|
|
; GFX10: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:sgpr(s64) = G_CONSTANT i64 4096
|
|
%2:sgpr(p1) = G_PTR_ADD %0, %1
|
|
%3:vgpr(p1) = COPY %2
|
|
%4:vgpr(s32) = G_LOAD %3 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: load_global_s32_from_sgpr_base_offset_4097
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_base_offset_4097
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
|
|
; GFX9: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], 1, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_base_offset_4097
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4096, implicit $exec
|
|
; GFX10: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], 1, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:sgpr(s64) = G_CONSTANT i64 4097
|
|
%2:sgpr(p1) = G_PTR_ADD %0, %1
|
|
%3:vgpr(p1) = COPY %2
|
|
%4:vgpr(s32) = G_LOAD %3 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: load_global_s32_from_sgpr_base_offset_neg4097
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_base_offset_neg4097
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294963199
|
|
; GFX9: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
|
|
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX9: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX9: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX9: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX9: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX9: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX9: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX9: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX9: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX9: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_base_offset_neg4097
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294963199
|
|
; GFX10: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX10: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX10: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX10: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX10: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX10: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:sgpr(s64) = G_CONSTANT i64 -4097
|
|
%2:sgpr(p1) = G_PTR_ADD %0, %1
|
|
%3:vgpr(p1) = COPY %2
|
|
%4:vgpr(s32) = G_LOAD %3 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: load_global_s32_from_sgpr_base_offset_2049
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_base_offset_2049
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX9: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], 2049, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_base_offset_2049
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2048, implicit $exec
|
|
; GFX10: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], 1, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:sgpr(s64) = G_CONSTANT i64 2049
|
|
%2:sgpr(p1) = G_PTR_ADD %0, %1
|
|
%3:vgpr(p1) = COPY %2
|
|
%4:vgpr(s32) = G_LOAD %3 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: load_global_s32_from_sgpr_base_offset_neg2049
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_base_offset_neg2049
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX9: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], -2049, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_base_offset_neg2049
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4294965247
|
|
; GFX10: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX10: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX10: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX10: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX10: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX10: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:sgpr(s64) = G_CONSTANT i64 -2049
|
|
%2:sgpr(p1) = G_PTR_ADD %0, %1
|
|
%3:vgpr(p1) = COPY %2
|
|
%4:vgpr(s32) = G_LOAD %3 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
---
|
|
name: load_global_s32_from_sgpr_base_offset_4294967295
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_base_offset_4294967295
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294963200, implicit $exec
|
|
; GFX9: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], 4095, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_base_offset_4294967295
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4294965248, implicit $exec
|
|
; GFX10: [[GLOBAL_LOAD_DWORD_SADDR:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD_SADDR [[COPY]], [[V_MOV_B32_e32_]], 2047, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD_SADDR]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:sgpr(s64) = G_CONSTANT i64 4294967295
|
|
%2:sgpr(p1) = G_PTR_ADD %0, %1
|
|
%3:vgpr(p1) = COPY %2
|
|
%4:vgpr(s32) = G_LOAD %3 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
---
|
|
name: load_global_s32_from_sgpr_base_offset_4294967296
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_base_offset_4294967296
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
|
|
; GFX9: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 1
|
|
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX9: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX9: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX9: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX9: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX9: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX9: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX9: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX9: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX9: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_base_offset_4294967296
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
|
|
; GFX10: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 1
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX10: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX10: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX10: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX10: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX10: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:sgpr(s64) = G_CONSTANT i64 4294967296
|
|
%2:sgpr(p1) = G_PTR_ADD %0, %1
|
|
%3:vgpr(p1) = COPY %2
|
|
%4:vgpr(s32) = G_LOAD %3 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: load_global_s32_from_sgpr_base_offset_4294971390
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_base_offset_4294971390
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4094
|
|
; GFX9: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 1
|
|
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX9: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX9: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX9: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX9: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX9: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX9: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX9: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX9: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX9: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_base_offset_4294971390
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 4094
|
|
; GFX10: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 1
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX10: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX10: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX10: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX10: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX10: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:sgpr(s64) = G_CONSTANT i64 4294971390
|
|
%2:sgpr(p1) = G_PTR_ADD %0, %1
|
|
%3:vgpr(p1) = COPY %2
|
|
%4:vgpr(s32) = G_LOAD %3 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: load_global_s32_from_sgpr_base_offset_neg4294967295
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_base_offset_neg4294967295
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 1
|
|
; GFX9: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
|
|
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX9: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX9: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX9: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX9: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX9: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX9: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX9: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX9: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX9: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_base_offset_neg4294967295
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 1
|
|
; GFX10: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX10: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX10: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX10: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX10: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX10: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:sgpr(s64) = G_CONSTANT i64 -4294967295
|
|
%2:sgpr(p1) = G_PTR_ADD %0, %1
|
|
%3:vgpr(p1) = COPY %2
|
|
%4:vgpr(s32) = G_LOAD %3 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
---
|
|
name: load_global_s32_from_sgpr_base_offset_neg4294967296
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1
|
|
|
|
; GFX9-LABEL: name: load_global_s32_from_sgpr_base_offset_neg4294967296
|
|
; GFX9: liveins: $sgpr0_sgpr1
|
|
; GFX9: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX9: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
|
|
; GFX9: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
|
|
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX9: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX9: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX9: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX9: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX9: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX9: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX9: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX9: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX9: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
; GFX10-LABEL: name: load_global_s32_from_sgpr_base_offset_neg4294967296
|
|
; GFX10: liveins: $sgpr0_sgpr1
|
|
; GFX10: [[COPY:%[0-9]+]]:sreg_64 = COPY $sgpr0_sgpr1
|
|
; GFX10: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
|
|
; GFX10: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -1
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[S_MOV_B32_]], %subreg.sub0, [[S_MOV_B32_1]], %subreg.sub1
|
|
; GFX10: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX10: [[S_ADD_U32_:%[0-9]+]]:sreg_32 = S_ADD_U32 [[COPY1]], [[COPY2]], implicit-def $scc
|
|
; GFX10: [[S_ADDC_U32_:%[0-9]+]]:sreg_32 = S_ADDC_U32 [[COPY3]], [[COPY4]], implicit-def $scc, implicit $scc
|
|
; GFX10: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[S_ADD_U32_]], %subreg.sub0, [[S_ADDC_U32_]], %subreg.sub1
|
|
; GFX10: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY5]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:sgpr(p1) = COPY $sgpr0_sgpr1
|
|
%1:sgpr(s64) = G_CONSTANT i64 -4294967296
|
|
%2:sgpr(p1) = G_PTR_ADD %0, %1
|
|
%3:vgpr(p1) = COPY %2
|
|
%4:vgpr(s32) = G_LOAD %3 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: load_global_s32_from_copy_undef_sgpr
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
; GFX9-LABEL: name: load_global_s32_from_copy_undef_sgpr
|
|
; GFX9: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY [[DEF]]
|
|
; GFX9: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
; GFX10-LABEL: name: load_global_s32_from_copy_undef_sgpr
|
|
; GFX10: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY [[DEF]]
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[COPY]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:sgpr(p1) = G_IMPLICIT_DEF
|
|
%1:vgpr(p1) = COPY %0
|
|
%2:vgpr(s32) = G_LOAD %1 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %2
|
|
|
|
...
|
|
|
|
---
|
|
name: load_global_s32_from_undef_vgpr
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
|
|
body: |
|
|
bb.0:
|
|
; GFX9-LABEL: name: load_global_s32_from_undef_vgpr
|
|
; GFX9: [[DEF:%[0-9]+]]:vreg_64 = IMPLICIT_DEF
|
|
; GFX9: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[DEF]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX9: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
; GFX10-LABEL: name: load_global_s32_from_undef_vgpr
|
|
; GFX10: [[DEF:%[0-9]+]]:vreg_64 = IMPLICIT_DEF
|
|
; GFX10: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[DEF]], 0, 0, 0, 0, implicit $exec :: (load 4, addrspace 1)
|
|
; GFX10: $vgpr0 = COPY [[GLOBAL_LOAD_DWORD]]
|
|
%0:vgpr(p1) = G_IMPLICIT_DEF
|
|
%1:vgpr(s32) = G_LOAD %0 :: (load 4, align 4, addrspace 1)
|
|
$vgpr0 = COPY %1
|
|
|
|
...
|