You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
694 lines
44 KiB
694 lines
44 KiB
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
|
|
# RUN: llc -march=amdgcn -mcpu=hawaii -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX7 %s
|
|
# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX7 %s
|
|
# 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: flat_atomicrmw_add_s32
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s32
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX7: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX7: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s32
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX9: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX9: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s32
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX10: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX10: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s32) = COPY $vgpr2
|
|
%2:vgpr(s32) = G_ATOMICRMW_ADD %0, %1 :: (load store seq_cst 4, addrspace 0)
|
|
$vgpr0 = COPY %2
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s32_nortn
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s32_nortn
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX7: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s32_nortn
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX9: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s32_nortn
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX10: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s32) = COPY $vgpr2
|
|
%2:vgpr(s32) = G_ATOMICRMW_ADD %0, %1 :: (load store seq_cst 4, addrspace 0)
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s32_offset2047
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s32_offset2047
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2047, implicit $exec
|
|
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX7: [[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX7: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %10, %subreg.sub1
|
|
; GFX7: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX7: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s32_offset2047
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX9: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 2047, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX9: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s32_offset2047
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2047, implicit $exec
|
|
; GFX10: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX10: %10:vgpr_32, dead %12:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX10: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX10: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s32) = COPY $vgpr2
|
|
%2:vgpr(s64) = G_CONSTANT i64 2047
|
|
%3:vgpr(p0) = G_PTR_ADD %0, %2
|
|
%4:vgpr(s32) = G_ATOMICRMW_ADD %3, %1 :: (load store seq_cst 4, addrspace 0)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s32_offset2047_nortn
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s32_offset2047_nortn
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2047, implicit $exec
|
|
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX7: [[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX7: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %10, %subreg.sub1
|
|
; GFX7: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s32_offset2047_nortn
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX9: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 2047, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s32_offset2047_nortn
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2047, implicit $exec
|
|
; GFX10: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX10: %10:vgpr_32, dead %12:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX10: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s32) = COPY $vgpr2
|
|
%2:vgpr(s64) = G_CONSTANT i64 2047
|
|
%3:vgpr(p0) = G_PTR_ADD %0, %2
|
|
%4:vgpr(s32) = G_ATOMICRMW_ADD %3, %1 :: (load store seq_cst 4, addrspace 0)
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s32_offset2048
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s32_offset2048
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2048, implicit $exec
|
|
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX7: [[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX7: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %10, %subreg.sub1
|
|
; GFX7: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX7: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s32_offset2048
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX9: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 2048, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX9: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s32_offset2048
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2048, implicit $exec
|
|
; GFX10: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX10: %10:vgpr_32, dead %12:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX10: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX10: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s32) = COPY $vgpr2
|
|
%2:vgpr(s64) = G_CONSTANT i64 2048
|
|
%3:vgpr(p0) = G_PTR_ADD %0, %2
|
|
%4:vgpr(s32) = G_ATOMICRMW_ADD %3, %1 :: (load store seq_cst 4, addrspace 0)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s32_offset2048_nortn
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s32_offset2048_nortn
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2048, implicit $exec
|
|
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX7: [[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX7: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %10, %subreg.sub1
|
|
; GFX7: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s32_offset2048_nortn
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX9: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 2048, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s32_offset2048_nortn
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2048, implicit $exec
|
|
; GFX10: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX10: %10:vgpr_32, dead %12:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX10: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s32) = COPY $vgpr2
|
|
%2:vgpr(s64) = G_CONSTANT i64 2048
|
|
%3:vgpr(p0) = G_PTR_ADD %0, %2
|
|
%4:vgpr(s32) = G_ATOMICRMW_ADD %3, %1 :: (load store seq_cst 4, addrspace 0)
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s32_offset4095
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s32_offset4095
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
|
|
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX7: [[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX7: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %10, %subreg.sub1
|
|
; GFX7: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX7: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s32_offset4095
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX9: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 4095, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX9: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s32_offset4095
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; 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_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX10: %10:vgpr_32, dead %12:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX10: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX10: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s32) = COPY $vgpr2
|
|
%2:vgpr(s64) = G_CONSTANT i64 4095
|
|
%3:vgpr(p0) = G_PTR_ADD %0, %2
|
|
%4:vgpr(s32) = G_ATOMICRMW_ADD %3, %1 :: (load store seq_cst 4, addrspace 0)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s32_offset4095_nortn
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s32_offset4095_nortn
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
|
|
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX7: [[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX7: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %10, %subreg.sub1
|
|
; GFX7: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s32_offset4095_nortn
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX9: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[COPY]], [[COPY1]], 4095, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s32_offset4095_nortn
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; 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_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX10: %10:vgpr_32, dead %12:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX10: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s32) = COPY $vgpr2
|
|
%2:vgpr(s64) = G_CONSTANT i64 4095
|
|
%3:vgpr(p0) = G_PTR_ADD %0, %2
|
|
%4:vgpr(s32) = G_ATOMICRMW_ADD %3, %1 :: (load store seq_cst 4, addrspace 0)
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s32_offset4097
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s32_offset4097
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4097, implicit $exec
|
|
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX7: [[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX7: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %10, %subreg.sub1
|
|
; GFX7: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX7: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s32_offset4097
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4097, implicit $exec
|
|
; GFX9: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX9: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX9: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX9: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX9: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX9: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX9: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX9: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s32_offset4097
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4097, implicit $exec
|
|
; GFX10: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX10: %10:vgpr_32, dead %12:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX10: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX10: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_RTN]]
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s32) = COPY $vgpr2
|
|
%2:vgpr(s64) = G_CONSTANT i64 4097
|
|
%3:vgpr(p0) = G_PTR_ADD %0, %2
|
|
%4:vgpr(s32) = G_ATOMICRMW_ADD %3, %1 :: (load store seq_cst 4, addrspace 0)
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s32_offset4097_nortn
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s32_offset4097_nortn
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4097, implicit $exec
|
|
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX7: [[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX7: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %10, %subreg.sub1
|
|
; GFX7: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s32_offset4097_nortn
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX9: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4097, implicit $exec
|
|
; GFX9: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX9: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX9: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX9: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX9: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX9: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX9: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX9: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s32_offset4097_nortn
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX10: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4097, implicit $exec
|
|
; GFX10: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX10: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX10: %10:vgpr_32, dead %12:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX10: [[FLAT_ATOMIC_ADD_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 4)
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s32) = COPY $vgpr2
|
|
%2:vgpr(s64) = G_CONSTANT i64 4097
|
|
%3:vgpr(p0) = G_PTR_ADD %0, %2
|
|
%4:vgpr(s32) = G_ATOMICRMW_ADD %3, %1 :: (load store seq_cst 4, addrspace 0)
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s64
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s64
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; GFX7: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
; GFX7: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_ADD_X2_RTN]]
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s64
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; GFX9: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
; GFX9: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_ADD_X2_RTN]]
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s64
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; GFX10: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
; GFX10: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_ADD_X2_RTN]]
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s64) = COPY $vgpr2_vgpr3
|
|
%2:vgpr(s64) = G_ATOMICRMW_ADD %0, %1 :: (load store seq_cst 8, addrspace 0)
|
|
$vgpr0_vgpr1 = COPY %2
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s64_nortn
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s64_nortn
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; GFX7: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s64_nortn
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; GFX9: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s64_nortn
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; GFX10: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[COPY]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s64) = COPY $vgpr2_vgpr3
|
|
%2:vgpr(s64) = G_ATOMICRMW_ADD %0, %1 :: (load store seq_cst 8, addrspace 0)
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s64_offset4095
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s64_offset4095
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
|
|
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX7: [[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX7: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %10, %subreg.sub1
|
|
; GFX7: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
; GFX7: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_ADD_X2_RTN]]
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s64_offset4095
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; GFX9: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[COPY]], [[COPY1]], 4095, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
; GFX9: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_ADD_X2_RTN]]
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s64_offset4095
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; 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_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX10: %10:vgpr_32, dead %12:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX10: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
; GFX10: $vgpr0_vgpr1 = COPY [[FLAT_ATOMIC_ADD_X2_RTN]]
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s64) = COPY $vgpr2_vgpr3
|
|
%2:vgpr(s64) = G_CONSTANT i64 4095
|
|
%3:vgpr(p0) = G_PTR_ADD %0, %2
|
|
%4:vgpr(s64) = G_ATOMICRMW_ADD %3, %1 :: (load store seq_cst 8, addrspace 0)
|
|
$vgpr0_vgpr1 = COPY %4
|
|
|
|
...
|
|
|
|
---
|
|
name: flat_atomicrmw_add_s64_offset4095_nortn
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
body: |
|
|
bb.0:
|
|
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
|
|
; GFX7-LABEL: name: flat_atomicrmw_add_s64_offset4095_nortn
|
|
; GFX7: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX7: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX7: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; GFX7: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 4095, implicit $exec
|
|
; GFX7: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX7: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX7: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX7: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX7: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
|
|
; GFX7: [[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX7: %10:vgpr_32, dead %12:sreg_64_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], killed [[V_ADD_CO_U32_e64_1]], 0, implicit $exec
|
|
; GFX7: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e64_]], %subreg.sub0, %10, %subreg.sub1
|
|
; GFX7: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
; GFX9-LABEL: name: flat_atomicrmw_add_s64_offset4095_nortn
|
|
; GFX9: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX9: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; GFX9: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[COPY]], [[COPY1]], 4095, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
; GFX10-LABEL: name: flat_atomicrmw_add_s64_offset4095_nortn
|
|
; GFX10: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
|
|
; GFX10: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
|
|
; GFX10: [[COPY1:%[0-9]+]]:vreg_64 = COPY $vgpr2_vgpr3
|
|
; 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_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
|
|
; GFX10: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub0
|
|
; GFX10: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub0
|
|
; GFX10: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[COPY]].sub1
|
|
; GFX10: [[COPY5:%[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 [[COPY2]], [[COPY3]], 0, implicit $exec
|
|
; GFX10: %10:vgpr_32, dead %12:sreg_32_xm0_xexec = V_ADDC_U32_e64 [[COPY4]], [[COPY5]], 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, %10, %subreg.sub1
|
|
; GFX10: [[FLAT_ATOMIC_ADD_X2_RTN:%[0-9]+]]:vreg_64 = FLAT_ATOMIC_ADD_X2_RTN [[REG_SEQUENCE1]], [[COPY1]], 0, 1, 0, implicit $exec, implicit $flat_scr :: (load store seq_cst 8)
|
|
%0:vgpr(p0) = COPY $vgpr0_vgpr1
|
|
%1:vgpr(s64) = COPY $vgpr2_vgpr3
|
|
%2:vgpr(s64) = G_CONSTANT i64 4095
|
|
%3:vgpr(p0) = G_PTR_ADD %0, %2
|
|
%4:vgpr(s64) = G_ATOMICRMW_ADD %3, %1 :: (load store seq_cst 8, addrspace 0)
|
|
|
|
...
|