[llvm-branch-commits] [llvm] AMDGPU: Mark dead carry-out when rewriting scalar carry op to VALU (PR #225738)
Matt Arsenault via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Wed Sep 23 06:26:51 PDT 2026
https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/225738
>From c21cfc864f519a99331b86ef1db86e3e87432ffc Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Wed, 23 Sep 2026 12:26:51 +0200
Subject: [PATCH] AMDGPU: Mark dead carry-out when rewriting scalar carry op to
VALU
If an operation is expanded with a dead scc def, the resultant vcc def will
also be dead, so preserve the dead flag. Reduces implicit reliance on
LiveVariables recomputing dead flags later.
Co-authored-by: Claude (Opus 4.8) <noreply at anthropic.com>
---
llvm/lib/Target/AMDGPU/SIInstrInfo.cpp | 22 +++-
.../test/CodeGen/AMDGPU/change-scc-to-vcc.mir | 4 +-
.../move-scalar-carry-to-valu-dead-vcc.mir | 116 ++++++++++++++++++
.../CodeGen/AMDGPU/move-to-valu-addsubu64.ll | 10 +-
.../CodeGen/AMDGPU/opt-sgpr-to-vgpr-copy.mir | 4 +-
...si-instr-info-correct-implicit-operands.ll | 2 +-
6 files changed, 144 insertions(+), 14 deletions(-)
create mode 100644 llvm/test/CodeGen/AMDGPU/move-scalar-carry-to-valu-dead-vcc.mir
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index f584e7d67dd73f..add5d1df83d8ed 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -8920,13 +8920,19 @@ void SIInstrInfo::moveToVALUImpl(
// Remove any references to SCC. Vector instructions can't read from it, and
// We're just about to add the implicit use / defs of VCC, and we don't want
// both.
+ bool DeadSCCDef = false;
for (MachineOperand &Op : Inst.implicit_operands()) {
if (Op.getReg() == AMDGPU::SCC) {
// Only propagate through live-def of SCC.
- if (Op.isDef() && !Op.isDead())
- addSCCDefUsersToVALUWorklist(Op, Inst, Worklist);
- if (Op.isUse())
- addSCCDefsToVALUWorklist(NewInstr, Worklist);
+ if (Op.isDef()) {
+ if (Op.isDead())
+ DeadSCCDef = true;
+ else
+ addSCCDefUsersToVALUWorklist(Op, Inst, Worklist);
+ continue;
+ }
+
+ addSCCDefsToVALUWorklist(NewInstr, Worklist);
}
}
Inst.eraseFromParent();
@@ -8942,6 +8948,14 @@ void SIInstrInfo::moveToVALUImpl(
}
fixImplicitOperands(*NewInstr);
+ if (DeadSCCDef) {
+ // A scalar op with a dead SCC def lowers to a VALU op whose VCC def will
+ // also be dead.
+ if (MachineOperand *VCCDef =
+ NewInstr->findRegisterDefOperand(RI.getVCC(), &RI))
+ VCCDef->setIsDead();
+ }
+
// Legalize the operands
legalizeOperands(*NewInstr, MDT);
if (NewDstReg)
diff --git a/llvm/test/CodeGen/AMDGPU/change-scc-to-vcc.mir b/llvm/test/CodeGen/AMDGPU/change-scc-to-vcc.mir
index 37f5a3b29f2add..7ca5754c7a0cf9 100644
--- a/llvm/test/CodeGen/AMDGPU/change-scc-to-vcc.mir
+++ b/llvm/test/CodeGen/AMDGPU/change-scc-to-vcc.mir
@@ -22,7 +22,7 @@ body: |
; GCN-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY killed [[DEF3]]
; GCN-NEXT: [[V_ADDC_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 killed [[DEF2]], [[COPY1]], implicit-def $vcc_lo, implicit $vcc_lo, implicit $exec
; GCN-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_2]]
- ; GCN-NEXT: [[V_ADDC_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 [[V_MUL_HI_U32_U24_e64_]], [[COPY2]], implicit-def $vcc_lo, implicit $vcc_lo, implicit $exec
+ ; GCN-NEXT: [[V_ADDC_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 [[V_MUL_HI_U32_U24_e64_]], [[COPY2]], implicit-def dead $vcc_lo, implicit $vcc_lo, implicit $exec
%0:sreg_32 = S_MOV_B32 681
%1:sreg_32 = IMPLICIT_DEF
%2:sreg_32 = IMPLICIT_DEF
@@ -55,7 +55,7 @@ body: |
; GCN-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY killed [[DEF4]]
; GCN-NEXT: [[V_ADDC_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 killed [[DEF3]], [[COPY]], implicit-def $vcc_lo, implicit $vcc_lo, implicit $exec
; GCN-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[DEF5]]
- ; GCN-NEXT: [[V_ADDC_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 killed [[V_ADDC_U32_e32_1]], [[COPY1]], implicit-def $vcc_lo, implicit $vcc_lo, implicit $exec
+ ; GCN-NEXT: [[V_ADDC_U32_e32_1:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 killed [[V_ADDC_U32_e32_1]], [[COPY1]], implicit-def dead $vcc_lo, implicit $vcc_lo, implicit $exec
%0:vgpr_32 = IMPLICIT_DEF
%1:vgpr_32 = IMPLICIT_DEF
%2:sreg_32 = IMPLICIT_DEF
diff --git a/llvm/test/CodeGen/AMDGPU/move-scalar-carry-to-valu-dead-vcc.mir b/llvm/test/CodeGen/AMDGPU/move-scalar-carry-to-valu-dead-vcc.mir
new file mode 100644
index 00000000000000..00ad65fac0fc43
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/move-scalar-carry-to-valu-dead-vcc.mir
@@ -0,0 +1,116 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=amdgpu8.03 -run-pass=si-fix-sgpr-copies -o - %s | FileCheck %s
+
+# When moveToVALU lowers a scalar op with a dead SCC def to a VALU op
+# the equivalent VCC def should inherit the dead flag.
+
+---
+name: s_add_i32_dead_carry
+tracksRegLiveness: true
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: s_add_i32_dead_carry
+ ; CHECK: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_ADD_CO_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_CO_U32_e32 [[DEF1]], [[DEF]], implicit-def dead $vcc, implicit $exec
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY [[V_ADD_CO_U32_e32_]]
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[COPY]]
+ %0:vgpr_32 = IMPLICIT_DEF
+ %1:sreg_32 = IMPLICIT_DEF
+ %2:sreg_32 = COPY %0:vgpr_32
+ %3:sreg_32 = S_ADD_I32 %1:sreg_32, %2:sreg_32, implicit-def dead $scc
+ %4:vgpr_32 = COPY %3:sreg_32
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: s_sub_i32_dead_carry
+tracksRegLiveness: true
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: s_sub_i32_dead_carry
+ ; CHECK: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_SUB_CO_U32_e32_:%[0-9]+]]:vgpr_32 = V_SUB_CO_U32_e32 [[DEF1]], [[DEF]], implicit-def dead $vcc, implicit $exec
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY [[V_SUB_CO_U32_e32_]]
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[COPY]]
+ %0:vgpr_32 = IMPLICIT_DEF
+ %1:sreg_32 = IMPLICIT_DEF
+ %2:sreg_32 = COPY %0:vgpr_32
+ %3:sreg_32 = S_SUB_I32 %1:sreg_32, %2:sreg_32, implicit-def dead $scc
+ %4:vgpr_32 = COPY %3:sreg_32
+ S_ENDPGM 0, implicit %4
+...
+
+---
+name: s_addc_u32_dead_carry
+tracksRegLiveness: true
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: s_addc_u32_dead_carry
+ ; CHECK: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF2:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_ADD_CO_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_CO_U32_e32 [[DEF2]].sub0, [[DEF]], implicit-def $vcc, implicit $exec
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY [[DEF2]].sub1
+ ; CHECK-NEXT: [[V_ADDC_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 [[COPY]], [[DEF1]], implicit-def dead $vcc, implicit $vcc, implicit $exec
+ ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_ADD_CO_U32_e32_]], %subreg.sub0, [[V_ADDC_U32_e32_]], %subreg.sub1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[COPY1]]
+ %0:vgpr_32 = IMPLICIT_DEF
+ %1:vgpr_32 = IMPLICIT_DEF
+ %2:sreg_64 = IMPLICIT_DEF
+ %3:sreg_32 = COPY %0:vgpr_32
+ %4:sreg_32 = COPY %1:vgpr_32
+ %5:sreg_32 = S_ADD_U32 %2.sub0:sreg_64, %3:sreg_32, implicit-def $scc
+ %6:sreg_32 = S_ADDC_U32 %2.sub1:sreg_64, %4:sreg_32, implicit-def dead $scc, implicit $scc
+ %7:sreg_64 = REG_SEQUENCE %5:sreg_32, %subreg.sub0, %6:sreg_32, %subreg.sub1
+ %8:vgpr_32 = COPY %7.sub1:sreg_64
+ S_ENDPGM 0, implicit %8
+...
+
+---
+name: s_subb_u32_dead_carry
+tracksRegLiveness: true
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: s_subb_u32_dead_carry
+ ; CHECK: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF2:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_SUB_CO_U32_e32_:%[0-9]+]]:vgpr_32 = V_SUB_CO_U32_e32 [[DEF2]].sub0, [[DEF]], implicit-def $vcc, implicit $exec
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY [[DEF2]].sub1
+ ; CHECK-NEXT: [[V_SUBB_U32_e32_:%[0-9]+]]:vgpr_32 = V_SUBB_U32_e32 [[COPY]], [[DEF1]], implicit-def dead $vcc, implicit $vcc, implicit $exec
+ ; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_SUB_CO_U32_e32_]], %subreg.sub0, [[V_SUBB_U32_e32_]], %subreg.sub1
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY [[REG_SEQUENCE]].sub1
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[COPY1]]
+ %0:vgpr_32 = IMPLICIT_DEF
+ %1:vgpr_32 = IMPLICIT_DEF
+ %2:sreg_64 = IMPLICIT_DEF
+ %3:sreg_32 = COPY %0:vgpr_32
+ %4:sreg_32 = COPY %1:vgpr_32
+ %5:sreg_32 = S_SUB_U32 %2.sub0:sreg_64, %3:sreg_32, implicit-def $scc
+ %6:sreg_32 = S_SUBB_U32 %2.sub1:sreg_64, %4:sreg_32, implicit-def dead $scc, implicit $scc
+ %7:sreg_64 = REG_SEQUENCE %5:sreg_32, %subreg.sub0, %6:sreg_32, %subreg.sub1
+ %8:vgpr_32 = COPY %7.sub1:sreg_64
+ S_ENDPGM 0, implicit %8
+...
+
+---
+name: s_and_b32_dead_scc_no_carry
+tracksRegLiveness: true
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: s_and_b32_dead_scc_no_carry
+ ; CHECK: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[DEF1:%[0-9]+]]:sreg_32 = IMPLICIT_DEF
+ ; CHECK-NEXT: [[V_AND_B32_e64_:%[0-9]+]]:vgpr_32 = V_AND_B32_e64 [[DEF1]], [[DEF]], implicit $exec
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY [[V_AND_B32_e64_]]
+ ; CHECK-NEXT: S_ENDPGM 0, implicit [[COPY]]
+ %0:vgpr_32 = IMPLICIT_DEF
+ %1:sreg_32 = IMPLICIT_DEF
+ %2:sreg_32 = COPY %0:vgpr_32
+ %3:sreg_32 = S_AND_B32 %1:sreg_32, %2:sreg_32, implicit-def dead $scc
+ %4:vgpr_32 = COPY %3:sreg_32
+ S_ENDPGM 0, implicit %4
+...
diff --git a/llvm/test/CodeGen/AMDGPU/move-to-valu-addsubu64.ll b/llvm/test/CodeGen/AMDGPU/move-to-valu-addsubu64.ll
index 82303c687c9093..cb8d039bf0ddbd 100644
--- a/llvm/test/CodeGen/AMDGPU/move-to-valu-addsubu64.ll
+++ b/llvm/test/CodeGen/AMDGPU/move-to-valu-addsubu64.ll
@@ -13,7 +13,7 @@ define amdgpu_kernel void @add_reg_imm(ptr addrspace(1) %ptr) {
; CHECK-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 28744523
; CHECK-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -1395630315
; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE killed [[S_MOV_B32_1]], %subreg.sub0, killed [[S_MOV_B32_]], %subreg.sub1
- ; CHECK-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[GLOBAL_LOAD_DWORDX2_SADDR]], killed [[REG_SEQUENCE]], implicit-def $vcc_lo, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[GLOBAL_LOAD_DWORDX2_SADDR]], killed [[REG_SEQUENCE]], implicit-def dead $vcc_lo, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY [[V_ADD_U]]
; CHECK-NEXT: GLOBAL_STORE_DWORDX2_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s64) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
@@ -33,7 +33,7 @@ define amdgpu_kernel void @add_reg_reg(ptr addrspace(1) %ptr) {
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_DWORDX2_SADDR:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s64) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[GLOBAL_LOAD_DWORDX2_SADDR1:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s64) from %ir.ptr.load, addrspace 1)
- ; CHECK-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[GLOBAL_LOAD_DWORDX2_SADDR]], [[GLOBAL_LOAD_DWORDX2_SADDR1]], implicit-def $vcc_lo, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[GLOBAL_LOAD_DWORDX2_SADDR]], [[GLOBAL_LOAD_DWORDX2_SADDR1]], implicit-def dead $vcc_lo, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY [[V_ADD_U]]
; CHECK-NEXT: GLOBAL_STORE_DWORDX2_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s64) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
@@ -56,7 +56,7 @@ define amdgpu_kernel void @sub_reg_imm(ptr addrspace(1) %ptr) {
; CHECK-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 -28744524
; CHECK-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 1395630315
; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE killed [[S_MOV_B32_1]], %subreg.sub0, killed [[S_MOV_B32_]], %subreg.sub1
- ; CHECK-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[GLOBAL_LOAD_DWORDX2_SADDR]], killed [[REG_SEQUENCE]], implicit-def $vcc_lo, implicit $exec
+ ; CHECK-NEXT: [[V_ADD_U:%[0-9]+]]:vreg_64 = V_ADD_U64_PSEUDO [[GLOBAL_LOAD_DWORDX2_SADDR]], killed [[REG_SEQUENCE]], implicit-def dead $vcc_lo, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY [[V_ADD_U]]
; CHECK-NEXT: GLOBAL_STORE_DWORDX2_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s64) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
@@ -78,7 +78,7 @@ define amdgpu_kernel void @sub_imm_reg(ptr addrspace(1) %ptr) {
; CHECK-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 28744523
; CHECK-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -1395630315
; CHECK-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE killed [[S_MOV_B32_1]], %subreg.sub0, killed [[S_MOV_B32_]], %subreg.sub1
- ; CHECK-NEXT: [[V_SUB_U:%[0-9]+]]:vreg_64 = V_SUB_U64_PSEUDO killed [[REG_SEQUENCE]], [[GLOBAL_LOAD_DWORDX2_SADDR]], implicit-def $vcc_lo, implicit $exec
+ ; CHECK-NEXT: [[V_SUB_U:%[0-9]+]]:vreg_64 = V_SUB_U64_PSEUDO killed [[REG_SEQUENCE]], [[GLOBAL_LOAD_DWORDX2_SADDR]], implicit-def dead $vcc_lo, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY [[V_SUB_U]]
; CHECK-NEXT: GLOBAL_STORE_DWORDX2_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s64) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
@@ -98,7 +98,7 @@ define amdgpu_kernel void @sub_reg_reg(ptr addrspace(1) %ptr) {
; CHECK-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; CHECK-NEXT: [[GLOBAL_LOAD_DWORDX2_SADDR:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s64) from %ir.ptr.load, addrspace 1)
; CHECK-NEXT: [[GLOBAL_LOAD_DWORDX2_SADDR1:%[0-9]+]]:vreg_64 = GLOBAL_LOAD_DWORDX2_SADDR [[S_LOAD_DWORDX2_IMM]], [[V_MOV_B32_e32_]], 0, 0, implicit $exec :: (volatile "amdgpu-noclobber" load (s64) from %ir.ptr.load, addrspace 1)
- ; CHECK-NEXT: [[V_SUB_U:%[0-9]+]]:vreg_64 = V_SUB_U64_PSEUDO [[GLOBAL_LOAD_DWORDX2_SADDR]], [[GLOBAL_LOAD_DWORDX2_SADDR1]], implicit-def $vcc_lo, implicit $exec
+ ; CHECK-NEXT: [[V_SUB_U:%[0-9]+]]:vreg_64 = V_SUB_U64_PSEUDO [[GLOBAL_LOAD_DWORDX2_SADDR]], [[GLOBAL_LOAD_DWORDX2_SADDR1]], implicit-def dead $vcc_lo, implicit $exec
; CHECK-NEXT: [[COPY1:%[0-9]+]]:vreg_64 = COPY [[V_SUB_U]]
; CHECK-NEXT: GLOBAL_STORE_DWORDX2_SADDR [[V_MOV_B32_e32_]], killed [[COPY1]], [[S_LOAD_DWORDX2_IMM]], 0, 0, implicit $exec :: (store (s64) into %ir.ptr.load, addrspace 1)
; CHECK-NEXT: S_ENDPGM 0
diff --git a/llvm/test/CodeGen/AMDGPU/opt-sgpr-to-vgpr-copy.mir b/llvm/test/CodeGen/AMDGPU/opt-sgpr-to-vgpr-copy.mir
index 6b3ef6c3b4547e..d5736fc9446c27 100644
--- a/llvm/test/CodeGen/AMDGPU/opt-sgpr-to-vgpr-copy.mir
+++ b/llvm/test/CodeGen/AMDGPU/opt-sgpr-to-vgpr-copy.mir
@@ -96,7 +96,7 @@ body: |
; GCN-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, killed [[V_MOV_B32_e32_]], %subreg.sub1
; GCN-NEXT: [[V_ADD_CO_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_CO_U32_e32 [[S_LOAD_DWORDX2_IMM1]].sub0, [[REG_SEQUENCE]].sub0, implicit-def $vcc, implicit $exec
; GCN-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[S_LOAD_DWORDX2_IMM1]].sub1
- ; GCN-NEXT: [[V_ADDC_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 0, [[COPY3]], implicit-def $vcc, implicit $vcc, implicit $exec
+ ; GCN-NEXT: [[V_ADDC_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 0, [[COPY3]], implicit-def dead $vcc, implicit $vcc, implicit $exec
; GCN-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE killed [[V_ADD_CO_U32_e32_]], %subreg.sub0, killed [[V_ADDC_U32_e32_]], %subreg.sub1
; GCN-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 0
; GCN-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 1048576
@@ -372,7 +372,7 @@ body: |
; GCN-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, killed [[V_MOV_B32_e32_]], %subreg.sub1
; GCN-NEXT: [[V_ADD_CO_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_CO_U32_e32 [[S_LOAD_DWORDX2_IMM1]].sub0, [[REG_SEQUENCE]].sub0, implicit-def $vcc, implicit $exec
; GCN-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY [[S_LOAD_DWORDX2_IMM1]].sub1
- ; GCN-NEXT: [[V_ADDC_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 0, [[COPY3]], implicit-def $vcc, implicit $vcc, implicit $exec
+ ; GCN-NEXT: [[V_ADDC_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADDC_U32_e32 0, [[COPY3]], implicit-def dead $vcc, implicit $vcc, implicit $exec
; GCN-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE killed [[V_ADD_CO_U32_e32_]], %subreg.sub0, killed [[V_ADDC_U32_e32_]], %subreg.sub1
; GCN-NEXT: [[V_CMP_LT_U32_e64_:%[0-9]+]]:sreg_64 = V_CMP_LT_U32_e64 killed [[REG_SEQUENCE1]].sub0, 12, implicit $exec
; GCN-NEXT: [[SI_IF:%[0-9]+]]:sreg_64 = SI_IF killed [[V_CMP_LT_U32_e64_]], %bb.2, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
diff --git a/llvm/test/CodeGen/AMDGPU/si-instr-info-correct-implicit-operands.ll b/llvm/test/CodeGen/AMDGPU/si-instr-info-correct-implicit-operands.ll
index d5f1b1a3a60633..4cdf8054fb8789 100644
--- a/llvm/test/CodeGen/AMDGPU/si-instr-info-correct-implicit-operands.ll
+++ b/llvm/test/CodeGen/AMDGPU/si-instr-info-correct-implicit-operands.ll
@@ -3,7 +3,7 @@
; register operands in the correct order when modifying the opcode of an
; instruction to V_ADD_CO_U32_e32.
-; CHECK: %{{[0-9]+}}:vgpr_32 = V_ADD_CO_U32_e32 %{{[0-9]+}}, %{{[0-9]+}}, implicit-def $vcc, implicit $exec
+; CHECK: %{{[0-9]+}}:vgpr_32 = V_ADD_CO_U32_e32 %{{[0-9]+}}, %{{[0-9]+}}, implicit-def dead $vcc, implicit $exec
define amdgpu_kernel void @test(ptr addrspace(1) %out, ptr addrspace(1) %in) {
entry:
More information about the llvm-branch-commits
mailing list