[llvm] AMDGPU: Preserve carry-out dead flag in buildShrunkInst (PR #225798)
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Wed Sep 23 07:48:15 PDT 2026
https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/225798
Shrinking from the e64 to the e32 form drops the explicit carry-out def
in favor of an implicit vcc def. If the original carry out had a dead flag,
preserve it in the shrunk instruction's implicit-def operand.
Co-Authored-By: Claude Opus 5 <noreply at anthropic.com>
>From 436ea379935fee1fea532b9717de4afad1b13a5c Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Wed, 23 Sep 2026 16:42:02 +0200
Subject: [PATCH] AMDGPU: Preserve carry-out dead flag in buildShrunkInst
Shrinking from the e64 to the e32 form drops the explicit carry-out def
in favor of an implicit vcc def. If the original carry out had a dead flag,
preserve it in the shrunk instruction's implicit-def operand.
Co-Authored-By: Claude Opus 5 <noreply at anthropic.com>
---
llvm/lib/Target/AMDGPU/SIInstrInfo.cpp | 10 ++
.../Target/AMDGPU/SIShrinkInstructions.cpp | 4 -
.../test/CodeGen/AMDGPU/GlobalISel/saddsat.ll | 114 ++++++++----------
.../test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll | 114 ++++++++----------
...ld-immediate-operand-shrink-dead-carry.mir | 44 +++++++
...ead-vcc-lo.mir => shrink-dead-vcc-def.mir} | 17 +++
6 files changed, 169 insertions(+), 134 deletions(-)
create mode 100644 llvm/test/CodeGen/AMDGPU/fold-immediate-operand-shrink-dead-carry.mir
rename llvm/test/CodeGen/AMDGPU/{shrink-v-cmp-wave32-dead-vcc-lo.mir => shrink-dead-vcc-def.mir} (73%)
diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index f584e7d67dd73f..46fc6d12f8e701 100644
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -5278,6 +5278,16 @@ MachineInstr *SIInstrInfo::buildShrunkInst(MachineInstr &MI,
// FIXME: Losing implicit operands
fixImplicitOperands(*Inst32);
+
+ // The explicit carry/result def is dropped in favor of an implicit VCC def;
+ // preserve the dead flag.
+ const MachineOperand *OldSDst = getNamedOperand(MI, AMDGPU::OpName::sdst);
+ if (OldSDst && OldSDst->isDead()) {
+ if (MachineOperand *NewVCC =
+ Inst32->findRegisterDefOperand(RI.getVCC(), &RI))
+ NewVCC->setIsDead();
+ }
+
return Inst32;
}
diff --git a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
index 7b5db0f75b1e7e..1483fd43ad8b1d 100644
--- a/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
+++ b/llvm/lib/Target/AMDGPU/SIShrinkInstructions.cpp
@@ -1117,10 +1117,6 @@ bool SIShrinkInstructions::run(MachineFunction &MF) {
// Copy extra operands not present in the instruction definition.
copyExtraImplicitOps(*Inst32, MI);
- // Copy deadness from the old explicit vcc def to the new implicit def.
- if (SDst && SDst->isDead())
- Inst32->findRegisterDefOperand(VCCReg, /*TRI=*/nullptr)->setIsDead();
-
MI.eraseFromParent();
foldImmediates(*Inst32);
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/saddsat.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/saddsat.ll
index 3850141f089a77..aee2cd8428427a 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/saddsat.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/saddsat.ll
@@ -311,11 +311,10 @@ define i16 @v_saddsat_v2i8(i16 %lhs.arg, i16 %rhs.arg) {
; GFX6-NEXT: v_lshrrev_b32_e32 v2, 8, v0
; GFX6-NEXT: v_lshlrev_b32_e32 v0, 24, v0
; GFX6-NEXT: v_min_i32_e32 v6, 0, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v7, 1
; GFX6-NEXT: v_lshrrev_b32_e32 v3, 8, v1
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 24, v1
; GFX6-NEXT: v_max_i32_e32 v4, 0, v0
-; GFX6-NEXT: v_sub_i32_e32 v6, vcc, v7, v6
+; GFX6-NEXT: v_sub_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_sub_i32_e32 v4, vcc, 0x7fffffff, v4
; GFX6-NEXT: v_max_i32_e32 v1, v6, v1
; GFX6-NEXT: v_min_i32_e32 v1, v1, v4
@@ -591,13 +590,12 @@ define i32 @v_saddsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
; GFX6-NEXT: v_lshrrev_b32_e32 v4, 24, v0
; GFX6-NEXT: v_lshlrev_b32_e32 v0, 24, v0
; GFX6-NEXT: v_min_i32_e32 v10, 0, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v11, 1
; GFX6-NEXT: v_lshrrev_b32_e32 v5, 8, v1
; GFX6-NEXT: v_lshrrev_b32_e32 v6, 16, v1
; GFX6-NEXT: v_lshrrev_b32_e32 v7, 24, v1
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 24, v1
; GFX6-NEXT: v_max_i32_e32 v8, 0, v0
-; GFX6-NEXT: v_sub_i32_e32 v10, vcc, v11, v10
+; GFX6-NEXT: v_sub_i32_e32 v10, vcc, 0x80000000, v10
; GFX6-NEXT: v_sub_i32_e32 v8, vcc, 0x7fffffff, v8
; GFX6-NEXT: v_max_i32_e32 v1, v10, v1
; GFX6-NEXT: v_min_i32_e32 v1, v1, v8
@@ -607,7 +605,7 @@ define i32 @v_saddsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
; GFX6-NEXT: v_bfrev_b32_e32 v9, -2
; GFX6-NEXT: v_lshlrev_b32_e32 v2, 24, v5
; GFX6-NEXT: v_max_i32_e32 v5, 0, v1
-; GFX6-NEXT: v_sub_i32_e32 v8, vcc, v11, v8
+; GFX6-NEXT: v_sub_i32_e32 v8, vcc, 0x80000000, v8
; GFX6-NEXT: v_sub_i32_e32 v5, vcc, v9, v5
; GFX6-NEXT: v_max_i32_e32 v2, v8, v2
; GFX6-NEXT: v_min_i32_e32 v2, v2, v5
@@ -616,7 +614,7 @@ define i32 @v_saddsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 24, v6
; GFX6-NEXT: v_min_i32_e32 v6, 0, v2
; GFX6-NEXT: v_max_i32_e32 v5, 0, v2
-; GFX6-NEXT: v_sub_i32_e32 v6, vcc, v11, v6
+; GFX6-NEXT: v_sub_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_sub_i32_e32 v5, vcc, v9, v5
; GFX6-NEXT: v_max_i32_e32 v3, v6, v3
; GFX6-NEXT: v_min_i32_e32 v3, v3, v5
@@ -626,7 +624,7 @@ define i32 @v_saddsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
; GFX6-NEXT: v_ashrrev_i32_e32 v1, 24, v1
; GFX6-NEXT: v_lshlrev_b32_e32 v4, 24, v7
; GFX6-NEXT: v_max_i32_e32 v5, 0, v3
-; GFX6-NEXT: v_sub_i32_e32 v6, vcc, v11, v6
+; GFX6-NEXT: v_sub_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_ashrrev_i32_e32 v0, 24, v0
; GFX6-NEXT: v_sub_i32_e32 v5, vcc, v9, v5
; GFX6-NEXT: v_max_i32_e32 v4, v6, v4
@@ -1583,18 +1581,17 @@ define <3 x i32> @v_saddsat_v3i32(<3 x i32> %lhs, <3 x i32> %rhs) {
; GFX6-LABEL: v_saddsat_v3i32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT: v_min_i32_e32 v8, 0, v0
+; GFX6-NEXT: v_min_i32_e32 v7, 0, v0
; GFX6-NEXT: v_max_i32_e32 v6, 0, v0
-; GFX6-NEXT: v_sub_i32_e32 v8, vcc, 0x80000000, v8
+; GFX6-NEXT: v_sub_i32_e32 v7, vcc, 0x80000000, v7
; GFX6-NEXT: v_sub_i32_e32 v6, vcc, 0x7fffffff, v6
-; GFX6-NEXT: v_max_i32_e32 v3, v8, v3
+; GFX6-NEXT: v_max_i32_e32 v3, v7, v3
; GFX6-NEXT: v_min_i32_e32 v3, v3, v6
; GFX6-NEXT: v_min_i32_e32 v6, 0, v1
-; GFX6-NEXT: v_bfrev_b32_e32 v7, -2
; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v3
; GFX6-NEXT: v_max_i32_e32 v3, 0, v1
; GFX6-NEXT: v_sub_i32_e32 v6, vcc, 0x80000000, v6
-; GFX6-NEXT: v_sub_i32_e32 v3, vcc, v7, v3
+; GFX6-NEXT: v_sub_i32_e32 v3, vcc, 0x7fffffff, v3
; GFX6-NEXT: v_max_i32_e32 v4, v6, v4
; GFX6-NEXT: v_min_i32_e32 v3, v4, v3
; GFX6-NEXT: v_min_i32_e32 v4, 0, v2
@@ -1610,18 +1607,17 @@ define <3 x i32> @v_saddsat_v3i32(<3 x i32> %lhs, <3 x i32> %rhs) {
; GFX8-LABEL: v_saddsat_v3i32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT: v_min_i32_e32 v8, 0, v0
+; GFX8-NEXT: v_min_i32_e32 v7, 0, v0
; GFX8-NEXT: v_max_i32_e32 v6, 0, v0
-; GFX8-NEXT: v_sub_u32_e32 v8, vcc, 0x80000000, v8
+; GFX8-NEXT: v_sub_u32_e32 v7, vcc, 0x80000000, v7
; GFX8-NEXT: v_sub_u32_e32 v6, vcc, 0x7fffffff, v6
-; GFX8-NEXT: v_max_i32_e32 v3, v8, v3
+; GFX8-NEXT: v_max_i32_e32 v3, v7, v3
; GFX8-NEXT: v_min_i32_e32 v3, v3, v6
; GFX8-NEXT: v_min_i32_e32 v6, 0, v1
-; GFX8-NEXT: v_bfrev_b32_e32 v7, -2
; GFX8-NEXT: v_add_u32_e32 v0, vcc, v0, v3
; GFX8-NEXT: v_max_i32_e32 v3, 0, v1
; GFX8-NEXT: v_sub_u32_e32 v6, vcc, 0x80000000, v6
-; GFX8-NEXT: v_sub_u32_e32 v3, vcc, v7, v3
+; GFX8-NEXT: v_sub_u32_e32 v3, vcc, 0x7fffffff, v3
; GFX8-NEXT: v_max_i32_e32 v4, v6, v4
; GFX8-NEXT: v_min_i32_e32 v3, v4, v3
; GFX8-NEXT: v_min_i32_e32 v4, 0, v2
@@ -1734,26 +1730,24 @@ define <4 x i32> @v_saddsat_v4i32(<4 x i32> %lhs, <4 x i32> %rhs) {
; GFX6-LABEL: v_saddsat_v4i32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT: v_min_i32_e32 v10, 0, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v11, 1
+; GFX6-NEXT: v_min_i32_e32 v9, 0, v0
; GFX6-NEXT: v_max_i32_e32 v8, 0, v0
-; GFX6-NEXT: v_sub_i32_e32 v10, vcc, v11, v10
+; GFX6-NEXT: v_sub_i32_e32 v9, vcc, 0x80000000, v9
; GFX6-NEXT: v_sub_i32_e32 v8, vcc, 0x7fffffff, v8
-; GFX6-NEXT: v_max_i32_e32 v4, v10, v4
+; GFX6-NEXT: v_max_i32_e32 v4, v9, v4
; GFX6-NEXT: v_min_i32_e32 v4, v4, v8
; GFX6-NEXT: v_min_i32_e32 v8, 0, v1
-; GFX6-NEXT: v_bfrev_b32_e32 v9, -2
; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v4
; GFX6-NEXT: v_max_i32_e32 v4, 0, v1
; GFX6-NEXT: v_sub_i32_e32 v8, vcc, 0x80000000, v8
-; GFX6-NEXT: v_sub_i32_e32 v4, vcc, v9, v4
+; GFX6-NEXT: v_sub_i32_e32 v4, vcc, 0x7fffffff, v4
; GFX6-NEXT: v_max_i32_e32 v5, v8, v5
; GFX6-NEXT: v_min_i32_e32 v4, v5, v4
; GFX6-NEXT: v_min_i32_e32 v5, 0, v2
; GFX6-NEXT: v_add_i32_e32 v1, vcc, v1, v4
; GFX6-NEXT: v_max_i32_e32 v4, 0, v2
; GFX6-NEXT: v_sub_i32_e32 v5, vcc, 0x80000000, v5
-; GFX6-NEXT: v_sub_i32_e32 v4, vcc, v9, v4
+; GFX6-NEXT: v_sub_i32_e32 v4, vcc, 0x7fffffff, v4
; GFX6-NEXT: v_max_i32_e32 v5, v5, v6
; GFX6-NEXT: v_min_i32_e32 v4, v5, v4
; GFX6-NEXT: v_min_i32_e32 v5, 0, v3
@@ -1769,26 +1763,24 @@ define <4 x i32> @v_saddsat_v4i32(<4 x i32> %lhs, <4 x i32> %rhs) {
; GFX8-LABEL: v_saddsat_v4i32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT: v_min_i32_e32 v10, 0, v0
-; GFX8-NEXT: v_bfrev_b32_e32 v11, 1
+; GFX8-NEXT: v_min_i32_e32 v9, 0, v0
; GFX8-NEXT: v_max_i32_e32 v8, 0, v0
-; GFX8-NEXT: v_sub_u32_e32 v10, vcc, v11, v10
+; GFX8-NEXT: v_sub_u32_e32 v9, vcc, 0x80000000, v9
; GFX8-NEXT: v_sub_u32_e32 v8, vcc, 0x7fffffff, v8
-; GFX8-NEXT: v_max_i32_e32 v4, v10, v4
+; GFX8-NEXT: v_max_i32_e32 v4, v9, v4
; GFX8-NEXT: v_min_i32_e32 v4, v4, v8
; GFX8-NEXT: v_min_i32_e32 v8, 0, v1
-; GFX8-NEXT: v_bfrev_b32_e32 v9, -2
; GFX8-NEXT: v_add_u32_e32 v0, vcc, v0, v4
; GFX8-NEXT: v_max_i32_e32 v4, 0, v1
; GFX8-NEXT: v_sub_u32_e32 v8, vcc, 0x80000000, v8
-; GFX8-NEXT: v_sub_u32_e32 v4, vcc, v9, v4
+; GFX8-NEXT: v_sub_u32_e32 v4, vcc, 0x7fffffff, v4
; GFX8-NEXT: v_max_i32_e32 v5, v8, v5
; GFX8-NEXT: v_min_i32_e32 v4, v5, v4
; GFX8-NEXT: v_min_i32_e32 v5, 0, v2
; GFX8-NEXT: v_add_u32_e32 v1, vcc, v1, v4
; GFX8-NEXT: v_max_i32_e32 v4, 0, v2
; GFX8-NEXT: v_sub_u32_e32 v5, vcc, 0x80000000, v5
-; GFX8-NEXT: v_sub_u32_e32 v4, vcc, v9, v4
+; GFX8-NEXT: v_sub_u32_e32 v4, vcc, 0x7fffffff, v4
; GFX8-NEXT: v_max_i32_e32 v5, v5, v6
; GFX8-NEXT: v_min_i32_e32 v4, v5, v4
; GFX8-NEXT: v_min_i32_e32 v5, 0, v3
@@ -1922,33 +1914,31 @@ define <5 x i32> @v_saddsat_v5i32(<5 x i32> %lhs, <5 x i32> %rhs) {
; GFX6-LABEL: v_saddsat_v5i32:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-NEXT: v_min_i32_e32 v12, 0, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v13, 1
+; GFX6-NEXT: v_min_i32_e32 v11, 0, v0
; GFX6-NEXT: v_max_i32_e32 v10, 0, v0
-; GFX6-NEXT: v_sub_i32_e32 v12, vcc, v13, v12
+; GFX6-NEXT: v_sub_i32_e32 v11, vcc, 0x80000000, v11
; GFX6-NEXT: v_sub_i32_e32 v10, vcc, 0x7fffffff, v10
-; GFX6-NEXT: v_max_i32_e32 v5, v12, v5
+; GFX6-NEXT: v_max_i32_e32 v5, v11, v5
; GFX6-NEXT: v_min_i32_e32 v5, v5, v10
; GFX6-NEXT: v_min_i32_e32 v10, 0, v1
-; GFX6-NEXT: v_bfrev_b32_e32 v11, -2
; GFX6-NEXT: v_add_i32_e32 v0, vcc, v0, v5
; GFX6-NEXT: v_max_i32_e32 v5, 0, v1
-; GFX6-NEXT: v_sub_i32_e32 v10, vcc, v13, v10
-; GFX6-NEXT: v_sub_i32_e32 v5, vcc, v11, v5
+; GFX6-NEXT: v_sub_i32_e32 v10, vcc, 0x80000000, v10
+; GFX6-NEXT: v_sub_i32_e32 v5, vcc, 0x7fffffff, v5
; GFX6-NEXT: v_max_i32_e32 v6, v10, v6
; GFX6-NEXT: v_min_i32_e32 v5, v6, v5
; GFX6-NEXT: v_min_i32_e32 v6, 0, v2
; GFX6-NEXT: v_add_i32_e32 v1, vcc, v1, v5
; GFX6-NEXT: v_max_i32_e32 v5, 0, v2
; GFX6-NEXT: v_sub_i32_e32 v6, vcc, 0x80000000, v6
-; GFX6-NEXT: v_sub_i32_e32 v5, vcc, v11, v5
+; GFX6-NEXT: v_sub_i32_e32 v5, vcc, 0x7fffffff, v5
; GFX6-NEXT: v_max_i32_e32 v6, v6, v7
; GFX6-NEXT: v_min_i32_e32 v5, v6, v5
; GFX6-NEXT: v_min_i32_e32 v6, 0, v3
; GFX6-NEXT: v_add_i32_e32 v2, vcc, v2, v5
; GFX6-NEXT: v_max_i32_e32 v5, 0, v3
; GFX6-NEXT: v_sub_i32_e32 v6, vcc, 0x80000000, v6
-; GFX6-NEXT: v_sub_i32_e32 v5, vcc, v11, v5
+; GFX6-NEXT: v_sub_i32_e32 v5, vcc, 0x7fffffff, v5
; GFX6-NEXT: v_max_i32_e32 v6, v6, v8
; GFX6-NEXT: v_min_i32_e32 v5, v6, v5
; GFX6-NEXT: v_min_i32_e32 v6, 0, v4
@@ -1964,33 +1954,31 @@ define <5 x i32> @v_saddsat_v5i32(<5 x i32> %lhs, <5 x i32> %rhs) {
; GFX8-LABEL: v_saddsat_v5i32:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT: v_min_i32_e32 v12, 0, v0
-; GFX8-NEXT: v_bfrev_b32_e32 v13, 1
+; GFX8-NEXT: v_min_i32_e32 v11, 0, v0
; GFX8-NEXT: v_max_i32_e32 v10, 0, v0
-; GFX8-NEXT: v_sub_u32_e32 v12, vcc, v13, v12
+; GFX8-NEXT: v_sub_u32_e32 v11, vcc, 0x80000000, v11
; GFX8-NEXT: v_sub_u32_e32 v10, vcc, 0x7fffffff, v10
-; GFX8-NEXT: v_max_i32_e32 v5, v12, v5
+; GFX8-NEXT: v_max_i32_e32 v5, v11, v5
; GFX8-NEXT: v_min_i32_e32 v5, v5, v10
; GFX8-NEXT: v_min_i32_e32 v10, 0, v1
-; GFX8-NEXT: v_bfrev_b32_e32 v11, -2
; GFX8-NEXT: v_add_u32_e32 v0, vcc, v0, v5
; GFX8-NEXT: v_max_i32_e32 v5, 0, v1
-; GFX8-NEXT: v_sub_u32_e32 v10, vcc, v13, v10
-; GFX8-NEXT: v_sub_u32_e32 v5, vcc, v11, v5
+; GFX8-NEXT: v_sub_u32_e32 v10, vcc, 0x80000000, v10
+; GFX8-NEXT: v_sub_u32_e32 v5, vcc, 0x7fffffff, v5
; GFX8-NEXT: v_max_i32_e32 v6, v10, v6
; GFX8-NEXT: v_min_i32_e32 v5, v6, v5
; GFX8-NEXT: v_min_i32_e32 v6, 0, v2
; GFX8-NEXT: v_add_u32_e32 v1, vcc, v1, v5
; GFX8-NEXT: v_max_i32_e32 v5, 0, v2
; GFX8-NEXT: v_sub_u32_e32 v6, vcc, 0x80000000, v6
-; GFX8-NEXT: v_sub_u32_e32 v5, vcc, v11, v5
+; GFX8-NEXT: v_sub_u32_e32 v5, vcc, 0x7fffffff, v5
; GFX8-NEXT: v_max_i32_e32 v6, v6, v7
; GFX8-NEXT: v_min_i32_e32 v5, v6, v5
; GFX8-NEXT: v_min_i32_e32 v6, 0, v3
; GFX8-NEXT: v_add_u32_e32 v2, vcc, v2, v5
; GFX8-NEXT: v_max_i32_e32 v5, 0, v3
; GFX8-NEXT: v_sub_u32_e32 v6, vcc, 0x80000000, v6
-; GFX8-NEXT: v_sub_u32_e32 v5, vcc, v11, v5
+; GFX8-NEXT: v_sub_u32_e32 v5, vcc, 0x7fffffff, v5
; GFX8-NEXT: v_max_i32_e32 v6, v6, v8
; GFX8-NEXT: v_min_i32_e32 v5, v6, v5
; GFX8-NEXT: v_min_i32_e32 v6, 0, v4
@@ -3002,11 +2990,10 @@ define <2 x i16> @v_saddsat_v2i16(<2 x i16> %lhs, <2 x i16> %rhs) {
; GFX6-NEXT: v_lshrrev_b32_e32 v2, 16, v0
; GFX6-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; GFX6-NEXT: v_min_i32_e32 v6, 0, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v7, 1
; GFX6-NEXT: v_lshrrev_b32_e32 v3, 16, v1
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; GFX6-NEXT: v_max_i32_e32 v4, 0, v0
-; GFX6-NEXT: v_sub_i32_e32 v6, vcc, v7, v6
+; GFX6-NEXT: v_sub_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_sub_i32_e32 v4, vcc, 0x7fffffff, v4
; GFX6-NEXT: v_max_i32_e32 v1, v6, v1
; GFX6-NEXT: v_min_i32_e32 v1, v1, v4
@@ -3222,11 +3209,10 @@ define amdgpu_ps float @saddsat_v2i16_vs(<2 x i16> %lhs, <2 x i16> inreg %rhs) {
; GFX6-NEXT: v_lshrrev_b32_e32 v1, 16, v0
; GFX6-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; GFX6-NEXT: v_min_i32_e32 v4, 0, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v5, 1
; GFX6-NEXT: s_lshr_b32 s1, s0, 16
; GFX6-NEXT: s_lshl_b32 s0, s0, 16
; GFX6-NEXT: v_max_i32_e32 v2, 0, v0
-; GFX6-NEXT: v_sub_i32_e32 v4, vcc, v5, v4
+; GFX6-NEXT: v_sub_i32_e32 v4, vcc, 0x80000000, v4
; GFX6-NEXT: v_sub_i32_e32 v2, vcc, 0x7fffffff, v2
; GFX6-NEXT: v_max_i32_e32 v4, s0, v4
; GFX6-NEXT: v_min_i32_e32 v2, v4, v2
@@ -3302,11 +3288,10 @@ define <2 x float> @v_saddsat_v4i16(<4 x i16> %lhs, <4 x i16> %rhs) {
; GFX6-NEXT: v_lshrrev_b32_e32 v4, 16, v0
; GFX6-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; GFX6-NEXT: v_min_i32_e32 v10, 0, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v11, 1
; GFX6-NEXT: v_lshrrev_b32_e32 v6, 16, v2
; GFX6-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; GFX6-NEXT: v_max_i32_e32 v8, 0, v0
-; GFX6-NEXT: v_sub_i32_e32 v10, vcc, v11, v10
+; GFX6-NEXT: v_sub_i32_e32 v10, vcc, 0x80000000, v10
; GFX6-NEXT: v_sub_i32_e32 v8, vcc, 0x7fffffff, v8
; GFX6-NEXT: v_max_i32_e32 v2, v10, v2
; GFX6-NEXT: v_min_i32_e32 v2, v2, v8
@@ -3316,7 +3301,7 @@ define <2 x float> @v_saddsat_v4i16(<4 x i16> %lhs, <4 x i16> %rhs) {
; GFX6-NEXT: v_bfrev_b32_e32 v9, -2
; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v6
; GFX6-NEXT: v_max_i32_e32 v6, 0, v2
-; GFX6-NEXT: v_sub_i32_e32 v8, vcc, v11, v8
+; GFX6-NEXT: v_sub_i32_e32 v8, vcc, 0x80000000, v8
; GFX6-NEXT: v_lshrrev_b32_e32 v5, 16, v1
; GFX6-NEXT: v_sub_i32_e32 v6, vcc, v9, v6
; GFX6-NEXT: v_max_i32_e32 v4, v8, v4
@@ -3327,7 +3312,7 @@ define <2 x float> @v_saddsat_v4i16(<4 x i16> %lhs, <4 x i16> %rhs) {
; GFX6-NEXT: v_add_i32_e32 v2, vcc, v2, v4
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3
; GFX6-NEXT: v_max_i32_e32 v4, 0, v1
-; GFX6-NEXT: v_sub_i32_e32 v6, vcc, v11, v6
+; GFX6-NEXT: v_sub_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_sub_i32_e32 v4, vcc, v9, v4
; GFX6-NEXT: v_max_i32_e32 v3, v6, v3
; GFX6-NEXT: v_min_i32_e32 v3, v3, v4
@@ -3336,7 +3321,7 @@ define <2 x float> @v_saddsat_v4i16(<4 x i16> %lhs, <4 x i16> %rhs) {
; GFX6-NEXT: v_min_i32_e32 v6, 0, v3
; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v7
; GFX6-NEXT: v_max_i32_e32 v5, 0, v3
-; GFX6-NEXT: v_sub_i32_e32 v6, vcc, v11, v6
+; GFX6-NEXT: v_sub_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_sub_i32_e32 v5, vcc, v9, v5
; GFX6-NEXT: v_max_i32_e32 v4, v6, v4
; GFX6-NEXT: v_ashrrev_i32_e32 v2, 16, v2
@@ -3570,11 +3555,10 @@ define <3 x float> @v_saddsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_lshrrev_b32_e32 v6, 16, v0
; GFX6-NEXT: v_lshlrev_b32_e32 v0, 16, v0
; GFX6-NEXT: v_min_i32_e32 v14, 0, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v15, 1
; GFX6-NEXT: v_lshrrev_b32_e32 v9, 16, v3
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3
; GFX6-NEXT: v_max_i32_e32 v12, 0, v0
-; GFX6-NEXT: v_sub_i32_e32 v14, vcc, v15, v14
+; GFX6-NEXT: v_sub_i32_e32 v14, vcc, 0x80000000, v14
; GFX6-NEXT: v_sub_i32_e32 v12, vcc, 0x7fffffff, v12
; GFX6-NEXT: v_max_i32_e32 v3, v14, v3
; GFX6-NEXT: v_min_i32_e32 v3, v3, v12
@@ -3584,7 +3568,7 @@ define <3 x float> @v_saddsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_bfrev_b32_e32 v13, -2
; GFX6-NEXT: v_lshlrev_b32_e32 v6, 16, v9
; GFX6-NEXT: v_max_i32_e32 v9, 0, v3
-; GFX6-NEXT: v_sub_i32_e32 v12, vcc, v15, v12
+; GFX6-NEXT: v_sub_i32_e32 v12, vcc, 0x80000000, v12
; GFX6-NEXT: v_lshrrev_b32_e32 v7, 16, v1
; GFX6-NEXT: v_sub_i32_e32 v9, vcc, v13, v9
; GFX6-NEXT: v_max_i32_e32 v6, v12, v6
@@ -3595,7 +3579,7 @@ define <3 x float> @v_saddsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_add_i32_e32 v3, vcc, v3, v6
; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v4
; GFX6-NEXT: v_max_i32_e32 v6, 0, v1
-; GFX6-NEXT: v_sub_i32_e32 v9, vcc, v15, v9
+; GFX6-NEXT: v_sub_i32_e32 v9, vcc, 0x80000000, v9
; GFX6-NEXT: v_sub_i32_e32 v6, vcc, v13, v6
; GFX6-NEXT: v_max_i32_e32 v4, v9, v4
; GFX6-NEXT: v_min_i32_e32 v4, v4, v6
@@ -3604,7 +3588,7 @@ define <3 x float> @v_saddsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_min_i32_e32 v9, 0, v4
; GFX6-NEXT: v_lshlrev_b32_e32 v6, 16, v10
; GFX6-NEXT: v_max_i32_e32 v7, 0, v4
-; GFX6-NEXT: v_sub_i32_e32 v9, vcc, v15, v9
+; GFX6-NEXT: v_sub_i32_e32 v9, vcc, 0x80000000, v9
; GFX6-NEXT: v_lshrrev_b32_e32 v8, 16, v2
; GFX6-NEXT: v_sub_i32_e32 v7, vcc, v13, v7
; GFX6-NEXT: v_max_i32_e32 v6, v9, v6
@@ -3615,7 +3599,7 @@ define <3 x float> @v_saddsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_add_i32_e32 v4, vcc, v4, v6
; GFX6-NEXT: v_lshlrev_b32_e32 v5, 16, v5
; GFX6-NEXT: v_max_i32_e32 v6, 0, v2
-; GFX6-NEXT: v_sub_i32_e32 v7, vcc, v15, v7
+; GFX6-NEXT: v_sub_i32_e32 v7, vcc, 0x80000000, v7
; GFX6-NEXT: v_sub_i32_e32 v6, vcc, v13, v6
; GFX6-NEXT: v_max_i32_e32 v5, v7, v5
; GFX6-NEXT: v_min_i32_e32 v5, v5, v6
@@ -3625,7 +3609,7 @@ define <3 x float> @v_saddsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_ashrrev_i32_e32 v3, 16, v3
; GFX6-NEXT: v_lshlrev_b32_e32 v6, 16, v11
; GFX6-NEXT: v_max_i32_e32 v7, 0, v5
-; GFX6-NEXT: v_sub_i32_e32 v8, vcc, v15, v8
+; GFX6-NEXT: v_sub_i32_e32 v8, vcc, 0x80000000, v8
; GFX6-NEXT: v_ashrrev_i32_e32 v0, 16, v0
; GFX6-NEXT: v_sub_i32_e32 v7, vcc, v13, v7
; GFX6-NEXT: v_max_i32_e32 v6, v8, v6
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll
index fb824fa0ee11bf..e23337d7ae9ec5 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/ssubsat.ll
@@ -315,8 +315,7 @@ define i16 @v_ssubsat_v2i8(i16 %lhs.arg, i16 %rhs.arg) {
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 24, v1
; GFX6-NEXT: v_add_i32_e32 v4, vcc, 0x80000001, v4
; GFX6-NEXT: v_min_i32_e32 v6, -1, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v7, 1
-; GFX6-NEXT: v_add_i32_e32 v6, vcc, v6, v7
+; GFX6-NEXT: v_add_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_max_i32_e32 v1, v4, v1
; GFX6-NEXT: v_min_i32_e32 v1, v1, v6
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -597,8 +596,7 @@ define i32 @v_ssubsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 24, v1
; GFX6-NEXT: v_add_i32_e32 v8, vcc, 0x80000001, v8
; GFX6-NEXT: v_min_i32_e32 v10, -1, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v11, 1
-; GFX6-NEXT: v_add_i32_e32 v10, vcc, v10, v11
+; GFX6-NEXT: v_add_i32_e32 v10, vcc, 0x80000000, v10
; GFX6-NEXT: v_max_i32_e32 v1, v8, v1
; GFX6-NEXT: v_min_i32_e32 v1, v1, v10
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -608,7 +606,7 @@ define i32 @v_ssubsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
; GFX6-NEXT: v_max_i32_e32 v5, -1, v1
; GFX6-NEXT: v_add_i32_e32 v5, vcc, v5, v9
; GFX6-NEXT: v_min_i32_e32 v8, -1, v1
-; GFX6-NEXT: v_add_i32_e32 v8, vcc, v8, v11
+; GFX6-NEXT: v_add_i32_e32 v8, vcc, 0x80000000, v8
; GFX6-NEXT: v_max_i32_e32 v2, v5, v2
; GFX6-NEXT: v_min_i32_e32 v2, v2, v8
; GFX6-NEXT: v_sub_i32_e32 v1, vcc, v1, v2
@@ -617,7 +615,7 @@ define i32 @v_ssubsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 24, v6
; GFX6-NEXT: v_add_i32_e32 v5, vcc, v5, v9
; GFX6-NEXT: v_min_i32_e32 v6, -1, v2
-; GFX6-NEXT: v_add_i32_e32 v6, vcc, v6, v11
+; GFX6-NEXT: v_add_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_max_i32_e32 v3, v5, v3
; GFX6-NEXT: v_min_i32_e32 v3, v3, v6
; GFX6-NEXT: v_sub_i32_e32 v2, vcc, v2, v3
@@ -628,7 +626,7 @@ define i32 @v_ssubsat_v4i8(i32 %lhs.arg, i32 %rhs.arg) {
; GFX6-NEXT: v_add_i32_e32 v5, vcc, v5, v9
; GFX6-NEXT: v_min_i32_e32 v6, -1, v3
; GFX6-NEXT: v_ashrrev_i32_e32 v0, 24, v0
-; GFX6-NEXT: v_add_i32_e32 v6, vcc, v6, v11
+; GFX6-NEXT: v_add_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_max_i32_e32 v4, v5, v4
; GFX6-NEXT: v_and_b32_e32 v1, 0xff, v1
; GFX6-NEXT: v_ashrrev_i32_e32 v2, 24, v2
@@ -1585,14 +1583,13 @@ define <3 x i32> @v_ssubsat_v3i32(<3 x i32> %lhs, <3 x i32> %rhs) {
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_max_i32_e32 v6, -1, v0
; GFX6-NEXT: v_add_i32_e32 v6, vcc, 0x80000001, v6
-; GFX6-NEXT: v_min_i32_e32 v8, -1, v0
-; GFX6-NEXT: v_add_i32_e32 v8, vcc, 0x80000000, v8
+; GFX6-NEXT: v_min_i32_e32 v7, -1, v0
+; GFX6-NEXT: v_add_i32_e32 v7, vcc, 0x80000000, v7
; GFX6-NEXT: v_max_i32_e32 v3, v6, v3
-; GFX6-NEXT: v_min_i32_e32 v3, v3, v8
-; GFX6-NEXT: v_mov_b32_e32 v7, 0x80000001
+; GFX6-NEXT: v_min_i32_e32 v3, v3, v7
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
; GFX6-NEXT: v_max_i32_e32 v3, -1, v1
-; GFX6-NEXT: v_add_i32_e32 v3, vcc, v3, v7
+; GFX6-NEXT: v_add_i32_e32 v3, vcc, 0x80000001, v3
; GFX6-NEXT: v_min_i32_e32 v6, -1, v1
; GFX6-NEXT: v_add_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_max_i32_e32 v3, v3, v4
@@ -1612,14 +1609,13 @@ define <3 x i32> @v_ssubsat_v3i32(<3 x i32> %lhs, <3 x i32> %rhs) {
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_max_i32_e32 v6, -1, v0
; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x80000001, v6
-; GFX8-NEXT: v_min_i32_e32 v8, -1, v0
-; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x80000000, v8
+; GFX8-NEXT: v_min_i32_e32 v7, -1, v0
+; GFX8-NEXT: v_add_u32_e32 v7, vcc, 0x80000000, v7
; GFX8-NEXT: v_max_i32_e32 v3, v6, v3
-; GFX8-NEXT: v_min_i32_e32 v3, v3, v8
-; GFX8-NEXT: v_mov_b32_e32 v7, 0x80000001
+; GFX8-NEXT: v_min_i32_e32 v3, v3, v7
; GFX8-NEXT: v_sub_u32_e32 v0, vcc, v0, v3
; GFX8-NEXT: v_max_i32_e32 v3, -1, v1
-; GFX8-NEXT: v_add_u32_e32 v3, vcc, v3, v7
+; GFX8-NEXT: v_add_u32_e32 v3, vcc, 0x80000001, v3
; GFX8-NEXT: v_min_i32_e32 v6, -1, v1
; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x80000000, v6
; GFX8-NEXT: v_max_i32_e32 v3, v3, v4
@@ -1736,22 +1732,20 @@ define <4 x i32> @v_ssubsat_v4i32(<4 x i32> %lhs, <4 x i32> %rhs) {
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_max_i32_e32 v8, -1, v0
; GFX6-NEXT: v_add_i32_e32 v8, vcc, 0x80000001, v8
-; GFX6-NEXT: v_min_i32_e32 v10, -1, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v11, 1
-; GFX6-NEXT: v_add_i32_e32 v10, vcc, v10, v11
+; GFX6-NEXT: v_min_i32_e32 v9, -1, v0
+; GFX6-NEXT: v_add_i32_e32 v9, vcc, 0x80000000, v9
; GFX6-NEXT: v_max_i32_e32 v4, v8, v4
-; GFX6-NEXT: v_min_i32_e32 v4, v4, v10
-; GFX6-NEXT: v_mov_b32_e32 v9, 0x80000001
+; GFX6-NEXT: v_min_i32_e32 v4, v4, v9
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v0, v4
; GFX6-NEXT: v_max_i32_e32 v4, -1, v1
-; GFX6-NEXT: v_add_i32_e32 v4, vcc, v4, v9
+; GFX6-NEXT: v_add_i32_e32 v4, vcc, 0x80000001, v4
; GFX6-NEXT: v_min_i32_e32 v8, -1, v1
; GFX6-NEXT: v_add_i32_e32 v8, vcc, 0x80000000, v8
; GFX6-NEXT: v_max_i32_e32 v4, v4, v5
; GFX6-NEXT: v_min_i32_e32 v4, v4, v8
; GFX6-NEXT: v_sub_i32_e32 v1, vcc, v1, v4
; GFX6-NEXT: v_max_i32_e32 v4, -1, v2
-; GFX6-NEXT: v_add_i32_e32 v4, vcc, v4, v9
+; GFX6-NEXT: v_add_i32_e32 v4, vcc, 0x80000001, v4
; GFX6-NEXT: v_min_i32_e32 v5, -1, v2
; GFX6-NEXT: v_add_i32_e32 v5, vcc, 0x80000000, v5
; GFX6-NEXT: v_max_i32_e32 v4, v4, v6
@@ -1771,22 +1765,20 @@ define <4 x i32> @v_ssubsat_v4i32(<4 x i32> %lhs, <4 x i32> %rhs) {
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_max_i32_e32 v8, -1, v0
; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x80000001, v8
-; GFX8-NEXT: v_min_i32_e32 v10, -1, v0
-; GFX8-NEXT: v_bfrev_b32_e32 v11, 1
-; GFX8-NEXT: v_add_u32_e32 v10, vcc, v10, v11
+; GFX8-NEXT: v_min_i32_e32 v9, -1, v0
+; GFX8-NEXT: v_add_u32_e32 v9, vcc, 0x80000000, v9
; GFX8-NEXT: v_max_i32_e32 v4, v8, v4
-; GFX8-NEXT: v_min_i32_e32 v4, v4, v10
-; GFX8-NEXT: v_mov_b32_e32 v9, 0x80000001
+; GFX8-NEXT: v_min_i32_e32 v4, v4, v9
; GFX8-NEXT: v_sub_u32_e32 v0, vcc, v0, v4
; GFX8-NEXT: v_max_i32_e32 v4, -1, v1
-; GFX8-NEXT: v_add_u32_e32 v4, vcc, v4, v9
+; GFX8-NEXT: v_add_u32_e32 v4, vcc, 0x80000001, v4
; GFX8-NEXT: v_min_i32_e32 v8, -1, v1
; GFX8-NEXT: v_add_u32_e32 v8, vcc, 0x80000000, v8
; GFX8-NEXT: v_max_i32_e32 v4, v4, v5
; GFX8-NEXT: v_min_i32_e32 v4, v4, v8
; GFX8-NEXT: v_sub_u32_e32 v1, vcc, v1, v4
; GFX8-NEXT: v_max_i32_e32 v4, -1, v2
-; GFX8-NEXT: v_add_u32_e32 v4, vcc, v4, v9
+; GFX8-NEXT: v_add_u32_e32 v4, vcc, 0x80000001, v4
; GFX8-NEXT: v_min_i32_e32 v5, -1, v2
; GFX8-NEXT: v_add_u32_e32 v5, vcc, 0x80000000, v5
; GFX8-NEXT: v_max_i32_e32 v4, v4, v6
@@ -1924,29 +1916,27 @@ define <5 x i32> @v_ssubsat_v5i32(<5 x i32> %lhs, <5 x i32> %rhs) {
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: v_max_i32_e32 v10, -1, v0
; GFX6-NEXT: v_add_i32_e32 v10, vcc, 0x80000001, v10
-; GFX6-NEXT: v_min_i32_e32 v12, -1, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v13, 1
-; GFX6-NEXT: v_add_i32_e32 v12, vcc, v12, v13
+; GFX6-NEXT: v_min_i32_e32 v11, -1, v0
+; GFX6-NEXT: v_add_i32_e32 v11, vcc, 0x80000000, v11
; GFX6-NEXT: v_max_i32_e32 v5, v10, v5
-; GFX6-NEXT: v_min_i32_e32 v5, v5, v12
-; GFX6-NEXT: v_mov_b32_e32 v11, 0x80000001
+; GFX6-NEXT: v_min_i32_e32 v5, v5, v11
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v0, v5
; GFX6-NEXT: v_max_i32_e32 v5, -1, v1
-; GFX6-NEXT: v_add_i32_e32 v5, vcc, v5, v11
+; GFX6-NEXT: v_add_i32_e32 v5, vcc, 0x80000001, v5
; GFX6-NEXT: v_min_i32_e32 v10, -1, v1
-; GFX6-NEXT: v_add_i32_e32 v10, vcc, v10, v13
+; GFX6-NEXT: v_add_i32_e32 v10, vcc, 0x80000000, v10
; GFX6-NEXT: v_max_i32_e32 v5, v5, v6
; GFX6-NEXT: v_min_i32_e32 v5, v5, v10
; GFX6-NEXT: v_sub_i32_e32 v1, vcc, v1, v5
; GFX6-NEXT: v_max_i32_e32 v5, -1, v2
-; GFX6-NEXT: v_add_i32_e32 v5, vcc, v5, v11
+; GFX6-NEXT: v_add_i32_e32 v5, vcc, 0x80000001, v5
; GFX6-NEXT: v_min_i32_e32 v6, -1, v2
; GFX6-NEXT: v_add_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_max_i32_e32 v5, v5, v7
; GFX6-NEXT: v_min_i32_e32 v5, v5, v6
; GFX6-NEXT: v_sub_i32_e32 v2, vcc, v2, v5
; GFX6-NEXT: v_max_i32_e32 v5, -1, v3
-; GFX6-NEXT: v_add_i32_e32 v5, vcc, v5, v11
+; GFX6-NEXT: v_add_i32_e32 v5, vcc, 0x80000001, v5
; GFX6-NEXT: v_min_i32_e32 v6, -1, v3
; GFX6-NEXT: v_add_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_max_i32_e32 v5, v5, v8
@@ -1966,29 +1956,27 @@ define <5 x i32> @v_ssubsat_v5i32(<5 x i32> %lhs, <5 x i32> %rhs) {
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX8-NEXT: v_max_i32_e32 v10, -1, v0
; GFX8-NEXT: v_add_u32_e32 v10, vcc, 0x80000001, v10
-; GFX8-NEXT: v_min_i32_e32 v12, -1, v0
-; GFX8-NEXT: v_bfrev_b32_e32 v13, 1
-; GFX8-NEXT: v_add_u32_e32 v12, vcc, v12, v13
+; GFX8-NEXT: v_min_i32_e32 v11, -1, v0
+; GFX8-NEXT: v_add_u32_e32 v11, vcc, 0x80000000, v11
; GFX8-NEXT: v_max_i32_e32 v5, v10, v5
-; GFX8-NEXT: v_min_i32_e32 v5, v5, v12
-; GFX8-NEXT: v_mov_b32_e32 v11, 0x80000001
+; GFX8-NEXT: v_min_i32_e32 v5, v5, v11
; GFX8-NEXT: v_sub_u32_e32 v0, vcc, v0, v5
; GFX8-NEXT: v_max_i32_e32 v5, -1, v1
-; GFX8-NEXT: v_add_u32_e32 v5, vcc, v5, v11
+; GFX8-NEXT: v_add_u32_e32 v5, vcc, 0x80000001, v5
; GFX8-NEXT: v_min_i32_e32 v10, -1, v1
-; GFX8-NEXT: v_add_u32_e32 v10, vcc, v10, v13
+; GFX8-NEXT: v_add_u32_e32 v10, vcc, 0x80000000, v10
; GFX8-NEXT: v_max_i32_e32 v5, v5, v6
; GFX8-NEXT: v_min_i32_e32 v5, v5, v10
; GFX8-NEXT: v_sub_u32_e32 v1, vcc, v1, v5
; GFX8-NEXT: v_max_i32_e32 v5, -1, v2
-; GFX8-NEXT: v_add_u32_e32 v5, vcc, v5, v11
+; GFX8-NEXT: v_add_u32_e32 v5, vcc, 0x80000001, v5
; GFX8-NEXT: v_min_i32_e32 v6, -1, v2
; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x80000000, v6
; GFX8-NEXT: v_max_i32_e32 v5, v5, v7
; GFX8-NEXT: v_min_i32_e32 v5, v5, v6
; GFX8-NEXT: v_sub_u32_e32 v2, vcc, v2, v5
; GFX8-NEXT: v_max_i32_e32 v5, -1, v3
-; GFX8-NEXT: v_add_u32_e32 v5, vcc, v5, v11
+; GFX8-NEXT: v_add_u32_e32 v5, vcc, 0x80000001, v5
; GFX8-NEXT: v_min_i32_e32 v6, -1, v3
; GFX8-NEXT: v_add_u32_e32 v6, vcc, 0x80000000, v6
; GFX8-NEXT: v_max_i32_e32 v5, v5, v8
@@ -3006,8 +2994,7 @@ define <2 x i16> @v_ssubsat_v2i16(<2 x i16> %lhs, <2 x i16> %rhs) {
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; GFX6-NEXT: v_add_i32_e32 v4, vcc, 0x80000001, v4
; GFX6-NEXT: v_min_i32_e32 v6, -1, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v7, 1
-; GFX6-NEXT: v_add_i32_e32 v6, vcc, v6, v7
+; GFX6-NEXT: v_add_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_max_i32_e32 v1, v4, v1
; GFX6-NEXT: v_min_i32_e32 v1, v1, v6
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -3226,8 +3213,7 @@ define amdgpu_ps float @ssubsat_v2i16_vs(<2 x i16> %lhs, <2 x i16> inreg %rhs) {
; GFX6-NEXT: s_lshl_b32 s0, s0, 16
; GFX6-NEXT: v_add_i32_e32 v2, vcc, 0x80000001, v2
; GFX6-NEXT: v_min_i32_e32 v4, -1, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v5, 1
-; GFX6-NEXT: v_add_i32_e32 v4, vcc, v4, v5
+; GFX6-NEXT: v_add_i32_e32 v4, vcc, 0x80000000, v4
; GFX6-NEXT: v_max_i32_e32 v2, s0, v2
; GFX6-NEXT: v_min_i32_e32 v2, v2, v4
; GFX6-NEXT: v_lshlrev_b32_e32 v1, 16, v1
@@ -3306,8 +3292,7 @@ define <2 x float> @v_ssubsat_v4i16(<4 x i16> %lhs, <4 x i16> %rhs) {
; GFX6-NEXT: v_lshlrev_b32_e32 v2, 16, v2
; GFX6-NEXT: v_add_i32_e32 v8, vcc, 0x80000001, v8
; GFX6-NEXT: v_min_i32_e32 v10, -1, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v11, 1
-; GFX6-NEXT: v_add_i32_e32 v10, vcc, v10, v11
+; GFX6-NEXT: v_add_i32_e32 v10, vcc, 0x80000000, v10
; GFX6-NEXT: v_max_i32_e32 v2, v8, v2
; GFX6-NEXT: v_min_i32_e32 v2, v2, v10
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v0, v2
@@ -3317,7 +3302,7 @@ define <2 x float> @v_ssubsat_v4i16(<4 x i16> %lhs, <4 x i16> %rhs) {
; GFX6-NEXT: v_max_i32_e32 v6, -1, v2
; GFX6-NEXT: v_add_i32_e32 v6, vcc, v6, v9
; GFX6-NEXT: v_min_i32_e32 v8, -1, v2
-; GFX6-NEXT: v_add_i32_e32 v8, vcc, v8, v11
+; GFX6-NEXT: v_add_i32_e32 v8, vcc, 0x80000000, v8
; GFX6-NEXT: v_max_i32_e32 v4, v6, v4
; GFX6-NEXT: v_lshrrev_b32_e32 v5, 16, v1
; GFX6-NEXT: v_min_i32_e32 v4, v4, v8
@@ -3328,7 +3313,7 @@ define <2 x float> @v_ssubsat_v4i16(<4 x i16> %lhs, <4 x i16> %rhs) {
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3
; GFX6-NEXT: v_add_i32_e32 v4, vcc, v4, v9
; GFX6-NEXT: v_min_i32_e32 v6, -1, v1
-; GFX6-NEXT: v_add_i32_e32 v6, vcc, v6, v11
+; GFX6-NEXT: v_add_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_max_i32_e32 v3, v4, v3
; GFX6-NEXT: v_min_i32_e32 v3, v3, v6
; GFX6-NEXT: v_sub_i32_e32 v1, vcc, v1, v3
@@ -3337,7 +3322,7 @@ define <2 x float> @v_ssubsat_v4i16(<4 x i16> %lhs, <4 x i16> %rhs) {
; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v7
; GFX6-NEXT: v_add_i32_e32 v5, vcc, v5, v9
; GFX6-NEXT: v_min_i32_e32 v6, -1, v3
-; GFX6-NEXT: v_add_i32_e32 v6, vcc, v6, v11
+; GFX6-NEXT: v_add_i32_e32 v6, vcc, 0x80000000, v6
; GFX6-NEXT: v_max_i32_e32 v4, v5, v4
; GFX6-NEXT: v_ashrrev_i32_e32 v2, 16, v2
; GFX6-NEXT: v_min_i32_e32 v4, v4, v6
@@ -3574,8 +3559,7 @@ define <3 x float> @v_ssubsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_lshlrev_b32_e32 v3, 16, v3
; GFX6-NEXT: v_add_i32_e32 v12, vcc, 0x80000001, v12
; GFX6-NEXT: v_min_i32_e32 v14, -1, v0
-; GFX6-NEXT: v_bfrev_b32_e32 v15, 1
-; GFX6-NEXT: v_add_i32_e32 v14, vcc, v14, v15
+; GFX6-NEXT: v_add_i32_e32 v14, vcc, 0x80000000, v14
; GFX6-NEXT: v_max_i32_e32 v3, v12, v3
; GFX6-NEXT: v_min_i32_e32 v3, v3, v14
; GFX6-NEXT: v_sub_i32_e32 v0, vcc, v0, v3
@@ -3585,7 +3569,7 @@ define <3 x float> @v_ssubsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_max_i32_e32 v9, -1, v3
; GFX6-NEXT: v_add_i32_e32 v9, vcc, v9, v13
; GFX6-NEXT: v_min_i32_e32 v12, -1, v3
-; GFX6-NEXT: v_add_i32_e32 v12, vcc, v12, v15
+; GFX6-NEXT: v_add_i32_e32 v12, vcc, 0x80000000, v12
; GFX6-NEXT: v_max_i32_e32 v6, v9, v6
; GFX6-NEXT: v_lshrrev_b32_e32 v7, 16, v1
; GFX6-NEXT: v_min_i32_e32 v6, v6, v12
@@ -3596,7 +3580,7 @@ define <3 x float> @v_ssubsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_lshlrev_b32_e32 v4, 16, v4
; GFX6-NEXT: v_add_i32_e32 v6, vcc, v6, v13
; GFX6-NEXT: v_min_i32_e32 v9, -1, v1
-; GFX6-NEXT: v_add_i32_e32 v9, vcc, v9, v15
+; GFX6-NEXT: v_add_i32_e32 v9, vcc, 0x80000000, v9
; GFX6-NEXT: v_max_i32_e32 v4, v6, v4
; GFX6-NEXT: v_min_i32_e32 v4, v4, v9
; GFX6-NEXT: v_sub_i32_e32 v1, vcc, v1, v4
@@ -3605,7 +3589,7 @@ define <3 x float> @v_ssubsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_lshlrev_b32_e32 v6, 16, v10
; GFX6-NEXT: v_add_i32_e32 v7, vcc, v7, v13
; GFX6-NEXT: v_min_i32_e32 v9, -1, v4
-; GFX6-NEXT: v_add_i32_e32 v9, vcc, v9, v15
+; GFX6-NEXT: v_add_i32_e32 v9, vcc, 0x80000000, v9
; GFX6-NEXT: v_max_i32_e32 v6, v7, v6
; GFX6-NEXT: v_lshrrev_b32_e32 v8, 16, v2
; GFX6-NEXT: v_min_i32_e32 v6, v6, v9
@@ -3616,7 +3600,7 @@ define <3 x float> @v_ssubsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_lshlrev_b32_e32 v5, 16, v5
; GFX6-NEXT: v_add_i32_e32 v6, vcc, v6, v13
; GFX6-NEXT: v_min_i32_e32 v7, -1, v2
-; GFX6-NEXT: v_add_i32_e32 v7, vcc, v7, v15
+; GFX6-NEXT: v_add_i32_e32 v7, vcc, 0x80000000, v7
; GFX6-NEXT: v_max_i32_e32 v5, v6, v5
; GFX6-NEXT: v_min_i32_e32 v5, v5, v7
; GFX6-NEXT: v_sub_i32_e32 v2, vcc, v2, v5
@@ -3627,7 +3611,7 @@ define <3 x float> @v_ssubsat_v6i16(<6 x i16> %lhs, <6 x i16> %rhs) {
; GFX6-NEXT: v_add_i32_e32 v7, vcc, v7, v13
; GFX6-NEXT: v_min_i32_e32 v8, -1, v5
; GFX6-NEXT: v_ashrrev_i32_e32 v0, 16, v0
-; GFX6-NEXT: v_add_i32_e32 v8, vcc, v8, v15
+; GFX6-NEXT: v_add_i32_e32 v8, vcc, 0x80000000, v8
; GFX6-NEXT: v_max_i32_e32 v6, v7, v6
; GFX6-NEXT: v_and_b32_e32 v3, 0xffff, v3
; GFX6-NEXT: v_ashrrev_i32_e32 v4, 16, v4
diff --git a/llvm/test/CodeGen/AMDGPU/fold-immediate-operand-shrink-dead-carry.mir b/llvm/test/CodeGen/AMDGPU/fold-immediate-operand-shrink-dead-carry.mir
new file mode 100644
index 00000000000000..02429c968d8d36
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/fold-immediate-operand-shrink-dead-carry.mir
@@ -0,0 +1,44 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=amdgpu7.00-amd-amdhsa -run-pass=si-fold-operands %s -o - | FileCheck -check-prefix=GCN %s
+
+# Make sure the dead flag on the vreg def is retained on the implicit
+# vcc def on the shrunk instruction.
+
+---
+name: shrink_dead_vreg_carry
+tracksRegLiveness: true
+
+body: |
+ bb.0:
+ ; GCN-LABEL: name: shrink_dead_vreg_carry
+ ; GCN: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 12345
+ ; GCN-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GCN-NEXT: [[V_ADD_CO_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_CO_U32_e32 [[S_MOV_B32_]], [[DEF]], implicit-def dead $vcc, implicit $exec
+ ; GCN-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GCN-NEXT: S_ENDPGM 0, implicit [[V_ADD_CO_U32_e32_]]
+ %0:sreg_32_xm0 = S_MOV_B32 12345
+ %1:vgpr_32 = IMPLICIT_DEF
+ %2:vgpr_32, dead %3:sreg_64 = V_ADD_CO_U32_e64 %0, %1, 0, implicit $exec
+ S_ENDPGM 0, implicit %2
+
+...
+
+---
+name: shrink_live_vreg_carry
+tracksRegLiveness: true
+
+body: |
+ bb.0:
+ ; GCN-LABEL: name: shrink_live_vreg_carry
+ ; GCN: [[S_MOV_B32_:%[0-9]+]]:sreg_32_xm0 = S_MOV_B32 12345
+ ; GCN-NEXT: [[DEF:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GCN-NEXT: [[V_ADD_CO_U32_e32_:%[0-9]+]]:vgpr_32 = V_ADD_CO_U32_e32 [[S_MOV_B32_]], [[DEF]], implicit-def $vcc, implicit $exec
+ ; GCN-NEXT: [[COPY:%[0-9]+]]:sreg_64 = COPY killed $vcc
+ ; GCN-NEXT: [[DEF1:%[0-9]+]]:vgpr_32 = IMPLICIT_DEF
+ ; GCN-NEXT: S_ENDPGM 0, implicit [[V_ADD_CO_U32_e32_]], implicit [[COPY]]
+ %0:sreg_32_xm0 = S_MOV_B32 12345
+ %1:vgpr_32 = IMPLICIT_DEF
+ %2:vgpr_32, %3:sreg_64 = V_ADD_CO_U32_e64 %0, %1, 0, implicit $exec
+ S_ENDPGM 0, implicit %2, implicit %3
+
+...
diff --git a/llvm/test/CodeGen/AMDGPU/shrink-v-cmp-wave32-dead-vcc-lo.mir b/llvm/test/CodeGen/AMDGPU/shrink-dead-vcc-def.mir
similarity index 73%
rename from llvm/test/CodeGen/AMDGPU/shrink-v-cmp-wave32-dead-vcc-lo.mir
rename to llvm/test/CodeGen/AMDGPU/shrink-dead-vcc-def.mir
index a13c1a2bf7fdeb..18ad708e3deffa 100644
--- a/llvm/test/CodeGen/AMDGPU/shrink-v-cmp-wave32-dead-vcc-lo.mir
+++ b/llvm/test/CodeGen/AMDGPU/shrink-dead-vcc-def.mir
@@ -53,3 +53,20 @@ body: |
S_SETPC_B64_return undef $sgpr30_sgpr31, implicit $vcc_lo
...
+
+# The carry-in is also vcc, so the shrunk form defines and uses it.
+---
+name: shrink_addc_vcc_lo_dead
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $vgpr0, $vgpr1, $vcc_lo
+ ; CHECK-LABEL: name: shrink_addc_vcc_lo_dead
+ ; CHECK: liveins: $vgpr0, $vgpr1, $vcc_lo
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: renamable $vgpr0 = V_ADDC_U32_e32 $vgpr0, $vgpr1, implicit-def dead $vcc_lo, implicit killed $vcc_lo, implicit $exec
+ ; CHECK-NEXT: S_ENDPGM 0, implicit killed $vgpr0
+ renamable $vgpr0, dead renamable $vcc_lo = V_ADDC_U32_e64 $vgpr0, $vgpr1, killed $vcc_lo, 0, implicit $exec
+ S_ENDPGM 0, implicit killed $vgpr0
+
+...
More information about the llvm-commits
mailing list