[llvm] [LV] Limits the splat operations be hoisted must not be defined by a recipe. (PR #117138)
Mel Chen via llvm-commits
llvm-commits at lists.llvm.org
Tue Feb 18 01:24:08 PST 2025
https://github.com/Mel-Chen updated https://github.com/llvm/llvm-project/pull/117138
>From 586e88334a0a3cc3025314c588aac4c691243b04 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Thu, 21 Nov 2024 02:05:59 -0800
Subject: [PATCH 1/7] [LV] Limits the splat operations be hoisted must not be
defined by a recipe.
---
llvm/lib/Transforms/Vectorize/VPlan.cpp | 2 +-
.../AArch64/extractvalue-no-scalarization-required.ll | 4 ++--
.../LoopVectorize/AArch64/sve-widen-extractvalue.ll | 2 +-
.../LoopVectorize/X86/epilog-vectorization-inductions.ll | 8 ++++----
4 files changed, 8 insertions(+), 8 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index 1332e50252978..a37154d08b4e8 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -264,7 +264,7 @@ Value *VPTransformState::get(VPValue *Def, bool NeedsScalar) {
return Data.VPV2Vector[Def];
auto GetBroadcastInstrs = [this, Def](Value *V) {
- bool SafeToHoist = Def->isDefinedOutsideLoopRegions();
+ bool SafeToHoist = !Def->hasDefiningRecipe();
if (VF.isScalar())
return V;
// Place the code for broadcasting invariant variables in the new preheader.
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/extractvalue-no-scalarization-required.ll b/llvm/test/Transforms/LoopVectorize/AArch64/extractvalue-no-scalarization-required.ll
index f969be80c3c00..ab21794baa924 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/extractvalue-no-scalarization-required.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/extractvalue-no-scalarization-required.ll
@@ -23,9 +23,9 @@
; Check that the extractvalue operands are actually free in vector code.
; FORCED: [[E1:%.+]] = extractvalue { i64, i64 } %sv, 0
-; FORCED-NEXT: [[E2:%.+]] = extractvalue { i64, i64 } %sv, 1
; FORCED-NEXT: %broadcast.splatinsert = insertelement <2 x i64> poison, i64 [[E1]], i64 0
; FORCED-NEXT: %broadcast.splat = shufflevector <2 x i64> %broadcast.splatinsert, <2 x i64> poison, <2 x i32> zeroinitializer
+; FORCED-NEXT: [[E2:%.+]] = extractvalue { i64, i64 } %sv, 1
; FORCED-NEXT: %broadcast.splatinsert1 = insertelement <2 x i64> poison, i64 [[E2]], i64 0
; FORCED-NEXT: %broadcast.splat2 = shufflevector <2 x i64> %broadcast.splatinsert1, <2 x i64> poison, <2 x i32> zeroinitializer
; FORCED-NEXT: [[ADD:%.+]] = add <2 x i64> %broadcast.splat, %broadcast.splat2
@@ -75,9 +75,9 @@ declare float @powf(float, float) readnone nounwind
; FORCED-LABEL: define void @test_getVectorCallCost
; FORCED: [[E1:%.+]] = extractvalue { float, float } %sv, 0
-; FORCED-NEXT: [[E2:%.+]] = extractvalue { float, float } %sv, 1
; FORCED-NEXT: %broadcast.splatinsert = insertelement <2 x float> poison, float [[E1]], i64 0
; FORCED-NEXT: %broadcast.splat = shufflevector <2 x float> %broadcast.splatinsert, <2 x float> poison, <2 x i32> zeroinitializer
+; FORCED-NEXT: [[E2:%.+]] = extractvalue { float, float } %sv, 1
; FORCED-NEXT: %broadcast.splatinsert1 = insertelement <2 x float> poison, float [[E2]], i64 0
; FORCED-NEXT: %broadcast.splat2 = shufflevector <2 x float> %broadcast.splatinsert1, <2 x float> poison, <2 x i32> zeroinitializer
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-extractvalue.ll b/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-extractvalue.ll
index 91dd6e475ec47..473fabfc9fecc 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-extractvalue.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-extractvalue.ll
@@ -5,9 +5,9 @@ target triple = "aarch64-unknown-linux-gnu"
define void @widen_extractvalue(ptr %dst, {i64, i64} %sv) #0 {
; CHECK-LABEL: @widen_extractvalue(
; CHECK: [[EXTRACT0:%.*]] = extractvalue { i64, i64 } [[SV:%.*]], 0
-; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { i64, i64 } [[SV]], 1
; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[EXTRACT0]], i64 0
; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[DOTSPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
+; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { i64, i64 } [[SV]], 1
; CHECK-NEXT: [[DOTSPLATINSERT1:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[EXTRACT1]], i64 0
; CHECK-NEXT: [[DOTSPLAT2:%.*]] = shufflevector <vscale x 2 x i64> [[DOTSPLATINSERT1]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; CHECK: [[ADD:%.*]] = add <vscale x 2 x i64> [[DOTSPLAT]], [[DOTSPLAT2]]
diff --git a/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll b/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll
index 7614be0bd3fce..22ebc691b8bd5 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll
@@ -137,6 +137,8 @@ define void @test_induction_step_needs_expansion(ptr noalias %j, ptr %k, i64 %l,
; CHECK-LABEL: @test_induction_step_needs_expansion(
; CHECK-NEXT: iter.check:
; CHECK-NEXT: [[TMP0:%.*]] = sub i16 0, [[OFF:%.*]]
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i16> poison, i16 [[TMP0]], i64 0
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[BROADCAST_SPLATINSERT]], <16 x i16> poison, <16 x i32> zeroinitializer
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[L:%.*]], 8
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
; CHECK: vector.main.loop.iter.check:
@@ -147,15 +149,13 @@ define void @test_induction_step_needs_expansion(ptr noalias %j, ptr %k, i64 %l,
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[L]], [[N_MOD_VF]]
; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i16
; CHECK-NEXT: [[IND_END:%.*]] = mul i16 [[DOTCAST]], [[TMP0]]
-; CHECK-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <16 x i16> poison, i16 [[TMP0]], i64 0
-; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[DOTSPLATINSERT2]], <16 x i16> poison, <16 x i32> zeroinitializer
; CHECK-NEXT: [[TMP1:%.*]] = mul <16 x i16> splat (i16 16), [[TMP2]]
; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <16 x i16> poison, i16 [[TMP0]], i64 0
; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <16 x i16> [[DOTSPLATINSERT]], <16 x i16> poison, <16 x i32> zeroinitializer
; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>, [[DOTSPLAT]]
; CHECK-NEXT: [[INDUCTION:%.*]] = add <16 x i16> zeroinitializer, [[TMP11]]
-; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i16> poison, i16 [[OFF]], i64 0
-; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <16 x i16> [[BROADCAST_SPLATINSERT]], <16 x i16> poison, <16 x i32> zeroinitializer
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <16 x i16> poison, i16 [[OFF]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <16 x i16> [[BROADCAST_SPLATINSERT2]], <16 x i16> poison, <16 x i32> zeroinitializer
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
>From c552c49b4265532a9671223b501e59c686265f76 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Tue, 26 Nov 2024 07:37:48 -0800
Subject: [PATCH 2/7] Use VPDom
---
llvm/lib/Transforms/Vectorize/VPlan.cpp | 8 +++++++-
llvm/lib/Transforms/Vectorize/VPlanDominatorTree.h | 2 ++
2 files changed, 9 insertions(+), 1 deletion(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index a37154d08b4e8..8833610e10566 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -19,6 +19,7 @@
#include "VPlan.h"
#include "LoopVectorizationPlanner.h"
#include "VPlanCFG.h"
+#include "VPlanDominatorTree.h"
#include "VPlanHelpers.h"
#include "VPlanPatternMatch.h"
#include "VPlanTransforms.h"
@@ -264,7 +265,12 @@ Value *VPTransformState::get(VPValue *Def, bool NeedsScalar) {
return Data.VPV2Vector[Def];
auto GetBroadcastInstrs = [this, Def](Value *V) {
- bool SafeToHoist = !Def->hasDefiningRecipe();
+ VPDominatorTree VPDT(*Plan);
+ bool SafeToHoist =
+ !Def->hasDefiningRecipe() ||
+ VPDT.properlyDominates(Def->getDefiningRecipe()->getParent(),
+ Plan->getVectorPreheader());
+
if (VF.isScalar())
return V;
// Place the code for broadcasting invariant variables in the new preheader.
diff --git a/llvm/lib/Transforms/Vectorize/VPlanDominatorTree.h b/llvm/lib/Transforms/Vectorize/VPlanDominatorTree.h
index 787cd940c4e26..995c6b8b2c2fb 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanDominatorTree.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanDominatorTree.h
@@ -39,8 +39,10 @@ class VPDominatorTree : public DominatorTreeBase<VPBlockBase, false> {
public:
VPDominatorTree() = default;
+ explicit VPDominatorTree(VPlan &Plan) { recalculate(Plan); }
/// Returns true if \p A properly dominates \p B.
+ using Base::properlyDominates;
bool properlyDominates(const VPRecipeBase *A, const VPRecipeBase *B);
};
>From 119300cc13472e6b839639f5b679cd708f0e5ad9 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Tue, 26 Nov 2024 11:45:33 -0800
Subject: [PATCH 3/7] Put VPDT as pointer into State
---
llvm/lib/Transforms/Vectorize/VPlan.cpp | 9 +++++----
llvm/lib/Transforms/Vectorize/VPlanHelpers.h | 4 ++++
2 files changed, 9 insertions(+), 4 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index 8833610e10566..e61b1827f4266 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -222,7 +222,9 @@ VPTransformState::VPTransformState(const TargetTransformInfo *TTI,
Loop *CurrentParentLoop, Type *CanonicalIVTy)
: TTI(TTI), VF(VF), CFG(DT), LI(LI), Builder(Builder), ILV(ILV), Plan(Plan),
CurrentParentLoop(CurrentParentLoop), LVer(nullptr),
- TypeAnalysis(CanonicalIVTy) {}
+ TypeAnalysis(CanonicalIVTy), VPDT(new VPDominatorTree(*Plan)) {}
+
+VPTransformState::~VPTransformState() { delete VPDT; }
Value *VPTransformState::get(VPValue *Def, const VPLane &Lane) {
if (Def->isLiveIn())
@@ -265,11 +267,10 @@ Value *VPTransformState::get(VPValue *Def, bool NeedsScalar) {
return Data.VPV2Vector[Def];
auto GetBroadcastInstrs = [this, Def](Value *V) {
- VPDominatorTree VPDT(*Plan);
bool SafeToHoist =
!Def->hasDefiningRecipe() ||
- VPDT.properlyDominates(Def->getDefiningRecipe()->getParent(),
- Plan->getVectorPreheader());
+ VPDT->properlyDominates(Def->getDefiningRecipe()->getParent(),
+ Plan->getVectorPreheader());
if (VF.isScalar())
return V;
diff --git a/llvm/lib/Transforms/Vectorize/VPlanHelpers.h b/llvm/lib/Transforms/Vectorize/VPlanHelpers.h
index cd1ad9bec91f7..13bf0ade042e1 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanHelpers.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanHelpers.h
@@ -38,6 +38,7 @@ class VPRegionBlock;
class VPlan;
class Value;
class LoopVersioning;
+class VPDominatorTree;
/// Returns a calculation for the total number of elements for a given \p VF.
/// For fixed width vectors this value is a constant, whereas for scalable
@@ -351,6 +352,9 @@ struct VPTransformState {
/// VPlan-based type analysis.
VPTypeAnalysis TypeAnalysis;
+
+ /// VPlan-based dominator tree.
+ VPDominatorTree *VPDT = nullptr;
};
/// Struct to hold various analysis needed for cost computations.
>From 1c463f45159a271a0730c9305c0936b458b17275 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Tue, 7 Jan 2025 04:31:56 -0800
Subject: [PATCH 4/7] Recalculate VPDT in VPlan::execute
---
llvm/lib/Transforms/Vectorize/VPlan.cpp | 4 ++++
.../LoopVectorize/X86/epilog-vectorization-inductions.ll | 4 ++--
2 files changed, 6 insertions(+), 2 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index e61b1827f4266..6daa3309a2ea8 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -994,6 +994,10 @@ void VPlan::execute(VPTransformState *State) {
State->CFG.PrevVPBB = nullptr;
State->CFG.ExitBB = State->CFG.PrevBB->getSingleSuccessor();
+ // Update VPDominatorTree since VPBasicBlock may be removed after State wsa
+ // constucted.
+ State->VPDT->recalculate(*this);
+
// Disconnect VectorPreHeader from ExitBB in both the CFG and DT.
BasicBlock *VectorPreHeader = State->CFG.PrevBB;
cast<BranchInst>(VectorPreHeader->getTerminator())->setSuccessor(0, nullptr);
diff --git a/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll b/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll
index 22ebc691b8bd5..623021b711910 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll
@@ -137,8 +137,6 @@ define void @test_induction_step_needs_expansion(ptr noalias %j, ptr %k, i64 %l,
; CHECK-LABEL: @test_induction_step_needs_expansion(
; CHECK-NEXT: iter.check:
; CHECK-NEXT: [[TMP0:%.*]] = sub i16 0, [[OFF:%.*]]
-; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i16> poison, i16 [[TMP0]], i64 0
-; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[BROADCAST_SPLATINSERT]], <16 x i16> poison, <16 x i32> zeroinitializer
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[L:%.*]], 8
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
; CHECK: vector.main.loop.iter.check:
@@ -149,6 +147,8 @@ define void @test_induction_step_needs_expansion(ptr noalias %j, ptr %k, i64 %l,
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[L]], [[N_MOD_VF]]
; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i16
; CHECK-NEXT: [[IND_END:%.*]] = mul i16 [[DOTCAST]], [[TMP0]]
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i16> poison, i16 [[TMP0]], i64 0
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[BROADCAST_SPLATINSERT]], <16 x i16> poison, <16 x i32> zeroinitializer
; CHECK-NEXT: [[TMP1:%.*]] = mul <16 x i16> splat (i16 16), [[TMP2]]
; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <16 x i16> poison, i16 [[TMP0]], i64 0
; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <16 x i16> [[DOTSPLATINSERT]], <16 x i16> poison, <16 x i32> zeroinitializer
>From 2e28c82fc73f8261ffb443a10d379ee8d474ef36 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Wed, 5 Feb 2025 05:35:44 -0800
Subject: [PATCH 5/7] Rebase
---
llvm/lib/Transforms/Vectorize/VPlanHelpers.h | 1 +
1 file changed, 1 insertion(+)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanHelpers.h b/llvm/lib/Transforms/Vectorize/VPlanHelpers.h
index 13bf0ade042e1..b55954d72fa7d 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanHelpers.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanHelpers.h
@@ -199,6 +199,7 @@ struct VPTransformState {
LoopInfo *LI, DominatorTree *DT, IRBuilderBase &Builder,
InnerLoopVectorizer *ILV, VPlan *Plan,
Loop *CurrentParentLoop, Type *CanonicalIVTy);
+ ~VPTransformState();
/// Target Transform Info.
const TargetTransformInfo *TTI;
>From bb6ca2e50478f13f602cc5b5caae8a93960f47a5 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Wed, 5 Feb 2025 05:53:39 -0800
Subject: [PATCH 6/7] Use VPDominatorTree instead of VPDominatorTree pointer
---
llvm/lib/Transforms/Vectorize/VPlan.cpp | 14 ++++++--------
llvm/lib/Transforms/Vectorize/VPlanHelpers.h | 5 ++---
2 files changed, 8 insertions(+), 11 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index 6daa3309a2ea8..8b0dfe7967f72 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -222,9 +222,7 @@ VPTransformState::VPTransformState(const TargetTransformInfo *TTI,
Loop *CurrentParentLoop, Type *CanonicalIVTy)
: TTI(TTI), VF(VF), CFG(DT), LI(LI), Builder(Builder), ILV(ILV), Plan(Plan),
CurrentParentLoop(CurrentParentLoop), LVer(nullptr),
- TypeAnalysis(CanonicalIVTy), VPDT(new VPDominatorTree(*Plan)) {}
-
-VPTransformState::~VPTransformState() { delete VPDT; }
+ TypeAnalysis(CanonicalIVTy), VPDT(*Plan) {}
Value *VPTransformState::get(VPValue *Def, const VPLane &Lane) {
if (Def->isLiveIn())
@@ -269,8 +267,8 @@ Value *VPTransformState::get(VPValue *Def, bool NeedsScalar) {
auto GetBroadcastInstrs = [this, Def](Value *V) {
bool SafeToHoist =
!Def->hasDefiningRecipe() ||
- VPDT->properlyDominates(Def->getDefiningRecipe()->getParent(),
- Plan->getVectorPreheader());
+ VPDT.properlyDominates(Def->getDefiningRecipe()->getParent(),
+ Plan->getVectorPreheader());
if (VF.isScalar())
return V;
@@ -994,9 +992,9 @@ void VPlan::execute(VPTransformState *State) {
State->CFG.PrevVPBB = nullptr;
State->CFG.ExitBB = State->CFG.PrevBB->getSingleSuccessor();
- // Update VPDominatorTree since VPBasicBlock may be removed after State wsa
- // constucted.
- State->VPDT->recalculate(*this);
+ // Update VPDominatorTree since VPBasicBlock may be removed after State was
+ // constructed.
+ State->VPDT.recalculate(*this);
// Disconnect VectorPreHeader from ExitBB in both the CFG and DT.
BasicBlock *VectorPreHeader = State->CFG.PrevBB;
diff --git a/llvm/lib/Transforms/Vectorize/VPlanHelpers.h b/llvm/lib/Transforms/Vectorize/VPlanHelpers.h
index b55954d72fa7d..1cfd1abeab3de 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanHelpers.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanHelpers.h
@@ -16,6 +16,7 @@
#define LLVM_TRANSFORMS_VECTORIZE_VPLANHELPERS_H
#include "VPlanAnalysis.h"
+#include "VPlanDominatorTree.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
@@ -38,7 +39,6 @@ class VPRegionBlock;
class VPlan;
class Value;
class LoopVersioning;
-class VPDominatorTree;
/// Returns a calculation for the total number of elements for a given \p VF.
/// For fixed width vectors this value is a constant, whereas for scalable
@@ -199,7 +199,6 @@ struct VPTransformState {
LoopInfo *LI, DominatorTree *DT, IRBuilderBase &Builder,
InnerLoopVectorizer *ILV, VPlan *Plan,
Loop *CurrentParentLoop, Type *CanonicalIVTy);
- ~VPTransformState();
/// Target Transform Info.
const TargetTransformInfo *TTI;
@@ -355,7 +354,7 @@ struct VPTransformState {
VPTypeAnalysis TypeAnalysis;
/// VPlan-based dominator tree.
- VPDominatorTree *VPDT = nullptr;
+ VPDominatorTree VPDT;
};
/// Struct to hold various analysis needed for cost computations.
>From 126c398009baa296b8cef623192c6f2b5b936ddc Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Mon, 17 Feb 2025 23:18:31 -0800
Subject: [PATCH 7/7] Restore the unchange test
---
.../LoopVectorize/X86/epilog-vectorization-inductions.ll | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll b/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll
index 623021b711910..7614be0bd3fce 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/epilog-vectorization-inductions.ll
@@ -147,15 +147,15 @@ define void @test_induction_step_needs_expansion(ptr noalias %j, ptr %k, i64 %l,
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[L]], [[N_MOD_VF]]
; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i16
; CHECK-NEXT: [[IND_END:%.*]] = mul i16 [[DOTCAST]], [[TMP0]]
-; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i16> poison, i16 [[TMP0]], i64 0
-; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[BROADCAST_SPLATINSERT]], <16 x i16> poison, <16 x i32> zeroinitializer
+; CHECK-NEXT: [[DOTSPLATINSERT2:%.*]] = insertelement <16 x i16> poison, i16 [[TMP0]], i64 0
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <16 x i16> [[DOTSPLATINSERT2]], <16 x i16> poison, <16 x i32> zeroinitializer
; CHECK-NEXT: [[TMP1:%.*]] = mul <16 x i16> splat (i16 16), [[TMP2]]
; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <16 x i16> poison, i16 [[TMP0]], i64 0
; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <16 x i16> [[DOTSPLATINSERT]], <16 x i16> poison, <16 x i32> zeroinitializer
; CHECK-NEXT: [[TMP11:%.*]] = mul <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>, [[DOTSPLAT]]
; CHECK-NEXT: [[INDUCTION:%.*]] = add <16 x i16> zeroinitializer, [[TMP11]]
-; CHECK-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <16 x i16> poison, i16 [[OFF]], i64 0
-; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <16 x i16> [[BROADCAST_SPLATINSERT2]], <16 x i16> poison, <16 x i32> zeroinitializer
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i16> poison, i16 [[OFF]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <16 x i16> [[BROADCAST_SPLATINSERT]], <16 x i16> poison, <16 x i32> zeroinitializer
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
More information about the llvm-commits
mailing list