[llvm] [VPlan] Explicitly handle scalar pointer inductions. (PR #83068)
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Mon Mar 11 06:57:38 PDT 2024
https://github.com/fhahn updated https://github.com/llvm/llvm-project/pull/83068
>From f4dabdfaa66744ecfca4c0a57472a357db9715d9 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Wed, 31 Jan 2024 14:02:38 +0000
Subject: [PATCH 01/13] [VPlan] Update VPInst::onlyFirstLaneUsed to check
users.
A VPInstruction only has its first lane used if all users use its first
lane only. Use vputils::onlyFirstLaneUsed to continue checking the
recipe's users to handle more cases.
Besides allowing additional introduction of scalar steps when
interleaving in some cases, this also enables using an Add VPInstruction
to model the increment.
---
llvm/lib/Transforms/Vectorize/VPlan.cpp | 4 +-
llvm/lib/Transforms/Vectorize/VPlan.h | 20 +--
.../lib/Transforms/Vectorize/VPlanRecipes.cpp | 18 ++
.../pr45679-fold-tail-by-masking.ll | 160 +++++++++---------
.../tail-folding-vectorization-factor-1.ll | 66 ++++----
5 files changed, 129 insertions(+), 139 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index a1bd6aaf0e5512..1ca2cfef447f63 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -1397,9 +1397,9 @@ void VPSlotTracker::assignSlots(const VPBasicBlock *VPBB) {
assignSlot(Def);
}
-bool vputils::onlyFirstLaneUsed(VPValue *Def) {
+bool vputils::onlyFirstLaneUsed(const VPValue *Def) {
return all_of(Def->users(),
- [Def](VPUser *U) { return U->onlyFirstLaneUsed(Def); });
+ [Def](const VPUser *U) { return U->onlyFirstLaneUsed(Def); });
}
bool vputils::onlyFirstPartUsed(VPValue *Def) {
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 20792cb9ac7c1c..30dc521947b3b9 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1256,23 +1256,7 @@ class VPInstruction : public VPRecipeWithIRFlags {
}
}
- /// Returns true if the recipe only uses the first lane of operand \p Op.
- bool onlyFirstLaneUsed(const VPValue *Op) const override {
- assert(is_contained(operands(), Op) &&
- "Op must be an operand of the recipe");
- if (getOperand(0) != Op)
- return false;
- switch (getOpcode()) {
- default:
- return false;
- case VPInstruction::ActiveLaneMask:
- case VPInstruction::CalculateTripCountMinusVF:
- case VPInstruction::CanonicalIVIncrementForPart:
- case VPInstruction::BranchOnCount:
- return true;
- };
- llvm_unreachable("switch should return");
- }
+ bool onlyFirstLaneUsed(const VPValue *Op) const override;
/// Returns true if the recipe only uses the first part of operand \p Op.
bool onlyFirstPartUsed(const VPValue *Op) const override {
@@ -3385,7 +3369,7 @@ class VPlanSlp {
namespace vputils {
/// Returns true if only the first lane of \p Def is used.
-bool onlyFirstLaneUsed(VPValue *Def);
+bool onlyFirstLaneUsed(const VPValue *Def);
/// Returns true if only the first part of \p Def is used.
bool onlyFirstPartUsed(VPValue *Def);
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index e51184b0dd1fe5..21b8d1eb77bf98 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -515,6 +515,24 @@ void VPInstruction::execute(VPTransformState &State) {
State.set(this, GeneratedValue, Part);
}
}
+bool VPInstruction::onlyFirstLaneUsed(const VPValue *Op) const {
+ assert(is_contained(operands(), Op) && "Op must be an operand of the recipe");
+ if (Instruction::isBinaryOp(getOpcode()))
+ return vputils::onlyFirstLaneUsed(this);
+
+ switch (getOpcode()) {
+ default:
+ return false;
+ case Instruction::ICmp:
+ return vputils::onlyFirstLaneUsed(this);
+ case VPInstruction::ActiveLaneMask:
+ case VPInstruction::CalculateTripCountMinusVF:
+ case VPInstruction::CanonicalIVIncrementForPart:
+ case VPInstruction::BranchOnCount:
+ return getOperand(0) == Op;
+ };
+ llvm_unreachable("switch should return");
+}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
void VPInstruction::dump() const {
diff --git a/llvm/test/Transforms/LoopVectorize/pr45679-fold-tail-by-masking.ll b/llvm/test/Transforms/LoopVectorize/pr45679-fold-tail-by-masking.ll
index e81fb66239bd4b..f05ec30619c5da 100644
--- a/llvm/test/Transforms/LoopVectorize/pr45679-fold-tail-by-masking.ll
+++ b/llvm/test/Transforms/LoopVectorize/pr45679-fold-tail-by-masking.ll
@@ -67,7 +67,7 @@ define void @pr45679(ptr %A) optsize {
; CHECK-NEXT: store i32 13, ptr [[ARRAYIDX]], align 1
; CHECK-NEXT: [[RIVPLUS1]] = add nuw nsw i32 [[RIV]], 1
; CHECK-NEXT: [[COND:%.*]] = icmp eq i32 [[RIVPLUS1]], 14
-; CHECK-NEXT: br i1 [[COND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP2:![0-9]+]]
+; CHECK-NEXT: br i1 [[COND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK: exit:
; CHECK-NEXT: ret void
;
@@ -129,7 +129,7 @@ define void @pr45679(ptr %A) optsize {
; VF2UF2-NEXT: store i32 13, ptr [[ARRAYIDX]], align 1
; VF2UF2-NEXT: [[RIVPLUS1]] = add nuw nsw i32 [[RIV]], 1
; VF2UF2-NEXT: [[COND:%.*]] = icmp eq i32 [[RIVPLUS1]], 14
-; VF2UF2-NEXT: br i1 [[COND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP2:![0-9]+]]
+; VF2UF2-NEXT: br i1 [[COND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
; VF2UF2: exit:
; VF2UF2-NEXT: ret void
;
@@ -139,46 +139,42 @@ define void @pr45679(ptr %A) optsize {
; VF1UF4: vector.ph:
; VF1UF4-NEXT: br label [[VECTOR_BODY:%.*]]
; VF1UF4: vector.body:
-; VF1UF4-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE12:%.*]] ]
-; VF1UF4-NEXT: [[VEC_IV:%.*]] = add i32 [[INDEX]], 0
-; VF1UF4-NEXT: [[VEC_IV4:%.*]] = add i32 [[INDEX]], 1
-; VF1UF4-NEXT: [[VEC_IV5:%.*]] = add i32 [[INDEX]], 2
-; VF1UF4-NEXT: [[VEC_IV6:%.*]] = add i32 [[INDEX]], 3
-; VF1UF4-NEXT: [[TMP0:%.*]] = icmp ule i32 [[VEC_IV]], 13
-; VF1UF4-NEXT: [[TMP1:%.*]] = icmp ule i32 [[VEC_IV4]], 13
-; VF1UF4-NEXT: [[TMP2:%.*]] = icmp ule i32 [[VEC_IV5]], 13
-; VF1UF4-NEXT: [[TMP3:%.*]] = icmp ule i32 [[VEC_IV6]], 13
-; VF1UF4-NEXT: br i1 [[TMP0]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
+; VF1UF4-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE6:%.*]] ]
+; VF1UF4-NEXT: [[TMP0:%.*]] = add i32 [[INDEX]], 0
+; VF1UF4-NEXT: [[TMP1:%.*]] = add i32 [[INDEX]], 1
+; VF1UF4-NEXT: [[TMP2:%.*]] = add i32 [[INDEX]], 2
+; VF1UF4-NEXT: [[TMP3:%.*]] = add i32 [[INDEX]], 3
+; VF1UF4-NEXT: [[TMP4:%.*]] = icmp ule i32 [[TMP0]], 13
+; VF1UF4-NEXT: [[TMP5:%.*]] = icmp ule i32 [[TMP1]], 13
+; VF1UF4-NEXT: [[TMP6:%.*]] = icmp ule i32 [[TMP2]], 13
+; VF1UF4-NEXT: [[TMP7:%.*]] = icmp ule i32 [[TMP3]], 13
+; VF1UF4-NEXT: br i1 [[TMP4]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; VF1UF4: pred.store.if:
-; VF1UF4-NEXT: [[INDUCTION:%.*]] = add i32 [[INDEX]], 0
-; VF1UF4-NEXT: [[TMP4:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 [[INDUCTION]]
-; VF1UF4-NEXT: store i32 13, ptr [[TMP4]], align 1
+; VF1UF4-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i32 [[TMP0]]
+; VF1UF4-NEXT: store i32 13, ptr [[TMP8]], align 1
; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE]]
; VF1UF4: pred.store.continue:
-; VF1UF4-NEXT: br i1 [[TMP1]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]]
-; VF1UF4: pred.store.if4:
-; VF1UF4-NEXT: [[INDUCTION1:%.*]] = add i32 [[INDEX]], 1
-; VF1UF4-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[INDUCTION1]]
-; VF1UF4-NEXT: store i32 13, ptr [[TMP5]], align 1
-; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE8]]
-; VF1UF4: pred.store.continue5:
-; VF1UF4-NEXT: br i1 [[TMP2]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10:%.*]]
-; VF1UF4: pred.store.if6:
-; VF1UF4-NEXT: [[INDUCTION2:%.*]] = add i32 [[INDEX]], 2
-; VF1UF4-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[INDUCTION2]]
-; VF1UF4-NEXT: store i32 13, ptr [[TMP6]], align 1
-; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE10]]
-; VF1UF4: pred.store.continue7:
-; VF1UF4-NEXT: br i1 [[TMP3]], label [[PRED_STORE_IF11:%.*]], label [[PRED_STORE_CONTINUE12]]
-; VF1UF4: pred.store.if8:
-; VF1UF4-NEXT: [[INDUCTION3:%.*]] = add i32 [[INDEX]], 3
-; VF1UF4-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[INDUCTION3]]
-; VF1UF4-NEXT: store i32 13, ptr [[TMP7]], align 1
-; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE12]]
-; VF1UF4: pred.store.continue9:
+; VF1UF4-NEXT: br i1 [[TMP5]], label [[PRED_STORE_IF1:%.*]], label [[PRED_STORE_CONTINUE2:%.*]]
+; VF1UF4: pred.store.if1:
+; VF1UF4-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[TMP1]]
+; VF1UF4-NEXT: store i32 13, ptr [[TMP9]], align 1
+; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE2]]
+; VF1UF4: pred.store.continue2:
+; VF1UF4-NEXT: br i1 [[TMP6]], label [[PRED_STORE_IF3:%.*]], label [[PRED_STORE_CONTINUE4:%.*]]
+; VF1UF4: pred.store.if3:
+; VF1UF4-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[TMP2]]
+; VF1UF4-NEXT: store i32 13, ptr [[TMP10]], align 1
+; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE4]]
+; VF1UF4: pred.store.continue4:
+; VF1UF4-NEXT: br i1 [[TMP7]], label [[PRED_STORE_IF5:%.*]], label [[PRED_STORE_CONTINUE6]]
+; VF1UF4: pred.store.if5:
+; VF1UF4-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, ptr [[A]], i32 [[TMP3]]
+; VF1UF4-NEXT: store i32 13, ptr [[TMP11]], align 1
+; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE6]]
+; VF1UF4: pred.store.continue6:
; VF1UF4-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 4
-; VF1UF4-NEXT: [[TMP8:%.*]] = icmp eq i32 [[INDEX_NEXT]], 16
-; VF1UF4-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; VF1UF4-NEXT: [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], 16
+; VF1UF4-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; VF1UF4: middle.block:
; VF1UF4-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
; VF1UF4: scalar.ph:
@@ -190,7 +186,7 @@ define void @pr45679(ptr %A) optsize {
; VF1UF4-NEXT: store i32 13, ptr [[ARRAYIDX]], align 1
; VF1UF4-NEXT: [[RIVPLUS1]] = add nuw nsw i32 [[RIV]], 1
; VF1UF4-NEXT: [[COND:%.*]] = icmp eq i32 [[RIVPLUS1]], 14
-; VF1UF4-NEXT: br i1 [[COND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP2:![0-9]+]]
+; VF1UF4-NEXT: br i1 [[COND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
; VF1UF4: exit:
; VF1UF4-NEXT: ret void
;
@@ -356,54 +352,50 @@ define void @load_variant(ptr noalias %a, ptr noalias %b) {
; VF1UF4: vector.ph:
; VF1UF4-NEXT: br label [[VECTOR_BODY:%.*]]
; VF1UF4: vector.body:
-; VF1UF4-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE12:%.*]] ]
-; VF1UF4-NEXT: [[VEC_IV:%.*]] = add i64 [[INDEX]], 0
-; VF1UF4-NEXT: [[VEC_IV4:%.*]] = add i64 [[INDEX]], 1
-; VF1UF4-NEXT: [[VEC_IV5:%.*]] = add i64 [[INDEX]], 2
-; VF1UF4-NEXT: [[VEC_IV6:%.*]] = add i64 [[INDEX]], 3
-; VF1UF4-NEXT: [[TMP0:%.*]] = icmp ule i64 [[VEC_IV]], 13
-; VF1UF4-NEXT: [[TMP1:%.*]] = icmp ule i64 [[VEC_IV4]], 13
-; VF1UF4-NEXT: [[TMP2:%.*]] = icmp ule i64 [[VEC_IV5]], 13
-; VF1UF4-NEXT: [[TMP3:%.*]] = icmp ule i64 [[VEC_IV6]], 13
-; VF1UF4-NEXT: br i1 [[TMP0]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
+; VF1UF4-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE6:%.*]] ]
+; VF1UF4-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; VF1UF4-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
+; VF1UF4-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2
+; VF1UF4-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3
+; VF1UF4-NEXT: [[TMP4:%.*]] = icmp ule i64 [[TMP0]], 13
+; VF1UF4-NEXT: [[TMP5:%.*]] = icmp ule i64 [[TMP1]], 13
+; VF1UF4-NEXT: [[TMP6:%.*]] = icmp ule i64 [[TMP2]], 13
+; VF1UF4-NEXT: [[TMP7:%.*]] = icmp ule i64 [[TMP3]], 13
+; VF1UF4-NEXT: br i1 [[TMP4]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; VF1UF4: pred.store.if:
-; VF1UF4-NEXT: [[INDUCTION:%.*]] = add i64 [[INDEX]], 0
-; VF1UF4-NEXT: [[TMP4:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDUCTION]]
-; VF1UF4-NEXT: [[TMP5:%.*]] = load i64, ptr [[TMP4]], align 8
-; VF1UF4-NEXT: store i64 [[TMP5]], ptr [[B:%.*]], align 8
+; VF1UF4-NEXT: [[TMP8:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP0]]
+; VF1UF4-NEXT: [[TMP9:%.*]] = load i64, ptr [[TMP8]], align 8
+; VF1UF4-NEXT: store i64 [[TMP9]], ptr [[B:%.*]], align 8
; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE]]
; VF1UF4: pred.store.continue:
-; VF1UF4-NEXT: [[TMP6:%.*]] = phi i64 [ poison, [[VECTOR_BODY]] ], [ [[TMP5]], [[PRED_STORE_IF]] ]
-; VF1UF4-NEXT: br i1 [[TMP1]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]]
-; VF1UF4: pred.store.if4:
-; VF1UF4-NEXT: [[INDUCTION1:%.*]] = add i64 [[INDEX]], 1
-; VF1UF4-NEXT: [[TMP7:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDUCTION1]]
-; VF1UF4-NEXT: [[TMP8:%.*]] = load i64, ptr [[TMP7]], align 8
-; VF1UF4-NEXT: store i64 [[TMP8]], ptr [[B]], align 8
-; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE8]]
-; VF1UF4: pred.store.continue5:
-; VF1UF4-NEXT: [[TMP9:%.*]] = phi i64 [ poison, [[PRED_STORE_CONTINUE]] ], [ [[TMP8]], [[PRED_STORE_IF7]] ]
-; VF1UF4-NEXT: br i1 [[TMP2]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10:%.*]]
-; VF1UF4: pred.store.if6:
-; VF1UF4-NEXT: [[INDUCTION2:%.*]] = add i64 [[INDEX]], 2
-; VF1UF4-NEXT: [[TMP10:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDUCTION2]]
-; VF1UF4-NEXT: [[TMP11:%.*]] = load i64, ptr [[TMP10]], align 8
-; VF1UF4-NEXT: store i64 [[TMP11]], ptr [[B]], align 8
-; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE10]]
-; VF1UF4: pred.store.continue7:
-; VF1UF4-NEXT: [[TMP12:%.*]] = phi i64 [ poison, [[PRED_STORE_CONTINUE8]] ], [ [[TMP11]], [[PRED_STORE_IF9]] ]
-; VF1UF4-NEXT: br i1 [[TMP3]], label [[PRED_STORE_IF11:%.*]], label [[PRED_STORE_CONTINUE12]]
-; VF1UF4: pred.store.if8:
-; VF1UF4-NEXT: [[INDUCTION3:%.*]] = add i64 [[INDEX]], 3
-; VF1UF4-NEXT: [[TMP13:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[INDUCTION3]]
-; VF1UF4-NEXT: [[TMP14:%.*]] = load i64, ptr [[TMP13]], align 8
-; VF1UF4-NEXT: store i64 [[TMP14]], ptr [[B]], align 8
-; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE12]]
-; VF1UF4: pred.store.continue9:
-; VF1UF4-NEXT: [[TMP15:%.*]] = phi i64 [ poison, [[PRED_STORE_CONTINUE10]] ], [ [[TMP14]], [[PRED_STORE_IF11]] ]
+; VF1UF4-NEXT: [[TMP10:%.*]] = phi i64 [ poison, [[VECTOR_BODY]] ], [ [[TMP9]], [[PRED_STORE_IF]] ]
+; VF1UF4-NEXT: br i1 [[TMP5]], label [[PRED_STORE_IF1:%.*]], label [[PRED_STORE_CONTINUE2:%.*]]
+; VF1UF4: pred.store.if1:
+; VF1UF4-NEXT: [[TMP11:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP1]]
+; VF1UF4-NEXT: [[TMP12:%.*]] = load i64, ptr [[TMP11]], align 8
+; VF1UF4-NEXT: store i64 [[TMP12]], ptr [[B]], align 8
+; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE2]]
+; VF1UF4: pred.store.continue2:
+; VF1UF4-NEXT: [[TMP13:%.*]] = phi i64 [ poison, [[PRED_STORE_CONTINUE]] ], [ [[TMP12]], [[PRED_STORE_IF1]] ]
+; VF1UF4-NEXT: br i1 [[TMP6]], label [[PRED_STORE_IF3:%.*]], label [[PRED_STORE_CONTINUE4:%.*]]
+; VF1UF4: pred.store.if3:
+; VF1UF4-NEXT: [[TMP14:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP2]]
+; VF1UF4-NEXT: [[TMP15:%.*]] = load i64, ptr [[TMP14]], align 8
+; VF1UF4-NEXT: store i64 [[TMP15]], ptr [[B]], align 8
+; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE4]]
+; VF1UF4: pred.store.continue4:
+; VF1UF4-NEXT: [[TMP16:%.*]] = phi i64 [ poison, [[PRED_STORE_CONTINUE2]] ], [ [[TMP15]], [[PRED_STORE_IF3]] ]
+; VF1UF4-NEXT: br i1 [[TMP7]], label [[PRED_STORE_IF5:%.*]], label [[PRED_STORE_CONTINUE6]]
+; VF1UF4: pred.store.if5:
+; VF1UF4-NEXT: [[TMP17:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP3]]
+; VF1UF4-NEXT: [[TMP18:%.*]] = load i64, ptr [[TMP17]], align 8
+; VF1UF4-NEXT: store i64 [[TMP18]], ptr [[B]], align 8
+; VF1UF4-NEXT: br label [[PRED_STORE_CONTINUE6]]
+; VF1UF4: pred.store.continue6:
+; VF1UF4-NEXT: [[TMP19:%.*]] = phi i64 [ poison, [[PRED_STORE_CONTINUE4]] ], [ [[TMP18]], [[PRED_STORE_IF5]] ]
; VF1UF4-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4
-; VF1UF4-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16
-; VF1UF4-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; VF1UF4-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16
+; VF1UF4-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; VF1UF4: middle.block:
; VF1UF4-NEXT: br i1 true, label [[FOR_END:%.*]], label [[SCALAR_PH]]
; VF1UF4: scalar.ph:
@@ -416,7 +408,7 @@ define void @load_variant(ptr noalias %a, ptr noalias %b) {
; VF1UF4-NEXT: store i64 [[V]], ptr [[B]], align 8
; VF1UF4-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
; VF1UF4-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 14
-; VF1UF4-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; VF1UF4-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
; VF1UF4: for.end:
; VF1UF4-NEXT: ret void
;
diff --git a/llvm/test/Transforms/LoopVectorize/tail-folding-vectorization-factor-1.ll b/llvm/test/Transforms/LoopVectorize/tail-folding-vectorization-factor-1.ll
index c07512644f721e..0c659a550b31e1 100644
--- a/llvm/test/Transforms/LoopVectorize/tail-folding-vectorization-factor-1.ll
+++ b/llvm/test/Transforms/LoopVectorize/tail-folding-vectorization-factor-1.ll
@@ -16,43 +16,39 @@ define void @VF1-VPlanExe(ptr %dst) {
; CHECK: vector.ph:
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE9:%.*]] ]
-; CHECK-NEXT: [[VEC_IV:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[VEC_IV1:%.*]] = add i64 [[INDEX]], 1
-; CHECK-NEXT: [[VEC_IV2:%.*]] = add i64 [[INDEX]], 2
-; CHECK-NEXT: [[VEC_IV3:%.*]] = add i64 [[INDEX]], 3
-; CHECK-NEXT: [[TMP0:%.*]] = icmp ule i64 [[VEC_IV]], 14
-; CHECK-NEXT: [[TMP1:%.*]] = icmp ule i64 [[VEC_IV1]], 14
-; CHECK-NEXT: [[TMP2:%.*]] = icmp ule i64 [[VEC_IV2]], 14
-; CHECK-NEXT: [[TMP3:%.*]] = icmp ule i64 [[VEC_IV3]], 14
-; CHECK-NEXT: br i1 [[TMP0]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE6:%.*]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2
+; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3
+; CHECK-NEXT: [[TMP4:%.*]] = icmp ule i64 [[TMP0]], 14
+; CHECK-NEXT: [[TMP5:%.*]] = icmp ule i64 [[TMP1]], 14
+; CHECK-NEXT: [[TMP6:%.*]] = icmp ule i64 [[TMP2]], 14
+; CHECK-NEXT: [[TMP7:%.*]] = icmp ule i64 [[TMP3]], 14
+; CHECK-NEXT: br i1 [[TMP4]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; CHECK: pred.store.if:
-; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i32, ptr [[DST:%.*]], i64 [[TMP4]]
-; CHECK-NEXT: store i32 0, ptr [[TMP5]], align 4
+; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[DST:%.*]], i64 [[TMP0]]
+; CHECK-NEXT: store i32 0, ptr [[TMP8]], align 4
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]]
; CHECK: pred.store.continue:
-; CHECK-NEXT: br i1 [[TMP1]], label [[PRED_STORE_IF4:%.*]], label [[PRED_STORE_CONTINUE5:%.*]]
-; CHECK: pred.store.if4:
-; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 1
-; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[DST]], i64 [[TMP6]]
-; CHECK-NEXT: store i32 0, ptr [[TMP7]], align 4
-; CHECK-NEXT: br label [[PRED_STORE_CONTINUE5]]
-; CHECK: pred.store.continue5:
-; CHECK-NEXT: br i1 [[TMP2]], label [[PRED_STORE_IF6:%.*]], label [[PRED_STORE_CONTINUE7:%.*]]
-; CHECK: pred.store.if6:
-; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[DST]], i64 [[TMP8]]
+; CHECK-NEXT: br i1 [[TMP5]], label [[PRED_STORE_IF1:%.*]], label [[PRED_STORE_CONTINUE2:%.*]]
+; CHECK: pred.store.if1:
+; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[DST]], i64 [[TMP1]]
; CHECK-NEXT: store i32 0, ptr [[TMP9]], align 4
-; CHECK-NEXT: br label [[PRED_STORE_CONTINUE7]]
-; CHECK: pred.store.continue7:
-; CHECK-NEXT: br i1 [[TMP3]], label [[PRED_STORE_IF8:%.*]], label [[PRED_STORE_CONTINUE9]]
-; CHECK: pred.store.if8:
-; CHECK-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], 3
-; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, ptr [[DST]], i64 [[TMP10]]
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE2]]
+; CHECK: pred.store.continue2:
+; CHECK-NEXT: br i1 [[TMP6]], label [[PRED_STORE_IF3:%.*]], label [[PRED_STORE_CONTINUE4:%.*]]
+; CHECK: pred.store.if3:
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i32, ptr [[DST]], i64 [[TMP2]]
+; CHECK-NEXT: store i32 0, ptr [[TMP10]], align 4
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE4]]
+; CHECK: pred.store.continue4:
+; CHECK-NEXT: br i1 [[TMP7]], label [[PRED_STORE_IF5:%.*]], label [[PRED_STORE_CONTINUE6]]
+; CHECK: pred.store.if5:
+; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, ptr [[DST]], i64 [[TMP3]]
; CHECK-NEXT: store i32 0, ptr [[TMP11]], align 4
-; CHECK-NEXT: br label [[PRED_STORE_CONTINUE9]]
-; CHECK: pred.store.continue9:
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE6]]
+; CHECK: pred.store.continue6:
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16
; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
@@ -69,7 +65,7 @@ define void @VF1-VPlanExe(ptr %dst) {
; CHECK-NEXT: store i32 0, ptr [[DST_PTR]], align 4
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 15
-; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP2:![0-9]+]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
;
entry:
br label %for.body
@@ -138,7 +134,7 @@ define void @VF1-VPWidenCanonicalIVRecipeExe(ptr %ptr1) {
; CHECK: pred.store.continue12:
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16
-; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
@@ -151,7 +147,7 @@ define void @VF1-VPWidenCanonicalIVRecipeExe(ptr %ptr1) {
; CHECK-NEXT: store double 0.000000e+00, ptr [[ADDR]], align 8
; CHECK-NEXT: [[PTR]] = getelementptr inbounds double, ptr [[ADDR]], i64 1
; CHECK-NEXT: [[COND:%.*]] = icmp eq ptr [[PTR]], [[PTR2]]
-; CHECK-NEXT: br i1 [[COND]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: br i1 [[COND]], label [[FOR_COND_CLEANUP]], label [[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
;
entry:
%ptr2 = getelementptr inbounds double, ptr %ptr1, i64 15
>From b08e8920ea23f2b4ce3bfe85af8878c81dfce28f Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Wed, 31 Jan 2024 15:08:29 +0000
Subject: [PATCH 02/13] [VPlan] Consistently use (Part, 0) for first lane
scalar values
At the moment, some VPInstructions create only a single scalar value,
but use VPTransformatState's 'vector' storage for this value. Those
values are effectively uniform-per-VF (or in some cases
uniform-across-VF-and-UF). Using the vector/per-part storage doesn't
interact well with other recipes, that more accurately using (Part,
Lane) to look up scalar values and prevents VPInstructions creating
scalars from interacting with other recipes working with scalars.
This PR tries to unify handling of scalars by using (Part, 0) for scalar
values where only the first lane is demanded. This allows using
VPInstructions with other recipes like VPScalarCastRecipe and is also
needed when using VPInstructions in more cases otuside the vector loop
region to generate scalars.
The patch is still a bit rough around the edges, but hopefully serves as
start for a discussion how to model more scalar recipes. A potential
alternative would be to split off the opcodes that generate scalars only
to a separate recipe.
---
llvm/lib/Transforms/Vectorize/VPlan.cpp | 18 ++++--
llvm/lib/Transforms/Vectorize/VPlan.h | 7 +++
.../lib/Transforms/Vectorize/VPlanRecipes.cpp | 36 ++++++++---
.../LoopVectorize/AArch64/masked-call.ll | 25 +++-----
.../LoopVectorize/first-order-recurrence.ll | 62 +++++++++----------
5 files changed, 87 insertions(+), 61 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index 1ca2cfef447f63..7ed07fe5f413a8 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -213,8 +213,13 @@ VPBasicBlock::iterator VPBasicBlock::getFirstNonPhi() {
}
Value *VPTransformState::get(VPValue *Def, const VPIteration &Instance) {
- if (Def->isLiveIn())
- return Def->getLiveInIRValue();
+ if (Def->isLiveIn()) {
+ if (Value *V = Def->getLiveInIRValue())
+ return V;
+ if (hasScalarValue(Def, VPIteration(0, 0))) {
+ return Data.PerPartScalars[Def][0][0];
+ }
+ }
if (hasScalarValue(Def, Instance)) {
return Data
@@ -794,7 +799,7 @@ void VPlan::prepareToExecute(Value *TripCountV, Value *VectorTripCountV,
// FIXME: Model VF * UF computation completely in VPlan.
State.set(&VFxUF,
createStepForVF(Builder, TripCountV->getType(), State.VF, State.UF),
- 0);
+ VPIteration(0, 0));
// When vectorizing the epilogue loop, the canonical induction start value
// needs to be changed from zero to the value after the main vector loop.
@@ -883,8 +888,11 @@ void VPlan::execute(VPTransformState *State) {
for (unsigned Part = 0; Part < LastPartForNewPhi; ++Part) {
Value *Phi = State->get(PhiR, Part);
- Value *Val = State->get(PhiR->getBackedgeValue(),
- SinglePartNeeded ? State->UF - 1 : Part);
+ Value *Val =
+ isa<VPCanonicalIVPHIRecipe>(PhiR)
+ ? State->get(PhiR->getBackedgeValue(), VPIteration(Part, 0))
+ : State->get(PhiR->getBackedgeValue(),
+ SinglePartNeeded ? State->UF - 1 : Part);
cast<PHINode>(Phi)->addIncoming(Val, VectorLatchBB);
}
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 30dc521947b3b9..97035146a2f4dc 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1388,6 +1388,13 @@ class VPScalarCastRecipe : public VPSingleDefRecipe {
/// Returns the result type of the cast.
Type *getResultType() const { return ResultTy; }
+
+ bool onlyFirstLaneUsed(const VPValue *Op) const override {
+ // At the moment, only scalar codegen is implemented.
+ assert(is_contained(operands(), Op) &&
+ "Op must be an operand of the recipe");
+ return true;
+ }
};
/// A recipe for widening Call instructions.
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 21b8d1eb77bf98..77f2cf899b0855 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -279,11 +279,17 @@ Value *VPInstruction::generateInstruction(VPTransformState &State,
Builder.SetCurrentDebugLocation(getDebugLoc());
if (Instruction::isBinaryOp(getOpcode())) {
+ bool OnlyFirstLaneUsed = vputils::onlyFirstLaneUsed(this);
if (Part != 0 && vputils::onlyFirstPartUsed(this))
- return State.get(this, 0);
-
- Value *A = State.get(getOperand(0), Part);
- Value *B = State.get(getOperand(1), Part);
+ return OnlyFirstLaneUsed ? State.get(this, VPIteration(0, 0))
+ : State.get(this, 0);
+
+ Value *A = OnlyFirstLaneUsed
+ ? State.get(getOperand(0), VPIteration(Part, 0))
+ : State.get(getOperand(0), Part);
+ Value *B = OnlyFirstLaneUsed
+ ? State.get(getOperand(1), VPIteration(Part, 0))
+ : State.get(getOperand(1), Part);
auto *Res =
Builder.CreateBinOp((Instruction::BinaryOps)getOpcode(), A, B, Name);
if (auto *I = dyn_cast<Instruction>(Res))
@@ -385,8 +391,8 @@ Value *VPInstruction::generateInstruction(VPTransformState &State,
if (Part != 0)
return nullptr;
// First create the compare.
- Value *IV = State.get(getOperand(0), Part);
- Value *TC = State.get(getOperand(1), Part);
+ Value *IV = State.get(getOperand(0), VPIteration(0, 0));
+ Value *TC = State.get(getOperand(1), VPIteration(0, 0));
Value *Cond = Builder.CreateICmpEQ(IV, TC);
// Now create the branch.
@@ -407,7 +413,7 @@ Value *VPInstruction::generateInstruction(VPTransformState &State,
}
case VPInstruction::ComputeReductionResult: {
if (Part != 0)
- return State.get(this, 0);
+ return State.get(this, VPIteration(0, 0));
// FIXME: The cross-recipe dependency on VPReductionPHIRecipe is temporary
// and will be removed by breaking up the recipe further.
@@ -512,7 +518,17 @@ void VPInstruction::execute(VPTransformState &State) {
if (!hasResult())
continue;
assert(GeneratedValue && "generateInstruction must produce a value");
- State.set(this, GeneratedValue, Part);
+ if (GeneratedValue->getType()->isVectorTy())
+ State.set(this, GeneratedValue, Part);
+ else {
+ if (getOpcode() == VPInstruction::ComputeReductionResult) {
+ State.set(this, GeneratedValue, VPIteration(Part, 0));
+ } else {
+ assert((State.VF.isScalar() || vputils::onlyFirstLaneUsed(this)) &&
+ "scalar value but not only first lane used");
+ State.set(this, GeneratedValue, VPIteration(Part, 0));
+ }
+ }
}
}
bool VPInstruction::onlyFirstLaneUsed(const VPValue *Op) const {
@@ -525,11 +541,13 @@ bool VPInstruction::onlyFirstLaneUsed(const VPValue *Op) const {
return false;
case Instruction::ICmp:
return vputils::onlyFirstLaneUsed(this);
+ case VPInstruction::ComputeReductionResult:
+ return true;
case VPInstruction::ActiveLaneMask:
case VPInstruction::CalculateTripCountMinusVF:
case VPInstruction::CanonicalIVIncrementForPart:
case VPInstruction::BranchOnCount:
- return getOperand(0) == Op;
+ return true;
};
llvm_unreachable("switch should return");
}
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll b/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll
index 1e79c3e1e8dc2d..2ef55742ffc0b1 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll
@@ -99,7 +99,7 @@ define void @test_widen(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK_ENTRY1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[TMP8]], i64 1025)
; TFA_INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]]
; TFA_INTERLEAVE: vector.body:
-; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT4:%.*]], [[VECTOR_BODY]] ]
+; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT5:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr i64, ptr [[B:%.*]], i64 [[INDEX]]
@@ -116,8 +116,7 @@ define void @test_widen(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[TMP18:%.*]] = getelementptr inbounds i64, ptr [[TMP15]], i64 [[TMP17]]
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[TMP13]], ptr [[TMP15]], i32 8, <vscale x 2 x i1> [[ACTIVE_LANE_MASK]])
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[TMP14]], ptr [[TMP18]], i32 8, <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]])
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT:%.*]] = add i64 [[INDEX]], [[TMP6]]
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT4]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP6]]
; TFA_INTERLEAVE-NEXT: [[TMP19:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP20:%.*]] = mul i64 [[TMP19]], 2
; TFA_INTERLEAVE-NEXT: [[TMP21:%.*]] = add i64 [[INDEX_NEXT]], [[TMP20]]
@@ -254,7 +253,7 @@ define void @test_if_then(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK_ENTRY1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[TMP8]], i64 1025)
; TFA_INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]]
; TFA_INTERLEAVE: vector.body:
-; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT5:%.*]], [[VECTOR_BODY]] ]
+; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT6:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDEX]]
@@ -283,8 +282,7 @@ define void @test_if_then(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[TMP28:%.*]] = getelementptr inbounds i64, ptr [[TMP25]], i64 [[TMP27]]
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[PREDPHI]], ptr [[TMP25]], i32 8, <vscale x 2 x i1> [[TMP23]])
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[PREDPHI4]], ptr [[TMP28]], i32 8, <vscale x 2 x i1> [[TMP24]])
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT:%.*]] = add i64 [[INDEX]], [[TMP6]]
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT5]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP6]]
; TFA_INTERLEAVE-NEXT: [[TMP29:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP30:%.*]] = mul i64 [[TMP29]], 2
; TFA_INTERLEAVE-NEXT: [[TMP31:%.*]] = add i64 [[INDEX_NEXT]], [[TMP30]]
@@ -437,7 +435,7 @@ define void @test_widen_if_then_else(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK_ENTRY1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[TMP8]], i64 1025)
; TFA_INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]]
; TFA_INTERLEAVE: vector.body:
-; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT5:%.*]], [[VECTOR_BODY]] ]
+; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT6:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDEX]]
@@ -468,8 +466,7 @@ define void @test_widen_if_then_else(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[TMP30:%.*]] = getelementptr inbounds i64, ptr [[TMP27]], i64 [[TMP29]]
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[PREDPHI]], ptr [[TMP27]], i32 8, <vscale x 2 x i1> [[TMP25]])
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[PREDPHI4]], ptr [[TMP30]], i32 8, <vscale x 2 x i1> [[TMP26]])
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT:%.*]] = add i64 [[INDEX]], [[TMP6]]
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT5]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP6]]
; TFA_INTERLEAVE-NEXT: [[TMP31:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP32:%.*]] = mul i64 [[TMP31]], 2
; TFA_INTERLEAVE-NEXT: [[TMP33:%.*]] = add i64 [[INDEX_NEXT]], [[TMP32]]
@@ -771,7 +768,7 @@ define void @test_widen_optmask(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK_ENTRY1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[TMP8]], i64 1025)
; TFA_INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]]
; TFA_INTERLEAVE: vector.body:
-; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT4:%.*]], [[VECTOR_BODY]] ]
+; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT5:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr i64, ptr [[B:%.*]], i64 [[INDEX]]
@@ -788,8 +785,7 @@ define void @test_widen_optmask(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[TMP18:%.*]] = getelementptr inbounds i64, ptr [[TMP15]], i64 [[TMP17]]
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[TMP13]], ptr [[TMP15]], i32 8, <vscale x 2 x i1> [[ACTIVE_LANE_MASK]])
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[TMP14]], ptr [[TMP18]], i32 8, <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]])
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT:%.*]] = add i64 [[INDEX]], [[TMP6]]
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT4]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP6]]
; TFA_INTERLEAVE-NEXT: [[TMP19:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP20:%.*]] = mul i64 [[TMP19]], 2
; TFA_INTERLEAVE-NEXT: [[TMP21:%.*]] = add i64 [[INDEX_NEXT]], [[TMP20]]
@@ -970,7 +966,7 @@ define double @test_widen_fmuladd_and_call(ptr noalias %a, ptr readnone %b, doub
; TFA_INTERLEAVE-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x double> [[BROADCAST_SPLATINSERT]], <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer
; TFA_INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]]
; TFA_INTERLEAVE: vector.body:
-; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT4:%.*]], [[VECTOR_BODY]] ]
+; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT5:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[VEC_PHI:%.*]] = phi double [ 0.000000e+00, [[ENTRY]] ], [ [[TMP26:%.*]], [[VECTOR_BODY]] ]
@@ -996,8 +992,7 @@ define double @test_widen_fmuladd_and_call(ptr noalias %a, ptr readnone %b, doub
; TFA_INTERLEAVE-NEXT: [[TMP24:%.*]] = call double @llvm.vector.reduce.fadd.nxv2f64(double [[VEC_PHI]], <vscale x 2 x double> [[TMP23]])
; TFA_INTERLEAVE-NEXT: [[TMP25:%.*]] = select <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]], <vscale x 2 x double> [[TMP14]], <vscale x 2 x double> shufflevector (<vscale x 2 x double> insertelement (<vscale x 2 x double> poison, double -0.000000e+00, i64 0), <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer)
; TFA_INTERLEAVE-NEXT: [[TMP26]] = call double @llvm.vector.reduce.fadd.nxv2f64(double [[TMP24]], <vscale x 2 x double> [[TMP25]])
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT:%.*]] = add i64 [[INDEX]], [[TMP6]]
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT4]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP6]]
; TFA_INTERLEAVE-NEXT: [[TMP27:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP28:%.*]] = mul i64 [[TMP27]], 2
; TFA_INTERLEAVE-NEXT: [[TMP29:%.*]] = add i64 [[INDEX_NEXT]], [[TMP28]]
diff --git a/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll b/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll
index b451d4b4e5462d..f0a5fc1592532d 100644
--- a/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll
+++ b/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll
@@ -3271,47 +3271,45 @@ define i32 @sink_into_replication_region_multiple(ptr %x, i32 %y) {
; UNROLL-NO-VF-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i32 [[TMP1]], 1
; UNROLL-NO-VF-NEXT: br label [[VECTOR_BODY:%.*]]
; UNROLL-NO-VF: vector.body:
-; UNROLL-NO-VF-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE7:%.*]] ]
-; UNROLL-NO-VF-NEXT: [[VECTOR_RECUR:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[PRED_STORE_CONTINUE7]] ]
-; UNROLL-NO-VF-NEXT: [[VEC_PHI:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP10:%.*]], [[PRED_STORE_CONTINUE7]] ]
-; UNROLL-NO-VF-NEXT: [[VEC_PHI2:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP11:%.*]], [[PRED_STORE_CONTINUE7]] ]
+; UNROLL-NO-VF-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE6:%.*]] ]
+; UNROLL-NO-VF-NEXT: [[VECTOR_RECUR:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP11:%.*]], [[PRED_STORE_CONTINUE6]] ]
+; UNROLL-NO-VF-NEXT: [[VEC_PHI:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP12:%.*]], [[PRED_STORE_CONTINUE6]] ]
+; UNROLL-NO-VF-NEXT: [[VEC_PHI2:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP13:%.*]], [[PRED_STORE_CONTINUE6]] ]
; UNROLL-NO-VF-NEXT: [[OFFSET_IDX:%.*]] = sub i32 [[Y]], [[INDEX]]
; UNROLL-NO-VF-NEXT: [[TMP2:%.*]] = add i32 [[OFFSET_IDX]], 0
; UNROLL-NO-VF-NEXT: [[TMP3:%.*]] = add i32 [[OFFSET_IDX]], -1
-; UNROLL-NO-VF-NEXT: [[VEC_IV:%.*]] = add i32 [[INDEX]], 0
-; UNROLL-NO-VF-NEXT: [[VEC_IV3:%.*]] = add i32 [[INDEX]], 1
-; UNROLL-NO-VF-NEXT: [[TMP4:%.*]] = icmp ule i32 [[VEC_IV]], [[TRIP_COUNT_MINUS_1]]
-; UNROLL-NO-VF-NEXT: [[TMP5:%.*]] = icmp ule i32 [[VEC_IV3]], [[TRIP_COUNT_MINUS_1]]
-; UNROLL-NO-VF-NEXT: br i1 [[TMP4]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
+; UNROLL-NO-VF-NEXT: [[TMP4:%.*]] = add i32 [[INDEX]], 0
+; UNROLL-NO-VF-NEXT: [[TMP5:%.*]] = add i32 [[INDEX]], 1
+; UNROLL-NO-VF-NEXT: [[TMP6:%.*]] = icmp ule i32 [[TMP4]], [[TRIP_COUNT_MINUS_1]]
+; UNROLL-NO-VF-NEXT: [[TMP7:%.*]] = icmp ule i32 [[TMP5]], [[TRIP_COUNT_MINUS_1]]
+; UNROLL-NO-VF-NEXT: br i1 [[TMP6]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
; UNROLL-NO-VF: pred.udiv.if:
-; UNROLL-NO-VF-NEXT: [[TMP6:%.*]] = udiv i32 219220132, [[TMP2]]
+; UNROLL-NO-VF-NEXT: [[TMP8:%.*]] = udiv i32 219220132, [[TMP2]]
; UNROLL-NO-VF-NEXT: br label [[PRED_UDIV_CONTINUE]]
; UNROLL-NO-VF: pred.udiv.continue:
-; UNROLL-NO-VF-NEXT: [[TMP7:%.*]] = phi i32 [ poison, [[VECTOR_BODY]] ], [ [[TMP6]], [[PRED_UDIV_IF]] ]
-; UNROLL-NO-VF-NEXT: br i1 [[TMP5]], label [[PRED_UDIV_IF4:%.*]], label [[PRED_UDIV_CONTINUE5:%.*]]
-; UNROLL-NO-VF: pred.udiv.if4:
-; UNROLL-NO-VF-NEXT: [[TMP8:%.*]] = udiv i32 219220132, [[TMP3]]
-; UNROLL-NO-VF-NEXT: br label [[PRED_UDIV_CONTINUE5]]
-; UNROLL-NO-VF: pred.udiv.continue5:
-; UNROLL-NO-VF-NEXT: [[TMP9]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE]] ], [ [[TMP8]], [[PRED_UDIV_IF4]] ]
-; UNROLL-NO-VF-NEXT: [[TMP10]] = add i32 [[VEC_PHI]], [[VECTOR_RECUR]]
-; UNROLL-NO-VF-NEXT: [[TMP11]] = add i32 [[VEC_PHI2]], [[TMP7]]
-; UNROLL-NO-VF-NEXT: br i1 [[TMP4]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
+; UNROLL-NO-VF-NEXT: [[TMP9:%.*]] = phi i32 [ poison, [[VECTOR_BODY]] ], [ [[TMP8]], [[PRED_UDIV_IF]] ]
+; UNROLL-NO-VF-NEXT: br i1 [[TMP7]], label [[PRED_UDIV_IF3:%.*]], label [[PRED_UDIV_CONTINUE4:%.*]]
+; UNROLL-NO-VF: pred.udiv.if3:
+; UNROLL-NO-VF-NEXT: [[TMP10:%.*]] = udiv i32 219220132, [[TMP3]]
+; UNROLL-NO-VF-NEXT: br label [[PRED_UDIV_CONTINUE4]]
+; UNROLL-NO-VF: pred.udiv.continue4:
+; UNROLL-NO-VF-NEXT: [[TMP11]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE]] ], [ [[TMP10]], [[PRED_UDIV_IF3]] ]
+; UNROLL-NO-VF-NEXT: [[TMP12]] = add i32 [[VEC_PHI]], [[VECTOR_RECUR]]
+; UNROLL-NO-VF-NEXT: [[TMP13]] = add i32 [[VEC_PHI2]], [[TMP9]]
+; UNROLL-NO-VF-NEXT: br i1 [[TMP6]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; UNROLL-NO-VF: pred.store.if:
-; UNROLL-NO-VF-NEXT: [[TMP12:%.*]] = add i32 [[INDEX]], 0
-; UNROLL-NO-VF-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[X:%.*]], i32 [[TMP12]]
-; UNROLL-NO-VF-NEXT: store i32 [[TMP2]], ptr [[TMP13]], align 4
+; UNROLL-NO-VF-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[X:%.*]], i32 [[TMP4]]
+; UNROLL-NO-VF-NEXT: store i32 [[TMP2]], ptr [[TMP14]], align 4
; UNROLL-NO-VF-NEXT: br label [[PRED_STORE_CONTINUE]]
; UNROLL-NO-VF: pred.store.continue:
-; UNROLL-NO-VF-NEXT: br i1 [[TMP5]], label [[PRED_STORE_IF6:%.*]], label [[PRED_STORE_CONTINUE7]]
-; UNROLL-NO-VF: pred.store.if6:
-; UNROLL-NO-VF-NEXT: [[TMP14:%.*]] = add i32 [[INDEX]], 1
-; UNROLL-NO-VF-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[X]], i32 [[TMP14]]
+; UNROLL-NO-VF-NEXT: br i1 [[TMP7]], label [[PRED_STORE_IF5:%.*]], label [[PRED_STORE_CONTINUE6]]
+; UNROLL-NO-VF: pred.store.if5:
+; UNROLL-NO-VF-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[X]], i32 [[TMP5]]
; UNROLL-NO-VF-NEXT: store i32 [[TMP3]], ptr [[TMP15]], align 4
-; UNROLL-NO-VF-NEXT: br label [[PRED_STORE_CONTINUE7]]
-; UNROLL-NO-VF: pred.store.continue7:
-; UNROLL-NO-VF-NEXT: [[TMP16:%.*]] = select i1 [[TMP4]], i32 [[TMP10]], i32 [[VEC_PHI]]
-; UNROLL-NO-VF-NEXT: [[TMP17:%.*]] = select i1 [[TMP5]], i32 [[TMP11]], i32 [[VEC_PHI2]]
+; UNROLL-NO-VF-NEXT: br label [[PRED_STORE_CONTINUE6]]
+; UNROLL-NO-VF: pred.store.continue6:
+; UNROLL-NO-VF-NEXT: [[TMP16:%.*]] = select i1 [[TMP6]], i32 [[TMP12]], i32 [[VEC_PHI]]
+; UNROLL-NO-VF-NEXT: [[TMP17:%.*]] = select i1 [[TMP7]], i32 [[TMP13]], i32 [[VEC_PHI2]]
; UNROLL-NO-VF-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 2
; UNROLL-NO-VF-NEXT: [[TMP18:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
; UNROLL-NO-VF-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !prof [[PROF27]], !llvm.loop [[LOOP31:![0-9]+]]
@@ -3319,7 +3317,7 @@ define i32 @sink_into_replication_region_multiple(ptr %x, i32 %y) {
; UNROLL-NO-VF-NEXT: [[BIN_RDX:%.*]] = add i32 [[TMP17]], [[TMP16]]
; UNROLL-NO-VF-NEXT: br i1 true, label [[BB1:%.*]], label [[SCALAR_PH]]
; UNROLL-NO-VF: scalar.ph:
-; UNROLL-NO-VF-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[BB:%.*]] ], [ [[TMP9]], [[MIDDLE_BLOCK]] ]
+; UNROLL-NO-VF-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[BB:%.*]] ], [ [[TMP11]], [[MIDDLE_BLOCK]] ]
; UNROLL-NO-VF-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[Y]], [[BB]] ]
; UNROLL-NO-VF-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[BB]] ]
; UNROLL-NO-VF-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[BB]] ], [ [[BIN_RDX]], [[MIDDLE_BLOCK]] ]
>From 172dbf6846b57c3ba963d6e8b67115e5191b19e0 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Sat, 3 Feb 2024 16:23:32 +0000
Subject: [PATCH 03/13] !fixup fix merge
---
.../lib/Transforms/Vectorize/VPlanRecipes.cpp | 3 +-
.../LoopVectorize/first-order-recurrence.ll | 62 ++++++++++---------
2 files changed, 33 insertions(+), 32 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 2829bb8bb95d8b..b1a0ebb5863f7e 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -544,7 +544,6 @@ bool VPInstruction::onlyFirstLaneUsed(const VPValue *Op) const {
return vputils::onlyFirstLaneUsed(this);
case VPInstruction::ComputeReductionResult:
return true;
-
case VPInstruction::ActiveLaneMask:
case VPInstruction::CalculateTripCountMinusVF:
case VPInstruction::CanonicalIVIncrementForPart:
@@ -562,7 +561,7 @@ void VPInstruction::dump() const {
void VPInstruction::print(raw_ostream &O, const Twine &Indent,
VPSlotTracker &SlotTracker) const {
- O F Indent << "EMIT ";
+ O << Indent << "EMIT ";
if (hasResult()) {
printAsOperand(O, SlotTracker);
diff --git a/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll b/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll
index f0a5fc1592532d..b451d4b4e5462d 100644
--- a/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll
+++ b/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll
@@ -3271,45 +3271,47 @@ define i32 @sink_into_replication_region_multiple(ptr %x, i32 %y) {
; UNROLL-NO-VF-NEXT: [[TRIP_COUNT_MINUS_1:%.*]] = sub i32 [[TMP1]], 1
; UNROLL-NO-VF-NEXT: br label [[VECTOR_BODY:%.*]]
; UNROLL-NO-VF: vector.body:
-; UNROLL-NO-VF-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE6:%.*]] ]
-; UNROLL-NO-VF-NEXT: [[VECTOR_RECUR:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP11:%.*]], [[PRED_STORE_CONTINUE6]] ]
-; UNROLL-NO-VF-NEXT: [[VEC_PHI:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP12:%.*]], [[PRED_STORE_CONTINUE6]] ]
-; UNROLL-NO-VF-NEXT: [[VEC_PHI2:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP13:%.*]], [[PRED_STORE_CONTINUE6]] ]
+; UNROLL-NO-VF-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE7:%.*]] ]
+; UNROLL-NO-VF-NEXT: [[VECTOR_RECUR:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[PRED_STORE_CONTINUE7]] ]
+; UNROLL-NO-VF-NEXT: [[VEC_PHI:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP10:%.*]], [[PRED_STORE_CONTINUE7]] ]
+; UNROLL-NO-VF-NEXT: [[VEC_PHI2:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP11:%.*]], [[PRED_STORE_CONTINUE7]] ]
; UNROLL-NO-VF-NEXT: [[OFFSET_IDX:%.*]] = sub i32 [[Y]], [[INDEX]]
; UNROLL-NO-VF-NEXT: [[TMP2:%.*]] = add i32 [[OFFSET_IDX]], 0
; UNROLL-NO-VF-NEXT: [[TMP3:%.*]] = add i32 [[OFFSET_IDX]], -1
-; UNROLL-NO-VF-NEXT: [[TMP4:%.*]] = add i32 [[INDEX]], 0
-; UNROLL-NO-VF-NEXT: [[TMP5:%.*]] = add i32 [[INDEX]], 1
-; UNROLL-NO-VF-NEXT: [[TMP6:%.*]] = icmp ule i32 [[TMP4]], [[TRIP_COUNT_MINUS_1]]
-; UNROLL-NO-VF-NEXT: [[TMP7:%.*]] = icmp ule i32 [[TMP5]], [[TRIP_COUNT_MINUS_1]]
-; UNROLL-NO-VF-NEXT: br i1 [[TMP6]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
+; UNROLL-NO-VF-NEXT: [[VEC_IV:%.*]] = add i32 [[INDEX]], 0
+; UNROLL-NO-VF-NEXT: [[VEC_IV3:%.*]] = add i32 [[INDEX]], 1
+; UNROLL-NO-VF-NEXT: [[TMP4:%.*]] = icmp ule i32 [[VEC_IV]], [[TRIP_COUNT_MINUS_1]]
+; UNROLL-NO-VF-NEXT: [[TMP5:%.*]] = icmp ule i32 [[VEC_IV3]], [[TRIP_COUNT_MINUS_1]]
+; UNROLL-NO-VF-NEXT: br i1 [[TMP4]], label [[PRED_UDIV_IF:%.*]], label [[PRED_UDIV_CONTINUE:%.*]]
; UNROLL-NO-VF: pred.udiv.if:
-; UNROLL-NO-VF-NEXT: [[TMP8:%.*]] = udiv i32 219220132, [[TMP2]]
+; UNROLL-NO-VF-NEXT: [[TMP6:%.*]] = udiv i32 219220132, [[TMP2]]
; UNROLL-NO-VF-NEXT: br label [[PRED_UDIV_CONTINUE]]
; UNROLL-NO-VF: pred.udiv.continue:
-; UNROLL-NO-VF-NEXT: [[TMP9:%.*]] = phi i32 [ poison, [[VECTOR_BODY]] ], [ [[TMP8]], [[PRED_UDIV_IF]] ]
-; UNROLL-NO-VF-NEXT: br i1 [[TMP7]], label [[PRED_UDIV_IF3:%.*]], label [[PRED_UDIV_CONTINUE4:%.*]]
-; UNROLL-NO-VF: pred.udiv.if3:
-; UNROLL-NO-VF-NEXT: [[TMP10:%.*]] = udiv i32 219220132, [[TMP3]]
-; UNROLL-NO-VF-NEXT: br label [[PRED_UDIV_CONTINUE4]]
-; UNROLL-NO-VF: pred.udiv.continue4:
-; UNROLL-NO-VF-NEXT: [[TMP11]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE]] ], [ [[TMP10]], [[PRED_UDIV_IF3]] ]
-; UNROLL-NO-VF-NEXT: [[TMP12]] = add i32 [[VEC_PHI]], [[VECTOR_RECUR]]
-; UNROLL-NO-VF-NEXT: [[TMP13]] = add i32 [[VEC_PHI2]], [[TMP9]]
-; UNROLL-NO-VF-NEXT: br i1 [[TMP6]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
+; UNROLL-NO-VF-NEXT: [[TMP7:%.*]] = phi i32 [ poison, [[VECTOR_BODY]] ], [ [[TMP6]], [[PRED_UDIV_IF]] ]
+; UNROLL-NO-VF-NEXT: br i1 [[TMP5]], label [[PRED_UDIV_IF4:%.*]], label [[PRED_UDIV_CONTINUE5:%.*]]
+; UNROLL-NO-VF: pred.udiv.if4:
+; UNROLL-NO-VF-NEXT: [[TMP8:%.*]] = udiv i32 219220132, [[TMP3]]
+; UNROLL-NO-VF-NEXT: br label [[PRED_UDIV_CONTINUE5]]
+; UNROLL-NO-VF: pred.udiv.continue5:
+; UNROLL-NO-VF-NEXT: [[TMP9]] = phi i32 [ poison, [[PRED_UDIV_CONTINUE]] ], [ [[TMP8]], [[PRED_UDIV_IF4]] ]
+; UNROLL-NO-VF-NEXT: [[TMP10]] = add i32 [[VEC_PHI]], [[VECTOR_RECUR]]
+; UNROLL-NO-VF-NEXT: [[TMP11]] = add i32 [[VEC_PHI2]], [[TMP7]]
+; UNROLL-NO-VF-NEXT: br i1 [[TMP4]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; UNROLL-NO-VF: pred.store.if:
-; UNROLL-NO-VF-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[X:%.*]], i32 [[TMP4]]
-; UNROLL-NO-VF-NEXT: store i32 [[TMP2]], ptr [[TMP14]], align 4
+; UNROLL-NO-VF-NEXT: [[TMP12:%.*]] = add i32 [[INDEX]], 0
+; UNROLL-NO-VF-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[X:%.*]], i32 [[TMP12]]
+; UNROLL-NO-VF-NEXT: store i32 [[TMP2]], ptr [[TMP13]], align 4
; UNROLL-NO-VF-NEXT: br label [[PRED_STORE_CONTINUE]]
; UNROLL-NO-VF: pred.store.continue:
-; UNROLL-NO-VF-NEXT: br i1 [[TMP7]], label [[PRED_STORE_IF5:%.*]], label [[PRED_STORE_CONTINUE6]]
-; UNROLL-NO-VF: pred.store.if5:
-; UNROLL-NO-VF-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[X]], i32 [[TMP5]]
+; UNROLL-NO-VF-NEXT: br i1 [[TMP5]], label [[PRED_STORE_IF6:%.*]], label [[PRED_STORE_CONTINUE7]]
+; UNROLL-NO-VF: pred.store.if6:
+; UNROLL-NO-VF-NEXT: [[TMP14:%.*]] = add i32 [[INDEX]], 1
+; UNROLL-NO-VF-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[X]], i32 [[TMP14]]
; UNROLL-NO-VF-NEXT: store i32 [[TMP3]], ptr [[TMP15]], align 4
-; UNROLL-NO-VF-NEXT: br label [[PRED_STORE_CONTINUE6]]
-; UNROLL-NO-VF: pred.store.continue6:
-; UNROLL-NO-VF-NEXT: [[TMP16:%.*]] = select i1 [[TMP6]], i32 [[TMP12]], i32 [[VEC_PHI]]
-; UNROLL-NO-VF-NEXT: [[TMP17:%.*]] = select i1 [[TMP7]], i32 [[TMP13]], i32 [[VEC_PHI2]]
+; UNROLL-NO-VF-NEXT: br label [[PRED_STORE_CONTINUE7]]
+; UNROLL-NO-VF: pred.store.continue7:
+; UNROLL-NO-VF-NEXT: [[TMP16:%.*]] = select i1 [[TMP4]], i32 [[TMP10]], i32 [[VEC_PHI]]
+; UNROLL-NO-VF-NEXT: [[TMP17:%.*]] = select i1 [[TMP5]], i32 [[TMP11]], i32 [[VEC_PHI2]]
; UNROLL-NO-VF-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 2
; UNROLL-NO-VF-NEXT: [[TMP18:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
; UNROLL-NO-VF-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !prof [[PROF27]], !llvm.loop [[LOOP31:![0-9]+]]
@@ -3317,7 +3319,7 @@ define i32 @sink_into_replication_region_multiple(ptr %x, i32 %y) {
; UNROLL-NO-VF-NEXT: [[BIN_RDX:%.*]] = add i32 [[TMP17]], [[TMP16]]
; UNROLL-NO-VF-NEXT: br i1 true, label [[BB1:%.*]], label [[SCALAR_PH]]
; UNROLL-NO-VF: scalar.ph:
-; UNROLL-NO-VF-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[BB:%.*]] ], [ [[TMP11]], [[MIDDLE_BLOCK]] ]
+; UNROLL-NO-VF-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i32 [ 0, [[BB:%.*]] ], [ [[TMP9]], [[MIDDLE_BLOCK]] ]
; UNROLL-NO-VF-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[Y]], [[BB]] ]
; UNROLL-NO-VF-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[BB]] ]
; UNROLL-NO-VF-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[BB]] ], [ [[BIN_RDX]], [[MIDDLE_BLOCK]] ]
>From 916a7d222c5d86decc5287e423297d834672ede7 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Mon, 29 Jan 2024 14:56:06 +0000
Subject: [PATCH 04/13] [VPlan] Explicitly handle scalar pointer inductions.
Add a new PtrAdd opcode to VPInstruction that corresponds to
IRBuilder::CreatePtrAdd, which creates a GEP with source element type
i8.
This is then used to model scalarizing VPWidenPointerInductionRecipe by
introducing scalar-steps to model the index increment followed by a
PtrAdd.
Note that PtrAdd needs to be able to generate code for only the first
lane or for all lanes. This may warrant introducing a separate recipe
for scalarizing that can be created without relying on the underlying
IR.
---
.../Transforms/Vectorize/LoopVectorize.cpp | 35 +--
llvm/lib/Transforms/Vectorize/VPlan.cpp | 6 +-
llvm/lib/Transforms/Vectorize/VPlan.h | 7 +
.../Transforms/Vectorize/VPlanAnalysis.cpp | 2 +
.../lib/Transforms/Vectorize/VPlanRecipes.cpp | 24 ++
.../Transforms/Vectorize/VPlanTransforms.cpp | 46 ++-
.../epilog-vectorization-widen-inductions.ll | 100 ++++---
.../LoopVectorize/AArch64/sve-epilog-vect.ll | 148 +++++-----
.../AArch64/sve-live-out-pointer-induction.ll | 74 ++---
.../LoopVectorize/AArch64/sve-widen-gep.ll | 12 +-
.../LoopVectorize/AArch64/sve-widen-phi.ll | 49 ++--
.../LoopVectorize/X86/gather_scatter.ll | 154 +++++------
.../X86/interleave-opaque-pointers.ll | 19 +-
.../LoopVectorize/X86/small-size.ll | 236 ++++++++--------
.../LoopVectorize/consecutive-ptr-uniforms.ll | 18 +-
.../LoopVectorize/first-order-recurrence.ll | 261 +++++++++---------
.../LoopVectorize/pointer-induction-unroll.ll | 61 ++--
.../LoopVectorize/pointer-induction.ll | 99 ++++---
.../tail-folding-vectorization-factor-1.ll | 25 +-
.../vplan-sink-scalars-and-merge.ll | 8 +-
20 files changed, 686 insertions(+), 698 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 55466720aa11f9..4ee878358f9bcd 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -9233,42 +9233,11 @@ void VPWidenPointerInductionRecipe::execute(VPTransformState &State) {
"Not a pointer induction according to InductionDescriptor!");
assert(cast<PHINode>(getUnderlyingInstr())->getType()->isPointerTy() &&
"Unexpected type.");
+ assert(!onlyScalarsGenerated(State.VF.isScalable()) &&
+ "Recipe should have been replaced");
auto *IVR = getParent()->getPlan()->getCanonicalIV();
PHINode *CanonicalIV = cast<PHINode>(State.get(IVR, 0));
-
- if (onlyScalarsGenerated(State.VF.isScalable())) {
- // This is the normalized GEP that starts counting at zero.
- Value *PtrInd = State.Builder.CreateSExtOrTrunc(
- CanonicalIV, IndDesc.getStep()->getType());
- // Determine the number of scalars we need to generate for each unroll
- // iteration. If the instruction is uniform, we only need to generate the
- // first lane. Otherwise, we generate all VF values.
- bool IsUniform = vputils::onlyFirstLaneUsed(this);
- assert((IsUniform || !State.VF.isScalable()) &&
- "Cannot scalarize a scalable VF");
- unsigned Lanes = IsUniform ? 1 : State.VF.getFixedValue();
-
- for (unsigned Part = 0; Part < State.UF; ++Part) {
- Value *PartStart =
- createStepForVF(State.Builder, PtrInd->getType(), State.VF, Part);
-
- for (unsigned Lane = 0; Lane < Lanes; ++Lane) {
- Value *Idx = State.Builder.CreateAdd(
- PartStart, ConstantInt::get(PtrInd->getType(), Lane));
- Value *GlobalIdx = State.Builder.CreateAdd(PtrInd, Idx);
-
- Value *Step = State.get(getOperand(1), VPIteration(Part, Lane));
- Value *SclrGep = emitTransformedIndex(
- State.Builder, GlobalIdx, IndDesc.getStartValue(), Step,
- IndDesc.getKind(), IndDesc.getInductionBinOp());
- SclrGep->setName("next.gep");
- State.set(this, SclrGep, VPIteration(Part, Lane));
- }
- }
- return;
- }
-
Type *PhiType = IndDesc.getStep()->getType();
// Build a pointer phi
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index e9c499c5cddf6a..a1a5f0aff12fb2 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -857,11 +857,7 @@ void VPlan::execute(VPTransformState *State) {
Phi = cast<PHINode>(State->get(R.getVPSingleValue(), 0));
} else {
auto *WidenPhi = cast<VPWidenPointerInductionRecipe>(&R);
- // TODO: Split off the case that all users of a pointer phi are scalar
- // from the VPWidenPointerInductionRecipe.
- if (WidenPhi->onlyScalarsGenerated(State->VF.isScalable()))
- continue;
-
+ assert(!WidenPhi->onlyScalarsGenerated(State->VF.isScalable()));
auto *GEP = cast<GetElementPtrInst>(State->get(WidenPhi, 0));
Phi = cast<PHINode>(GEP->getPointerOperand());
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 520e45201c6c58..699445b78b3e09 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1156,6 +1156,7 @@ class VPInstruction : public VPRecipeWithIRFlags {
BranchOnCount,
BranchOnCond,
ComputeReductionResult,
+ PtrAdd,
};
private:
@@ -2503,6 +2504,12 @@ class VPDerivedIVRecipe : public VPSingleDefRecipe {
dyn_cast_or_null<FPMathOperator>(IndDesc.getInductionBinOp()),
Start, CanonicalIV, Step) {}
+ VPDerivedIVRecipe(InductionDescriptor::InductionKind Kind, VPValue *Start,
+ VPCanonicalIVPHIRecipe *CanonicalIV, VPValue *Step,
+ FPMathOperator *FPBinOp)
+ : VPSingleDefRecipe(VPDef::VPDerivedIVSC, {Start, CanonicalIV, Step}),
+ Kind(Kind), FPBinOp(FPBinOp) {}
+
~VPDerivedIVRecipe() override = default;
VPRecipeBase *clone() override {
diff --git a/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp b/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
index 515dc41a55ea1b..bd2f65935e4798 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanAnalysis.cpp
@@ -43,6 +43,8 @@ Type *VPTypeAnalysis::inferScalarTypeForRecipe(const VPInstruction *R) {
CachedTypes[OtherV] = ResTy;
return ResTy;
}
+ case VPInstruction::PtrAdd:
+ return inferScalarType(R->getOperand(0));
default:
break;
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index b1a0ebb5863f7e..1659b37dcb7e95 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -127,6 +127,7 @@ bool VPRecipeBase::mayHaveSideEffects() const {
case VPInstruction::Not:
case VPInstruction::CalculateTripCountMinusVF:
case VPInstruction::CanonicalIVIncrementForPart:
+ case VPInstruction::PtrAdd:
return false;
default:
return true;
@@ -489,6 +490,23 @@ Value *VPInstruction::generateInstruction(VPTransformState &State,
return ReducedPartRdx;
}
+ case VPInstruction::PtrAdd: {
+ if (vputils::onlyFirstLaneUsed(this)) {
+ auto *P = Builder.CreatePtrAdd(
+ State.get(getOperand(0), VPIteration(Part, 0)),
+ State.get(getOperand(1), VPIteration(Part, 0)), Name);
+ State.set(this, P, VPIteration(Part, 0));
+ } else {
+ for (unsigned Lane = 0; Lane != State.VF.getKnownMinValue(); ++Lane) {
+ Value *P = Builder.CreatePtrAdd(
+ State.get(getOperand(0), VPIteration(Part, Lane)),
+ State.get(getOperand(1), VPIteration(Part, Lane)), Name);
+
+ State.set(this, P, VPIteration(Part, Lane));
+ }
+ }
+ return nullptr;
+ }
default:
llvm_unreachable("Unsupported opcode for instruction");
}
@@ -515,6 +533,8 @@ void VPInstruction::execute(VPTransformState &State) {
State.Builder.setFastMathFlags(getFastMathFlags());
for (unsigned Part = 0; Part < State.UF; ++Part) {
Value *GeneratedValue = generateInstruction(State, Part);
+ if (!GeneratedValue)
+ continue;
if (!hasResult())
continue;
assert(GeneratedValue && "generateInstruction must produce a value");
@@ -540,6 +560,7 @@ bool VPInstruction::onlyFirstLaneUsed(const VPValue *Op) const {
default:
return false;
case Instruction::ICmp:
+ case VPInstruction::PtrAdd:
// TODO: Cover additional opcodes.
return vputils::onlyFirstLaneUsed(this);
case VPInstruction::ComputeReductionResult:
@@ -599,6 +620,9 @@ void VPInstruction::print(raw_ostream &O, const Twine &Indent,
case VPInstruction::ComputeReductionResult:
O << "compute-reduction-result";
break;
+ case VPInstruction::PtrAdd:
+ O << "ptradd";
+ break;
default:
O << Instruction::getOpcodeName(getOpcode());
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 71f5285f90236b..6f5c7090455ca6 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -489,15 +489,18 @@ void VPlanTransforms::removeDeadRecipes(VPlan &Plan) {
}
}
-static VPValue *createScalarIVSteps(VPlan &Plan, const InductionDescriptor &ID,
+static VPValue *createScalarIVSteps(VPlan &Plan,
+ InductionDescriptor::InductionKind Kind,
ScalarEvolution &SE, Instruction *TruncI,
VPValue *StartV, VPValue *Step,
- VPBasicBlock::iterator IP) {
+ Instruction::BinaryOps InductionOpcode,
+ VPBasicBlock::iterator IP,
+ FPMathOperator *FPBinOp = nullptr) {
VPBasicBlock *HeaderVPBB = Plan.getVectorLoopRegion()->getEntryBasicBlock();
VPCanonicalIVPHIRecipe *CanonicalIV = Plan.getCanonicalIV();
VPSingleDefRecipe *BaseIV = CanonicalIV;
- if (!CanonicalIV->isCanonical(ID.getKind(), StartV, Step)) {
- BaseIV = new VPDerivedIVRecipe(ID, StartV, CanonicalIV, Step);
+ if (!CanonicalIV->isCanonical(Kind, StartV, Step)) {
+ BaseIV = new VPDerivedIVRecipe(Kind, StartV, CanonicalIV, Step, FPBinOp);
HeaderVPBB->insert(BaseIV, IP);
}
@@ -526,7 +529,9 @@ static VPValue *createScalarIVSteps(VPlan &Plan, const InductionDescriptor &ID,
VecPreheader->appendRecipe(Step->getDefiningRecipe());
}
- VPScalarIVStepsRecipe *Steps = new VPScalarIVStepsRecipe(ID, BaseIV, Step);
+ VPScalarIVStepsRecipe *Steps = new VPScalarIVStepsRecipe(
+ BaseIV, Step, InductionOpcode,
+ FPBinOp ? FPBinOp->getFastMathFlags() : FastMathFlags());
HeaderVPBB->insert(Steps, IP);
return Steps;
}
@@ -537,6 +542,30 @@ void VPlanTransforms::optimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
bool HasOnlyVectorVFs = !Plan.hasVF(ElementCount::getFixed(1));
VPBasicBlock::iterator InsertPt = HeaderVPBB->getFirstNonPhi();
for (VPRecipeBase &Phi : HeaderVPBB->phis()) {
+ if (auto *PtrIV = dyn_cast<VPWidenPointerInductionRecipe>(&Phi)) {
+ if (!PtrIV->onlyScalarsGenerated(Plan.hasScalableVF()))
+ continue;
+
+ const InductionDescriptor &ID = PtrIV->getInductionDescriptor();
+ VPValue *StartV = Plan.getVPValueOrAddLiveIn(
+ ConstantInt::get(ID.getStep()->getType(), 0));
+ VPValue *StepV = PtrIV->getOperand(1);
+ VPRecipeBase *Steps =
+ createScalarIVSteps(Plan, InductionDescriptor::IK_IntInduction, SE,
+ nullptr, StartV, StepV, Instruction::Add,
+ InsertPt)
+ ->getDefiningRecipe();
+
+ auto *Recipe =
+ new VPInstruction(VPInstruction::PtrAdd,
+ {PtrIV->getStartValue(), Steps->getVPSingleValue()},
+ PtrIV->getDebugLoc(), "next.gep");
+
+ Recipe->insertAfter(Steps);
+ PtrIV->replaceAllUsesWith(Recipe);
+ continue;
+ }
+
auto *WideIV = dyn_cast<VPWidenIntOrFpInductionRecipe>(&Phi);
if (!WideIV)
continue;
@@ -546,9 +575,10 @@ void VPlanTransforms::optimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
continue;
const InductionDescriptor &ID = WideIV->getInductionDescriptor();
- VPValue *Steps = createScalarIVSteps(Plan, ID, SE, WideIV->getTruncInst(),
- WideIV->getStartValue(),
- WideIV->getStepValue(), InsertPt);
+ VPValue *Steps = createScalarIVSteps(
+ Plan, ID.getKind(), SE, WideIV->getTruncInst(), WideIV->getStartValue(),
+ WideIV->getStepValue(), ID.getInductionOpcode(), InsertPt,
+ dyn_cast_or_null<FPMathOperator>(ID.getInductionBinOp()));
// Update scalar users of IV to use Step instead.
if (!HasOnlyVectorVFs)
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/epilog-vectorization-widen-inductions.ll b/llvm/test/Transforms/LoopVectorize/AArch64/epilog-vectorization-widen-inductions.ll
index 24c59fdb47b613..00ec396107dcb5 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/epilog-vectorization-widen-inductions.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/epilog-vectorization-widen-inductions.ll
@@ -11,76 +11,74 @@ define void @test_widen_ptr_induction(ptr %ptr.start.1) {
; CHECK: vector.main.loop.iter.check:
; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
-; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[PTR_START_1:%.*]], i64 10000
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP0]]
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
-; CHECK-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP1]]
-; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x ptr> poison, ptr [[NEXT_GEP]], i32 0
-; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x ptr> [[TMP2]], ptr [[NEXT_GEP1]], i32 1
-; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 2
-; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP4]]
-; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 3
-; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP5]]
-; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x ptr> poison, ptr [[NEXT_GEP2]], i32 0
-; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x ptr> [[TMP6]], ptr [[NEXT_GEP3]], i32 1
-; CHECK-NEXT: [[TMP8:%.*]] = icmp ne <2 x ptr> [[TMP3]], zeroinitializer
-; CHECK-NEXT: [[TMP9:%.*]] = icmp ne <2 x ptr> [[TMP7]], zeroinitializer
-; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i1> [[TMP8]], i32 0
-; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP10]])
-; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i1> [[TMP8]], i32 1
-; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]])
-; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x i1> [[TMP9]], i32 0
-; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP12]])
-; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i1> [[TMP9]], i32 1
-; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP13]])
-; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 0
-; CHECK-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 2
-; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[TMP14]], align 1
-; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[TMP15]], align 1
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2
+; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[PTR_START_1:%.*]], i64 [[TMP0]]
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP1]]
+; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP4]], i32 0
+; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x ptr> [[TMP6]], ptr [[TMP5]], i32 1
+; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP2]]
+; CHECK-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP3]]
+; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP8]], i32 0
+; CHECK-NEXT: [[TMP11:%.*]] = insertelement <2 x ptr> [[TMP10]], ptr [[TMP9]], i32 1
+; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <2 x ptr> [[TMP7]], zeroinitializer
+; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <2 x ptr> [[TMP11]], zeroinitializer
+; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i1> [[TMP12]], i32 0
+; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP14]])
+; CHECK-NEXT: [[TMP15:%.*]] = extractelement <2 x i1> [[TMP12]], i32 1
+; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP15]])
+; CHECK-NEXT: [[TMP16:%.*]] = extractelement <2 x i1> [[TMP13]], i32 0
+; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP16]])
+; CHECK-NEXT: [[TMP17:%.*]] = extractelement <2 x i1> [[TMP13]], i32 1
+; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP17]])
+; CHECK-NEXT: [[TMP18:%.*]] = getelementptr i8, ptr [[TMP4]], i32 0
+; CHECK-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[TMP4]], i32 2
+; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[TMP18]], align 1
+; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[TMP19]], align 1
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
-; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10000
-; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], {{!llvm.loop ![0-9]+}}
+; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10000
+; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], {{!llvm.loop ![0-9]+}}
; CHECK: middle.block:
; CHECK-NEXT: br i1 false, label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
; CHECK: vec.epilog.iter.check:
-; CHECK-NEXT: [[IND_END6:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 10000
+; CHECK-NEXT: [[IND_END1:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 10000
; CHECK-NEXT: br i1 true, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
; CHECK: vec.epilog.ph:
-; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PTR_START_1]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 10000, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
-; CHECK-NEXT: [[IND_END5:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 10000
+; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 10000
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; CHECK: vec.epilog.vector.body:
-; CHECK-NEXT: [[INDEX8:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT11:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP17:%.*]] = add i64 [[INDEX8]], 0
-; CHECK-NEXT: [[NEXT_GEP9:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP17]]
-; CHECK-NEXT: [[TMP18:%.*]] = add i64 [[INDEX8]], 1
-; CHECK-NEXT: [[NEXT_GEP10:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP18]]
-; CHECK-NEXT: [[TMP19:%.*]] = insertelement <2 x ptr> poison, ptr [[NEXT_GEP9]], i32 0
-; CHECK-NEXT: [[TMP20:%.*]] = insertelement <2 x ptr> [[TMP19]], ptr [[NEXT_GEP10]], i32 1
-; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <2 x ptr> [[TMP20]], zeroinitializer
-; CHECK-NEXT: [[TMP22:%.*]] = extractelement <2 x i1> [[TMP21]], i32 0
-; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP22]])
-; CHECK-NEXT: [[TMP23:%.*]] = extractelement <2 x i1> [[TMP21]], i32 1
-; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP23]])
-; CHECK-NEXT: [[TMP24:%.*]] = getelementptr i8, ptr [[NEXT_GEP9]], i32 0
-; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[TMP24]], align 1
-; CHECK-NEXT: [[INDEX_NEXT11]] = add nuw i64 [[INDEX8]], 2
-; CHECK-NEXT: [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT11]], 10000
-; CHECK-NEXT: br i1 [[TMP25]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], {{!llvm.loop ![0-9]+}}
+; CHECK-NEXT: [[INDEX3:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT4:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP21:%.*]] = add i64 [[INDEX3]], 0
+; CHECK-NEXT: [[TMP22:%.*]] = add i64 [[INDEX3]], 1
+; CHECK-NEXT: [[TMP23:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP21]]
+; CHECK-NEXT: [[TMP24:%.*]] = getelementptr i8, ptr [[PTR_START_1]], i64 [[TMP22]]
+; CHECK-NEXT: [[TMP25:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP23]], i32 0
+; CHECK-NEXT: [[TMP26:%.*]] = insertelement <2 x ptr> [[TMP25]], ptr [[TMP24]], i32 1
+; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <2 x ptr> [[TMP26]], zeroinitializer
+; CHECK-NEXT: [[TMP28:%.*]] = extractelement <2 x i1> [[TMP27]], i32 0
+; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP28]])
+; CHECK-NEXT: [[TMP29:%.*]] = extractelement <2 x i1> [[TMP27]], i32 1
+; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP29]])
+; CHECK-NEXT: [[TMP30:%.*]] = getelementptr i8, ptr [[TMP23]], i32 0
+; CHECK-NEXT: store <2 x i8> zeroinitializer, ptr [[TMP30]], align 1
+; CHECK-NEXT: [[INDEX_NEXT4]] = add nuw i64 [[INDEX3]], 2
+; CHECK-NEXT: [[TMP31:%.*]] = icmp eq i64 [[INDEX_NEXT4]], 10000
+; CHECK-NEXT: br i1 [[TMP31]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], {{!llvm.loop ![0-9]+}}
; CHECK: vec.epilog.middle.block:
; CHECK-NEXT: br i1 false, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
; CHECK: vec.epilog.scalar.ph:
-; CHECK-NEXT: [[BC_RESUME_VAL4:%.*]] = phi i64 [ 10000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 10000, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
-; CHECK-NEXT: [[BC_RESUME_VAL7:%.*]] = phi ptr [ [[IND_END5]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END6]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PTR_START_1]], [[ITER_CHECK]] ]
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 10000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 10000, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
+; CHECK-NEXT: [[BC_RESUME_VAL2:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END1]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PTR_START_1]], [[ITER_CHECK]] ]
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL4]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[BC_RESUME_VAL7]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[PTR_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[BC_RESUME_VAL2]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[PTR_IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: [[CMP_I_I_I_I:%.*]] = icmp ne ptr [[PTR_IV]], null
; CHECK-NEXT: tail call void @llvm.assume(i1 [[CMP_I_I_I_I]])
; CHECK-NEXT: store i8 0, ptr [[PTR_IV]], align 1
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/sve-epilog-vect.ll b/llvm/test/Transforms/LoopVectorize/AArch64/sve-epilog-vect.ll
index 24d2127ee171a0..12889c2acc8e5f 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/sve-epilog-vect.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/sve-epilog-vect.ll
@@ -146,13 +146,13 @@ define void @main_vf_vscale_x_16(ptr %A) #0 {
; CHECK-VF8-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; CHECK-VF8-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; CHECK-VF8: vec.epilog.vector.body:
-; CHECK-VF8-NEXT: [[INDEX2:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT3:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
-; CHECK-VF8-NEXT: [[TMP19:%.*]] = add i64 [[INDEX2]], 0
+; CHECK-VF8-NEXT: [[INDEX1:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT2:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
+; CHECK-VF8-NEXT: [[TMP19:%.*]] = add i64 [[INDEX1]], 0
; CHECK-VF8-NEXT: [[TMP20:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP19]]
; CHECK-VF8-NEXT: [[TMP21:%.*]] = getelementptr inbounds i8, ptr [[TMP20]], i32 0
; CHECK-VF8-NEXT: store <8 x i8> <i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>, ptr [[TMP21]], align 1
-; CHECK-VF8-NEXT: [[INDEX_NEXT3]] = add nuw i64 [[INDEX2]], 8
-; CHECK-VF8-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT3]], 1024
+; CHECK-VF8-NEXT: [[INDEX_NEXT2]] = add nuw i64 [[INDEX1]], 8
+; CHECK-VF8-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT2]], 1024
; CHECK-VF8-NEXT: br i1 [[TMP22]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK-VF8: vec.epilog.middle.block:
; CHECK-VF8-NEXT: br i1 true, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
@@ -246,13 +246,13 @@ define void @main_vf_vscale_x_2(ptr %A) #0 vscale_range(8, 8) {
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; CHECK: vec.epilog.vector.body:
-; CHECK-NEXT: [[INDEX2:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT3:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP19:%.*]] = add i64 [[INDEX2]], 0
+; CHECK-NEXT: [[INDEX1:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT2:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP19:%.*]] = add i64 [[INDEX1]], 0
; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP19]]
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i64, ptr [[TMP20]], i32 0
; CHECK-NEXT: store <8 x i64> <i64 1, i64 1, i64 1, i64 1, i64 1, i64 1, i64 1, i64 1>, ptr [[TMP21]], align 1
-; CHECK-NEXT: [[INDEX_NEXT3]] = add nuw i64 [[INDEX2]], 8
-; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT3]], 1024
+; CHECK-NEXT: [[INDEX_NEXT2]] = add nuw i64 [[INDEX1]], 8
+; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT2]], 1024
; CHECK-NEXT: br i1 [[TMP22]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK: vec.epilog.middle.block:
; CHECK-NEXT: br i1 true, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
@@ -315,13 +315,13 @@ define void @main_vf_vscale_x_2(ptr %A) #0 vscale_range(8, 8) {
; CHECK-VF8-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; CHECK-VF8-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; CHECK-VF8: vec.epilog.vector.body:
-; CHECK-VF8-NEXT: [[INDEX2:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT3:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
-; CHECK-VF8-NEXT: [[TMP19:%.*]] = add i64 [[INDEX2]], 0
+; CHECK-VF8-NEXT: [[INDEX1:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT2:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
+; CHECK-VF8-NEXT: [[TMP19:%.*]] = add i64 [[INDEX1]], 0
; CHECK-VF8-NEXT: [[TMP20:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP19]]
; CHECK-VF8-NEXT: [[TMP21:%.*]] = getelementptr inbounds i64, ptr [[TMP20]], i32 0
; CHECK-VF8-NEXT: store <8 x i64> <i64 1, i64 1, i64 1, i64 1, i64 1, i64 1, i64 1, i64 1>, ptr [[TMP21]], align 1
-; CHECK-VF8-NEXT: [[INDEX_NEXT3]] = add nuw i64 [[INDEX2]], 8
-; CHECK-VF8-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT3]], 1024
+; CHECK-VF8-NEXT: [[INDEX_NEXT2]] = add nuw i64 [[INDEX1]], 8
+; CHECK-VF8-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT2]], 1024
; CHECK-VF8-NEXT: br i1 [[TMP22]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK-VF8: vec.epilog.middle.block:
; CHECK-VF8-NEXT: br i1 true, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
@@ -374,66 +374,65 @@ define void @test_pr57912_pointer_induction(ptr %start) #0 {
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 10000, [[N_MOD_VF]]
; CHECK-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP7:%.*]] = mul i64 [[TMP6]], 32
-; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START:%.*]], i64 [[N_VEC]]
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP8]]
; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP10:%.*]] = mul i64 [[TMP9]], 16
; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[TMP10]], 0
-; CHECK-NEXT: [[TMP12:%.*]] = add i64 [[INDEX]], [[TMP11]]
-; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP12]]
-; CHECK-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 0
-; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP15:%.*]] = mul i64 [[TMP14]], 16
-; CHECK-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 [[TMP15]]
-; CHECK-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP13]], align 1
+; CHECK-NEXT: [[TMP12:%.*]] = mul i64 [[TMP11]], 1
+; CHECK-NEXT: [[TMP13:%.*]] = add i64 [[INDEX]], [[TMP12]]
+; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[START:%.*]], i64 [[TMP8]]
+; CHECK-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP13]]
+; CHECK-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[TMP14]], i32 0
+; CHECK-NEXT: [[TMP17:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP18:%.*]] = mul i64 [[TMP17]], 16
+; CHECK-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[TMP14]], i64 [[TMP18]]
; CHECK-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP16]], align 1
+; CHECK-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP19]], align 1
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP7]]
-; CHECK-NEXT: [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
+; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 10000, [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
; CHECK: vec.epilog.iter.check:
-; CHECK-NEXT: [[IND_END7:%.*]] = getelementptr i8, ptr [[START]], i64 [[N_VEC]]
+; CHECK-NEXT: [[IND_END4:%.*]] = getelementptr i8, ptr [[START]], i64 [[N_VEC]]
; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 10000, [[N_VEC]]
-; CHECK-NEXT: [[TMP18:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP19:%.*]] = mul i64 [[TMP18]], 8
-; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], [[TMP19]]
+; CHECK-NEXT: [[TMP21:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP22:%.*]] = mul i64 [[TMP21]], 8
+; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], [[TMP22]]
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
; CHECK: vec.epilog.ph:
-; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
-; CHECK-NEXT: [[TMP20:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP21:%.*]] = mul i64 [[TMP20]], 8
-; CHECK-NEXT: [[N_MOD_VF3:%.*]] = urem i64 10000, [[TMP21]]
-; CHECK-NEXT: [[N_VEC4:%.*]] = sub i64 10000, [[N_MOD_VF3]]
-; CHECK-NEXT: [[IND_END6:%.*]] = getelementptr i8, ptr [[START]], i64 [[N_VEC4]]
-; CHECK-NEXT: [[TMP22:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP23:%.*]] = mul i64 [[TMP22]], 8
+; CHECK-NEXT: [[TMP23:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP24:%.*]] = mul i64 [[TMP23]], 8
+; CHECK-NEXT: [[N_MOD_VF2:%.*]] = urem i64 10000, [[TMP24]]
+; CHECK-NEXT: [[N_VEC3:%.*]] = sub i64 10000, [[N_MOD_VF2]]
+; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START]], i64 [[N_VEC3]]
+; CHECK-NEXT: [[TMP25:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP26:%.*]] = mul i64 [[TMP25]], 8
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; CHECK: vec.epilog.vector.body:
-; CHECK-NEXT: [[INDEX10:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT12:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP24:%.*]] = add i64 [[INDEX10]], 0
-; CHECK-NEXT: [[NEXT_GEP11:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP24]]
-; CHECK-NEXT: [[TMP25:%.*]] = getelementptr i8, ptr [[NEXT_GEP11]], i32 0
-; CHECK-NEXT: store <vscale x 8 x i8> zeroinitializer, ptr [[TMP25]], align 1
-; CHECK-NEXT: [[INDEX_NEXT12]] = add nuw i64 [[INDEX10]], [[TMP23]]
-; CHECK-NEXT: [[TMP26:%.*]] = icmp eq i64 [[INDEX_NEXT12]], [[N_VEC4]]
-; CHECK-NEXT: br i1 [[TMP26]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
+; CHECK-NEXT: [[INDEX7:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT8:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP27:%.*]] = add i64 [[INDEX7]], 0
+; CHECK-NEXT: [[TMP28:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP27]]
+; CHECK-NEXT: [[TMP29:%.*]] = getelementptr i8, ptr [[TMP28]], i32 0
+; CHECK-NEXT: store <vscale x 8 x i8> zeroinitializer, ptr [[TMP29]], align 1
+; CHECK-NEXT: [[INDEX_NEXT8]] = add nuw i64 [[INDEX7]], [[TMP26]]
+; CHECK-NEXT: [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT8]], [[N_VEC3]]
+; CHECK-NEXT: br i1 [[TMP30]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
; CHECK: vec.epilog.middle.block:
-; CHECK-NEXT: [[CMP_N9:%.*]] = icmp eq i64 10000, [[N_VEC4]]
-; CHECK-NEXT: br i1 [[CMP_N9]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
+; CHECK-NEXT: [[CMP_N6:%.*]] = icmp eq i64 10000, [[N_VEC3]]
+; CHECK-NEXT: br i1 [[CMP_N6]], label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
; CHECK: vec.epilog.scalar.ph:
-; CHECK-NEXT: [[BC_RESUME_VAL5:%.*]] = phi i64 [ [[N_VEC4]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
-; CHECK-NEXT: [[BC_RESUME_VAL8:%.*]] = phi ptr [ [[IND_END6]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END7]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[ITER_CHECK]] ]
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC3]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
+; CHECK-NEXT: [[BC_RESUME_VAL5:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END4]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[ITER_CHECK]] ]
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL5]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[BC_RESUME_VAL8]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[PTR_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[BC_RESUME_VAL5]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[PTR_IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-NEXT: store i8 0, ptr [[PTR_IV]], align 1
; CHECK-NEXT: [[PTR_IV_NEXT]] = getelementptr inbounds i8, ptr [[PTR_IV]], i64 1
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
@@ -457,57 +456,56 @@ define void @test_pr57912_pointer_induction(ptr %start) #0 {
; CHECK-VF8-NEXT: [[N_VEC:%.*]] = sub i64 10000, [[N_MOD_VF]]
; CHECK-VF8-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-VF8-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 32
-; CHECK-VF8-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START:%.*]], i64 [[N_VEC]]
; CHECK-VF8-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK-VF8: vector.body:
; CHECK-VF8-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-VF8-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 0
-; CHECK-VF8-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP6]]
; CHECK-VF8-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-VF8-NEXT: [[TMP8:%.*]] = mul i64 [[TMP7]], 16
; CHECK-VF8-NEXT: [[TMP9:%.*]] = add i64 [[TMP8]], 0
-; CHECK-VF8-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], [[TMP9]]
-; CHECK-VF8-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP10]]
-; CHECK-VF8-NEXT: [[TMP11:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i32 0
-; CHECK-VF8-NEXT: [[TMP12:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-VF8-NEXT: [[TMP13:%.*]] = mul i64 [[TMP12]], 16
-; CHECK-VF8-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 [[TMP13]]
-; CHECK-VF8-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP11]], align 1
+; CHECK-VF8-NEXT: [[TMP10:%.*]] = mul i64 [[TMP9]], 1
+; CHECK-VF8-NEXT: [[TMP11:%.*]] = add i64 [[INDEX]], [[TMP10]]
+; CHECK-VF8-NEXT: [[TMP12:%.*]] = getelementptr i8, ptr [[START:%.*]], i64 [[TMP6]]
+; CHECK-VF8-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP11]]
+; CHECK-VF8-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP12]], i32 0
+; CHECK-VF8-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-VF8-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 16
+; CHECK-VF8-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[TMP12]], i64 [[TMP16]]
; CHECK-VF8-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP14]], align 1
+; CHECK-VF8-NEXT: store <vscale x 16 x i8> zeroinitializer, ptr [[TMP17]], align 1
; CHECK-VF8-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
-; CHECK-VF8-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-VF8-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
+; CHECK-VF8-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-VF8-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK-VF8: middle.block:
; CHECK-VF8-NEXT: [[CMP_N:%.*]] = icmp eq i64 10000, [[N_VEC]]
; CHECK-VF8-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
; CHECK-VF8: vec.epilog.iter.check:
-; CHECK-VF8-NEXT: [[IND_END4:%.*]] = getelementptr i8, ptr [[START]], i64 [[N_VEC]]
+; CHECK-VF8-NEXT: [[IND_END1:%.*]] = getelementptr i8, ptr [[START]], i64 [[N_VEC]]
; CHECK-VF8-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 10000, [[N_VEC]]
; CHECK-VF8-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 8
; CHECK-VF8-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
; CHECK-VF8: vec.epilog.ph:
-; CHECK-VF8-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; CHECK-VF8-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
-; CHECK-VF8-NEXT: [[IND_END3:%.*]] = getelementptr i8, ptr [[START]], i64 10000
+; CHECK-VF8-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START]], i64 10000
; CHECK-VF8-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; CHECK-VF8: vec.epilog.vector.body:
-; CHECK-VF8-NEXT: [[INDEX7:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT9:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
-; CHECK-VF8-NEXT: [[TMP16:%.*]] = add i64 [[INDEX7]], 0
-; CHECK-VF8-NEXT: [[NEXT_GEP8:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP16]]
-; CHECK-VF8-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[NEXT_GEP8]], i32 0
-; CHECK-VF8-NEXT: store <8 x i8> zeroinitializer, ptr [[TMP17]], align 1
-; CHECK-VF8-NEXT: [[INDEX_NEXT9]] = add nuw i64 [[INDEX7]], 8
-; CHECK-VF8-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT9]], 10000
-; CHECK-VF8-NEXT: br i1 [[TMP18]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
+; CHECK-VF8-NEXT: [[INDEX3:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT4:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
+; CHECK-VF8-NEXT: [[TMP19:%.*]] = add i64 [[INDEX3]], 0
+; CHECK-VF8-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP19]]
+; CHECK-VF8-NEXT: [[TMP21:%.*]] = getelementptr i8, ptr [[TMP20]], i32 0
+; CHECK-VF8-NEXT: store <8 x i8> zeroinitializer, ptr [[TMP21]], align 1
+; CHECK-VF8-NEXT: [[INDEX_NEXT4]] = add nuw i64 [[INDEX3]], 8
+; CHECK-VF8-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT4]], 10000
+; CHECK-VF8-NEXT: br i1 [[TMP22]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
; CHECK-VF8: vec.epilog.middle.block:
; CHECK-VF8-NEXT: br i1 true, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
; CHECK-VF8: vec.epilog.scalar.ph:
-; CHECK-VF8-NEXT: [[BC_RESUME_VAL2:%.*]] = phi i64 [ 10000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
-; CHECK-VF8-NEXT: [[BC_RESUME_VAL5:%.*]] = phi ptr [ [[IND_END3]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END4]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[ITER_CHECK]] ]
+; CHECK-VF8-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 10000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
+; CHECK-VF8-NEXT: [[BC_RESUME_VAL2:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END1]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[START]], [[ITER_CHECK]] ]
; CHECK-VF8-NEXT: br label [[LOOP:%.*]]
; CHECK-VF8: loop:
-; CHECK-VF8-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL2]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
-; CHECK-VF8-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[BC_RESUME_VAL5]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[PTR_IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-VF8-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-VF8-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[BC_RESUME_VAL2]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[PTR_IV_NEXT:%.*]], [[LOOP]] ]
; CHECK-VF8-NEXT: store i8 0, ptr [[PTR_IV]], align 1
; CHECK-VF8-NEXT: [[PTR_IV_NEXT]] = getelementptr inbounds i8, ptr [[PTR_IV]], i64 1
; CHECK-VF8-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/sve-live-out-pointer-induction.ll b/llvm/test/Transforms/LoopVectorize/AArch64/sve-live-out-pointer-induction.ll
index cfb0f9e59ecbc5..8b64d7a083662e 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/sve-live-out-pointer-induction.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/sve-live-out-pointer-induction.ll
@@ -23,54 +23,54 @@ define ptr @test(ptr %start.1, ptr %start.2, ptr %end) {
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START_1:%.*]], i64 [[TMP8]]
; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[N_VEC]], 8
; CHECK-NEXT: [[IND_END3:%.*]] = getelementptr i8, ptr [[START_2]], i64 [[TMP9]]
-; CHECK-NEXT: [[TMP33:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP34:%.*]] = mul i64 [[TMP33]], 4
+; CHECK-NEXT: [[TMP10:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP11:%.*]] = mul i64 [[TMP10]], 4
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[START_1]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP10:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP11:%.*]] = mul i64 [[TMP10]], 2
-; CHECK-NEXT: [[TMP12:%.*]] = mul i64 [[TMP11]], 2
-; CHECK-NEXT: [[TMP13:%.*]] = mul i64 8, [[TMP12]]
-; CHECK-NEXT: [[TMP14:%.*]] = mul i64 [[TMP11]], 0
-; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP14]], i64 0
+; CHECK-NEXT: [[TMP12:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP13:%.*]] = mul i64 [[TMP12]], 2
+; CHECK-NEXT: [[TMP14:%.*]] = mul i64 [[TMP13]], 2
+; CHECK-NEXT: [[TMP15:%.*]] = mul i64 8, [[TMP14]]
+; CHECK-NEXT: [[TMP16:%.*]] = mul i64 [[TMP13]], 0
+; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP16]], 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: [[TMP15:%.*]] = call <vscale x 2 x i64> @llvm.experimental.stepvector.nxv2i64()
-; CHECK-NEXT: [[TMP16:%.*]] = add <vscale x 2 x i64> [[DOTSPLAT]], [[TMP15]]
-; CHECK-NEXT: [[VECTOR_GEP:%.*]] = mul <vscale x 2 x i64> [[TMP16]], shufflevector (<vscale x 2 x i64> insertelement (<vscale x 2 x i64> poison, i64 8, i64 0), <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer)
-; CHECK-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <vscale x 2 x i64> [[VECTOR_GEP]]
-; CHECK-NEXT: [[TMP18:%.*]] = mul i64 [[TMP11]], 1
-; CHECK-NEXT: [[DOTSPLATINSERT5:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP18]], i64 0
+; CHECK-NEXT: [[TMP17:%.*]] = call <vscale x 2 x i64> @llvm.experimental.stepvector.nxv2i64()
+; CHECK-NEXT: [[TMP18:%.*]] = add <vscale x 2 x i64> [[DOTSPLAT]], [[TMP17]]
+; CHECK-NEXT: [[VECTOR_GEP:%.*]] = mul <vscale x 2 x i64> [[TMP18]], shufflevector (<vscale x 2 x i64> insertelement (<vscale x 2 x i64> poison, i64 8, i64 0), <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer)
+; CHECK-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <vscale x 2 x i64> [[VECTOR_GEP]]
+; CHECK-NEXT: [[TMP20:%.*]] = mul i64 [[TMP13]], 1
+; CHECK-NEXT: [[DOTSPLATINSERT5:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP20]], i64 0
; CHECK-NEXT: [[DOTSPLAT6:%.*]] = shufflevector <vscale x 2 x i64> [[DOTSPLATINSERT5]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
-; CHECK-NEXT: [[TMP19:%.*]] = call <vscale x 2 x i64> @llvm.experimental.stepvector.nxv2i64()
-; CHECK-NEXT: [[TMP20:%.*]] = add <vscale x 2 x i64> [[DOTSPLAT6]], [[TMP19]]
-; CHECK-NEXT: [[VECTOR_GEP7:%.*]] = mul <vscale x 2 x i64> [[TMP20]], shufflevector (<vscale x 2 x i64> insertelement (<vscale x 2 x i64> poison, i64 8, i64 0), <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer)
-; CHECK-NEXT: [[TMP21:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <vscale x 2 x i64> [[VECTOR_GEP7]]
-; CHECK-NEXT: [[TMP22:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[TMP23:%.*]] = mul i64 [[TMP22]], 8
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START_2]], i64 [[TMP23]]
-; CHECK-NEXT: [[TMP24:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP25:%.*]] = mul i64 [[TMP24]], 2
-; CHECK-NEXT: [[TMP26:%.*]] = add i64 [[TMP25]], 0
-; CHECK-NEXT: [[TMP27:%.*]] = add i64 [[INDEX]], [[TMP26]]
+; CHECK-NEXT: [[TMP21:%.*]] = call <vscale x 2 x i64> @llvm.experimental.stepvector.nxv2i64()
+; CHECK-NEXT: [[TMP22:%.*]] = add <vscale x 2 x i64> [[DOTSPLAT6]], [[TMP21]]
+; CHECK-NEXT: [[VECTOR_GEP7:%.*]] = mul <vscale x 2 x i64> [[TMP22]], shufflevector (<vscale x 2 x i64> insertelement (<vscale x 2 x i64> poison, i64 8, i64 0), <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer)
+; CHECK-NEXT: [[TMP23:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <vscale x 2 x i64> [[VECTOR_GEP7]]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8
+; CHECK-NEXT: [[TMP24:%.*]] = add i64 [[OFFSET_IDX]], 0
+; CHECK-NEXT: [[TMP25:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP26:%.*]] = mul i64 [[TMP25]], 2
+; CHECK-NEXT: [[TMP27:%.*]] = add i64 [[TMP26]], 0
; CHECK-NEXT: [[TMP28:%.*]] = mul i64 [[TMP27]], 8
-; CHECK-NEXT: [[NEXT_GEP8:%.*]] = getelementptr i8, ptr [[START_2]], i64 [[TMP28]]
-; CHECK-NEXT: [[TMP29:%.*]] = getelementptr i64, ptr [[NEXT_GEP]], i32 0
-; CHECK-NEXT: [[TMP30:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP31:%.*]] = mul i64 [[TMP30]], 2
-; CHECK-NEXT: [[TMP32:%.*]] = getelementptr i64, ptr [[NEXT_GEP]], i64 [[TMP31]]
-; CHECK-NEXT: store <vscale x 2 x i64> zeroinitializer, ptr [[TMP29]], align 8
+; CHECK-NEXT: [[TMP29:%.*]] = add i64 [[OFFSET_IDX]], [[TMP28]]
+; CHECK-NEXT: [[TMP30:%.*]] = getelementptr i8, ptr [[START_2]], i64 [[TMP24]]
+; CHECK-NEXT: [[TMP31:%.*]] = getelementptr i8, ptr [[START_2]], i64 [[TMP29]]
+; CHECK-NEXT: [[TMP32:%.*]] = getelementptr i64, ptr [[TMP30]], i32 0
+; CHECK-NEXT: [[TMP33:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP34:%.*]] = mul i64 [[TMP33]], 2
+; CHECK-NEXT: [[TMP35:%.*]] = getelementptr i64, ptr [[TMP30]], i64 [[TMP34]]
; CHECK-NEXT: store <vscale x 2 x i64> zeroinitializer, ptr [[TMP32]], align 8
-; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP34]]
-; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 [[TMP13]]
-; CHECK-NEXT: [[TMP35:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP35]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-NEXT: store <vscale x 2 x i64> zeroinitializer, ptr [[TMP35]], align 8
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP11]]
+; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 [[TMP15]]
+; CHECK-NEXT: [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP36]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
; CHECK-NEXT: [[CMO:%.*]] = sub i64 [[N_VEC]], 1
-; CHECK-NEXT: [[TMP36:%.*]] = mul i64 [[CMO]], 8
-; CHECK-NEXT: [[IND_ESCAPE:%.*]] = getelementptr i8, ptr [[START_1]], i64 [[TMP36]]
+; CHECK-NEXT: [[TMP37:%.*]] = mul i64 [[CMO]], 8
+; CHECK-NEXT: [[IND_ESCAPE:%.*]] = getelementptr i8, ptr [[START_1]], i64 [[TMP37]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[START_1]], [[ENTRY:%.*]] ]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-gep.ll b/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-gep.ll
index 899fcce5c02aba..3bab341e1c248a 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-gep.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-gep.ll
@@ -19,10 +19,12 @@ target triple = "aarch64-unknown-linux-gnu"
; CHECK-NEXT: <x1> vector loop: {
; CHECK-NEXT: vector.body:
; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
-; CHECK-NEXT: EMIT ir<%ptr.iv.1> = WIDEN-POINTER-INDUCTION ir<%start.1>, 8
; CHECK-NEXT: EMIT ir<%ptr.iv.2> = WIDEN-POINTER-INDUCTION ir<%start.2>, 1
+; CHECK-NEXT: vp<[[PTR_IDX:%.+]]> = DERIVED-IV ir<0> + vp<[[CAN_IV]]> * ir<8>
+; CHECK-NEXT: vp<[[PTR_IDX_STEPS:%.+]]> = SCALAR-STEPS vp<[[PTR_IDX]]>, ir<8>
+; CHECK-NEXT: EMIT vp<[[PTR_IV_1:%.+]]> = ptradd ir<%start.1>, vp<[[PTR_IDX_STEPS]]>
; CHECK-NEXT: WIDEN-GEP Var[Inv] ir<%ptr.iv.2.next> = getelementptr inbounds ir<%ptr.iv.2>, ir<1>
-; CHECK-NEXT: vp<[[VEC_PTR:%.+]]> = vector-pointer ir<%ptr.iv.1>
+; CHECK-NEXT: vp<[[VEC_PTR:%.+]]> = vector-pointer vp<[[PTR_IV_1]]>
; CHECK-NEXT: WIDEN store vp<[[VEC_PTR]]>, ir<%ptr.iv.2.next>
; CHECK-NEXT: vp<[[VEC_PTR2:%.+]]> = vector-pointer ir<%ptr.iv.2>
; CHECK-NEXT: WIDEN ir<%lv> = load vp<[[VEC_PTR2]]>
@@ -59,9 +61,6 @@ define void @pointer_induction_used_as_vector(ptr noalias %start.1, ptr noalias
; CHECK: vector.body:
; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[START_2]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[TMP6:%.*]] = mul i64 [[TMP5]], 8
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START_1]], i64 [[TMP6]]
; CHECK-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP8:%.*]] = mul i64 [[TMP7]], 2
; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 1
@@ -73,6 +72,9 @@ define void @pointer_induction_used_as_vector(ptr noalias %start.1, ptr noalias
; CHECK-NEXT: [[TMP13:%.*]] = add <vscale x 2 x i64> [[DOTSPLAT]], [[TMP12]]
; CHECK-NEXT: [[VECTOR_GEP:%.*]] = mul <vscale x 2 x i64> [[TMP13]], shufflevector (<vscale x 2 x i64> insertelement (<vscale x 2 x i64> poison, i64 1, i64 0), <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer)
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <vscale x 2 x i64> [[VECTOR_GEP]]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8
+; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 0
+; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START_1]], i64 [[TMP5]]
; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i8, <vscale x 2 x ptr> [[TMP14]], i64 1
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr ptr, ptr [[NEXT_GEP]], i32 0
; CHECK-NEXT: store <vscale x 2 x ptr> [[TMP15]], ptr [[TMP16]], align 8
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-phi.ll b/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-phi.ll
index 7226048c478d48..126ceac7325a44 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-phi.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-phi.ll
@@ -32,21 +32,20 @@ define void @widen_ptr_phi_unrolled(ptr noalias nocapture %a, ptr noalias nocapt
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[INDEX]], 3
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[C]], i64 [[TMP6]]
-; CHECK-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64()
-; CHECK-NEXT: [[TMP8:%.*]] = shl nuw nsw i64 [[TMP7]], 5
-; CHECK-NEXT: [[TMP9:%.*]] = shl i64 [[INDEX]], 3
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[C]], i64 [[TMP8]]
-; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[TMP10]], i64 [[TMP9]]
-; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 8 x i32>, ptr [[NEXT_GEP]], align 4
-; CHECK-NEXT: [[WIDE_VEC3:%.*]] = load <vscale x 8 x i32>, ptr [[NEXT_GEP2]], align 4
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 3
+; CHECK-NEXT: [[TMP6:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP7:%.*]] = shl nuw nsw i64 [[TMP6]], 5
+; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[C]], i64 [[OFFSET_IDX]]
+; CHECK-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[C]], i64 [[OFFSET_IDX]]
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[TMP9]], i64 [[TMP7]]
+; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 8 x i32>, ptr [[TMP8]], align 4
+; CHECK-NEXT: [[WIDE_VEC2:%.*]] = load <vscale x 8 x i32>, ptr [[TMP10]], align 4
; CHECK-NEXT: [[STRIDED_VEC:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.experimental.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> [[WIDE_VEC]])
; CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC]], 0
; CHECK-NEXT: [[TMP12:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC]], 1
-; CHECK-NEXT: [[STRIDED_VEC4:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.experimental.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> [[WIDE_VEC3]])
-; CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC4]], 0
-; CHECK-NEXT: [[TMP14:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC4]], 1
+; CHECK-NEXT: [[STRIDED_VEC3:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.experimental.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> [[WIDE_VEC2]])
+; CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC3]], 0
+; CHECK-NEXT: [[TMP14:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[STRIDED_VEC3]], 1
; CHECK-NEXT: [[TMP15:%.*]] = add nsw <vscale x 4 x i32> [[TMP11]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 1, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer)
; CHECK-NEXT: [[TMP16:%.*]] = add nsw <vscale x 4 x i32> [[TMP13]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 1, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer)
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]]
@@ -148,21 +147,21 @@ define void @widen_2ptrs_phi_unrolled(ptr noalias nocapture %dst, ptr noalias no
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP7:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP7]]
-; CHECK-NEXT: [[TMP8:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP8]]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 2
+; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[OFFSET_IDX]]
+; CHECK-NEXT: [[OFFSET_IDX4:%.*]] = shl i64 [[INDEX]], 2
+; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[DST]], i64 [[OFFSET_IDX4]]
; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP10:%.*]] = shl nuw nsw i64 [[TMP9]], 2
-; CHECK-NEXT: [[TMP11:%.*]] = getelementptr i32, ptr [[NEXT_GEP]], i64 [[TMP10]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[NEXT_GEP]], align 4
-; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <vscale x 4 x i32>, ptr [[TMP11]], align 4
+; CHECK-NEXT: [[TMP11:%.*]] = getelementptr i32, ptr [[TMP7]], i64 [[TMP10]]
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 4 x i32>, ptr [[TMP7]], align 4
+; CHECK-NEXT: [[WIDE_LOAD5:%.*]] = load <vscale x 4 x i32>, ptr [[TMP11]], align 4
; CHECK-NEXT: [[TMP12:%.*]] = shl nsw <vscale x 4 x i32> [[WIDE_LOAD]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 1, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer)
-; CHECK-NEXT: [[TMP13:%.*]] = shl nsw <vscale x 4 x i32> [[WIDE_LOAD7]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 1, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer)
+; CHECK-NEXT: [[TMP13:%.*]] = shl nsw <vscale x 4 x i32> [[WIDE_LOAD5]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 1, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer)
; CHECK-NEXT: [[TMP14:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP15:%.*]] = shl nuw nsw i64 [[TMP14]], 2
-; CHECK-NEXT: [[TMP16:%.*]] = getelementptr i32, ptr [[NEXT_GEP5]], i64 [[TMP15]]
-; CHECK-NEXT: store <vscale x 4 x i32> [[TMP12]], ptr [[NEXT_GEP5]], align 4
+; CHECK-NEXT: [[TMP16:%.*]] = getelementptr i32, ptr [[TMP8]], i64 [[TMP15]]
+; CHECK-NEXT: store <vscale x 4 x i32> [[TMP12]], ptr [[TMP8]], align 4
; CHECK-NEXT: store <vscale x 4 x i32> [[TMP13]], ptr [[TMP16]], align 4
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]
; CHECK-NEXT: [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
@@ -246,12 +245,12 @@ define i32 @pointer_iv_mixed(ptr noalias %a, ptr noalias %b, i64 %n) #0 {
; CHECK-NEXT: [[TMP9:%.*]] = call <vscale x 2 x i64> @llvm.experimental.stepvector.nxv2i64()
; CHECK-NEXT: [[VECTOR_GEP:%.*]] = shl <vscale x 2 x i64> [[TMP9]], shufflevector (<vscale x 2 x i64> insertelement (<vscale x 2 x i64> poison, i64 2, i64 0), <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer)
; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <vscale x 2 x i64> [[VECTOR_GEP]]
-; CHECK-NEXT: [[TMP11:%.*]] = shl i64 [[INDEX]], 3
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP11]]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 3
+; CHECK-NEXT: [[TMP11:%.*]] = getelementptr i8, ptr [[B]], i64 [[OFFSET_IDX]]
; CHECK-NEXT: [[TMP12:%.*]] = extractelement <vscale x 2 x ptr> [[TMP10]], i64 0
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 2 x i32>, ptr [[TMP12]], align 8
; CHECK-NEXT: [[TMP13]] = add <vscale x 2 x i32> [[WIDE_LOAD]], [[VEC_PHI]]
-; CHECK-NEXT: store <vscale x 2 x ptr> [[TMP10]], ptr [[NEXT_GEP]], align 8
+; CHECK-NEXT: store <vscale x 2 x ptr> [[TMP10]], ptr [[TMP11]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]
; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 [[TMP8]]
; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
diff --git a/llvm/test/Transforms/LoopVectorize/X86/gather_scatter.ll b/llvm/test/Transforms/LoopVectorize/X86/gather_scatter.ll
index e9541c1ee035f2..6516b05ab4ede9 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/gather_scatter.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/gather_scatter.ll
@@ -639,87 +639,84 @@ define void @test_gather_not_profitable_pr48429(i32 %d, ptr readonly noalias %pt
; AVX512: vector.ph:
; AVX512-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 16
; AVX512-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]
-; AVX512-NEXT: [[TMP13:%.*]] = mul i64 [[N_VEC]], 4
-; AVX512-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP13]]
-; AVX512-NEXT: [[TMP14:%.*]] = mul i64 [[N_VEC]], 64
-; AVX512-NEXT: [[IND_END9:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP14]]
+; AVX512-NEXT: [[TMP13:%.*]] = mul i64 [[N_VEC]], 64
+; AVX512-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP13]]
; AVX512-NEXT: br label [[VECTOR_BODY:%.*]]
; AVX512: vector.body:
; AVX512-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[DEST]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]
; AVX512-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; AVX512-NEXT: [[TMP15:%.*]] = add i64 [[INDEX]], 0
-; AVX512-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 4
-; AVX512-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP16]]
-; AVX512-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <16 x i64> <i64 0, i64 64, i64 128, i64 192, i64 256, i64 320, i64 384, i64 448, i64 512, i64 576, i64 640, i64 704, i64 768, i64 832, i64 896, i64 960>
-; AVX512-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, ptr [[NEXT_GEP]], i64 [[IDXPROM]]
-; AVX512-NEXT: [[TMP19:%.*]] = getelementptr inbounds float, ptr [[TMP18]], i32 0
-; AVX512-NEXT: [[WIDE_LOAD:%.*]] = load <16 x float>, ptr [[TMP19]], align 4, !alias.scope !8
-; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0(<16 x float> [[WIDE_LOAD]], <16 x ptr> [[TMP17]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !11, !noalias !13
-; AVX512-NEXT: [[TMP20:%.*]] = getelementptr float, ptr [[NEXT_GEP]], i32 0
-; AVX512-NEXT: [[WIDE_LOAD8:%.*]] = load <16 x float>, ptr [[TMP20]], align 4, !alias.scope !15
-; AVX512-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, <16 x ptr> [[TMP17]], i64 1
-; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0(<16 x float> [[WIDE_LOAD8]], <16 x ptr> [[TMP21]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !11, !noalias !13
+; AVX512-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <16 x i64> <i64 0, i64 64, i64 128, i64 192, i64 256, i64 320, i64 384, i64 448, i64 512, i64 576, i64 640, i64 704, i64 768, i64 832, i64 896, i64 960>
+; AVX512-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 4
+; AVX512-NEXT: [[TMP15:%.*]] = add i64 [[OFFSET_IDX]], 0
+; AVX512-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP15]]
+; AVX512-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, ptr [[TMP16]], i64 [[IDXPROM]]
+; AVX512-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, ptr [[TMP17]], i32 0
+; AVX512-NEXT: [[WIDE_LOAD:%.*]] = load <16 x float>, ptr [[TMP18]], align 4, !alias.scope [[META8:![0-9]+]]
+; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0(<16 x float> [[WIDE_LOAD]], <16 x ptr> [[TMP14]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope [[META11:![0-9]+]], !noalias [[META13:![0-9]+]]
+; AVX512-NEXT: [[TMP19:%.*]] = getelementptr float, ptr [[TMP16]], i32 0
+; AVX512-NEXT: [[WIDE_LOAD8:%.*]] = load <16 x float>, ptr [[TMP19]], align 4, !alias.scope [[META15:![0-9]+]]
+; AVX512-NEXT: [[TMP20:%.*]] = getelementptr inbounds float, <16 x ptr> [[TMP14]], i64 1
+; AVX512-NEXT: call void @llvm.masked.scatter.v16f32.v16p0(<16 x float> [[WIDE_LOAD8]], <16 x ptr> [[TMP20]], i32 4, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope [[META11]], !noalias [[META13]]
; AVX512-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
; AVX512-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 1024
-; AVX512-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; AVX512-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
+; AVX512-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; AVX512-NEXT: br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; AVX512: middle.block:
; AVX512-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
; AVX512-NEXT: br i1 [[CMP_N]], label [[FOR_END]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
; AVX512: vec.epilog.iter.check:
-; AVX512-NEXT: [[TMP23:%.*]] = mul i64 [[N_VEC]], 64
-; AVX512-NEXT: [[IND_END17:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP23]]
-; AVX512-NEXT: [[TMP24:%.*]] = mul i64 [[N_VEC]], 4
-; AVX512-NEXT: [[IND_END14:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP24]]
+; AVX512-NEXT: [[TMP22:%.*]] = mul i64 [[N_VEC]], 64
+; AVX512-NEXT: [[IND_END15:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP22]]
+; AVX512-NEXT: [[TMP23:%.*]] = mul i64 [[N_VEC]], 4
+; AVX512-NEXT: [[IND_END12:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP23]]
; AVX512-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[TMP3]], [[N_VEC]]
; AVX512-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 8
; AVX512-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
; AVX512: vec.epilog.ph:
-; AVX512-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PTR]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
-; AVX512-NEXT: [[BC_RESUME_VAL10:%.*]] = phi ptr [ [[IND_END9]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[DEST]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
+; AVX512-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[DEST]], [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; AVX512-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
-; AVX512-NEXT: [[N_MOD_VF11:%.*]] = urem i64 [[TMP3]], 8
-; AVX512-NEXT: [[N_VEC12:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF11]]
-; AVX512-NEXT: [[TMP25:%.*]] = mul i64 [[N_VEC12]], 4
-; AVX512-NEXT: [[IND_END13:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP25]]
-; AVX512-NEXT: [[TMP26:%.*]] = mul i64 [[N_VEC12]], 64
-; AVX512-NEXT: [[IND_END16:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP26]]
+; AVX512-NEXT: [[N_MOD_VF9:%.*]] = urem i64 [[TMP3]], 8
+; AVX512-NEXT: [[N_VEC10:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF9]]
+; AVX512-NEXT: [[TMP24:%.*]] = mul i64 [[N_VEC10]], 4
+; AVX512-NEXT: [[IND_END11:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP24]]
+; AVX512-NEXT: [[TMP25:%.*]] = mul i64 [[N_VEC10]], 64
+; AVX512-NEXT: [[IND_END14:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP25]]
; AVX512-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; AVX512: vec.epilog.vector.body:
-; AVX512-NEXT: [[POINTER_PHI22:%.*]] = phi ptr [ [[BC_RESUME_VAL10]], [[VEC_EPILOG_PH]] ], [ [[PTR_IND23:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
-; AVX512-NEXT: [[INDEX20:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT26:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
-; AVX512-NEXT: [[TMP27:%.*]] = add i64 [[INDEX20]], 0
-; AVX512-NEXT: [[TMP28:%.*]] = mul i64 [[TMP27]], 4
-; AVX512-NEXT: [[NEXT_GEP21:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP28]]
-; AVX512-NEXT: [[TMP29:%.*]] = getelementptr i8, ptr [[POINTER_PHI22]], <8 x i64> <i64 0, i64 64, i64 128, i64 192, i64 256, i64 320, i64 384, i64 448>
-; AVX512-NEXT: [[TMP30:%.*]] = getelementptr inbounds float, ptr [[NEXT_GEP21]], i64 [[IDXPROM]]
-; AVX512-NEXT: [[TMP31:%.*]] = getelementptr inbounds float, ptr [[TMP30]], i32 0
-; AVX512-NEXT: [[WIDE_LOAD24:%.*]] = load <8 x float>, ptr [[TMP31]], align 4, !alias.scope !17
-; AVX512-NEXT: call void @llvm.masked.scatter.v8f32.v8p0(<8 x float> [[WIDE_LOAD24]], <8 x ptr> [[TMP29]], i32 4, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !20, !noalias !22
-; AVX512-NEXT: [[TMP32:%.*]] = getelementptr float, ptr [[NEXT_GEP21]], i32 0
-; AVX512-NEXT: [[WIDE_LOAD25:%.*]] = load <8 x float>, ptr [[TMP32]], align 4, !alias.scope !24
-; AVX512-NEXT: [[TMP33:%.*]] = getelementptr inbounds float, <8 x ptr> [[TMP29]], i64 1
-; AVX512-NEXT: call void @llvm.masked.scatter.v8f32.v8p0(<8 x float> [[WIDE_LOAD25]], <8 x ptr> [[TMP33]], i32 4, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope !20, !noalias !22
-; AVX512-NEXT: [[INDEX_NEXT26]] = add nuw i64 [[INDEX20]], 8
-; AVX512-NEXT: [[PTR_IND23]] = getelementptr i8, ptr [[POINTER_PHI22]], i64 512
-; AVX512-NEXT: [[TMP34:%.*]] = icmp eq i64 [[INDEX_NEXT26]], [[N_VEC12]]
-; AVX512-NEXT: br i1 [[TMP34]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP25:![0-9]+]]
+; AVX512-NEXT: [[POINTER_PHI19:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[PTR_IND20:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
+; AVX512-NEXT: [[INDEX18:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT24:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
+; AVX512-NEXT: [[TMP26:%.*]] = getelementptr i8, ptr [[POINTER_PHI19]], <8 x i64> <i64 0, i64 64, i64 128, i64 192, i64 256, i64 320, i64 384, i64 448>
+; AVX512-NEXT: [[OFFSET_IDX21:%.*]] = mul i64 [[INDEX18]], 4
+; AVX512-NEXT: [[TMP27:%.*]] = add i64 [[OFFSET_IDX21]], 0
+; AVX512-NEXT: [[TMP28:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP27]]
+; AVX512-NEXT: [[TMP29:%.*]] = getelementptr inbounds float, ptr [[TMP28]], i64 [[IDXPROM]]
+; AVX512-NEXT: [[TMP30:%.*]] = getelementptr inbounds float, ptr [[TMP29]], i32 0
+; AVX512-NEXT: [[WIDE_LOAD22:%.*]] = load <8 x float>, ptr [[TMP30]], align 4, !alias.scope [[META17:![0-9]+]]
+; AVX512-NEXT: call void @llvm.masked.scatter.v8f32.v8p0(<8 x float> [[WIDE_LOAD22]], <8 x ptr> [[TMP26]], i32 4, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope [[META20:![0-9]+]], !noalias [[META22:![0-9]+]]
+; AVX512-NEXT: [[TMP31:%.*]] = getelementptr float, ptr [[TMP28]], i32 0
+; AVX512-NEXT: [[WIDE_LOAD23:%.*]] = load <8 x float>, ptr [[TMP31]], align 4, !alias.scope [[META24:![0-9]+]]
+; AVX512-NEXT: [[TMP32:%.*]] = getelementptr inbounds float, <8 x ptr> [[TMP26]], i64 1
+; AVX512-NEXT: call void @llvm.masked.scatter.v8f32.v8p0(<8 x float> [[WIDE_LOAD23]], <8 x ptr> [[TMP32]], i32 4, <8 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>), !alias.scope [[META20]], !noalias [[META22]]
+; AVX512-NEXT: [[INDEX_NEXT24]] = add nuw i64 [[INDEX18]], 8
+; AVX512-NEXT: [[PTR_IND20]] = getelementptr i8, ptr [[POINTER_PHI19]], i64 512
+; AVX512-NEXT: [[TMP33:%.*]] = icmp eq i64 [[INDEX_NEXT24]], [[N_VEC10]]
+; AVX512-NEXT: br i1 [[TMP33]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP25:![0-9]+]]
; AVX512: vec.epilog.middle.block:
-; AVX512-NEXT: [[CMP_N19:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC12]]
-; AVX512-NEXT: br i1 [[CMP_N19]], label [[FOR_END]], label [[VEC_EPILOG_SCALAR_PH]]
+; AVX512-NEXT: [[CMP_N17:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC10]]
+; AVX512-NEXT: br i1 [[CMP_N17]], label [[FOR_END]], label [[VEC_EPILOG_SCALAR_PH]]
; AVX512: vec.epilog.scalar.ph:
-; AVX512-NEXT: [[BC_RESUME_VAL15:%.*]] = phi ptr [ [[IND_END13]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END14]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PTR]], [[VECTOR_MEMCHECK]] ], [ [[PTR]], [[ITER_CHECK]] ]
-; AVX512-NEXT: [[BC_RESUME_VAL18:%.*]] = phi ptr [ [[IND_END16]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END17]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[DEST]], [[VECTOR_MEMCHECK]] ], [ [[DEST]], [[ITER_CHECK]] ]
+; AVX512-NEXT: [[BC_RESUME_VAL13:%.*]] = phi ptr [ [[IND_END11]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END12]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PTR]], [[VECTOR_MEMCHECK]] ], [ [[PTR]], [[ITER_CHECK]] ]
+; AVX512-NEXT: [[BC_RESUME_VAL16:%.*]] = phi ptr [ [[IND_END14]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END15]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[DEST]], [[VECTOR_MEMCHECK]] ], [ [[DEST]], [[ITER_CHECK]] ]
; AVX512-NEXT: br label [[FOR_BODY:%.*]]
; AVX512: for.body:
-; AVX512-NEXT: [[PTR_ADDR_012:%.*]] = phi ptr [ [[BC_RESUME_VAL15]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ]
-; AVX512-NEXT: [[DEST_ADDR_011:%.*]] = phi ptr [ [[BC_RESUME_VAL18]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[ADD_PTR6:%.*]], [[FOR_BODY]] ]
+; AVX512-NEXT: [[PTR_ADDR_012:%.*]] = phi ptr [ [[BC_RESUME_VAL13]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ]
+; AVX512-NEXT: [[DEST_ADDR_011:%.*]] = phi ptr [ [[BC_RESUME_VAL16]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[ADD_PTR6:%.*]], [[FOR_BODY]] ]
; AVX512-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[PTR_ADDR_012]], i64 [[IDXPROM]]
-; AVX512-NEXT: [[TMP35:%.*]] = load float, ptr [[ARRAYIDX]], align 4
-; AVX512-NEXT: store float [[TMP35]], ptr [[DEST_ADDR_011]], align 4
-; AVX512-NEXT: [[TMP36:%.*]] = load float, ptr [[PTR_ADDR_012]], align 4
+; AVX512-NEXT: [[TMP34:%.*]] = load float, ptr [[ARRAYIDX]], align 4
+; AVX512-NEXT: store float [[TMP34]], ptr [[DEST_ADDR_011]], align 4
+; AVX512-NEXT: [[TMP35:%.*]] = load float, ptr [[PTR_ADDR_012]], align 4
; AVX512-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds float, ptr [[DEST_ADDR_011]], i64 1
-; AVX512-NEXT: store float [[TMP36]], ptr [[ARRAYIDX5]], align 4
+; AVX512-NEXT: store float [[TMP35]], ptr [[ARRAYIDX5]], align 4
; AVX512-NEXT: [[INCDEC_PTR]] = getelementptr inbounds float, ptr [[PTR_ADDR_012]], i64 1
; AVX512-NEXT: [[ADD_PTR6]] = getelementptr inbounds float, ptr [[DEST_ADDR_011]], i64 16
; AVX512-NEXT: [[CMP_NOT:%.*]] = icmp eq ptr [[INCDEC_PTR]], [[ADD_PTR]]
@@ -774,30 +771,29 @@ define void @test_gather_not_profitable_pr48429(i32 %d, ptr readonly noalias %pt
; FVW2-NEXT: br label [[VECTOR_BODY:%.*]]
; FVW2: vector.body:
; FVW2-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; FVW2-NEXT: [[TMP15:%.*]] = add i64 [[INDEX]], 0
-; FVW2-NEXT: [[TMP16:%.*]] = mul i64 [[TMP15]], 4
-; FVW2-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP16]]
-; FVW2-NEXT: [[TMP17:%.*]] = add i64 [[INDEX]], 0
-; FVW2-NEXT: [[TMP18:%.*]] = mul i64 [[TMP17]], 64
-; FVW2-NEXT: [[NEXT_GEP9:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP18]]
-; FVW2-NEXT: [[TMP19:%.*]] = add i64 [[INDEX]], 1
-; FVW2-NEXT: [[TMP20:%.*]] = mul i64 [[TMP19]], 64
-; FVW2-NEXT: [[NEXT_GEP10:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP20]]
-; FVW2-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, ptr [[NEXT_GEP]], i64 [[IDXPROM]]
+; FVW2-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 4
+; FVW2-NEXT: [[TMP15:%.*]] = add i64 [[OFFSET_IDX]], 0
+; FVW2-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[PTR]], i64 [[TMP15]]
+; FVW2-NEXT: [[OFFSET_IDX9:%.*]] = mul i64 [[INDEX]], 64
+; FVW2-NEXT: [[TMP17:%.*]] = add i64 [[OFFSET_IDX9]], 0
+; FVW2-NEXT: [[TMP18:%.*]] = add i64 [[OFFSET_IDX9]], 64
+; FVW2-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP17]]
+; FVW2-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[DEST]], i64 [[TMP18]]
+; FVW2-NEXT: [[TMP21:%.*]] = getelementptr inbounds float, ptr [[TMP16]], i64 [[IDXPROM]]
; FVW2-NEXT: [[TMP22:%.*]] = getelementptr inbounds float, ptr [[TMP21]], i32 0
-; FVW2-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP22]], align 4, !alias.scope !8
+; FVW2-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP22]], align 4, !alias.scope [[META8:![0-9]+]]
; FVW2-NEXT: [[TMP23:%.*]] = extractelement <2 x float> [[WIDE_LOAD]], i32 0
-; FVW2-NEXT: store float [[TMP23]], ptr [[NEXT_GEP9]], align 4, !alias.scope !11, !noalias !13
+; FVW2-NEXT: store float [[TMP23]], ptr [[TMP19]], align 4, !alias.scope [[META11:![0-9]+]], !noalias [[META13:![0-9]+]]
; FVW2-NEXT: [[TMP24:%.*]] = extractelement <2 x float> [[WIDE_LOAD]], i32 1
-; FVW2-NEXT: store float [[TMP24]], ptr [[NEXT_GEP10]], align 4, !alias.scope !11, !noalias !13
-; FVW2-NEXT: [[TMP25:%.*]] = getelementptr float, ptr [[NEXT_GEP]], i32 0
-; FVW2-NEXT: [[WIDE_LOAD11:%.*]] = load <2 x float>, ptr [[TMP25]], align 4, !alias.scope !15
-; FVW2-NEXT: [[TMP26:%.*]] = getelementptr inbounds float, ptr [[NEXT_GEP9]], i64 1
-; FVW2-NEXT: [[TMP27:%.*]] = getelementptr inbounds float, ptr [[NEXT_GEP10]], i64 1
-; FVW2-NEXT: [[TMP28:%.*]] = extractelement <2 x float> [[WIDE_LOAD11]], i32 0
-; FVW2-NEXT: store float [[TMP28]], ptr [[TMP26]], align 4, !alias.scope !11, !noalias !13
-; FVW2-NEXT: [[TMP29:%.*]] = extractelement <2 x float> [[WIDE_LOAD11]], i32 1
-; FVW2-NEXT: store float [[TMP29]], ptr [[TMP27]], align 4, !alias.scope !11, !noalias !13
+; FVW2-NEXT: store float [[TMP24]], ptr [[TMP20]], align 4, !alias.scope [[META11]], !noalias [[META13]]
+; FVW2-NEXT: [[TMP25:%.*]] = getelementptr float, ptr [[TMP16]], i32 0
+; FVW2-NEXT: [[WIDE_LOAD10:%.*]] = load <2 x float>, ptr [[TMP25]], align 4, !alias.scope [[META15:![0-9]+]]
+; FVW2-NEXT: [[TMP26:%.*]] = getelementptr inbounds float, ptr [[TMP19]], i64 1
+; FVW2-NEXT: [[TMP27:%.*]] = getelementptr inbounds float, ptr [[TMP20]], i64 1
+; FVW2-NEXT: [[TMP28:%.*]] = extractelement <2 x float> [[WIDE_LOAD10]], i32 0
+; FVW2-NEXT: store float [[TMP28]], ptr [[TMP26]], align 4, !alias.scope [[META11]], !noalias [[META13]]
+; FVW2-NEXT: [[TMP29:%.*]] = extractelement <2 x float> [[WIDE_LOAD10]], i32 1
+; FVW2-NEXT: store float [[TMP29]], ptr [[TMP27]], align 4, !alias.scope [[META11]], !noalias [[META13]]
; FVW2-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
; FVW2-NEXT: [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; FVW2-NEXT: br i1 [[TMP30]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
diff --git a/llvm/test/Transforms/LoopVectorize/X86/interleave-opaque-pointers.ll b/llvm/test/Transforms/LoopVectorize/X86/interleave-opaque-pointers.ll
index 022912f3b8552e..1b0118e137e735 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/interleave-opaque-pointers.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/interleave-opaque-pointers.ll
@@ -25,15 +25,14 @@ define void @test_pr55375_interleave_opaque_ptr(ptr %start, ptr %end) {
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[TMP6:%.*]] = mul i64 [[TMP5]], 16
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP6]]
-; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[INDEX]], 1
-; CHECK-NEXT: [[TMP8:%.*]] = mul i64 [[TMP7]], 16
-; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP8]]
-; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x ptr> poison, ptr [[NEXT_GEP]], i32 0
-; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x ptr> [[TMP9]], ptr [[NEXT_GEP3]], i32 1
-; CHECK-NEXT: [[TMP11:%.*]] = getelementptr ptr, ptr [[NEXT_GEP]], i32 0
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 16
+; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 0
+; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 16
+; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP5]]
+; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP6]]
+; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x ptr> poison, ptr [[TMP7]], i32 0
+; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x ptr> [[TMP9]], ptr [[TMP8]], i32 1
+; CHECK-NEXT: [[TMP11:%.*]] = getelementptr ptr, ptr [[TMP7]], i32 0
; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <2 x ptr> zeroinitializer, <2 x ptr> [[TMP10]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
; CHECK-NEXT: [[INTERLEAVED_VEC:%.*]] = shufflevector <4 x ptr> [[TMP12]], <4 x ptr> poison, <4 x i32> <i32 0, i32 2, i32 1, i32 3>
; CHECK-NEXT: store <4 x ptr> [[INTERLEAVED_VEC]], ptr [[TMP11]], align 8
@@ -53,7 +52,7 @@ define void @test_pr55375_interleave_opaque_ptr(ptr %start, ptr %end) {
; CHECK-NEXT: store ptr null, ptr [[IV]], align 8
; CHECK-NEXT: [[IV_NEXT]] = getelementptr inbounds [[PAIR]], ptr [[IV]], i64 1
; CHECK-NEXT: [[EC:%.*]] = icmp eq ptr [[IV_NEXT]], [[END]]
-; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP2:![0-9]+]]
+; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK: exit:
; CHECK-NEXT: ret void
;
diff --git a/llvm/test/Transforms/LoopVectorize/X86/small-size.ll b/llvm/test/Transforms/LoopVectorize/X86/small-size.ll
index be83329d30fef9..317f2f89732149 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/small-size.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/small-size.ll
@@ -273,64 +273,58 @@ define void @example3(i32 %n, ptr noalias nocapture %p, ptr noalias nocapture %q
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE19:%.*]] ]
-; CHECK-NEXT: [[BROADCAST_SPLATINSERT12:%.*]] = insertelement <4 x i64> poison, i64 [[INDEX]], i64 0
-; CHECK-NEXT: [[BROADCAST_SPLAT13:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT12]], <4 x i64> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT: [[VEC_IV:%.*]] = or disjoint <4 x i64> [[BROADCAST_SPLAT13]], <i64 0, i64 1, i64 2, i64 3>
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE20:%.*]] ]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 2
+; CHECK-NEXT: [[OFFSET_IDX8:%.*]] = shl i64 [[INDEX]], 2
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT13:%.*]] = insertelement <4 x i64> poison, i64 [[INDEX]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT14:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT13]], <4 x i64> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT: [[VEC_IV:%.*]] = or disjoint <4 x i64> [[BROADCAST_SPLAT14]], <i64 0, i64 1, i64 2, i64 3>
; CHECK-NEXT: [[TMP3:%.*]] = icmp ule <4 x i64> [[VEC_IV]], [[BROADCAST_SPLAT]]
; CHECK-NEXT: [[TMP4:%.*]] = extractelement <4 x i1> [[TMP3]], i64 0
; CHECK-NEXT: br i1 [[TMP4]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; CHECK: pred.store.if:
-; CHECK-NEXT: [[TMP5:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[P:%.*]], i64 [[TMP5]]
-; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[NEXT_GEP8:%.*]] = getelementptr i8, ptr [[Q:%.*]], i64 [[TMP6]]
-; CHECK-NEXT: [[TMP7:%.*]] = load i32, ptr [[NEXT_GEP8]], align 16
-; CHECK-NEXT: store i32 [[TMP7]], ptr [[NEXT_GEP]], align 16
+; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[P:%.*]], i64 [[OFFSET_IDX]]
+; CHECK-NEXT: [[NEXT_GEP9:%.*]] = getelementptr i8, ptr [[Q:%.*]], i64 [[OFFSET_IDX8]]
+; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr [[NEXT_GEP9]], align 16
+; CHECK-NEXT: store i32 [[TMP5]], ptr [[NEXT_GEP]], align 16
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]]
; CHECK: pred.store.continue:
-; CHECK-NEXT: [[TMP8:%.*]] = extractelement <4 x i1> [[TMP3]], i64 1
-; CHECK-NEXT: br i1 [[TMP8]], label [[PRED_STORE_IF14:%.*]], label [[PRED_STORE_CONTINUE15:%.*]]
-; CHECK: pred.store.if14:
-; CHECK-NEXT: [[TMP9:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP10:%.*]] = or disjoint i64 [[TMP9]], 4
-; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP10]]
-; CHECK-NEXT: [[TMP11:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP12:%.*]] = or disjoint i64 [[TMP11]], 4
-; CHECK-NEXT: [[NEXT_GEP9:%.*]] = getelementptr i8, ptr [[Q]], i64 [[TMP12]]
-; CHECK-NEXT: [[TMP13:%.*]] = load i32, ptr [[NEXT_GEP9]], align 16
-; CHECK-NEXT: store i32 [[TMP13]], ptr [[NEXT_GEP5]], align 16
-; CHECK-NEXT: br label [[PRED_STORE_CONTINUE15]]
-; CHECK: pred.store.continue15:
-; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x i1> [[TMP3]], i64 2
-; CHECK-NEXT: br i1 [[TMP14]], label [[PRED_STORE_IF16:%.*]], label [[PRED_STORE_CONTINUE17:%.*]]
-; CHECK: pred.store.if16:
-; CHECK-NEXT: [[TMP15:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP16:%.*]] = or disjoint i64 [[TMP15]], 8
-; CHECK-NEXT: [[NEXT_GEP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP16]]
-; CHECK-NEXT: [[TMP17:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP18:%.*]] = or disjoint i64 [[TMP17]], 8
-; CHECK-NEXT: [[NEXT_GEP10:%.*]] = getelementptr i8, ptr [[Q]], i64 [[TMP18]]
-; CHECK-NEXT: [[TMP19:%.*]] = load i32, ptr [[NEXT_GEP10]], align 16
-; CHECK-NEXT: store i32 [[TMP19]], ptr [[NEXT_GEP6]], align 16
-; CHECK-NEXT: br label [[PRED_STORE_CONTINUE17]]
-; CHECK: pred.store.continue17:
-; CHECK-NEXT: [[TMP20:%.*]] = extractelement <4 x i1> [[TMP3]], i64 3
-; CHECK-NEXT: br i1 [[TMP20]], label [[PRED_STORE_IF18:%.*]], label [[PRED_STORE_CONTINUE19]]
-; CHECK: pred.store.if18:
-; CHECK-NEXT: [[TMP21:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP22:%.*]] = or disjoint i64 [[TMP21]], 12
-; CHECK-NEXT: [[NEXT_GEP7:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP22]]
-; CHECK-NEXT: [[TMP23:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP24:%.*]] = or disjoint i64 [[TMP23]], 12
-; CHECK-NEXT: [[NEXT_GEP11:%.*]] = getelementptr i8, ptr [[Q]], i64 [[TMP24]]
-; CHECK-NEXT: [[TMP25:%.*]] = load i32, ptr [[NEXT_GEP11]], align 16
-; CHECK-NEXT: store i32 [[TMP25]], ptr [[NEXT_GEP7]], align 16
-; CHECK-NEXT: br label [[PRED_STORE_CONTINUE19]]
-; CHECK: pred.store.continue19:
+; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x i1> [[TMP3]], i64 1
+; CHECK-NEXT: br i1 [[TMP6]], label [[PRED_STORE_IF15:%.*]], label [[PRED_STORE_CONTINUE16:%.*]]
+; CHECK: pred.store.if15:
+; CHECK-NEXT: [[TMP7:%.*]] = or disjoint i64 [[OFFSET_IDX]], 4
+; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP7]]
+; CHECK-NEXT: [[TMP8:%.*]] = or disjoint i64 [[OFFSET_IDX8]], 4
+; CHECK-NEXT: [[NEXT_GEP10:%.*]] = getelementptr i8, ptr [[Q]], i64 [[TMP8]]
+; CHECK-NEXT: [[TMP9:%.*]] = load i32, ptr [[NEXT_GEP10]], align 16
+; CHECK-NEXT: store i32 [[TMP9]], ptr [[NEXT_GEP5]], align 16
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE16]]
+; CHECK: pred.store.continue16:
+; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x i1> [[TMP3]], i64 2
+; CHECK-NEXT: br i1 [[TMP10]], label [[PRED_STORE_IF17:%.*]], label [[PRED_STORE_CONTINUE18:%.*]]
+; CHECK: pred.store.if17:
+; CHECK-NEXT: [[TMP11:%.*]] = or disjoint i64 [[OFFSET_IDX]], 8
+; CHECK-NEXT: [[NEXT_GEP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP11]]
+; CHECK-NEXT: [[TMP12:%.*]] = or disjoint i64 [[OFFSET_IDX8]], 8
+; CHECK-NEXT: [[NEXT_GEP11:%.*]] = getelementptr i8, ptr [[Q]], i64 [[TMP12]]
+; CHECK-NEXT: [[TMP13:%.*]] = load i32, ptr [[NEXT_GEP11]], align 16
+; CHECK-NEXT: store i32 [[TMP13]], ptr [[NEXT_GEP6]], align 16
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE18]]
+; CHECK: pred.store.continue18:
+; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x i1> [[TMP3]], i64 3
+; CHECK-NEXT: br i1 [[TMP14]], label [[PRED_STORE_IF19:%.*]], label [[PRED_STORE_CONTINUE20]]
+; CHECK: pred.store.if19:
+; CHECK-NEXT: [[TMP15:%.*]] = or disjoint i64 [[OFFSET_IDX]], 12
+; CHECK-NEXT: [[NEXT_GEP7:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP15]]
+; CHECK-NEXT: [[TMP16:%.*]] = or disjoint i64 [[OFFSET_IDX8]], 12
+; CHECK-NEXT: [[NEXT_GEP12:%.*]] = getelementptr i8, ptr [[Q]], i64 [[TMP16]]
+; CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr [[NEXT_GEP12]], align 16
+; CHECK-NEXT: store i32 [[TMP17]], ptr [[NEXT_GEP7]], align 16
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE20]]
+; CHECK: pred.store.continue20:
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4
-; CHECK-NEXT: [[TMP26:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP26]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
+; CHECK-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: br i1 true, label [[DOT_CRIT_EDGE_LOOPEXIT:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
@@ -410,24 +404,24 @@ define void @example23b(ptr noalias nocapture %src, ptr noalias nocapture %dst)
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[INDEX]], 1
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[SRC:%.*]], i64 [[TMP1]]
-; CHECK-NEXT: [[TMP2:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[DST:%.*]], i64 [[TMP2]]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 1
+; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[SRC:%.*]], i64 [[OFFSET_IDX]]
+; CHECK-NEXT: [[OFFSET_IDX4:%.*]] = shl i64 [[INDEX]], 2
+; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[DST:%.*]], i64 [[OFFSET_IDX4]]
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i16>, ptr [[NEXT_GEP]], align 2
-; CHECK-NEXT: [[TMP3:%.*]] = zext <4 x i16> [[WIDE_LOAD]] to <4 x i32>
-; CHECK-NEXT: [[TMP4:%.*]] = shl nuw nsw <4 x i32> [[TMP3]], <i32 7, i32 7, i32 7, i32 7>
-; CHECK-NEXT: store <4 x i32> [[TMP4]], ptr [[NEXT_GEP4]], align 4
+; CHECK-NEXT: [[TMP1:%.*]] = zext <4 x i16> [[WIDE_LOAD]] to <4 x i32>
+; CHECK-NEXT: [[TMP2:%.*]] = shl nuw nsw <4 x i32> [[TMP1]], <i32 7, i32 7, i32 7, i32 7>
+; CHECK-NEXT: store <4 x i32> [[TMP2]], ptr [[NEXT_GEP5]], align 4
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
-; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
-; CHECK-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
+; CHECK-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 256
+; CHECK-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
; CHECK: middle.block:
-; CHECK-NEXT: br i1 true, label [[TMP7:%.*]], label [[SCALAR_PH]]
+; CHECK-NEXT: br i1 true, label [[TMP5:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
-; CHECK-NEXT: br label [[TMP6:%.*]]
-; CHECK: 6:
-; CHECK-NEXT: br i1 poison, label [[TMP7]], label [[TMP6]], !llvm.loop [[LOOP11:![0-9]+]]
-; CHECK: 7:
+; CHECK-NEXT: br label [[TMP4:%.*]]
+; CHECK: 4:
+; CHECK-NEXT: br i1 poison, label [[TMP5]], label [[TMP4]], !llvm.loop [[LOOP11:![0-9]+]]
+; CHECK: 5:
; CHECK-NEXT: ret void
;
br label %1
@@ -457,7 +451,9 @@ define void @example23c(ptr noalias nocapture %src, ptr noalias nocapture %dst)
; CHECK: vector.ph:
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE16:%.*]] ]
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE17:%.*]] ]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = shl i64 [[INDEX]], 1
+; CHECK-NEXT: [[OFFSET_IDX7:%.*]] = shl i64 [[INDEX]], 2
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[INDEX]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
; CHECK-NEXT: [[VEC_IV:%.*]] = or disjoint <4 x i64> [[BROADCAST_SPLAT]], <i64 0, i64 1, i64 2, i64 3>
@@ -465,71 +461,63 @@ define void @example23c(ptr noalias nocapture %src, ptr noalias nocapture %dst)
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <4 x i1> [[TMP1]], i64 0
; CHECK-NEXT: br i1 [[TMP2]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; CHECK: pred.store.if:
-; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[NEXT_GEP7:%.*]] = getelementptr i8, ptr [[DST:%.*]], i64 [[TMP3]]
-; CHECK-NEXT: [[TMP4:%.*]] = shl i64 [[INDEX]], 1
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[SRC:%.*]], i64 [[TMP4]]
-; CHECK-NEXT: [[TMP5:%.*]] = load i16, ptr [[NEXT_GEP]], align 2
-; CHECK-NEXT: [[TMP6:%.*]] = zext i16 [[TMP5]] to i32
-; CHECK-NEXT: [[TMP7:%.*]] = shl nuw nsw i32 [[TMP6]], 7
-; CHECK-NEXT: store i32 [[TMP7]], ptr [[NEXT_GEP7]], align 4
+; CHECK-NEXT: [[NEXT_GEP8:%.*]] = getelementptr i8, ptr [[DST:%.*]], i64 [[OFFSET_IDX7]]
+; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[SRC:%.*]], i64 [[OFFSET_IDX]]
+; CHECK-NEXT: [[TMP3:%.*]] = load i16, ptr [[NEXT_GEP]], align 2
+; CHECK-NEXT: [[TMP4:%.*]] = zext i16 [[TMP3]] to i32
+; CHECK-NEXT: [[TMP5:%.*]] = shl nuw nsw i32 [[TMP4]], 7
+; CHECK-NEXT: store i32 [[TMP5]], ptr [[NEXT_GEP8]], align 4
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]]
; CHECK: pred.store.continue:
-; CHECK-NEXT: [[TMP8:%.*]] = extractelement <4 x i1> [[TMP1]], i64 1
-; CHECK-NEXT: br i1 [[TMP8]], label [[PRED_STORE_IF11:%.*]], label [[PRED_STORE_CONTINUE12:%.*]]
-; CHECK: pred.store.if11:
-; CHECK-NEXT: [[TMP9:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP10:%.*]] = or disjoint i64 [[TMP9]], 4
-; CHECK-NEXT: [[NEXT_GEP8:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP10]]
-; CHECK-NEXT: [[TMP11:%.*]] = shl i64 [[INDEX]], 1
-; CHECK-NEXT: [[TMP12:%.*]] = or disjoint i64 [[TMP11]], 2
-; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP12]]
-; CHECK-NEXT: [[TMP13:%.*]] = load i16, ptr [[NEXT_GEP4]], align 2
-; CHECK-NEXT: [[TMP14:%.*]] = zext i16 [[TMP13]] to i32
-; CHECK-NEXT: [[TMP15:%.*]] = shl nuw nsw i32 [[TMP14]], 7
-; CHECK-NEXT: store i32 [[TMP15]], ptr [[NEXT_GEP8]], align 4
-; CHECK-NEXT: br label [[PRED_STORE_CONTINUE12]]
-; CHECK: pred.store.continue12:
-; CHECK-NEXT: [[TMP16:%.*]] = extractelement <4 x i1> [[TMP1]], i64 2
-; CHECK-NEXT: br i1 [[TMP16]], label [[PRED_STORE_IF13:%.*]], label [[PRED_STORE_CONTINUE14:%.*]]
-; CHECK: pred.store.if13:
-; CHECK-NEXT: [[TMP17:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP18:%.*]] = or disjoint i64 [[TMP17]], 8
-; CHECK-NEXT: [[NEXT_GEP9:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP18]]
-; CHECK-NEXT: [[TMP19:%.*]] = shl i64 [[INDEX]], 1
-; CHECK-NEXT: [[TMP20:%.*]] = or disjoint i64 [[TMP19]], 4
-; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP20]]
-; CHECK-NEXT: [[TMP21:%.*]] = load i16, ptr [[NEXT_GEP5]], align 2
+; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x i1> [[TMP1]], i64 1
+; CHECK-NEXT: br i1 [[TMP6]], label [[PRED_STORE_IF12:%.*]], label [[PRED_STORE_CONTINUE13:%.*]]
+; CHECK: pred.store.if12:
+; CHECK-NEXT: [[TMP7:%.*]] = or disjoint i64 [[OFFSET_IDX7]], 4
+; CHECK-NEXT: [[NEXT_GEP9:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP7]]
+; CHECK-NEXT: [[TMP8:%.*]] = or disjoint i64 [[OFFSET_IDX]], 2
+; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP8]]
+; CHECK-NEXT: [[TMP9:%.*]] = load i16, ptr [[NEXT_GEP4]], align 2
+; CHECK-NEXT: [[TMP10:%.*]] = zext i16 [[TMP9]] to i32
+; CHECK-NEXT: [[TMP11:%.*]] = shl nuw nsw i32 [[TMP10]], 7
+; CHECK-NEXT: store i32 [[TMP11]], ptr [[NEXT_GEP9]], align 4
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE13]]
+; CHECK: pred.store.continue13:
+; CHECK-NEXT: [[TMP12:%.*]] = extractelement <4 x i1> [[TMP1]], i64 2
+; CHECK-NEXT: br i1 [[TMP12]], label [[PRED_STORE_IF14:%.*]], label [[PRED_STORE_CONTINUE15:%.*]]
+; CHECK: pred.store.if14:
+; CHECK-NEXT: [[TMP13:%.*]] = or disjoint i64 [[OFFSET_IDX7]], 8
+; CHECK-NEXT: [[NEXT_GEP10:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP13]]
+; CHECK-NEXT: [[TMP14:%.*]] = or disjoint i64 [[OFFSET_IDX]], 4
+; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP14]]
+; CHECK-NEXT: [[TMP15:%.*]] = load i16, ptr [[NEXT_GEP5]], align 2
+; CHECK-NEXT: [[TMP16:%.*]] = zext i16 [[TMP15]] to i32
+; CHECK-NEXT: [[TMP17:%.*]] = shl nuw nsw i32 [[TMP16]], 7
+; CHECK-NEXT: store i32 [[TMP17]], ptr [[NEXT_GEP10]], align 4
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE15]]
+; CHECK: pred.store.continue15:
+; CHECK-NEXT: [[TMP18:%.*]] = extractelement <4 x i1> [[TMP1]], i64 3
+; CHECK-NEXT: br i1 [[TMP18]], label [[PRED_STORE_IF16:%.*]], label [[PRED_STORE_CONTINUE17]]
+; CHECK: pred.store.if16:
+; CHECK-NEXT: [[TMP19:%.*]] = or disjoint i64 [[OFFSET_IDX7]], 12
+; CHECK-NEXT: [[NEXT_GEP11:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP19]]
+; CHECK-NEXT: [[TMP20:%.*]] = or disjoint i64 [[OFFSET_IDX]], 6
+; CHECK-NEXT: [[NEXT_GEP6:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP20]]
+; CHECK-NEXT: [[TMP21:%.*]] = load i16, ptr [[NEXT_GEP6]], align 2
; CHECK-NEXT: [[TMP22:%.*]] = zext i16 [[TMP21]] to i32
; CHECK-NEXT: [[TMP23:%.*]] = shl nuw nsw i32 [[TMP22]], 7
-; CHECK-NEXT: store i32 [[TMP23]], ptr [[NEXT_GEP9]], align 4
-; CHECK-NEXT: br label [[PRED_STORE_CONTINUE14]]
-; CHECK: pred.store.continue14:
-; CHECK-NEXT: [[TMP24:%.*]] = extractelement <4 x i1> [[TMP1]], i64 3
-; CHECK-NEXT: br i1 [[TMP24]], label [[PRED_STORE_IF15:%.*]], label [[PRED_STORE_CONTINUE16]]
-; CHECK: pred.store.if15:
-; CHECK-NEXT: [[TMP25:%.*]] = shl i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP26:%.*]] = or disjoint i64 [[TMP25]], 12
-; CHECK-NEXT: [[NEXT_GEP10:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP26]]
-; CHECK-NEXT: [[TMP27:%.*]] = shl i64 [[INDEX]], 1
-; CHECK-NEXT: [[TMP28:%.*]] = or disjoint i64 [[TMP27]], 6
-; CHECK-NEXT: [[NEXT_GEP6:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP28]]
-; CHECK-NEXT: [[TMP29:%.*]] = load i16, ptr [[NEXT_GEP6]], align 2
-; CHECK-NEXT: [[TMP30:%.*]] = zext i16 [[TMP29]] to i32
-; CHECK-NEXT: [[TMP31:%.*]] = shl nuw nsw i32 [[TMP30]], 7
-; CHECK-NEXT: store i32 [[TMP31]], ptr [[NEXT_GEP10]], align 4
-; CHECK-NEXT: br label [[PRED_STORE_CONTINUE16]]
-; CHECK: pred.store.continue16:
+; CHECK-NEXT: store i32 [[TMP23]], ptr [[NEXT_GEP11]], align 4
+; CHECK-NEXT: br label [[PRED_STORE_CONTINUE17]]
+; CHECK: pred.store.continue17:
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4
-; CHECK-NEXT: [[TMP32:%.*]] = icmp eq i64 [[INDEX_NEXT]], 260
-; CHECK-NEXT: br i1 [[TMP32]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
+; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], 260
+; CHECK-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
; CHECK: middle.block:
-; CHECK-NEXT: br i1 true, label [[TMP34:%.*]], label [[SCALAR_PH]]
+; CHECK-NEXT: br i1 true, label [[TMP26:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
-; CHECK-NEXT: br label [[TMP33:%.*]]
-; CHECK: 33:
-; CHECK-NEXT: br i1 poison, label [[TMP34]], label [[TMP33]], !llvm.loop [[LOOP13:![0-9]+]]
-; CHECK: 34:
+; CHECK-NEXT: br label [[TMP25:%.*]]
+; CHECK: 25:
+; CHECK-NEXT: br i1 poison, label [[TMP26]], label [[TMP25]], !llvm.loop [[LOOP13:![0-9]+]]
+; CHECK: 26:
; CHECK-NEXT: ret void
;
br label %1
diff --git a/llvm/test/Transforms/LoopVectorize/consecutive-ptr-uniforms.ll b/llvm/test/Transforms/LoopVectorize/consecutive-ptr-uniforms.ll
index c7eaac315fd861..1e4e3d1e47be0e 100644
--- a/llvm/test/Transforms/LoopVectorize/consecutive-ptr-uniforms.ll
+++ b/llvm/test/Transforms/LoopVectorize/consecutive-ptr-uniforms.ll
@@ -311,15 +311,12 @@ for.end:
; INTER: vector.body
; INTER: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
; INTER: %[[I0:.+]] = shl i64 %index, 4
+; INTER: %[[I1:.+]] = or disjoint i64 %[[I0]], 16
+; INTER: %[[I2:.+]] = or disjoint i64 %[[I0]], 32
+; INTER: %[[I3:.+]] = or disjoint i64 %[[I0]], 48
; INTER: %next.gep = getelementptr i8, ptr %a, i64 %[[I0]]
-; INTER: %[[S1:.+]] = shl i64 %index, 4
-; INTER: %[[I1:.+]] = or disjoint i64 %[[S1]], 16
; INTER: %next.gep2 = getelementptr i8, ptr %a, i64 %[[I1]]
-; INTER: %[[S2:.+]] = shl i64 %index, 4
-; INTER: %[[I2:.+]] = or disjoint i64 %[[S2]], 32
; INTER: %next.gep3 = getelementptr i8, ptr %a, i64 %[[I2]]
-; INTER: %[[S3:.+]] = shl i64 %index, 4
-; INTER: %[[I3:.+]] = or disjoint i64 %[[S3]], 48
; INTER: %next.gep4 = getelementptr i8, ptr %a, i64 %[[I3]]
; INTER: br i1 {{.*}}, label %middle.block, label %vector.body
;
@@ -361,15 +358,12 @@ for.end:
; CHECK: vector.body
; CHECK: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
; CHECK: [[SHL1:%.+]] = shl i64 %index, 4
+; CHECK: %[[I1:.+]] = or disjoint i64 [[SHL1]], 16
+; CHECK: %[[I2:.+]] = or disjoint i64 [[SHL1]], 32
+; CHECK: %[[I3:.+]] = or disjoint i64 [[SHL1]], 48
; CHECK: %next.gep = getelementptr i8, ptr %a, i64 [[SHL1]]
-; CHECK: [[SHL2:%.+]] = shl i64 %index, 4
-; CHECK: %[[I1:.+]] = or disjoint i64 [[SHL2]], 16
; CHECK: %next.gep2 = getelementptr i8, ptr %a, i64 %[[I1]]
-; CHECK: [[SHL3:%.+]] = shl i64 %index, 4
-; CHECK: %[[I2:.+]] = or disjoint i64 [[SHL3]], 32
; CHECK: %next.gep3 = getelementptr i8, ptr %a, i64 %[[I2]]
-; CHECK: [[SHL4:%.+]] = shl i64 %index, 4
-; CHECK: %[[I3:.+]] = or disjoint i64 [[SHL4]], 48
; CHECK: %next.gep4 = getelementptr i8, ptr %a, i64 %[[I3]]
; CHECK: br i1 {{.*}}, label %middle.block, label %vector.body
;
diff --git a/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll b/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll
index b451d4b4e5462d..0a37e5ea0ca007 100644
--- a/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll
+++ b/llvm/test/Transforms/LoopVectorize/first-order-recurrence.ll
@@ -1537,92 +1537,85 @@ define i32 @PR33613(ptr %b, double %j, i32 %d) {
; UNROLL-NO-IC-NEXT: br label [[VECTOR_BODY:%.*]]
; UNROLL-NO-IC: vector.body:
; UNROLL-NO-IC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; UNROLL-NO-IC-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP48:%.*]], [[VECTOR_BODY]] ]
-; UNROLL-NO-IC-NEXT: [[VEC_PHI9:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP49:%.*]], [[VECTOR_BODY]] ]
-; UNROLL-NO-IC-NEXT: [[VECTOR_RECUR:%.*]] = phi <4 x double> [ [[VECTOR_RECUR_INIT]], [[VECTOR_PH]] ], [ [[TMP39:%.*]], [[VECTOR_BODY]] ]
-; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
-; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 200
-; UNROLL-NO-IC-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP1]]
-; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 1
-; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 200
-; UNROLL-NO-IC-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP3]]
-; UNROLL-NO-IC-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 2
-; UNROLL-NO-IC-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 200
-; UNROLL-NO-IC-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP5]]
-; UNROLL-NO-IC-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 3
-; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = mul i64 [[TMP6]], 200
-; UNROLL-NO-IC-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP7]]
-; UNROLL-NO-IC-NEXT: [[TMP8:%.*]] = add i64 [[INDEX]], 4
-; UNROLL-NO-IC-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 200
-; UNROLL-NO-IC-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP9]]
-; UNROLL-NO-IC-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], 5
-; UNROLL-NO-IC-NEXT: [[TMP11:%.*]] = mul i64 [[TMP10]], 200
-; UNROLL-NO-IC-NEXT: [[NEXT_GEP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP11]]
-; UNROLL-NO-IC-NEXT: [[TMP12:%.*]] = add i64 [[INDEX]], 6
-; UNROLL-NO-IC-NEXT: [[TMP13:%.*]] = mul i64 [[TMP12]], 200
-; UNROLL-NO-IC-NEXT: [[NEXT_GEP7:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP13]]
-; UNROLL-NO-IC-NEXT: [[TMP14:%.*]] = add i64 [[INDEX]], 7
-; UNROLL-NO-IC-NEXT: [[TMP15:%.*]] = mul i64 [[TMP14]], 200
-; UNROLL-NO-IC-NEXT: [[NEXT_GEP8:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP15]]
-; UNROLL-NO-IC-NEXT: [[TMP16:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP]], i64 [[IDXPROM]]
-; UNROLL-NO-IC-NEXT: [[TMP17:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP2]], i64 [[IDXPROM]]
-; UNROLL-NO-IC-NEXT: [[TMP18:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP3]], i64 [[IDXPROM]]
-; UNROLL-NO-IC-NEXT: [[TMP19:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP4]], i64 [[IDXPROM]]
-; UNROLL-NO-IC-NEXT: [[TMP20:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP5]], i64 [[IDXPROM]]
-; UNROLL-NO-IC-NEXT: [[TMP21:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP6]], i64 [[IDXPROM]]
-; UNROLL-NO-IC-NEXT: [[TMP22:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP7]], i64 [[IDXPROM]]
-; UNROLL-NO-IC-NEXT: [[TMP23:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP8]], i64 [[IDXPROM]]
-; UNROLL-NO-IC-NEXT: [[TMP24:%.*]] = load double, ptr [[TMP16]], align 8
-; UNROLL-NO-IC-NEXT: [[TMP25:%.*]] = load double, ptr [[TMP17]], align 8
-; UNROLL-NO-IC-NEXT: [[TMP26:%.*]] = load double, ptr [[TMP18]], align 8
-; UNROLL-NO-IC-NEXT: [[TMP27:%.*]] = load double, ptr [[TMP19]], align 8
+; UNROLL-NO-IC-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP40:%.*]], [[VECTOR_BODY]] ]
+; UNROLL-NO-IC-NEXT: [[VEC_PHI2:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP41:%.*]], [[VECTOR_BODY]] ]
+; UNROLL-NO-IC-NEXT: [[VECTOR_RECUR:%.*]] = phi <4 x double> [ [[VECTOR_RECUR_INIT]], [[VECTOR_PH]] ], [ [[TMP31:%.*]], [[VECTOR_BODY]] ]
+; UNROLL-NO-IC-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 200
+; UNROLL-NO-IC-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET_IDX]], 0
+; UNROLL-NO-IC-NEXT: [[TMP1:%.*]] = add i64 [[OFFSET_IDX]], 200
+; UNROLL-NO-IC-NEXT: [[TMP2:%.*]] = add i64 [[OFFSET_IDX]], 400
+; UNROLL-NO-IC-NEXT: [[TMP3:%.*]] = add i64 [[OFFSET_IDX]], 600
+; UNROLL-NO-IC-NEXT: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 800
+; UNROLL-NO-IC-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 1000
+; UNROLL-NO-IC-NEXT: [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 1200
+; UNROLL-NO-IC-NEXT: [[TMP7:%.*]] = add i64 [[OFFSET_IDX]], 1400
+; UNROLL-NO-IC-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP0]]
+; UNROLL-NO-IC-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP1]]
+; UNROLL-NO-IC-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP2]]
+; UNROLL-NO-IC-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP3]]
+; UNROLL-NO-IC-NEXT: [[NEXT_GEP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP4]]
+; UNROLL-NO-IC-NEXT: [[NEXT_GEP7:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP5]]
+; UNROLL-NO-IC-NEXT: [[NEXT_GEP8:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP6]]
+; UNROLL-NO-IC-NEXT: [[NEXT_GEP9:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP7]]
+; UNROLL-NO-IC-NEXT: [[TMP8:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP]], i64 [[IDXPROM]]
+; UNROLL-NO-IC-NEXT: [[TMP9:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP3]], i64 [[IDXPROM]]
+; UNROLL-NO-IC-NEXT: [[TMP10:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP4]], i64 [[IDXPROM]]
+; UNROLL-NO-IC-NEXT: [[TMP11:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP5]], i64 [[IDXPROM]]
+; UNROLL-NO-IC-NEXT: [[TMP12:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP6]], i64 [[IDXPROM]]
+; UNROLL-NO-IC-NEXT: [[TMP13:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP7]], i64 [[IDXPROM]]
+; UNROLL-NO-IC-NEXT: [[TMP14:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP8]], i64 [[IDXPROM]]
+; UNROLL-NO-IC-NEXT: [[TMP15:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP9]], i64 [[IDXPROM]]
+; UNROLL-NO-IC-NEXT: [[TMP16:%.*]] = load double, ptr [[TMP8]], align 8
+; UNROLL-NO-IC-NEXT: [[TMP17:%.*]] = load double, ptr [[TMP9]], align 8
+; UNROLL-NO-IC-NEXT: [[TMP18:%.*]] = load double, ptr [[TMP10]], align 8
+; UNROLL-NO-IC-NEXT: [[TMP19:%.*]] = load double, ptr [[TMP11]], align 8
+; UNROLL-NO-IC-NEXT: [[TMP20:%.*]] = insertelement <4 x double> poison, double [[TMP16]], i32 0
+; UNROLL-NO-IC-NEXT: [[TMP21:%.*]] = insertelement <4 x double> [[TMP20]], double [[TMP17]], i32 1
+; UNROLL-NO-IC-NEXT: [[TMP22:%.*]] = insertelement <4 x double> [[TMP21]], double [[TMP18]], i32 2
+; UNROLL-NO-IC-NEXT: [[TMP23:%.*]] = insertelement <4 x double> [[TMP22]], double [[TMP19]], i32 3
+; UNROLL-NO-IC-NEXT: [[TMP24:%.*]] = load double, ptr [[TMP12]], align 8
+; UNROLL-NO-IC-NEXT: [[TMP25:%.*]] = load double, ptr [[TMP13]], align 8
+; UNROLL-NO-IC-NEXT: [[TMP26:%.*]] = load double, ptr [[TMP14]], align 8
+; UNROLL-NO-IC-NEXT: [[TMP27:%.*]] = load double, ptr [[TMP15]], align 8
; UNROLL-NO-IC-NEXT: [[TMP28:%.*]] = insertelement <4 x double> poison, double [[TMP24]], i32 0
; UNROLL-NO-IC-NEXT: [[TMP29:%.*]] = insertelement <4 x double> [[TMP28]], double [[TMP25]], i32 1
; UNROLL-NO-IC-NEXT: [[TMP30:%.*]] = insertelement <4 x double> [[TMP29]], double [[TMP26]], i32 2
-; UNROLL-NO-IC-NEXT: [[TMP31:%.*]] = insertelement <4 x double> [[TMP30]], double [[TMP27]], i32 3
-; UNROLL-NO-IC-NEXT: [[TMP32:%.*]] = load double, ptr [[TMP20]], align 8
-; UNROLL-NO-IC-NEXT: [[TMP33:%.*]] = load double, ptr [[TMP21]], align 8
-; UNROLL-NO-IC-NEXT: [[TMP34:%.*]] = load double, ptr [[TMP22]], align 8
-; UNROLL-NO-IC-NEXT: [[TMP35:%.*]] = load double, ptr [[TMP23]], align 8
-; UNROLL-NO-IC-NEXT: [[TMP36:%.*]] = insertelement <4 x double> poison, double [[TMP32]], i32 0
-; UNROLL-NO-IC-NEXT: [[TMP37:%.*]] = insertelement <4 x double> [[TMP36]], double [[TMP33]], i32 1
-; UNROLL-NO-IC-NEXT: [[TMP38:%.*]] = insertelement <4 x double> [[TMP37]], double [[TMP34]], i32 2
-; UNROLL-NO-IC-NEXT: [[TMP39]] = insertelement <4 x double> [[TMP38]], double [[TMP35]], i32 3
-; UNROLL-NO-IC-NEXT: [[TMP40:%.*]] = shufflevector <4 x double> [[VECTOR_RECUR]], <4 x double> [[TMP31]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
-; UNROLL-NO-IC-NEXT: [[TMP41:%.*]] = shufflevector <4 x double> [[TMP31]], <4 x double> [[TMP39]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
-; UNROLL-NO-IC-NEXT: [[TMP42:%.*]] = fmul <4 x double> [[TMP40]], [[TMP31]]
-; UNROLL-NO-IC-NEXT: [[TMP43:%.*]] = fmul <4 x double> [[TMP41]], [[TMP39]]
-; UNROLL-NO-IC-NEXT: [[TMP44:%.*]] = fcmp une <4 x double> [[TMP42]], zeroinitializer
-; UNROLL-NO-IC-NEXT: [[TMP45:%.*]] = fcmp une <4 x double> [[TMP43]], zeroinitializer
-; UNROLL-NO-IC-NEXT: [[TMP46:%.*]] = zext <4 x i1> [[TMP44]] to <4 x i32>
-; UNROLL-NO-IC-NEXT: [[TMP47:%.*]] = zext <4 x i1> [[TMP45]] to <4 x i32>
-; UNROLL-NO-IC-NEXT: [[TMP48]] = add <4 x i32> [[VEC_PHI]], [[TMP46]]
-; UNROLL-NO-IC-NEXT: [[TMP49]] = add <4 x i32> [[VEC_PHI9]], [[TMP47]]
+; UNROLL-NO-IC-NEXT: [[TMP31]] = insertelement <4 x double> [[TMP30]], double [[TMP27]], i32 3
+; UNROLL-NO-IC-NEXT: [[TMP32:%.*]] = shufflevector <4 x double> [[VECTOR_RECUR]], <4 x double> [[TMP23]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
+; UNROLL-NO-IC-NEXT: [[TMP33:%.*]] = shufflevector <4 x double> [[TMP23]], <4 x double> [[TMP31]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
+; UNROLL-NO-IC-NEXT: [[TMP34:%.*]] = fmul <4 x double> [[TMP32]], [[TMP23]]
+; UNROLL-NO-IC-NEXT: [[TMP35:%.*]] = fmul <4 x double> [[TMP33]], [[TMP31]]
+; UNROLL-NO-IC-NEXT: [[TMP36:%.*]] = fcmp une <4 x double> [[TMP34]], zeroinitializer
+; UNROLL-NO-IC-NEXT: [[TMP37:%.*]] = fcmp une <4 x double> [[TMP35]], zeroinitializer
+; UNROLL-NO-IC-NEXT: [[TMP38:%.*]] = zext <4 x i1> [[TMP36]] to <4 x i32>
+; UNROLL-NO-IC-NEXT: [[TMP39:%.*]] = zext <4 x i1> [[TMP37]] to <4 x i32>
+; UNROLL-NO-IC-NEXT: [[TMP40]] = add <4 x i32> [[VEC_PHI]], [[TMP38]]
+; UNROLL-NO-IC-NEXT: [[TMP41]] = add <4 x i32> [[VEC_PHI2]], [[TMP39]]
; UNROLL-NO-IC-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
-; UNROLL-NO-IC-NEXT: [[TMP50:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10240
-; UNROLL-NO-IC-NEXT: br i1 [[TMP50]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
+; UNROLL-NO-IC-NEXT: [[TMP42:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10240
+; UNROLL-NO-IC-NEXT: br i1 [[TMP42]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; UNROLL-NO-IC: middle.block:
-; UNROLL-NO-IC-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[TMP49]], [[TMP48]]
-; UNROLL-NO-IC-NEXT: [[TMP51:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
-; UNROLL-NO-IC-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x double> [[TMP39]], i32 3
+; UNROLL-NO-IC-NEXT: [[BIN_RDX:%.*]] = add <4 x i32> [[TMP41]], [[TMP40]]
+; UNROLL-NO-IC-NEXT: [[TMP43:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[BIN_RDX]])
+; UNROLL-NO-IC-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x double> [[TMP31]], i32 3
; UNROLL-NO-IC-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; UNROLL-NO-IC: scalar.ph:
; UNROLL-NO-IC-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi double [ [[J]], [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[B]], [[ENTRY]] ]
; UNROLL-NO-IC-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ 10240, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
-; UNROLL-NO-IC-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP51]], [[MIDDLE_BLOCK]] ]
+; UNROLL-NO-IC-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP43]], [[MIDDLE_BLOCK]] ]
; UNROLL-NO-IC-NEXT: br label [[FOR_BODY:%.*]]
; UNROLL-NO-IC: for.cond.cleanup:
-; UNROLL-NO-IC-NEXT: [[A_1_LCSSA:%.*]] = phi i32 [ [[A_1:%.*]], [[FOR_BODY]] ], [ [[TMP51]], [[MIDDLE_BLOCK]] ]
+; UNROLL-NO-IC-NEXT: [[A_1_LCSSA:%.*]] = phi i32 [ [[A_1:%.*]], [[FOR_BODY]] ], [ [[TMP43]], [[MIDDLE_BLOCK]] ]
; UNROLL-NO-IC-NEXT: ret i32 [[A_1_LCSSA]]
; UNROLL-NO-IC: for.body:
; UNROLL-NO-IC-NEXT: [[B_ADDR_012:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[ADD_PTR:%.*]], [[FOR_BODY]] ]
; UNROLL-NO-IC-NEXT: [[I_011:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC1:%.*]], [[FOR_BODY]] ]
; UNROLL-NO-IC-NEXT: [[A_010:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_1]], [[FOR_BODY]] ]
-; UNROLL-NO-IC-NEXT: [[SCALAR_RECUR:%.*]] = phi double [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[TMP52:%.*]], [[FOR_BODY]] ]
+; UNROLL-NO-IC-NEXT: [[SCALAR_RECUR:%.*]] = phi double [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[TMP44:%.*]], [[FOR_BODY]] ]
; UNROLL-NO-IC-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[B_ADDR_012]], i64 [[IDXPROM]]
-; UNROLL-NO-IC-NEXT: [[TMP52]] = load double, ptr [[ARRAYIDX]], align 8
-; UNROLL-NO-IC-NEXT: [[MUL:%.*]] = fmul double [[SCALAR_RECUR]], [[TMP52]]
+; UNROLL-NO-IC-NEXT: [[TMP44]] = load double, ptr [[ARRAYIDX]], align 8
+; UNROLL-NO-IC-NEXT: [[MUL:%.*]] = fmul double [[SCALAR_RECUR]], [[TMP44]]
; UNROLL-NO-IC-NEXT: [[TOBOOL:%.*]] = fcmp une double [[MUL]], 0.000000e+00
; UNROLL-NO-IC-NEXT: [[INC:%.*]] = zext i1 [[TOBOOL]] to i32
; UNROLL-NO-IC-NEXT: [[A_1]] = add nsw i32 [[A_010]], [[INC]]
@@ -1640,35 +1633,34 @@ define i32 @PR33613(ptr %b, double %j, i32 %d) {
; UNROLL-NO-VF-NEXT: br label [[VECTOR_BODY:%.*]]
; UNROLL-NO-VF: vector.body:
; UNROLL-NO-VF-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; UNROLL-NO-VF-NEXT: [[VEC_PHI:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP14:%.*]], [[VECTOR_BODY]] ]
-; UNROLL-NO-VF-NEXT: [[VEC_PHI3:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[VECTOR_BODY]] ]
-; UNROLL-NO-VF-NEXT: [[VECTOR_RECUR:%.*]] = phi double [ [[J:%.*]], [[VECTOR_PH]] ], [ [[TMP7:%.*]], [[VECTOR_BODY]] ]
-; UNROLL-NO-VF-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
-; UNROLL-NO-VF-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 200
-; UNROLL-NO-VF-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP1]]
-; UNROLL-NO-VF-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 1
-; UNROLL-NO-VF-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 200
-; UNROLL-NO-VF-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP3]]
-; UNROLL-NO-VF-NEXT: [[TMP4:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP]], i64 [[IDXPROM]]
-; UNROLL-NO-VF-NEXT: [[TMP5:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP2]], i64 [[IDXPROM]]
-; UNROLL-NO-VF-NEXT: [[TMP6:%.*]] = load double, ptr [[TMP4]], align 8
-; UNROLL-NO-VF-NEXT: [[TMP7]] = load double, ptr [[TMP5]], align 8
-; UNROLL-NO-VF-NEXT: [[TMP8:%.*]] = fmul double [[VECTOR_RECUR]], [[TMP6]]
-; UNROLL-NO-VF-NEXT: [[TMP9:%.*]] = fmul double [[TMP6]], [[TMP7]]
-; UNROLL-NO-VF-NEXT: [[TMP10:%.*]] = fcmp une double [[TMP8]], 0.000000e+00
-; UNROLL-NO-VF-NEXT: [[TMP11:%.*]] = fcmp une double [[TMP9]], 0.000000e+00
-; UNROLL-NO-VF-NEXT: [[TMP12:%.*]] = zext i1 [[TMP10]] to i32
-; UNROLL-NO-VF-NEXT: [[TMP13:%.*]] = zext i1 [[TMP11]] to i32
-; UNROLL-NO-VF-NEXT: [[TMP14]] = add i32 [[VEC_PHI]], [[TMP12]]
-; UNROLL-NO-VF-NEXT: [[TMP15]] = add i32 [[VEC_PHI3]], [[TMP13]]
+; UNROLL-NO-VF-NEXT: [[VEC_PHI:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP12:%.*]], [[VECTOR_BODY]] ]
+; UNROLL-NO-VF-NEXT: [[VEC_PHI2:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP13:%.*]], [[VECTOR_BODY]] ]
+; UNROLL-NO-VF-NEXT: [[VECTOR_RECUR:%.*]] = phi double [ [[J:%.*]], [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
+; UNROLL-NO-VF-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 200
+; UNROLL-NO-VF-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET_IDX]], 0
+; UNROLL-NO-VF-NEXT: [[TMP1:%.*]] = add i64 [[OFFSET_IDX]], 200
+; UNROLL-NO-VF-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP0]]
+; UNROLL-NO-VF-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP1]]
+; UNROLL-NO-VF-NEXT: [[TMP2:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP]], i64 [[IDXPROM]]
+; UNROLL-NO-VF-NEXT: [[TMP3:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP3]], i64 [[IDXPROM]]
+; UNROLL-NO-VF-NEXT: [[TMP4:%.*]] = load double, ptr [[TMP2]], align 8
+; UNROLL-NO-VF-NEXT: [[TMP5]] = load double, ptr [[TMP3]], align 8
+; UNROLL-NO-VF-NEXT: [[TMP6:%.*]] = fmul double [[VECTOR_RECUR]], [[TMP4]]
+; UNROLL-NO-VF-NEXT: [[TMP7:%.*]] = fmul double [[TMP4]], [[TMP5]]
+; UNROLL-NO-VF-NEXT: [[TMP8:%.*]] = fcmp une double [[TMP6]], 0.000000e+00
+; UNROLL-NO-VF-NEXT: [[TMP9:%.*]] = fcmp une double [[TMP7]], 0.000000e+00
+; UNROLL-NO-VF-NEXT: [[TMP10:%.*]] = zext i1 [[TMP8]] to i32
+; UNROLL-NO-VF-NEXT: [[TMP11:%.*]] = zext i1 [[TMP9]] to i32
+; UNROLL-NO-VF-NEXT: [[TMP12]] = add i32 [[VEC_PHI]], [[TMP10]]
+; UNROLL-NO-VF-NEXT: [[TMP13]] = add i32 [[VEC_PHI2]], [[TMP11]]
; UNROLL-NO-VF-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
-; UNROLL-NO-VF-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10240
-; UNROLL-NO-VF-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
+; UNROLL-NO-VF-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10240
+; UNROLL-NO-VF-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; UNROLL-NO-VF: middle.block:
-; UNROLL-NO-VF-NEXT: [[BIN_RDX:%.*]] = add i32 [[TMP15]], [[TMP14]]
+; UNROLL-NO-VF-NEXT: [[BIN_RDX:%.*]] = add i32 [[TMP13]], [[TMP12]]
; UNROLL-NO-VF-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; UNROLL-NO-VF: scalar.ph:
-; UNROLL-NO-VF-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi double [ [[J]], [[ENTRY:%.*]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ]
+; UNROLL-NO-VF-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi double [ [[J]], [[ENTRY:%.*]] ], [ [[TMP5]], [[MIDDLE_BLOCK]] ]
; UNROLL-NO-VF-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[B]], [[ENTRY]] ]
; UNROLL-NO-VF-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ 10240, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
; UNROLL-NO-VF-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[BIN_RDX]], [[MIDDLE_BLOCK]] ]
@@ -1680,10 +1672,10 @@ define i32 @PR33613(ptr %b, double %j, i32 %d) {
; UNROLL-NO-VF-NEXT: [[B_ADDR_012:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[ADD_PTR:%.*]], [[FOR_BODY]] ]
; UNROLL-NO-VF-NEXT: [[I_011:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC1:%.*]], [[FOR_BODY]] ]
; UNROLL-NO-VF-NEXT: [[A_010:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_1]], [[FOR_BODY]] ]
-; UNROLL-NO-VF-NEXT: [[SCALAR_RECUR:%.*]] = phi double [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[TMP17:%.*]], [[FOR_BODY]] ]
+; UNROLL-NO-VF-NEXT: [[SCALAR_RECUR:%.*]] = phi double [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[TMP15:%.*]], [[FOR_BODY]] ]
; UNROLL-NO-VF-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[B_ADDR_012]], i64 [[IDXPROM]]
-; UNROLL-NO-VF-NEXT: [[TMP17]] = load double, ptr [[ARRAYIDX]], align 8
-; UNROLL-NO-VF-NEXT: [[MUL:%.*]] = fmul double [[SCALAR_RECUR]], [[TMP17]]
+; UNROLL-NO-VF-NEXT: [[TMP15]] = load double, ptr [[ARRAYIDX]], align 8
+; UNROLL-NO-VF-NEXT: [[MUL:%.*]] = fmul double [[SCALAR_RECUR]], [[TMP15]]
; UNROLL-NO-VF-NEXT: [[TOBOOL:%.*]] = fcmp une double [[MUL]], 0.000000e+00
; UNROLL-NO-VF-NEXT: [[INC:%.*]] = zext i1 [[TOBOOL]] to i32
; UNROLL-NO-VF-NEXT: [[A_1]] = add nsw i32 [[A_010]], [[INC]]
@@ -1702,61 +1694,58 @@ define i32 @PR33613(ptr %b, double %j, i32 %d) {
; SINK-AFTER-NEXT: br label [[VECTOR_BODY:%.*]]
; SINK-AFTER: vector.body:
; SINK-AFTER-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; SINK-AFTER-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP24:%.*]], [[VECTOR_BODY]] ]
-; SINK-AFTER-NEXT: [[VECTOR_RECUR:%.*]] = phi <4 x double> [ [[VECTOR_RECUR_INIT]], [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ]
-; SINK-AFTER-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
-; SINK-AFTER-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 200
-; SINK-AFTER-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP1]]
-; SINK-AFTER-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 1
-; SINK-AFTER-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 200
-; SINK-AFTER-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP3]]
-; SINK-AFTER-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 2
-; SINK-AFTER-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 200
-; SINK-AFTER-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP5]]
-; SINK-AFTER-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 3
-; SINK-AFTER-NEXT: [[TMP7:%.*]] = mul i64 [[TMP6]], 200
-; SINK-AFTER-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP7]]
-; SINK-AFTER-NEXT: [[TMP8:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP]], i64 [[IDXPROM]]
-; SINK-AFTER-NEXT: [[TMP9:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP2]], i64 [[IDXPROM]]
-; SINK-AFTER-NEXT: [[TMP10:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP3]], i64 [[IDXPROM]]
-; SINK-AFTER-NEXT: [[TMP11:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP4]], i64 [[IDXPROM]]
-; SINK-AFTER-NEXT: [[TMP12:%.*]] = load double, ptr [[TMP8]], align 8
-; SINK-AFTER-NEXT: [[TMP13:%.*]] = load double, ptr [[TMP9]], align 8
-; SINK-AFTER-NEXT: [[TMP14:%.*]] = load double, ptr [[TMP10]], align 8
-; SINK-AFTER-NEXT: [[TMP15:%.*]] = load double, ptr [[TMP11]], align 8
-; SINK-AFTER-NEXT: [[TMP16:%.*]] = insertelement <4 x double> poison, double [[TMP12]], i32 0
-; SINK-AFTER-NEXT: [[TMP17:%.*]] = insertelement <4 x double> [[TMP16]], double [[TMP13]], i32 1
-; SINK-AFTER-NEXT: [[TMP18:%.*]] = insertelement <4 x double> [[TMP17]], double [[TMP14]], i32 2
-; SINK-AFTER-NEXT: [[TMP19]] = insertelement <4 x double> [[TMP18]], double [[TMP15]], i32 3
-; SINK-AFTER-NEXT: [[TMP20:%.*]] = shufflevector <4 x double> [[VECTOR_RECUR]], <4 x double> [[TMP19]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
-; SINK-AFTER-NEXT: [[TMP21:%.*]] = fmul <4 x double> [[TMP20]], [[TMP19]]
-; SINK-AFTER-NEXT: [[TMP22:%.*]] = fcmp une <4 x double> [[TMP21]], zeroinitializer
-; SINK-AFTER-NEXT: [[TMP23:%.*]] = zext <4 x i1> [[TMP22]] to <4 x i32>
-; SINK-AFTER-NEXT: [[TMP24]] = add <4 x i32> [[VEC_PHI]], [[TMP23]]
+; SINK-AFTER-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP20:%.*]], [[VECTOR_BODY]] ]
+; SINK-AFTER-NEXT: [[VECTOR_RECUR:%.*]] = phi <4 x double> [ [[VECTOR_RECUR_INIT]], [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[VECTOR_BODY]] ]
+; SINK-AFTER-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 200
+; SINK-AFTER-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET_IDX]], 0
+; SINK-AFTER-NEXT: [[TMP1:%.*]] = add i64 [[OFFSET_IDX]], 200
+; SINK-AFTER-NEXT: [[TMP2:%.*]] = add i64 [[OFFSET_IDX]], 400
+; SINK-AFTER-NEXT: [[TMP3:%.*]] = add i64 [[OFFSET_IDX]], 600
+; SINK-AFTER-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP0]]
+; SINK-AFTER-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP1]]
+; SINK-AFTER-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP2]]
+; SINK-AFTER-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP3]]
+; SINK-AFTER-NEXT: [[TMP4:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP]], i64 [[IDXPROM]]
+; SINK-AFTER-NEXT: [[TMP5:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP2]], i64 [[IDXPROM]]
+; SINK-AFTER-NEXT: [[TMP6:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP3]], i64 [[IDXPROM]]
+; SINK-AFTER-NEXT: [[TMP7:%.*]] = getelementptr inbounds double, ptr [[NEXT_GEP4]], i64 [[IDXPROM]]
+; SINK-AFTER-NEXT: [[TMP8:%.*]] = load double, ptr [[TMP4]], align 8
+; SINK-AFTER-NEXT: [[TMP9:%.*]] = load double, ptr [[TMP5]], align 8
+; SINK-AFTER-NEXT: [[TMP10:%.*]] = load double, ptr [[TMP6]], align 8
+; SINK-AFTER-NEXT: [[TMP11:%.*]] = load double, ptr [[TMP7]], align 8
+; SINK-AFTER-NEXT: [[TMP12:%.*]] = insertelement <4 x double> poison, double [[TMP8]], i32 0
+; SINK-AFTER-NEXT: [[TMP13:%.*]] = insertelement <4 x double> [[TMP12]], double [[TMP9]], i32 1
+; SINK-AFTER-NEXT: [[TMP14:%.*]] = insertelement <4 x double> [[TMP13]], double [[TMP10]], i32 2
+; SINK-AFTER-NEXT: [[TMP15]] = insertelement <4 x double> [[TMP14]], double [[TMP11]], i32 3
+; SINK-AFTER-NEXT: [[TMP16:%.*]] = shufflevector <4 x double> [[VECTOR_RECUR]], <4 x double> [[TMP15]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
+; SINK-AFTER-NEXT: [[TMP17:%.*]] = fmul <4 x double> [[TMP16]], [[TMP15]]
+; SINK-AFTER-NEXT: [[TMP18:%.*]] = fcmp une <4 x double> [[TMP17]], zeroinitializer
+; SINK-AFTER-NEXT: [[TMP19:%.*]] = zext <4 x i1> [[TMP18]] to <4 x i32>
+; SINK-AFTER-NEXT: [[TMP20]] = add <4 x i32> [[VEC_PHI]], [[TMP19]]
; SINK-AFTER-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
-; SINK-AFTER-NEXT: [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10240
-; SINK-AFTER-NEXT: br i1 [[TMP25]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
+; SINK-AFTER-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], 10240
+; SINK-AFTER-NEXT: br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; SINK-AFTER: middle.block:
-; SINK-AFTER-NEXT: [[TMP26:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP24]])
-; SINK-AFTER-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x double> [[TMP19]], i32 3
+; SINK-AFTER-NEXT: [[TMP22:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP20]])
+; SINK-AFTER-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x double> [[TMP15]], i32 3
; SINK-AFTER-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; SINK-AFTER: scalar.ph:
; SINK-AFTER-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi double [ [[J]], [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
; SINK-AFTER-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[B]], [[ENTRY]] ]
; SINK-AFTER-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i32 [ 10240, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
-; SINK-AFTER-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP26]], [[MIDDLE_BLOCK]] ]
+; SINK-AFTER-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP22]], [[MIDDLE_BLOCK]] ]
; SINK-AFTER-NEXT: br label [[FOR_BODY:%.*]]
; SINK-AFTER: for.cond.cleanup:
-; SINK-AFTER-NEXT: [[A_1_LCSSA:%.*]] = phi i32 [ [[A_1:%.*]], [[FOR_BODY]] ], [ [[TMP26]], [[MIDDLE_BLOCK]] ]
+; SINK-AFTER-NEXT: [[A_1_LCSSA:%.*]] = phi i32 [ [[A_1:%.*]], [[FOR_BODY]] ], [ [[TMP22]], [[MIDDLE_BLOCK]] ]
; SINK-AFTER-NEXT: ret i32 [[A_1_LCSSA]]
; SINK-AFTER: for.body:
; SINK-AFTER-NEXT: [[B_ADDR_012:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[ADD_PTR:%.*]], [[FOR_BODY]] ]
; SINK-AFTER-NEXT: [[I_011:%.*]] = phi i32 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[INC1:%.*]], [[FOR_BODY]] ]
; SINK-AFTER-NEXT: [[A_010:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[A_1]], [[FOR_BODY]] ]
-; SINK-AFTER-NEXT: [[SCALAR_RECUR:%.*]] = phi double [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[TMP27:%.*]], [[FOR_BODY]] ]
+; SINK-AFTER-NEXT: [[SCALAR_RECUR:%.*]] = phi double [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[TMP23:%.*]], [[FOR_BODY]] ]
; SINK-AFTER-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[B_ADDR_012]], i64 [[IDXPROM]]
-; SINK-AFTER-NEXT: [[TMP27]] = load double, ptr [[ARRAYIDX]], align 8
-; SINK-AFTER-NEXT: [[MUL:%.*]] = fmul double [[SCALAR_RECUR]], [[TMP27]]
+; SINK-AFTER-NEXT: [[TMP23]] = load double, ptr [[ARRAYIDX]], align 8
+; SINK-AFTER-NEXT: [[MUL:%.*]] = fmul double [[SCALAR_RECUR]], [[TMP23]]
; SINK-AFTER-NEXT: [[TOBOOL:%.*]] = fcmp une double [[MUL]], 0.000000e+00
; SINK-AFTER-NEXT: [[INC:%.*]] = zext i1 [[TOBOOL]] to i32
; SINK-AFTER-NEXT: [[A_1]] = add nsw i32 [[A_010]], [[INC]]
diff --git a/llvm/test/Transforms/LoopVectorize/pointer-induction-unroll.ll b/llvm/test/Transforms/LoopVectorize/pointer-induction-unroll.ll
index 410947704fead1..31d862a3438ae9 100644
--- a/llvm/test/Transforms/LoopVectorize/pointer-induction-unroll.ll
+++ b/llvm/test/Transforms/LoopVectorize/pointer-induction-unroll.ll
@@ -32,35 +32,36 @@ define void @non_constant_scalar_expansion(i32 %0, ptr %call) {
; STRIDED-NEXT: br label [[VECTOR_BODY:%.*]]
; STRIDED: vector.body:
; STRIDED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; STRIDED-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0
-; STRIDED-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], [[TMP1]]
-; STRIDED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr null, i64 [[TMP5]]
-; STRIDED-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 1
-; STRIDED-NEXT: [[TMP7:%.*]] = mul i64 [[TMP6]], [[TMP1]]
-; STRIDED-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr null, i64 [[TMP7]]
-; STRIDED-NEXT: [[TMP8:%.*]] = add i64 [[INDEX]], 2
-; STRIDED-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], [[TMP1]]
-; STRIDED-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP9]]
-; STRIDED-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], 3
-; STRIDED-NEXT: [[TMP11:%.*]] = mul i64 [[TMP10]], [[TMP1]]
-; STRIDED-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr null, i64 [[TMP11]]
; STRIDED-NEXT: [[DOTCAST:%.*]] = trunc i64 [[INDEX]] to i32
; STRIDED-NEXT: [[OFFSET_IDX:%.*]] = add i32 30, [[DOTCAST]]
-; STRIDED-NEXT: [[TMP12:%.*]] = add i32 [[OFFSET_IDX]], 0
-; STRIDED-NEXT: [[TMP13:%.*]] = add i32 [[OFFSET_IDX]], 1
-; STRIDED-NEXT: [[TMP14:%.*]] = add i32 [[OFFSET_IDX]], 2
-; STRIDED-NEXT: [[TMP15:%.*]] = add i32 [[OFFSET_IDX]], 3
-; STRIDED-NEXT: [[TMP16:%.*]] = getelementptr ptr, ptr [[CALL:%.*]], i32 [[TMP12]]
-; STRIDED-NEXT: [[TMP17:%.*]] = getelementptr ptr, ptr [[CALL]], i32 [[TMP13]]
-; STRIDED-NEXT: [[TMP18:%.*]] = getelementptr ptr, ptr [[CALL]], i32 [[TMP14]]
-; STRIDED-NEXT: [[TMP19:%.*]] = getelementptr ptr, ptr [[CALL]], i32 [[TMP15]]
-; STRIDED-NEXT: store ptr [[NEXT_GEP]], ptr [[TMP16]], align 4
-; STRIDED-NEXT: store ptr [[NEXT_GEP2]], ptr [[TMP17]], align 4
-; STRIDED-NEXT: store ptr [[NEXT_GEP3]], ptr [[TMP18]], align 4
-; STRIDED-NEXT: store ptr [[NEXT_GEP4]], ptr [[TMP19]], align 4
+; STRIDED-NEXT: [[TMP3:%.*]] = add i32 [[OFFSET_IDX]], 0
+; STRIDED-NEXT: [[TMP4:%.*]] = add i32 [[OFFSET_IDX]], 1
+; STRIDED-NEXT: [[TMP5:%.*]] = add i32 [[OFFSET_IDX]], 2
+; STRIDED-NEXT: [[TMP6:%.*]] = add i32 [[OFFSET_IDX]], 3
+; STRIDED-NEXT: [[OFFSET_IDX2:%.*]] = mul i64 [[INDEX]], [[TMP1]]
+; STRIDED-NEXT: [[TMP7:%.*]] = mul i64 0, [[TMP1]]
+; STRIDED-NEXT: [[TMP8:%.*]] = add i64 [[OFFSET_IDX2]], [[TMP7]]
+; STRIDED-NEXT: [[TMP9:%.*]] = mul i64 1, [[TMP1]]
+; STRIDED-NEXT: [[TMP10:%.*]] = add i64 [[OFFSET_IDX2]], [[TMP9]]
+; STRIDED-NEXT: [[TMP11:%.*]] = mul i64 2, [[TMP1]]
+; STRIDED-NEXT: [[TMP12:%.*]] = add i64 [[OFFSET_IDX2]], [[TMP11]]
+; STRIDED-NEXT: [[TMP13:%.*]] = mul i64 3, [[TMP1]]
+; STRIDED-NEXT: [[TMP14:%.*]] = add i64 [[OFFSET_IDX2]], [[TMP13]]
+; STRIDED-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr null, i64 [[TMP8]]
+; STRIDED-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr null, i64 [[TMP10]]
+; STRIDED-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr null, i64 [[TMP12]]
+; STRIDED-NEXT: [[TMP18:%.*]] = getelementptr i8, ptr null, i64 [[TMP14]]
+; STRIDED-NEXT: [[TMP19:%.*]] = getelementptr ptr, ptr [[CALL:%.*]], i32 [[TMP3]]
+; STRIDED-NEXT: [[TMP20:%.*]] = getelementptr ptr, ptr [[CALL]], i32 [[TMP4]]
+; STRIDED-NEXT: [[TMP21:%.*]] = getelementptr ptr, ptr [[CALL]], i32 [[TMP5]]
+; STRIDED-NEXT: [[TMP22:%.*]] = getelementptr ptr, ptr [[CALL]], i32 [[TMP6]]
+; STRIDED-NEXT: store ptr [[TMP15]], ptr [[TMP19]], align 4
+; STRIDED-NEXT: store ptr [[TMP16]], ptr [[TMP20]], align 4
+; STRIDED-NEXT: store ptr [[TMP17]], ptr [[TMP21]], align 4
+; STRIDED-NEXT: store ptr [[TMP18]], ptr [[TMP22]], align 4
; STRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
-; STRIDED-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4294967264
-; STRIDED-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; STRIDED-NEXT: [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4294967264
+; STRIDED-NEXT: br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; STRIDED: middle.block:
; STRIDED-NEXT: br i1 false, label [[FOR_END:%.*]], label [[SCALAR_PH]]
; STRIDED: scalar.ph:
@@ -68,13 +69,13 @@ define void @non_constant_scalar_expansion(i32 %0, ptr %call) {
; STRIDED-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ null, [[ENTRY]] ]
; STRIDED-NEXT: br label [[FOR_COND:%.*]]
; STRIDED: for.cond:
-; STRIDED-NEXT: [[TMP21:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_COND]] ]
+; STRIDED-NEXT: [[TMP24:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_COND]] ]
; STRIDED-NEXT: [[P_0:%.*]] = phi ptr [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[ADD_PTR:%.*]], [[FOR_COND]] ]
; STRIDED-NEXT: [[ADD_PTR]] = getelementptr i8, ptr [[P_0]], i32 [[MUL]]
-; STRIDED-NEXT: [[ARRAYIDX:%.*]] = getelementptr ptr, ptr [[CALL]], i32 [[TMP21]]
+; STRIDED-NEXT: [[ARRAYIDX:%.*]] = getelementptr ptr, ptr [[CALL]], i32 [[TMP24]]
; STRIDED-NEXT: store ptr [[P_0]], ptr [[ARRAYIDX]], align 4
-; STRIDED-NEXT: [[INC]] = add i32 [[TMP21]], 1
-; STRIDED-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[TMP21]], 0
+; STRIDED-NEXT: [[INC]] = add i32 [[TMP24]], 1
+; STRIDED-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[TMP24]], 0
; STRIDED-NEXT: br i1 [[TOBOOL_NOT]], label [[FOR_END]], label [[FOR_COND]], !llvm.loop [[LOOP3:![0-9]+]]
; STRIDED: for.end:
; STRIDED-NEXT: ret void
diff --git a/llvm/test/Transforms/LoopVectorize/pointer-induction.ll b/llvm/test/Transforms/LoopVectorize/pointer-induction.ll
index 27f6f7b1cb4e6a..35037968160c61 100644
--- a/llvm/test/Transforms/LoopVectorize/pointer-induction.ll
+++ b/llvm/test/Transforms/LoopVectorize/pointer-induction.ll
@@ -23,56 +23,53 @@ define void @a(ptr readnone %b) {
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE10:%.*]] ]
-; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], -1
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr null, i64 [[TMP3]]
-; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP]], i64 -1
-; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 0
-; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[TMP5]], i32 -3
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP6]], align 1
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = sub i64 0, [[INDEX]]
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[OFFSET_IDX]], 0
+; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr null, i64 [[TMP2]]
+; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP]], i64 -1
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[TMP3]], i32 0
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 -3
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP5]], align 1
; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <4 x i8> [[WIDE_LOAD]], <4 x i8> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT: [[TMP7:%.*]] = icmp eq <4 x i8> [[REVERSE]], zeroinitializer
-; CHECK-NEXT: [[TMP8:%.*]] = xor <4 x i1> [[TMP7]], <i1 true, i1 true, i1 true, i1 true>
-; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i1> [[TMP8]], i32 0
-; CHECK-NEXT: br i1 [[TMP9]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
+; CHECK-NEXT: [[TMP6:%.*]] = icmp eq <4 x i8> [[REVERSE]], zeroinitializer
+; CHECK-NEXT: [[TMP7:%.*]] = xor <4 x i1> [[TMP6]], <i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT: [[TMP8:%.*]] = extractelement <4 x i1> [[TMP7]], i32 0
+; CHECK-NEXT: br i1 [[TMP8]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; CHECK: pred.store.if:
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP]], i64 -1
-; CHECK-NEXT: store i8 95, ptr [[TMP10]], align 1
+; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP]], i64 -1
+; CHECK-NEXT: store i8 95, ptr [[TMP9]], align 1
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]]
; CHECK: pred.store.continue:
-; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x i1> [[TMP8]], i32 1
-; CHECK-NEXT: br i1 [[TMP11]], label [[PRED_STORE_IF5:%.*]], label [[PRED_STORE_CONTINUE6:%.*]]
+; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x i1> [[TMP7]], i32 1
+; CHECK-NEXT: br i1 [[TMP10]], label [[PRED_STORE_IF5:%.*]], label [[PRED_STORE_CONTINUE6:%.*]]
; CHECK: pred.store.if5:
-; CHECK-NEXT: [[TMP12:%.*]] = add i64 [[INDEX]], 1
-; CHECK-NEXT: [[TMP13:%.*]] = mul i64 [[TMP12]], -1
-; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr null, i64 [[TMP13]]
-; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP2]], i64 -1
-; CHECK-NEXT: store i8 95, ptr [[TMP14]], align 1
+; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[OFFSET_IDX]], -1
+; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr null, i64 [[TMP11]]
+; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP2]], i64 -1
+; CHECK-NEXT: store i8 95, ptr [[TMP12]], align 1
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE6]]
; CHECK: pred.store.continue6:
-; CHECK-NEXT: [[TMP15:%.*]] = extractelement <4 x i1> [[TMP8]], i32 2
-; CHECK-NEXT: br i1 [[TMP15]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]]
+; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x i1> [[TMP7]], i32 2
+; CHECK-NEXT: br i1 [[TMP13]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]]
; CHECK: pred.store.if7:
-; CHECK-NEXT: [[TMP16:%.*]] = add i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP17:%.*]] = mul i64 [[TMP16]], -1
-; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP17]]
-; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP3]], i64 -1
-; CHECK-NEXT: store i8 95, ptr [[TMP18]], align 1
+; CHECK-NEXT: [[TMP14:%.*]] = add i64 [[OFFSET_IDX]], -2
+; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP14]]
+; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP3]], i64 -1
+; CHECK-NEXT: store i8 95, ptr [[TMP15]], align 1
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE8]]
; CHECK: pred.store.continue8:
-; CHECK-NEXT: [[TMP19:%.*]] = extractelement <4 x i1> [[TMP8]], i32 3
-; CHECK-NEXT: br i1 [[TMP19]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10]]
+; CHECK-NEXT: [[TMP16:%.*]] = extractelement <4 x i1> [[TMP7]], i32 3
+; CHECK-NEXT: br i1 [[TMP16]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10]]
; CHECK: pred.store.if9:
-; CHECK-NEXT: [[TMP20:%.*]] = add i64 [[INDEX]], 3
-; CHECK-NEXT: [[TMP21:%.*]] = mul i64 [[TMP20]], -1
-; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr null, i64 [[TMP21]]
-; CHECK-NEXT: [[TMP22:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP4]], i64 -1
-; CHECK-NEXT: store i8 95, ptr [[TMP22]], align 1
+; CHECK-NEXT: [[TMP17:%.*]] = add i64 [[OFFSET_IDX]], -3
+; CHECK-NEXT: [[NEXT_GEP4:%.*]] = getelementptr i8, ptr null, i64 [[TMP17]]
+; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[NEXT_GEP4]], i64 -1
+; CHECK-NEXT: store i8 95, ptr [[TMP18]], align 1
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE10]]
; CHECK: pred.store.continue10:
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
-; CHECK-NEXT: [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]
@@ -86,8 +83,8 @@ define void @a(ptr readnone %b) {
; CHECK: for.body:
; CHECK-NEXT: [[C_05:%.*]] = phi ptr [ [[INCDEC_PTR:%.*]], [[IF_END:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds i8, ptr [[C_05]], i64 -1
-; CHECK-NEXT: [[TMP24:%.*]] = load i8, ptr [[INCDEC_PTR]], align 1
-; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i8 [[TMP24]], 0
+; CHECK-NEXT: [[TMP20:%.*]] = load i8, ptr [[INCDEC_PTR]], align 1
+; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i8 [[TMP20]], 0
; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label [[IF_END]], label [[IF_THEN:%.*]]
; CHECK: if.then:
; CHECK-NEXT: store i8 95, ptr [[INCDEC_PTR]], align 1
@@ -145,22 +142,22 @@ define void @pointer_induction_used_as_vector(ptr noalias %start.1, ptr noalias
; CHECK: vector.body:
; CHECK-NEXT: [[POINTER_PHI:%.*]] = phi ptr [ [[START_2]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[TMP2:%.*]] = mul i64 [[TMP1]], 8
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 0, i64 1, i64 2, i64 3>
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[OFFSET_IDX]], 0
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START_1]], i64 [[TMP2]]
-; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 0, i64 1, i64 2, i64 3>
-; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, <4 x ptr> [[TMP3]], i64 1
-; CHECK-NEXT: [[TMP5:%.*]] = getelementptr ptr, ptr [[NEXT_GEP]], i32 0
-; CHECK-NEXT: store <4 x ptr> [[TMP4]], ptr [[TMP5]], align 8
-; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x ptr> [[TMP3]], i32 0
-; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[TMP6]], i32 0
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP7]], align 1
-; CHECK-NEXT: [[TMP8:%.*]] = add <4 x i8> [[WIDE_LOAD]], <i8 1, i8 1, i8 1, i8 1>
-; CHECK-NEXT: store <4 x i8> [[TMP8]], ptr [[TMP7]], align 1
+; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i8, <4 x ptr> [[TMP1]], i64 1
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr ptr, ptr [[NEXT_GEP]], i32 0
+; CHECK-NEXT: store <4 x ptr> [[TMP3]], ptr [[TMP4]], align 8
+; CHECK-NEXT: [[TMP5:%.*]] = extractelement <4 x ptr> [[TMP1]], i32 0
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[TMP5]], i32 0
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i8>, ptr [[TMP6]], align 1
+; CHECK-NEXT: [[TMP7:%.*]] = add <4 x i8> [[WIDE_LOAD]], <i8 1, i8 1, i8 1, i8 1>
+; CHECK-NEXT: store <4 x i8> [[TMP7]], ptr [[TMP6]], align 1
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 4
-; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
diff --git a/llvm/test/Transforms/LoopVectorize/tail-folding-vectorization-factor-1.ll b/llvm/test/Transforms/LoopVectorize/tail-folding-vectorization-factor-1.ll
index 0c659a550b31e1..92ca77bc841ca1 100644
--- a/llvm/test/Transforms/LoopVectorize/tail-folding-vectorization-factor-1.ll
+++ b/llvm/test/Transforms/LoopVectorize/tail-folding-vectorization-factor-1.ll
@@ -92,6 +92,7 @@ define void @VF1-VPWidenCanonicalIVRecipeExe(ptr %ptr1) {
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_STORE_CONTINUE12:%.*]] ]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8
; CHECK-NEXT: [[VEC_IV:%.*]] = add i64 [[INDEX]], 0
; CHECK-NEXT: [[VEC_IV4:%.*]] = add i64 [[INDEX]], 1
; CHECK-NEXT: [[VEC_IV5:%.*]] = add i64 [[INDEX]], 2
@@ -102,39 +103,35 @@ define void @VF1-VPWidenCanonicalIVRecipeExe(ptr %ptr1) {
; CHECK-NEXT: [[TMP3:%.*]] = icmp ule i64 [[VEC_IV6]], 14
; CHECK-NEXT: br i1 [[TMP0]], label [[PRED_STORE_IF:%.*]], label [[PRED_STORE_CONTINUE:%.*]]
; CHECK: pred.store.if:
-; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0
-; CHECK-NEXT: [[TMP5:%.*]] = mul i64 [[TMP4]], 8
-; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PTR1]], i64 [[TMP5]]
+; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 0
+; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[PTR1]], i64 [[TMP4]]
; CHECK-NEXT: store double 0.000000e+00, ptr [[NEXT_GEP]], align 8
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE]]
; CHECK: pred.store.continue:
; CHECK-NEXT: br i1 [[TMP1]], label [[PRED_STORE_IF7:%.*]], label [[PRED_STORE_CONTINUE8:%.*]]
; CHECK: pred.store.if7:
-; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 1
-; CHECK-NEXT: [[TMP7:%.*]] = mul i64 [[TMP6]], 8
-; CHECK-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[PTR1]], i64 [[TMP7]]
+; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 8
+; CHECK-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[PTR1]], i64 [[TMP5]]
; CHECK-NEXT: store double 0.000000e+00, ptr [[NEXT_GEP1]], align 8
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE8]]
; CHECK: pred.store.continue8:
; CHECK-NEXT: br i1 [[TMP2]], label [[PRED_STORE_IF9:%.*]], label [[PRED_STORE_CONTINUE10:%.*]]
; CHECK: pred.store.if9:
-; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[TMP8]], 8
-; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[PTR1]], i64 [[TMP9]]
+; CHECK-NEXT: [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 16
+; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[PTR1]], i64 [[TMP6]]
; CHECK-NEXT: store double 0.000000e+00, ptr [[NEXT_GEP2]], align 8
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE10]]
; CHECK: pred.store.continue10:
; CHECK-NEXT: br i1 [[TMP3]], label [[PRED_STORE_IF11:%.*]], label [[PRED_STORE_CONTINUE12]]
; CHECK: pred.store.if11:
-; CHECK-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], 3
-; CHECK-NEXT: [[TMP11:%.*]] = mul i64 [[TMP10]], 8
-; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[PTR1]], i64 [[TMP11]]
+; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[OFFSET_IDX]], 24
+; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[PTR1]], i64 [[TMP7]]
; CHECK-NEXT: store double 0.000000e+00, ptr [[NEXT_GEP3]], align 8
; CHECK-NEXT: br label [[PRED_STORE_CONTINUE12]]
; CHECK: pred.store.continue12:
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4
-; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16
-; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16
+; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: br i1 true, label [[FOR_COND_CLEANUP:%.*]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
diff --git a/llvm/test/Transforms/LoopVectorize/vplan-sink-scalars-and-merge.ll b/llvm/test/Transforms/LoopVectorize/vplan-sink-scalars-and-merge.ll
index 9b9c3e704852a1..89b3a6da16c1ff 100644
--- a/llvm/test/Transforms/LoopVectorize/vplan-sink-scalars-and-merge.ll
+++ b/llvm/test/Transforms/LoopVectorize/vplan-sink-scalars-and-merge.ll
@@ -1113,8 +1113,10 @@ define void @ptr_induction_remove_dead_recipe(ptr %start, ptr %end) {
; CHECK-NEXT: <x1> vector loop: {
; CHECK-NEXT: vector.body:
; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION
-; CHECK-NEXT: EMIT ir<%ptr.iv> = WIDEN-POINTER-INDUCTION ir<%start>, -1
-; CHECK-NEXT: CLONE ir<%ptr.iv.next> = getelementptr inbounds ir<%ptr.iv>, ir<-1>
+; CHECK-NEXT: vp<[[DEV_IV:%.+]]> = DERIVED-IV ir<0> + vp<%3> * ir<-1>
+; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[DEV_IV]]>, ir<-1>
+; CHECK-NEXT: EMIT vp<[[PTR_IV:%.+]]> = ptradd ir<%start>, vp<[[STEPS]]>
+; CHECK-NEXT: CLONE ir<%ptr.iv.next> = getelementptr inbounds vp<[[PTR_IV]]>, ir<-1>
; CHECK-NEXT: vp<[[VEC_PTR:%.+]]> = vector-pointer (reverse) ir<%ptr.iv.next>
; CHECK-NEXT: WIDEN ir<%l> = load vp<[[VEC_PTR]]>
; CHECK-NEXT: WIDEN ir<%c.1> = icmp eq ir<%l>, ir<0>
@@ -1127,7 +1129,7 @@ define void @ptr_induction_remove_dead_recipe(ptr %start, ptr %end) {
; CHECK-NEXT: Successor(s): pred.store.if, pred.store.continue
; CHECK-EMPTY:
; CHECK-NEXT: pred.store.if:
-; CHECK-NEXT: REPLICATE ir<%ptr.iv.next> = getelementptr inbounds ir<%ptr.iv>, ir<-1>
+; CHECK-NEXT: REPLICATE ir<%ptr.iv.next> = getelementptr inbounds vp<[[PTR_IV]]>, ir<-1>
; CHECK-NEXT: REPLICATE store ir<95>, ir<%ptr.iv.next>
; CHECK-NEXT: Successor(s): pred.store.continue
; CHECK-EMPTY:
>From c6797e6cf65aa7b13a742d0a2412c895c8264651 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Wed, 7 Feb 2024 11:20:03 +0000
Subject: [PATCH 05/13] !fixup address latest comments, thanks!
---
llvm/lib/Transforms/Vectorize/VPlan.h | 2 +-
llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp | 12 +++---------
2 files changed, 4 insertions(+), 10 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 520e45201c6c58..2363683f1a188d 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1391,7 +1391,7 @@ class VPScalarCastRecipe : public VPSingleDefRecipe {
Type *getResultType() const { return ResultTy; }
bool onlyFirstLaneUsed(const VPValue *Op) const override {
- // At the moment, only scalar codegen is implemented.
+ // At the moment, only uniform codegen is implemented.
assert(is_contained(operands(), Op) &&
"Op must be an operand of the recipe");
return true;
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index b1a0ebb5863f7e..ae8afa58956958 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -521,13 +521,9 @@ void VPInstruction::execute(VPTransformState &State) {
if (GeneratedValue->getType()->isVectorTy())
State.set(this, GeneratedValue, Part);
else {
- if (getOpcode() == VPInstruction::ComputeReductionResult) {
- State.set(this, GeneratedValue, VPIteration(Part, 0));
- } else {
- assert((State.VF.isScalar() || vputils::onlyFirstLaneUsed(this)) &&
- "scalar value but not only first lane used");
- State.set(this, GeneratedValue, VPIteration(Part, 0));
- }
+ assert((getOpcode() == VPInstruction::ComputeReductionResult || State.VF.isScalar() || vputils::onlyFirstLaneUsed(this)) &&
+ "scalar value but not only first lane used");
+ State.set(this, GeneratedValue, VPIteration(Part, 0));
}
}
}
@@ -542,8 +538,6 @@ bool VPInstruction::onlyFirstLaneUsed(const VPValue *Op) const {
case Instruction::ICmp:
// TODO: Cover additional opcodes.
return vputils::onlyFirstLaneUsed(this);
- case VPInstruction::ComputeReductionResult:
- return true;
case VPInstruction::ActiveLaneMask:
case VPInstruction::CalculateTripCountMinusVF:
case VPInstruction::CanonicalIVIncrementForPart:
>From 53f293752addd8dff0a07ec9ab97f31a9bd31a23 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Wed, 7 Feb 2024 13:09:18 +0000
Subject: [PATCH 06/13] !fixup fix formatting
---
llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index ae8afa58956958..ed7cac5b809cfb 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -521,7 +521,8 @@ void VPInstruction::execute(VPTransformState &State) {
if (GeneratedValue->getType()->isVectorTy())
State.set(this, GeneratedValue, Part);
else {
- assert((getOpcode() == VPInstruction::ComputeReductionResult || State.VF.isScalar() || vputils::onlyFirstLaneUsed(this)) &&
+ assert((getOpcode() == VPInstruction::ComputeReductionResult ||
+ State.VF.isScalar() || vputils::onlyFirstLaneUsed(this)) &&
"scalar value but not only first lane used");
State.set(this, GeneratedValue, VPIteration(Part, 0));
}
>From e6d2db86aaf62a821f7ff1a9a9c821d9c038230e Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Wed, 7 Feb 2024 14:57:01 +0000
Subject: [PATCH 07/13] !fixup Address latest comments, thanks!
---
llvm/lib/Transforms/Vectorize/VPlan.cpp | 3 +-
llvm/lib/Transforms/Vectorize/VPlan.h | 14 +-
.../lib/Transforms/Vectorize/VPlanRecipes.cpp | 252 +++++++++++++++---
.../Transforms/Vectorize/VPlanTransforms.cpp | 2 +-
4 files changed, 227 insertions(+), 44 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index a1a5f0aff12fb2..b5df58bf548cb9 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -857,7 +857,8 @@ void VPlan::execute(VPTransformState *State) {
Phi = cast<PHINode>(State->get(R.getVPSingleValue(), 0));
} else {
auto *WidenPhi = cast<VPWidenPointerInductionRecipe>(&R);
- assert(!WidenPhi->onlyScalarsGenerated(State->VF.isScalable()));
+ assert(!WidenPhi->onlyScalarsGenerated(State->VF.isScalable()) &&
+ "recipe only generating scalars should have been replaced");
auto *GEP = cast<GetElementPtrInst>(State->get(WidenPhi, 0));
Phi = cast<PHINode>(GEP->getPointerOperand());
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 48e31ef6697254..8538d8f6511b29 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1170,7 +1170,8 @@ class VPInstruction : public VPRecipeWithIRFlags {
/// modeled instruction. \returns the generated value for \p Part.
/// In some cases an existing value is returned rather than a generated
/// one.
- Value *generateInstruction(VPTransformState &State, unsigned Part);
+ Value *generatePerPart(VPTransformState &State, unsigned Part);
+ Value *generatePerLane(VPTransformState &State, const VPIteration &Lane);
#if !defined(NDEBUG)
/// Return true if the VPInstruction is a floating point math operation, i.e.
@@ -2490,11 +2491,6 @@ class VPDerivedIVRecipe : public VPSingleDefRecipe {
/// for floating point inductions.
const FPMathOperator *FPBinOp;
- VPDerivedIVRecipe(InductionDescriptor::InductionKind Kind,
- const FPMathOperator *FPBinOp, VPValue *Start,
- VPCanonicalIVPHIRecipe *CanonicalIV, VPValue *Step)
- : VPSingleDefRecipe(VPDef::VPDerivedIVSC, {Start, CanonicalIV, Step}),
- Kind(Kind), FPBinOp(FPBinOp) {}
public:
VPDerivedIVRecipe(const InductionDescriptor &IndDesc, VPValue *Start,
@@ -2504,9 +2500,9 @@ class VPDerivedIVRecipe : public VPSingleDefRecipe {
dyn_cast_or_null<FPMathOperator>(IndDesc.getInductionBinOp()),
Start, CanonicalIV, Step) {}
- VPDerivedIVRecipe(InductionDescriptor::InductionKind Kind, VPValue *Start,
- VPCanonicalIVPHIRecipe *CanonicalIV, VPValue *Step,
- FPMathOperator *FPBinOp)
+ VPDerivedIVRecipe(InductionDescriptor::InductionKind Kind,
+ const FPMathOperator *FPBinOp, VPValue *Start,
+ VPCanonicalIVPHIRecipe *CanonicalIV, VPValue *Step)
: VPSingleDefRecipe(VPDef::VPDerivedIVSC, {Start, CanonicalIV, Step}),
Kind(Kind), FPBinOp(FPBinOp) {}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 71d82ef3aa6ce7..9ae4320dd6fdc6 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -274,8 +274,206 @@ VPInstruction::VPInstruction(unsigned Opcode,
assert(isFPMathOp() && "this op can't take fast-math flags");
}
-Value *VPInstruction::generateInstruction(VPTransformState &State,
- unsigned Part) {
+Value *VPInstruction::generatePerLane(VPTransformState &State,
+ const VPIteration &Lane) {
+ IRBuilderBase &Builder = State.Builder;
+ Builder.SetCurrentDebugLocation(getDebugLoc());
+
+ if (Instruction::isBinaryOp(getOpcode())) {
+ bool OnlyFirstLaneUsed = vputils::onlyFirstLaneUsed(this);
+ if (Lane.Part != 0 && vputils::onlyFirstPartUsed(this))
+ return OnlyFirstLaneUsed ? State.get(this, VPIteration(0, 0))
+ : State.get(this, 0);
+
+ Value *A = State.get(getOperand(0), Lane);
+ Value *B = State.get(getOperand(1), Lane);
+ auto *Res =
+ Builder.CreateBinOp((Instruction::BinaryOps)getOpcode(), A, B, Name);
+ if (auto *I = dyn_cast<Instruction>(Res))
+ setFlags(I);
+ return Res;
+ }
+
+ switch (getOpcode()) {
+ case VPInstruction::Not: {
+ Value *A = State.get(getOperand(0), Lane);
+ return Builder.CreateNot(A, Name);
+ }
+ case Instruction::Select: {
+ Value *Cond = State.get(getOperand(0), Lane);
+ Value *Op1 = State.get(getOperand(1), Lane);
+ Value *Op2 = State.get(getOperand(2), Lane);
+ return Builder.CreateSelect(Cond, Op1, Op2, Name);
+ }
+
+ case VPInstruction::ActiveLaneMask: {
+ // Get first lane of vector induction variable.
+ Value *VIVElem0 = State.get(getOperand(0), Lane);
+ // Get the original loop tripcount.
+ Value *ScalarTC = State.get(getOperand(1), Lane);
+
+ auto *Int1Ty = Type::getInt1Ty(Builder.getContext());
+ auto *PredTy = VectorType::get(Int1Ty, State.VF);
+ return Builder.CreateIntrinsic(Intrinsic::get_active_lane_mask,
+ {PredTy, ScalarTC->getType()},
+ {VIVElem0, ScalarTC}, nullptr, Name);
+ }
+ case VPInstruction::CalculateTripCountMinusVF: {
+ Value *ScalarTC = State.get(getOperand(0), {0, 0});
+ Value *Step =
+ createStepForVF(Builder, ScalarTC->getType(), State.VF, State.UF);
+ Value *Sub = Builder.CreateSub(ScalarTC, Step);
+ Value *Cmp =
+ Builder.CreateICmp(CmpInst::Predicate::ICMP_UGT, ScalarTC, Step);
+ Value *Zero = ConstantInt::get(ScalarTC->getType(), 0);
+ return Builder.CreateSelect(Cmp, Sub, Zero);
+ }
+ case VPInstruction::CanonicalIVIncrementForPart: {
+ auto *IV = State.get(getOperand(0), VPIteration(0, 0));
+ if (Lane.Part == 0)
+ return IV;
+
+ // The canonical IV is incremented by the vectorization factor (num of SIMD
+ // elements) times the unroll part.
+ Value *Step = createStepForVF(Builder, IV->getType(), State.VF, Lane.Part);
+ return Builder.CreateAdd(IV, Step, Name, hasNoUnsignedWrap(),
+ hasNoSignedWrap());
+ }
+
+ case VPInstruction::BranchOnCond: {
+ if (Lane.Part != 0)
+ return nullptr;
+
+ Value *Cond = State.get(getOperand(0), Lane);
+ VPRegionBlock *ParentRegion = getParent()->getParent();
+ VPBasicBlock *Header = ParentRegion->getEntryBasicBlock();
+
+ // Replace the temporary unreachable terminator with a new conditional
+ // branch, hooking it up to backward destination for exiting blocks now and
+ // to forward destination(s) later when they are created.
+ BranchInst *CondBr =
+ Builder.CreateCondBr(Cond, Builder.GetInsertBlock(), nullptr);
+
+ if (getParent()->isExiting())
+ CondBr->setSuccessor(1, State.CFG.VPBB2IRBB[Header]);
+
+ CondBr->setSuccessor(0, nullptr);
+ Builder.GetInsertBlock()->getTerminator()->eraseFromParent();
+ return CondBr;
+ }
+ case VPInstruction::BranchOnCount: {
+ if (Lane.Part != 0)
+ return nullptr;
+ // First create the compare.
+ Value *IV = State.get(getOperand(0), Lane);
+ Value *TC = State.get(getOperand(1), Lane);
+ Value *Cond = Builder.CreateICmpEQ(IV, TC);
+
+ // Now create the branch.
+ auto *Plan = getParent()->getPlan();
+ VPRegionBlock *TopRegion = Plan->getVectorLoopRegion();
+ VPBasicBlock *Header = TopRegion->getEntry()->getEntryBasicBlock();
+
+ // Replace the temporary unreachable terminator with a new conditional
+ // branch, hooking it up to backward destination (the header) now and to the
+ // forward destination (the exit/middle block) later when it is created.
+ // Note that CreateCondBr expects a valid BB as first argument, so we need
+ // to set it to nullptr later.
+ BranchInst *CondBr = Builder.CreateCondBr(Cond, Builder.GetInsertBlock(),
+ State.CFG.VPBB2IRBB[Header]);
+ CondBr->setSuccessor(0, nullptr);
+ Builder.GetInsertBlock()->getTerminator()->eraseFromParent();
+ return CondBr;
+ }
+ case VPInstruction::ComputeReductionResult: {
+ if (Lane.Part != 0)
+ return State.get(this, VPIteration(0, 0));
+
+ // FIXME: The cross-recipe dependency on VPReductionPHIRecipe is temporary
+ // and will be removed by breaking up the recipe further.
+ auto *PhiR = cast<VPReductionPHIRecipe>(getOperand(0));
+ auto *OrigPhi = cast<PHINode>(PhiR->getUnderlyingValue());
+ // Get its reduction variable descriptor.
+ const RecurrenceDescriptor &RdxDesc = PhiR->getRecurrenceDescriptor();
+
+ RecurKind RK = RdxDesc.getRecurrenceKind();
+
+ State.setDebugLocFrom(getDebugLoc());
+
+ VPValue *LoopExitingDef = getOperand(1);
+ Type *PhiTy = OrigPhi->getType();
+ VectorParts RdxParts(State.UF);
+ for (unsigned Part = 0; Part < State.UF; ++Part)
+ RdxParts[Part] = State.get(LoopExitingDef, Part);
+
+ // If the vector reduction can be performed in a smaller type, we truncate
+ // then extend the loop exit value to enable InstCombine to evaluate the
+ // entire expression in the smaller type.
+ // TODO: Handle this in truncateToMinBW.
+ if (State.VF.isVector() && PhiTy != RdxDesc.getRecurrenceType()) {
+ Type *RdxVecTy = VectorType::get(RdxDesc.getRecurrenceType(), State.VF);
+ for (unsigned Part = 0; Part < State.UF; ++Part)
+ RdxParts[Part] = Builder.CreateTrunc(RdxParts[Part], RdxVecTy);
+ }
+ // Reduce all of the unrolled parts into a single vector.
+ Value *ReducedPartRdx = RdxParts[0];
+ unsigned Op = RecurrenceDescriptor::getOpcode(RK);
+
+ if (PhiR->isOrdered()) {
+ ReducedPartRdx = RdxParts[State.UF - 1];
+ } else {
+ // Floating-point operations should have some FMF to enable the reduction.
+ IRBuilderBase::FastMathFlagGuard FMFG(Builder);
+ Builder.setFastMathFlags(RdxDesc.getFastMathFlags());
+ for (unsigned Part = 1; Part < State.UF; ++Part) {
+ Value *RdxPart = RdxParts[Part];
+ if (Op != Instruction::ICmp && Op != Instruction::FCmp)
+ ReducedPartRdx = Builder.CreateBinOp(
+ (Instruction::BinaryOps)Op, RdxPart, ReducedPartRdx, "bin.rdx");
+ else if (RecurrenceDescriptor::isAnyOfRecurrenceKind(RK)) {
+ TrackingVH<Value> ReductionStartValue =
+ RdxDesc.getRecurrenceStartValue();
+ ReducedPartRdx = createAnyOfOp(Builder, ReductionStartValue, RK,
+ ReducedPartRdx, RdxPart);
+ } else
+ ReducedPartRdx = createMinMaxOp(Builder, RK, ReducedPartRdx, RdxPart);
+ }
+ }
+
+ // Create the reduction after the loop. Note that inloop reductions create
+ // the target reduction in the loop using a Reduction recipe.
+ if (State.VF.isVector() && !PhiR->isInLoop()) {
+ ReducedPartRdx =
+ createTargetReduction(Builder, RdxDesc, ReducedPartRdx, OrigPhi);
+ // If the reduction can be performed in a smaller type, we need to extend
+ // the reduction to the wider type before we branch to the original loop.
+ if (PhiTy != RdxDesc.getRecurrenceType())
+ ReducedPartRdx = RdxDesc.isSigned()
+ ? Builder.CreateSExt(ReducedPartRdx, PhiTy)
+ : Builder.CreateZExt(ReducedPartRdx, PhiTy);
+ }
+
+ // If there were stores of the reduction value to a uniform memory address
+ // inside the loop, create the final store here.
+ if (StoreInst *SI = RdxDesc.IntermediateStore) {
+ auto *NewSI = Builder.CreateAlignedStore(
+ ReducedPartRdx, SI->getPointerOperand(), SI->getAlign());
+ propagateMetadata(NewSI, SI);
+ }
+
+ return ReducedPartRdx;
+ }
+ case VPInstruction::PtrAdd: {
+ auto *P = Builder.CreatePtrAdd(State.get(getOperand(0), Lane),
+ State.get(getOperand(1), Lane), Name);
+ return P;
+ }
+ default:
+ llvm_unreachable("Unsupported opcode for instruction");
+ }
+}
+
+Value *VPInstruction::generatePerPart(VPTransformState &State, unsigned Part) {
IRBuilderBase &Builder = State.Builder;
Builder.SetCurrentDebugLocation(getDebugLoc());
@@ -347,26 +545,6 @@ Value *VPInstruction::generateInstruction(VPTransformState &State,
Value *V2 = State.get(getOperand(1), Part);
return Builder.CreateVectorSplice(PartMinus1, V2, -1, Name);
}
- case VPInstruction::CalculateTripCountMinusVF: {
- Value *ScalarTC = State.get(getOperand(0), {0, 0});
- Value *Step =
- createStepForVF(Builder, ScalarTC->getType(), State.VF, State.UF);
- Value *Sub = Builder.CreateSub(ScalarTC, Step);
- Value *Cmp = Builder.CreateICmp(CmpInst::Predicate::ICMP_UGT, ScalarTC, Step);
- Value *Zero = ConstantInt::get(ScalarTC->getType(), 0);
- return Builder.CreateSelect(Cmp, Sub, Zero);
- }
- case VPInstruction::CanonicalIVIncrementForPart: {
- auto *IV = State.get(getOperand(0), VPIteration(0, 0));
- if (Part == 0)
- return IV;
-
- // The canonical IV is incremented by the vectorization factor (num of SIMD
- // elements) times the unroll part.
- Value *Step = createStepForVF(Builder, IV->getType(), State.VF, Part);
- return Builder.CreateAdd(IV, Step, Name, hasNoUnsignedWrap(),
- hasNoSignedWrap());
- }
case VPInstruction::BranchOnCond: {
if (Part != 0)
return nullptr;
@@ -532,20 +710,28 @@ void VPInstruction::execute(VPTransformState &State) {
if (hasFastMathFlags())
State.Builder.setFastMathFlags(getFastMathFlags());
for (unsigned Part = 0; Part < State.UF; ++Part) {
- Value *GeneratedValue = generateInstruction(State, Part);
- if (!GeneratedValue)
+ if (getOpcode() == VPInstruction::ComputeReductionResult ||
+ getOpcode() == VPInstruction::PtrAdd ||
+ vputils::onlyFirstLaneUsed(this)) {
+ unsigned NumLanes =
+ vputils::onlyFirstLaneUsed(this) ? 1 : State.VF.getKnownMinValue();
+ if (getOpcode() == VPInstruction::ComputeReductionResult)
+ NumLanes = 1;
+ for (unsigned Lane = 0; Lane != NumLanes; ++Lane) {
+ Value *P = generatePerLane(State, VPIteration(Part, Lane));
+ State.set(this, P, VPIteration(Part, Lane));
+ }
+
continue;
+ }
+
+ Value *GeneratedValue = generatePerPart(State, Part);
if (!hasResult())
continue;
- assert(GeneratedValue && "generateInstruction must produce a value");
- if (GeneratedValue->getType()->isVectorTy())
- State.set(this, GeneratedValue, Part);
- else {
- assert((getOpcode() == VPInstruction::ComputeReductionResult ||
- State.VF.isScalar() || vputils::onlyFirstLaneUsed(this)) &&
- "scalar value but not only first lane used");
- State.set(this, GeneratedValue, VPIteration(Part, 0));
- }
+ assert(GeneratedValue &&
+ (State.VF.isScalar() || GeneratedValue->getType()->isVectorTy()) &&
+ "generateInstruction must produce a vector value");
+ State.set(this, GeneratedValue, Part);
}
}
bool VPInstruction::onlyFirstLaneUsed(const VPValue *Op) const {
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 6f5c7090455ca6..b60a4d3a7f6127 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -500,7 +500,7 @@ static VPValue *createScalarIVSteps(VPlan &Plan,
VPCanonicalIVPHIRecipe *CanonicalIV = Plan.getCanonicalIV();
VPSingleDefRecipe *BaseIV = CanonicalIV;
if (!CanonicalIV->isCanonical(Kind, StartV, Step)) {
- BaseIV = new VPDerivedIVRecipe(Kind, StartV, CanonicalIV, Step, FPBinOp);
+ BaseIV = new VPDerivedIVRecipe(Kind, FPBinOp, StartV, CanonicalIV, Step);
HeaderVPBB->insert(BaseIV, IP);
}
>From 506533128533f053a21bc0c5b8fe723af2f6ce03 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Wed, 7 Feb 2024 21:28:04 +0000
Subject: [PATCH 08/13] !Fixup split generateInstruction into per-part and per
lane.
---
llvm/lib/Transforms/Vectorize/VPlan.h | 2 +
.../lib/Transforms/Vectorize/VPlanRecipes.cpp | 175 ++----------------
2 files changed, 15 insertions(+), 162 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 8538d8f6511b29..70d6032d1a0931 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1166,6 +1166,8 @@ class VPInstruction : public VPRecipeWithIRFlags {
/// An optional name that can be used for the generated IR instruction.
const std::string Name;
+ bool generatesScalars() const;
+
/// Utility method serving execute(): generates a single instance of the
/// modeled instruction. \returns the generated value for \p Part.
/// In some cases an existing value is returned rather than a generated
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 9ae4320dd6fdc6..e089f58f12ea18 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -274,17 +274,24 @@ VPInstruction::VPInstruction(unsigned Opcode,
assert(isFPMathOp() && "this op can't take fast-math flags");
}
+bool VPInstruction::generatesScalars() const {
+ if (Opcode == VPInstruction::FirstOrderRecurrenceSplice ||
+ Opcode == VPInstruction::ActiveLaneMask)
+ return false;
+
+ return Opcode == VPInstruction::ComputeReductionResult ||
+ Opcode == VPInstruction::PtrAdd || vputils::onlyFirstLaneUsed(this);
+}
+
Value *VPInstruction::generatePerLane(VPTransformState &State,
const VPIteration &Lane) {
IRBuilderBase &Builder = State.Builder;
Builder.SetCurrentDebugLocation(getDebugLoc());
- if (Instruction::isBinaryOp(getOpcode())) {
- bool OnlyFirstLaneUsed = vputils::onlyFirstLaneUsed(this);
- if (Lane.Part != 0 && vputils::onlyFirstPartUsed(this))
- return OnlyFirstLaneUsed ? State.get(this, VPIteration(0, 0))
- : State.get(this, 0);
+ if (Lane.Part > 0 && vputils::onlyFirstPartUsed(this))
+ return State.get(this, VPIteration(0, Lane.Lane.getKnownLane()));
+ if (Instruction::isBinaryOp(getOpcode())) {
Value *A = State.get(getOperand(0), Lane);
Value *B = State.get(getOperand(1), Lane);
auto *Res =
@@ -305,19 +312,6 @@ Value *VPInstruction::generatePerLane(VPTransformState &State,
Value *Op2 = State.get(getOperand(2), Lane);
return Builder.CreateSelect(Cond, Op1, Op2, Name);
}
-
- case VPInstruction::ActiveLaneMask: {
- // Get first lane of vector induction variable.
- Value *VIVElem0 = State.get(getOperand(0), Lane);
- // Get the original loop tripcount.
- Value *ScalarTC = State.get(getOperand(1), Lane);
-
- auto *Int1Ty = Type::getInt1Ty(Builder.getContext());
- auto *PredTy = VectorType::get(Int1Ty, State.VF);
- return Builder.CreateIntrinsic(Intrinsic::get_active_lane_mask,
- {PredTy, ScalarTC->getType()},
- {VIVElem0, ScalarTC}, nullptr, Name);
- }
case VPInstruction::CalculateTripCountMinusVF: {
Value *ScalarTC = State.get(getOperand(0), {0, 0});
Value *Step =
@@ -545,146 +539,6 @@ Value *VPInstruction::generatePerPart(VPTransformState &State, unsigned Part) {
Value *V2 = State.get(getOperand(1), Part);
return Builder.CreateVectorSplice(PartMinus1, V2, -1, Name);
}
- case VPInstruction::BranchOnCond: {
- if (Part != 0)
- return nullptr;
-
- Value *Cond = State.get(getOperand(0), VPIteration(Part, 0));
- VPRegionBlock *ParentRegion = getParent()->getParent();
- VPBasicBlock *Header = ParentRegion->getEntryBasicBlock();
-
- // Replace the temporary unreachable terminator with a new conditional
- // branch, hooking it up to backward destination for exiting blocks now and
- // to forward destination(s) later when they are created.
- BranchInst *CondBr =
- Builder.CreateCondBr(Cond, Builder.GetInsertBlock(), nullptr);
-
- if (getParent()->isExiting())
- CondBr->setSuccessor(1, State.CFG.VPBB2IRBB[Header]);
-
- CondBr->setSuccessor(0, nullptr);
- Builder.GetInsertBlock()->getTerminator()->eraseFromParent();
- return CondBr;
- }
- case VPInstruction::BranchOnCount: {
- if (Part != 0)
- return nullptr;
- // First create the compare.
- Value *IV = State.get(getOperand(0), VPIteration(0, 0));
- Value *TC = State.get(getOperand(1), VPIteration(0, 0));
- Value *Cond = Builder.CreateICmpEQ(IV, TC);
-
- // Now create the branch.
- auto *Plan = getParent()->getPlan();
- VPRegionBlock *TopRegion = Plan->getVectorLoopRegion();
- VPBasicBlock *Header = TopRegion->getEntry()->getEntryBasicBlock();
-
- // Replace the temporary unreachable terminator with a new conditional
- // branch, hooking it up to backward destination (the header) now and to the
- // forward destination (the exit/middle block) later when it is created.
- // Note that CreateCondBr expects a valid BB as first argument, so we need
- // to set it to nullptr later.
- BranchInst *CondBr = Builder.CreateCondBr(Cond, Builder.GetInsertBlock(),
- State.CFG.VPBB2IRBB[Header]);
- CondBr->setSuccessor(0, nullptr);
- Builder.GetInsertBlock()->getTerminator()->eraseFromParent();
- return CondBr;
- }
- case VPInstruction::ComputeReductionResult: {
- if (Part != 0)
- return State.get(this, VPIteration(0, 0));
-
- // FIXME: The cross-recipe dependency on VPReductionPHIRecipe is temporary
- // and will be removed by breaking up the recipe further.
- auto *PhiR = cast<VPReductionPHIRecipe>(getOperand(0));
- auto *OrigPhi = cast<PHINode>(PhiR->getUnderlyingValue());
- // Get its reduction variable descriptor.
- const RecurrenceDescriptor &RdxDesc = PhiR->getRecurrenceDescriptor();
-
- RecurKind RK = RdxDesc.getRecurrenceKind();
-
- State.setDebugLocFrom(getDebugLoc());
-
- VPValue *LoopExitingDef = getOperand(1);
- Type *PhiTy = OrigPhi->getType();
- VectorParts RdxParts(State.UF);
- for (unsigned Part = 0; Part < State.UF; ++Part)
- RdxParts[Part] = State.get(LoopExitingDef, Part);
-
- // If the vector reduction can be performed in a smaller type, we truncate
- // then extend the loop exit value to enable InstCombine to evaluate the
- // entire expression in the smaller type.
- // TODO: Handle this in truncateToMinBW.
- if (State.VF.isVector() && PhiTy != RdxDesc.getRecurrenceType()) {
- Type *RdxVecTy = VectorType::get(RdxDesc.getRecurrenceType(), State.VF);
- for (unsigned Part = 0; Part < State.UF; ++Part)
- RdxParts[Part] = Builder.CreateTrunc(RdxParts[Part], RdxVecTy);
- }
- // Reduce all of the unrolled parts into a single vector.
- Value *ReducedPartRdx = RdxParts[0];
- unsigned Op = RecurrenceDescriptor::getOpcode(RK);
-
- if (PhiR->isOrdered()) {
- ReducedPartRdx = RdxParts[State.UF - 1];
- } else {
- // Floating-point operations should have some FMF to enable the reduction.
- IRBuilderBase::FastMathFlagGuard FMFG(Builder);
- Builder.setFastMathFlags(RdxDesc.getFastMathFlags());
- for (unsigned Part = 1; Part < State.UF; ++Part) {
- Value *RdxPart = RdxParts[Part];
- if (Op != Instruction::ICmp && Op != Instruction::FCmp)
- ReducedPartRdx = Builder.CreateBinOp(
- (Instruction::BinaryOps)Op, RdxPart, ReducedPartRdx, "bin.rdx");
- else if (RecurrenceDescriptor::isAnyOfRecurrenceKind(RK)) {
- TrackingVH<Value> ReductionStartValue =
- RdxDesc.getRecurrenceStartValue();
- ReducedPartRdx = createAnyOfOp(Builder, ReductionStartValue, RK,
- ReducedPartRdx, RdxPart);
- } else
- ReducedPartRdx = createMinMaxOp(Builder, RK, ReducedPartRdx, RdxPart);
- }
- }
-
- // Create the reduction after the loop. Note that inloop reductions create
- // the target reduction in the loop using a Reduction recipe.
- if (State.VF.isVector() && !PhiR->isInLoop()) {
- ReducedPartRdx =
- createTargetReduction(Builder, RdxDesc, ReducedPartRdx, OrigPhi);
- // If the reduction can be performed in a smaller type, we need to extend
- // the reduction to the wider type before we branch to the original loop.
- if (PhiTy != RdxDesc.getRecurrenceType())
- ReducedPartRdx = RdxDesc.isSigned()
- ? Builder.CreateSExt(ReducedPartRdx, PhiTy)
- : Builder.CreateZExt(ReducedPartRdx, PhiTy);
- }
-
- // If there were stores of the reduction value to a uniform memory address
- // inside the loop, create the final store here.
- if (StoreInst *SI = RdxDesc.IntermediateStore) {
- auto *NewSI = Builder.CreateAlignedStore(
- ReducedPartRdx, SI->getPointerOperand(), SI->getAlign());
- propagateMetadata(NewSI, SI);
- }
-
- return ReducedPartRdx;
- }
- case VPInstruction::PtrAdd: {
- if (vputils::onlyFirstLaneUsed(this)) {
- auto *P = Builder.CreatePtrAdd(
- State.get(getOperand(0), VPIteration(Part, 0)),
- State.get(getOperand(1), VPIteration(Part, 0)), Name);
- State.set(this, P, VPIteration(Part, 0));
- } else {
- for (unsigned Lane = 0; Lane != State.VF.getKnownMinValue(); ++Lane) {
- Value *P = Builder.CreatePtrAdd(
- State.get(getOperand(0), VPIteration(Part, Lane)),
- State.get(getOperand(1), VPIteration(Part, Lane)), Name);
-
- State.set(this, P, VPIteration(Part, Lane));
- }
- }
- return nullptr;
- }
default:
llvm_unreachable("Unsupported opcode for instruction");
}
@@ -710,9 +564,7 @@ void VPInstruction::execute(VPTransformState &State) {
if (hasFastMathFlags())
State.Builder.setFastMathFlags(getFastMathFlags());
for (unsigned Part = 0; Part < State.UF; ++Part) {
- if (getOpcode() == VPInstruction::ComputeReductionResult ||
- getOpcode() == VPInstruction::PtrAdd ||
- vputils::onlyFirstLaneUsed(this)) {
+ if (generatesScalars()) {
unsigned NumLanes =
vputils::onlyFirstLaneUsed(this) ? 1 : State.VF.getKnownMinValue();
if (getOpcode() == VPInstruction::ComputeReductionResult)
@@ -721,7 +573,6 @@ void VPInstruction::execute(VPTransformState &State) {
Value *P = generatePerLane(State, VPIteration(Part, Lane));
State.set(this, P, VPIteration(Part, Lane));
}
-
continue;
}
>From f38d682089dd87df9fcfd9db6f2689367d015089 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Wed, 7 Feb 2024 22:03:16 +0000
Subject: [PATCH 09/13] !fixup address comments in VPlanTransforms.cpp, thanks!
---
.../Transforms/Vectorize/VPlanTransforms.cpp | 18 ++++++++++--------
.../lib/Transforms/Vectorize/VPlanTransforms.h | 1 +
2 files changed, 11 insertions(+), 8 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index b60a4d3a7f6127..8448c2bd3dd13c 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -491,11 +491,11 @@ void VPlanTransforms::removeDeadRecipes(VPlan &Plan) {
static VPValue *createScalarIVSteps(VPlan &Plan,
InductionDescriptor::InductionKind Kind,
+ Instruction::BinaryOps InductionOpcode,
+ FPMathOperator *FPBinOp,
ScalarEvolution &SE, Instruction *TruncI,
VPValue *StartV, VPValue *Step,
- Instruction::BinaryOps InductionOpcode,
- VPBasicBlock::iterator IP,
- FPMathOperator *FPBinOp = nullptr) {
+ VPBasicBlock::iterator IP) {
VPBasicBlock *HeaderVPBB = Plan.getVectorLoopRegion()->getEntryBasicBlock();
VPCanonicalIVPHIRecipe *CanonicalIV = Plan.getCanonicalIV();
VPSingleDefRecipe *BaseIV = CanonicalIV;
@@ -542,6 +542,8 @@ void VPlanTransforms::optimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
bool HasOnlyVectorVFs = !Plan.hasVF(ElementCount::getFixed(1));
VPBasicBlock::iterator InsertPt = HeaderVPBB->getFirstNonPhi();
for (VPRecipeBase &Phi : HeaderVPBB->phis()) {
+ // Replace wide pointer inductions which have only their scalars used by
+ // PtrAdd(IndStart, ScalarIVSteps (0, Step)).
if (auto *PtrIV = dyn_cast<VPWidenPointerInductionRecipe>(&Phi)) {
if (!PtrIV->onlyScalarsGenerated(Plan.hasScalableVF()))
continue;
@@ -551,8 +553,8 @@ void VPlanTransforms::optimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
ConstantInt::get(ID.getStep()->getType(), 0));
VPValue *StepV = PtrIV->getOperand(1);
VPRecipeBase *Steps =
- createScalarIVSteps(Plan, InductionDescriptor::IK_IntInduction, SE,
- nullptr, StartV, StepV, Instruction::Add,
+ createScalarIVSteps(Plan, InductionDescriptor::IK_IntInduction, Instruction::Add, nullptr, SE,
+ nullptr, StartV, StepV,
InsertPt)
->getDefiningRecipe();
@@ -576,9 +578,9 @@ void VPlanTransforms::optimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
const InductionDescriptor &ID = WideIV->getInductionDescriptor();
VPValue *Steps = createScalarIVSteps(
- Plan, ID.getKind(), SE, WideIV->getTruncInst(), WideIV->getStartValue(),
- WideIV->getStepValue(), ID.getInductionOpcode(), InsertPt,
- dyn_cast_or_null<FPMathOperator>(ID.getInductionBinOp()));
+ Plan, ID.getKind(), ID.getInductionOpcode(), dyn_cast_or_null<FPMathOperator>(ID.getInductionBinOp()), SE, WideIV->getTruncInst(), WideIV->getStartValue(),
+ WideIV->getStepValue(), InsertPt
+ );
// Update scalar users of IV to use Step instead.
if (!HasOnlyVectorVFs)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
index 3bf91115debb7d..c59f196e9b0506 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
@@ -107,6 +107,7 @@ struct VPlanTransforms {
/// provide them by building scalar steps off of the canonical scalar IV and
/// update the original IV's users. This is an optional optimization to reduce
/// the needs of vector extracts.
+ /// If all users of VPWidenPointerInductionRecipe only use its scalar values, replace it with a PtrAdd (IndStart, ScalarIVSteps (0, Step)).
static void optimizeInductions(VPlan &Plan, ScalarEvolution &SE);
/// Remove redundant EpxandSCEVRecipes in \p Plan's entry block by replacing
>From 133776f783c3d2ed1ea831c19518f9c5aa12f36c Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Mon, 26 Feb 2024 21:50:36 +0000
Subject: [PATCH 10/13] !fixup fix things after update to main.
---
llvm/lib/Transforms/Vectorize/VPlan.cpp | 9 +-
.../lib/Transforms/Vectorize/VPlanRecipes.cpp | 157 +++++++++++++++++-
.../Transforms/Vectorize/VPlanTransforms.cpp | 2 +
.../Transforms/Vectorize/VPlanTransforms.h | 8 -
4 files changed, 152 insertions(+), 24 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index bf1af312fe4031..a0810af195d380 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -221,13 +221,8 @@ VPTransformState::VPTransformState(ElementCount VF, unsigned UF, LoopInfo *LI,
TypeAnalysis(Plan->getCanonicalIV()->getScalarType(), Ctx) {}
Value *VPTransformState::get(VPValue *Def, const VPIteration &Instance) {
- if (Def->isLiveIn()) {
- if (Value *V = Def->getLiveInIRValue())
- return V;
- if (hasScalarValue(Def, VPIteration(0, 0))) {
- return Data.PerPartScalars[Def][0][0];
- }
- }
+ if (Def->isLiveIn())
+ return Def->getLiveInIRValue();
if (hasScalarValue(Def, Instance)) {
return Data
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index fd946cfb16c54f..6b767c163c7991 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -302,15 +302,10 @@ Value *VPInstruction::generatePerPart(VPTransformState &State, unsigned Part) {
if (Instruction::isBinaryOp(getOpcode())) {
bool OnlyFirstLaneUsed = vputils::onlyFirstLaneUsed(this);
if (Part != 0 && vputils::onlyFirstPartUsed(this))
- return OnlyFirstLaneUsed ? State.get(this, VPIteration(0, 0))
- : State.get(this, 0);
-
- Value *A = OnlyFirstLaneUsed
- ? State.get(getOperand(0), VPIteration(Part, 0))
- : State.get(getOperand(0), Part);
- Value *B = OnlyFirstLaneUsed
- ? State.get(getOperand(1), VPIteration(Part, 0))
- : State.get(getOperand(1), Part);
+ return State.get(this, 0, OnlyFirstLaneUsed);
+
+ Value *A = State.get(getOperand(0), Part, OnlyFirstLaneUsed);
+ Value *B = State.get(getOperand(1), Part, OnlyFirstLaneUsed);
auto *Res =
Builder.CreateBinOp((Instruction::BinaryOps)getOpcode(), A, B, Name);
if (auto *I = dyn_cast<Instruction>(Res))
@@ -367,6 +362,150 @@ Value *VPInstruction::generatePerPart(VPTransformState &State, unsigned Part) {
Value *V2 = State.get(getOperand(1), Part);
return Builder.CreateVectorSplice(PartMinus1, V2, -1, Name);
}
+ case VPInstruction::CalculateTripCountMinusVF: {
+ Value *ScalarTC = State.get(getOperand(0), {0, 0});
+ Value *Step =
+ createStepForVF(Builder, ScalarTC->getType(), State.VF, State.UF);
+ Value *Sub = Builder.CreateSub(ScalarTC, Step);
+ Value *Cmp =
+ Builder.CreateICmp(CmpInst::Predicate::ICMP_UGT, ScalarTC, Step);
+ Value *Zero = ConstantInt::get(ScalarTC->getType(), 0);
+ return Builder.CreateSelect(Cmp, Sub, Zero);
+ }
+ case VPInstruction::CanonicalIVIncrementForPart: {
+ auto *IV = State.get(getOperand(0), VPIteration(0, 0));
+ if (Part == 0)
+ return IV;
+
+ // The canonical IV is incremented by the vectorization factor (num of SIMD
+ // elements) times the unroll part.
+ Value *Step = createStepForVF(Builder, IV->getType(), State.VF, Part);
+ return Builder.CreateAdd(IV, Step, Name, hasNoUnsignedWrap(),
+ hasNoSignedWrap());
+ }
+ case VPInstruction::BranchOnCond: {
+ if (Part != 0)
+ return nullptr;
+
+ Value *Cond = State.get(getOperand(0), VPIteration(Part, 0));
+ VPRegionBlock *ParentRegion = getParent()->getParent();
+ VPBasicBlock *Header = ParentRegion->getEntryBasicBlock();
+
+ // Replace the temporary unreachable terminator with a new conditional
+ // branch, hooking it up to backward destination for exiting blocks now and
+ // to forward destination(s) later when they are created.
+ BranchInst *CondBr =
+ Builder.CreateCondBr(Cond, Builder.GetInsertBlock(), nullptr);
+
+ if (getParent()->isExiting())
+ CondBr->setSuccessor(1, State.CFG.VPBB2IRBB[Header]);
+
+ CondBr->setSuccessor(0, nullptr);
+ Builder.GetInsertBlock()->getTerminator()->eraseFromParent();
+ return CondBr;
+ }
+ case VPInstruction::BranchOnCount: {
+ if (Part != 0)
+ return nullptr;
+ // First create the compare.
+ Value *IV = State.get(getOperand(0), Part, /*IsScalar*/ true);
+ Value *TC = State.get(getOperand(1), Part, /*IsScalar*/ true);
+ Value *Cond = Builder.CreateICmpEQ(IV, TC);
+
+ // Now create the branch.
+ auto *Plan = getParent()->getPlan();
+ VPRegionBlock *TopRegion = Plan->getVectorLoopRegion();
+ VPBasicBlock *Header = TopRegion->getEntry()->getEntryBasicBlock();
+
+ // Replace the temporary unreachable terminator with a new conditional
+ // branch, hooking it up to backward destination (the header) now and to the
+ // forward destination (the exit/middle block) later when it is created.
+ // Note that CreateCondBr expects a valid BB as first argument, so we need
+ // to set it to nullptr later.
+ BranchInst *CondBr = Builder.CreateCondBr(Cond, Builder.GetInsertBlock(),
+ State.CFG.VPBB2IRBB[Header]);
+ CondBr->setSuccessor(0, nullptr);
+ Builder.GetInsertBlock()->getTerminator()->eraseFromParent();
+ return CondBr;
+ }
+ case VPInstruction::ComputeReductionResult: {
+ if (Part != 0)
+ return State.get(this, 0, /*IsScalar*/ true);
+
+ // FIXME: The cross-recipe dependency on VPReductionPHIRecipe is temporary
+ // and will be removed by breaking up the recipe further.
+ auto *PhiR = cast<VPReductionPHIRecipe>(getOperand(0));
+ auto *OrigPhi = cast<PHINode>(PhiR->getUnderlyingValue());
+ // Get its reduction variable descriptor.
+ const RecurrenceDescriptor &RdxDesc = PhiR->getRecurrenceDescriptor();
+
+ RecurKind RK = RdxDesc.getRecurrenceKind();
+
+ State.setDebugLocFrom(getDebugLoc());
+
+ VPValue *LoopExitingDef = getOperand(1);
+ Type *PhiTy = OrigPhi->getType();
+ VectorParts RdxParts(State.UF);
+ for (unsigned Part = 0; Part < State.UF; ++Part)
+ RdxParts[Part] = State.get(LoopExitingDef, Part, PhiR->isInLoop());
+
+ // If the vector reduction can be performed in a smaller type, we truncate
+ // then extend the loop exit value to enable InstCombine to evaluate the
+ // entire expression in the smaller type.
+ // TODO: Handle this in truncateToMinBW.
+ if (State.VF.isVector() && PhiTy != RdxDesc.getRecurrenceType()) {
+ Type *RdxVecTy = VectorType::get(RdxDesc.getRecurrenceType(), State.VF);
+ for (unsigned Part = 0; Part < State.UF; ++Part)
+ RdxParts[Part] = Builder.CreateTrunc(RdxParts[Part], RdxVecTy);
+ }
+ // Reduce all of the unrolled parts into a single vector.
+ Value *ReducedPartRdx = RdxParts[0];
+ unsigned Op = RecurrenceDescriptor::getOpcode(RK);
+
+ if (PhiR->isOrdered()) {
+ ReducedPartRdx = RdxParts[State.UF - 1];
+ } else {
+ // Floating-point operations should have some FMF to enable the reduction.
+ IRBuilderBase::FastMathFlagGuard FMFG(Builder);
+ Builder.setFastMathFlags(RdxDesc.getFastMathFlags());
+ for (unsigned Part = 1; Part < State.UF; ++Part) {
+ Value *RdxPart = RdxParts[Part];
+ if (Op != Instruction::ICmp && Op != Instruction::FCmp)
+ ReducedPartRdx = Builder.CreateBinOp(
+ (Instruction::BinaryOps)Op, RdxPart, ReducedPartRdx, "bin.rdx");
+ else if (RecurrenceDescriptor::isAnyOfRecurrenceKind(RK)) {
+ TrackingVH<Value> ReductionStartValue =
+ RdxDesc.getRecurrenceStartValue();
+ ReducedPartRdx = createAnyOfOp(Builder, ReductionStartValue, RK,
+ ReducedPartRdx, RdxPart);
+ } else
+ ReducedPartRdx = createMinMaxOp(Builder, RK, ReducedPartRdx, RdxPart);
+ }
+ }
+
+ // Create the reduction after the loop. Note that inloop reductions create
+ // the target reduction in the loop using a Reduction recipe.
+ if (State.VF.isVector() && !PhiR->isInLoop()) {
+ ReducedPartRdx =
+ createTargetReduction(Builder, RdxDesc, ReducedPartRdx, OrigPhi);
+ // If the reduction can be performed in a smaller type, we need to extend
+ // the reduction to the wider type before we branch to the original loop.
+ if (PhiTy != RdxDesc.getRecurrenceType())
+ ReducedPartRdx = RdxDesc.isSigned()
+ ? Builder.CreateSExt(ReducedPartRdx, PhiTy)
+ : Builder.CreateZExt(ReducedPartRdx, PhiTy);
+ }
+
+ // If there were stores of the reduction value to a uniform memory address
+ // inside the loop, create the final store here.
+ if (StoreInst *SI = RdxDesc.IntermediateStore) {
+ auto *NewSI = Builder.CreateAlignedStore(
+ ReducedPartRdx, SI->getPointerOperand(), SI->getAlign());
+ propagateMetadata(NewSI, SI);
+ }
+
+ return ReducedPartRdx;
+ }
default:
llvm_unreachable("Unsupported opcode for instruction");
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index e3b918b61c332d..a497384d639a9d 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -552,6 +552,8 @@ static VPValue *createScalarIVSteps(VPlan &Plan,
/// provide them by building scalar steps off of the canonical scalar IV and
/// update the original IV's users. This is an optional optimization to reduce
/// the needs of vector extracts.
+/// If all users of VPWidenPointerInductionRecipe only use its scalar values,
+/// replace it with a PtrAdd (IndStart, ScalarIVSteps (0, Step)).
static void optimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
SmallVector<VPRecipeBase *> ToRemove;
VPBasicBlock *HeaderVPBB = Plan.getVectorLoopRegion()->getEntryBasicBlock();
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
index 93d77a40d58e65..ff83c3f083b093 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.h
@@ -84,14 +84,6 @@ struct VPlanTransforms {
const MapVector<Instruction *, uint64_t> &MinBWs,
LLVMContext &Ctx);
-
- /// If any user of a VPWidenIntOrFpInductionRecipe needs scalar values,
- /// provide them by building scalar steps off of the canonical scalar IV and
- /// update the original IV's users. This is an optional optimization to reduce
- /// the needs of vector extracts.
- /// If all users of VPWidenPointerInductionRecipe only use its scalar values, replace it with a PtrAdd (IndStart, ScalarIVSteps (0, Step)).
- static void optimizeInductions(VPlan &Plan, ScalarEvolution &SE);
-
/// Drop poison flags from recipes that may generate a poison value that is
/// used after vectorization, even when their operands are not poison. Those
/// recipes meet the following conditions:
>From 0bb9f5cb4a2b736c765921a3eb5d0a4b11197b3e Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Fri, 1 Mar 2024 14:09:14 +0000
Subject: [PATCH 11/13] !fixup fix formatting.
---
llvm/lib/Transforms/Vectorize/VPlan.h | 1 -
llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp | 3 +--
llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp | 13 +++++++------
3 files changed, 8 insertions(+), 9 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 529bbba6aff932..1cd505676d6094 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -2492,7 +2492,6 @@ class VPDerivedIVRecipe : public VPSingleDefRecipe {
/// for floating point inductions.
const FPMathOperator *FPBinOp;
-
public:
VPDerivedIVRecipe(const InductionDescriptor &IndDesc, VPValue *Start,
VPCanonicalIVPHIRecipe *CanonicalIV, VPValue *Step)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 6b767c163c7991..db5298bdf8107d 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -275,8 +275,7 @@ VPInstruction::VPInstruction(unsigned Opcode,
}
bool VPInstruction::generatesScalars() const {
- return
- Opcode == VPInstruction::PtrAdd;
+ return Opcode == VPInstruction::PtrAdd;
}
Value *VPInstruction::generatePerLane(VPTransformState &State,
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index a497384d639a9d..3729b24c337c30 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -571,9 +571,9 @@ static void optimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
ConstantInt::get(ID.getStep()->getType(), 0));
VPValue *StepV = PtrIV->getOperand(1);
VPRecipeBase *Steps =
- createScalarIVSteps(Plan, InductionDescriptor::IK_IntInduction, Instruction::Add, nullptr, SE,
- nullptr, StartV, StepV,
- InsertPt)
+ createScalarIVSteps(Plan, InductionDescriptor::IK_IntInduction,
+ Instruction::Add, nullptr, SE, nullptr, StartV,
+ StepV, InsertPt)
->getDefiningRecipe();
auto *Recipe =
@@ -596,9 +596,10 @@ static void optimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
const InductionDescriptor &ID = WideIV->getInductionDescriptor();
VPValue *Steps = createScalarIVSteps(
- Plan, ID.getKind(), ID.getInductionOpcode(), dyn_cast_or_null<FPMathOperator>(ID.getInductionBinOp()), SE, WideIV->getTruncInst(), WideIV->getStartValue(),
- WideIV->getStepValue(), InsertPt
- );
+ Plan, ID.getKind(), ID.getInductionOpcode(),
+ dyn_cast_or_null<FPMathOperator>(ID.getInductionBinOp()), SE,
+ WideIV->getTruncInst(), WideIV->getStartValue(), WideIV->getStepValue(),
+ InsertPt);
// Update scalar users of IV to use Step instead.
if (!HasOnlyVectorVFs)
>From 1e4111193551d3519396082f8400c376537721f8 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Thu, 7 Mar 2024 21:39:05 +0000
Subject: [PATCH 12/13] !fixup address latest comments, thanks!
---
llvm/lib/Transforms/Vectorize/VPlan.cpp | 2 +-
llvm/lib/Transforms/Vectorize/VPlan.h | 16 +++++++++---
.../lib/Transforms/Vectorize/VPlanRecipes.cpp | 23 ++++++++++-------
.../Transforms/Vectorize/VPlanTransforms.cpp | 2 ++
.../LoopVectorize/AArch64/masked-call.ll | 25 +++++++++++--------
.../vplan-printing-before-execute.ll | 14 ++++++-----
6 files changed, 52 insertions(+), 30 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.cpp b/llvm/lib/Transforms/Vectorize/VPlan.cpp
index 1b818bb59caa4e..8c664081c40c15 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlan.cpp
@@ -861,7 +861,7 @@ void VPlan::execute(VPTransformState *State) {
} else {
auto *WidenPhi = cast<VPWidenPointerInductionRecipe>(&R);
assert(!WidenPhi->onlyScalarsGenerated(State->VF.isScalable()) &&
- "recipe only generating scalars should have been replaced");
+ "recipe generating only scalars should have been replaced");
auto *GEP = cast<GetElementPtrInst>(State->get(WidenPhi, 0));
Phi = cast<PHINode>(GEP->getPointerOperand());
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index a9cc969c54ab0f..22e193be19f70c 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1155,6 +1155,9 @@ class VPInstruction : public VPRecipeWithIRFlags {
BranchOnCount,
BranchOnCond,
ComputeReductionResult,
+ // Add an offset in bytes (second operand) to a base pointer (first
+ // operand). Only generates scalar valuse (either for the first lane only or
+ // for all lanes, depending on its uses).
PtrAdd,
};
@@ -1165,13 +1168,18 @@ class VPInstruction : public VPRecipeWithIRFlags {
/// An optional name that can be used for the generated IR instruction.
const std::string Name;
- bool generatesScalars() const;
+ /// Returns true if this VPInstruction generates scalar values only.
+ bool doesGenerateScalars() const;
- /// Utility method serving execute(): generates a single instance of the
- /// modeled instruction. \returns the generated value for \p Part.
- /// In some cases an existing value is returned rather than a generated
+ /// Utility methods serving execute(): generates a single instance of the
+ /// modeled instruction for a given part. \returns the generated value for \p
+ /// Part. In some cases an existing value is returned rather than a generated
/// one.
Value *generatePerPart(VPTransformState &State, unsigned Part);
+
+ /// Utility methods serving execute(): generates a scalar single instance of
+ /// the modeled instruction for a given lane. \returns the scalar generated
+ /// value for lane \p Lane.
Value *generatePerLane(VPTransformState &State, const VPIteration &Lane);
#if !defined(NDEBUG)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 70b9db22a5851a..351a12e778dfcc 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -274,7 +274,7 @@ VPInstruction::VPInstruction(unsigned Opcode,
assert(isFPMathOp() && "this op can't take fast-math flags");
}
-bool VPInstruction::generatesScalars() const {
+bool VPInstruction::doesGenerateScalars() const {
return Opcode == VPInstruction::PtrAdd;
}
@@ -536,12 +536,18 @@ void VPInstruction::execute(VPTransformState &State) {
if (hasFastMathFlags())
State.Builder.setFastMathFlags(getFastMathFlags());
for (unsigned Part = 0; Part < State.UF; ++Part) {
- if (generatesScalars()) {
- unsigned NumLanes =
- vputils::onlyFirstLaneUsed(this) ? 1 : State.VF.getKnownMinValue();
- if (getOpcode() == VPInstruction::ComputeReductionResult)
- NumLanes = 1;
- for (unsigned Lane = 0; Lane != NumLanes; ++Lane) {
+ bool OnlyFirstLaneDefined =
+ vputils::onlyFirstLaneUsed(this) ||
+ getOpcode() == VPInstruction::ComputeReductionResult;
+ if (doesGenerateScalars()) {
+ if (OnlyFirstLaneDefined) {
+ Value *P = generatePerLane(State, VPIteration(Part, 0));
+ State.set(this, P, Part, /*IsScalar*/ true);
+ continue;
+ }
+
+ for (unsigned Lane = 0, NumLanes = State.VF.getKnownMinValue();
+ Lane != NumLanes; ++Lane) {
Value *P = generatePerLane(State, VPIteration(Part, Lane));
State.set(this, P, VPIteration(Part, Lane));
}
@@ -555,8 +561,7 @@ void VPInstruction::execute(VPTransformState &State) {
bool IsVector = GeneratedValue->getType()->isVectorTy();
State.set(this, GeneratedValue, Part, !IsVector);
- assert((IsVector || getOpcode() == VPInstruction::ComputeReductionResult ||
- State.VF.isScalar() || vputils::onlyFirstLaneUsed(this)) &&
+ assert((IsVector || OnlyFirstLaneDefined || State.VF.isScalar()) &&
"scalar value but not only first lane used");
}
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 62f243a3422997..700556f79eb02d 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -586,6 +586,8 @@ static void optimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
continue;
}
+ // Replace widened induction with scalar steps for users that only use
+ // scalars.
auto *WideIV = dyn_cast<VPWidenIntOrFpInductionRecipe>(&Phi);
if (!WideIV)
continue;
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll b/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll
index 2ef55742ffc0b1..1e79c3e1e8dc2d 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/masked-call.ll
@@ -99,7 +99,7 @@ define void @test_widen(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK_ENTRY1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[TMP8]], i64 1025)
; TFA_INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]]
; TFA_INTERLEAVE: vector.body:
-; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT4:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT5:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr i64, ptr [[B:%.*]], i64 [[INDEX]]
@@ -116,7 +116,8 @@ define void @test_widen(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[TMP18:%.*]] = getelementptr inbounds i64, ptr [[TMP15]], i64 [[TMP17]]
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[TMP13]], ptr [[TMP15]], i32 8, <vscale x 2 x i1> [[ACTIVE_LANE_MASK]])
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[TMP14]], ptr [[TMP18]], i32 8, <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]])
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT:%.*]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT4]] = add i64 [[INDEX]], [[TMP6]]
; TFA_INTERLEAVE-NEXT: [[TMP19:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP20:%.*]] = mul i64 [[TMP19]], 2
; TFA_INTERLEAVE-NEXT: [[TMP21:%.*]] = add i64 [[INDEX_NEXT]], [[TMP20]]
@@ -253,7 +254,7 @@ define void @test_if_then(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK_ENTRY1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[TMP8]], i64 1025)
; TFA_INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]]
; TFA_INTERLEAVE: vector.body:
-; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT5:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT6:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDEX]]
@@ -282,7 +283,8 @@ define void @test_if_then(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[TMP28:%.*]] = getelementptr inbounds i64, ptr [[TMP25]], i64 [[TMP27]]
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[PREDPHI]], ptr [[TMP25]], i32 8, <vscale x 2 x i1> [[TMP23]])
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[PREDPHI4]], ptr [[TMP28]], i32 8, <vscale x 2 x i1> [[TMP24]])
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT:%.*]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT5]] = add i64 [[INDEX]], [[TMP6]]
; TFA_INTERLEAVE-NEXT: [[TMP29:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP30:%.*]] = mul i64 [[TMP29]], 2
; TFA_INTERLEAVE-NEXT: [[TMP31:%.*]] = add i64 [[INDEX_NEXT]], [[TMP30]]
@@ -435,7 +437,7 @@ define void @test_widen_if_then_else(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK_ENTRY1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[TMP8]], i64 1025)
; TFA_INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]]
; TFA_INTERLEAVE: vector.body:
-; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT5:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT6:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[INDEX]]
@@ -466,7 +468,8 @@ define void @test_widen_if_then_else(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[TMP30:%.*]] = getelementptr inbounds i64, ptr [[TMP27]], i64 [[TMP29]]
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[PREDPHI]], ptr [[TMP27]], i32 8, <vscale x 2 x i1> [[TMP25]])
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[PREDPHI4]], ptr [[TMP30]], i32 8, <vscale x 2 x i1> [[TMP26]])
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT:%.*]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT5]] = add i64 [[INDEX]], [[TMP6]]
; TFA_INTERLEAVE-NEXT: [[TMP31:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP32:%.*]] = mul i64 [[TMP31]], 2
; TFA_INTERLEAVE-NEXT: [[TMP33:%.*]] = add i64 [[INDEX_NEXT]], [[TMP32]]
@@ -768,7 +771,7 @@ define void @test_widen_optmask(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK_ENTRY1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i64(i64 [[TMP8]], i64 1025)
; TFA_INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]]
; TFA_INTERLEAVE: vector.body:
-; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT4:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT5:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[TMP9:%.*]] = getelementptr i64, ptr [[B:%.*]], i64 [[INDEX]]
@@ -785,7 +788,8 @@ define void @test_widen_optmask(ptr noalias %a, ptr readnone %b) #4 {
; TFA_INTERLEAVE-NEXT: [[TMP18:%.*]] = getelementptr inbounds i64, ptr [[TMP15]], i64 [[TMP17]]
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[TMP13]], ptr [[TMP15]], i32 8, <vscale x 2 x i1> [[ACTIVE_LANE_MASK]])
; TFA_INTERLEAVE-NEXT: call void @llvm.masked.store.nxv2i64.p0(<vscale x 2 x i64> [[TMP14]], ptr [[TMP18]], i32 8, <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]])
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT:%.*]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT4]] = add i64 [[INDEX]], [[TMP6]]
; TFA_INTERLEAVE-NEXT: [[TMP19:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP20:%.*]] = mul i64 [[TMP19]], 2
; TFA_INTERLEAVE-NEXT: [[TMP21:%.*]] = add i64 [[INDEX_NEXT]], [[TMP20]]
@@ -966,7 +970,7 @@ define double @test_widen_fmuladd_and_call(ptr noalias %a, ptr readnone %b, doub
; TFA_INTERLEAVE-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x double> [[BROADCAST_SPLATINSERT]], <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer
; TFA_INTERLEAVE-NEXT: br label [[VECTOR_BODY:%.*]]
; TFA_INTERLEAVE: vector.body:
-; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; TFA_INTERLEAVE-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDEX_NEXT4:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[ACTIVE_LANE_MASK2:%.*]] = phi <vscale x 2 x i1> [ [[ACTIVE_LANE_MASK_ENTRY1]], [[ENTRY]] ], [ [[ACTIVE_LANE_MASK_NEXT5:%.*]], [[VECTOR_BODY]] ]
; TFA_INTERLEAVE-NEXT: [[VEC_PHI:%.*]] = phi double [ 0.000000e+00, [[ENTRY]] ], [ [[TMP26:%.*]], [[VECTOR_BODY]] ]
@@ -992,7 +996,8 @@ define double @test_widen_fmuladd_and_call(ptr noalias %a, ptr readnone %b, doub
; TFA_INTERLEAVE-NEXT: [[TMP24:%.*]] = call double @llvm.vector.reduce.fadd.nxv2f64(double [[VEC_PHI]], <vscale x 2 x double> [[TMP23]])
; TFA_INTERLEAVE-NEXT: [[TMP25:%.*]] = select <vscale x 2 x i1> [[ACTIVE_LANE_MASK2]], <vscale x 2 x double> [[TMP14]], <vscale x 2 x double> shufflevector (<vscale x 2 x double> insertelement (<vscale x 2 x double> poison, double -0.000000e+00, i64 0), <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer)
; TFA_INTERLEAVE-NEXT: [[TMP26]] = call double @llvm.vector.reduce.fadd.nxv2f64(double [[TMP24]], <vscale x 2 x double> [[TMP25]])
-; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT:%.*]] = add i64 [[INDEX]], [[TMP6]]
+; TFA_INTERLEAVE-NEXT: [[INDEX_NEXT4]] = add i64 [[INDEX]], [[TMP6]]
; TFA_INTERLEAVE-NEXT: [[TMP27:%.*]] = call i64 @llvm.vscale.i64()
; TFA_INTERLEAVE-NEXT: [[TMP28:%.*]] = mul i64 [[TMP27]], 2
; TFA_INTERLEAVE-NEXT: [[TMP29:%.*]] = add i64 [[INDEX_NEXT]], [[TMP28]]
diff --git a/llvm/test/Transforms/LoopVectorize/vplan-printing-before-execute.ll b/llvm/test/Transforms/LoopVectorize/vplan-printing-before-execute.ll
index 1dddbfe20a2ed7..ca9dfdc6f6d29d 100644
--- a/llvm/test/Transforms/LoopVectorize/vplan-printing-before-execute.ll
+++ b/llvm/test/Transforms/LoopVectorize/vplan-printing-before-execute.ll
@@ -23,11 +23,12 @@ define void @test_tc_less_than_16(ptr %A, i64 %N) {
; CHECK-NEXT: <x1> vector loop: {
; CHECK-NEXT: vector.body:
; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION ir<0>, vp<[[CAN_IV_NEXT:%.+]]>
-; CHECK-NEXT: EMIT ir<%p.src> = WIDEN-POINTER-INDUCTION ir<%A>, 1
-; CHECK-NEXT: vp<[[VPTR:%.]]> = vector-pointer ir<%p.src>
+; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
+; CHECK-NEXT: EMIT vp<[[PADD:%.+]]> = ptradd ir<%A>, vp<[[STEPS]]>
+; CHECK-NEXT: vp<[[VPTR:%.]]> = vector-pointer vp<[[PADD]]>
; CHECK-NEXT: WIDEN ir<%l> = load vp<[[VPTR]]>
; CHECK-NEXT: WIDEN ir<%add> = add nsw ir<%l>, ir<10>
-; CHECK-NEXT: vp<[[VPTR2:%.+]]> = vector-pointer ir<%p.src>
+; CHECK-NEXT: vp<[[VPTR2:%.+]]> = vector-pointer vp<[[PADD]]>
; CHECK-NEXT: WIDEN store vp<[[VPTR2]]>, ir<%add>
; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT]]> = add nuw vp<[[CAN_IV:%.+]]>, vp<[[VFxUF]]>
; CHECK-NEXT: EMIT branch-on-count vp<[[CAN_IV_NEXT]]>, vp<[[VTC]]>
@@ -54,11 +55,12 @@ define void @test_tc_less_than_16(ptr %A, i64 %N) {
; CHECK-NEXT: <x1> vector loop: {
; CHECK-NEXT: vector.body:
; CHECK-NEXT: EMIT vp<[[CAN_IV:%.+]]> = CANONICAL-INDUCTION ir<0>, vp<[[CAN_IV_NEXT:%.+]]>
-; CHECK-NEXT: EMIT ir<%p.src> = WIDEN-POINTER-INDUCTION ir<%A>, 1
-; CHECK-NEXT: vp<[[VPTR:%.]]> = vector-pointer ir<%p.src>
+; CHECK-NEXT: vp<[[STEPS:%.+]]> = SCALAR-STEPS vp<[[CAN_IV]]>, ir<1>
+; CHECK-NEXT: EMIT vp<[[PADD:%.+]]> = ptradd ir<%A>, vp<[[STEPS]]>
+; CHECK-NEXT: vp<[[VPTR:%.]]> = vector-pointer vp<[[PADD]]>
; CHECK-NEXT: WIDEN ir<%l> = load vp<[[VPTR]]>
; CHECK-NEXT: WIDEN ir<%add> = add nsw ir<%l>, ir<10>
-; CHECK-NEXT: vp<[[VPTR2:%.+]]> = vector-pointer ir<%p.src>
+; CHECK-NEXT: vp<[[VPTR2:%.+]]> = vector-pointer vp<[[PADD]]>
; CHECK-NEXT: WIDEN store vp<[[VPTR2]]>, ir<%add>
; CHECK-NEXT: EMIT vp<[[CAN_IV_NEXT]]> = add nuw vp<[[CAN_IV:%.+]]>, vp<[[VFxUF]]>
; CHECK-NEXT: EMIT branch-on-cond ir<true>
>From 6f4516f8e0a44ea1b1f7ef28d33ce6245b290339 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Mon, 11 Mar 2024 11:39:38 +0000
Subject: [PATCH 13/13] fixup address comments.
---
llvm/lib/Transforms/Vectorize/VPlan.h | 8 ++-
.../lib/Transforms/Vectorize/VPlanRecipes.cpp | 58 ++++++++++++-------
.../Transforms/Vectorize/VPlanTransforms.cpp | 12 ++--
3 files changed, 49 insertions(+), 29 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 22e193be19f70c..95a64944c00c28 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -1168,8 +1168,12 @@ class VPInstruction : public VPRecipeWithIRFlags {
/// An optional name that can be used for the generated IR instruction.
const std::string Name;
- /// Returns true if this VPInstruction generates scalar values only.
- bool doesGenerateScalars() const;
+ /// Returns true if this VPInstruction generates scalar values for all lanes.
+ bool doesGeneratePerAllLanes() const;
+
+ /// Returns true if we can generate a scalar for the first lane only if
+ /// needed.
+ bool canGenerateScalarForFirstLane() const;
/// Utility methods serving execute(): generates a single instance of the
/// modeled instruction for a given part. \returns the generated value for \p
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 351a12e778dfcc..c6e28b8bd395ed 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -274,14 +274,30 @@ VPInstruction::VPInstruction(unsigned Opcode,
assert(isFPMathOp() && "this op can't take fast-math flags");
}
-bool VPInstruction::doesGenerateScalars() const {
- return Opcode == VPInstruction::PtrAdd;
+bool VPInstruction::doesGeneratePerAllLanes() const {
+ return Opcode == VPInstruction::PtrAdd && !vputils::onlyFirstLaneUsed(this);
+}
+
+bool VPInstruction::canGenerateScalarForFirstLane() const {
+ if (Instruction::isBinaryOp(getOpcode()))
+ return true;
+
+ switch (Opcode) {
+ case VPInstruction::BranchOnCond:
+ case VPInstruction::BranchOnCount:
+ case VPInstruction::CalculateTripCountMinusVF:
+ case VPInstruction::CanonicalIVIncrementForPart:
+ case VPInstruction::ComputeReductionResult:
+ case VPInstruction::PtrAdd:
+ return true;
+ default:
+ return false;
+ }
}
Value *VPInstruction::generatePerLane(VPTransformState &State,
const VPIteration &Lane) {
IRBuilderBase &Builder = State.Builder;
- Builder.SetCurrentDebugLocation(getDebugLoc());
switch (getOpcode()) {
case VPInstruction::PtrAdd: {
@@ -289,14 +305,16 @@ Value *VPInstruction::generatePerLane(VPTransformState &State,
State.get(getOperand(1), Lane), Name);
return P;
}
- default:
- llvm_unreachable("Unsupported opcode for instruction");
+ default: {
+ Value *Res = generatePerPart(State, Lane.Part);
+ assert(!hasResult() || !Res->getType()->isVectorTy());
+ return Res;
+ }
}
}
Value *VPInstruction::generatePerPart(VPTransformState &State, unsigned Part) {
IRBuilderBase &Builder = State.Builder;
- Builder.SetCurrentDebugLocation(getDebugLoc());
if (Instruction::isBinaryOp(getOpcode())) {
bool OnlyFirstLaneUsed = vputils::onlyFirstLaneUsed(this);
@@ -535,17 +553,13 @@ void VPInstruction::execute(VPTransformState &State) {
"Recipe not a FPMathOp but has fast-math flags?");
if (hasFastMathFlags())
State.Builder.setFastMathFlags(getFastMathFlags());
+ State.Builder.SetCurrentDebugLocation(getDebugLoc());
+ bool OnlyGenerateFirstLane =
+ canGenerateScalarForFirstLane() &&
+ (vputils::onlyFirstLaneUsed(this) ||
+ getOpcode() == VPInstruction::ComputeReductionResult);
for (unsigned Part = 0; Part < State.UF; ++Part) {
- bool OnlyFirstLaneDefined =
- vputils::onlyFirstLaneUsed(this) ||
- getOpcode() == VPInstruction::ComputeReductionResult;
- if (doesGenerateScalars()) {
- if (OnlyFirstLaneDefined) {
- Value *P = generatePerLane(State, VPIteration(Part, 0));
- State.set(this, P, Part, /*IsScalar*/ true);
- continue;
- }
-
+ if (doesGeneratePerAllLanes()) {
for (unsigned Lane = 0, NumLanes = State.VF.getKnownMinValue();
Lane != NumLanes; ++Lane) {
Value *P = generatePerLane(State, VPIteration(Part, Lane));
@@ -554,15 +568,17 @@ void VPInstruction::execute(VPTransformState &State) {
continue;
}
- Value *GeneratedValue = generatePerPart(State, Part);
+ Value *GeneratedValue = OnlyGenerateFirstLane
+ ? generatePerLane(State, VPIteration(Part, 0))
+ : generatePerPart(State, Part);
if (!hasResult())
continue;
assert(GeneratedValue && "generateInstruction must produce a value");
- bool IsVector = GeneratedValue->getType()->isVectorTy();
- State.set(this, GeneratedValue, Part, !IsVector);
- assert((IsVector || OnlyFirstLaneDefined || State.VF.isScalar()) &&
- "scalar value but not only first lane used");
+ State.set(this, GeneratedValue, Part, /*IsScalar*/ OnlyGenerateFirstLane);
+ assert((GeneratedValue->getType()->isVectorTy() == !OnlyGenerateFirstLane ||
+ State.VF.isScalar()) &&
+ "scalar value but not only first lane defined");
}
}
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 700556f79eb02d..15dda511ae0c03 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -548,13 +548,13 @@ static VPValue *createScalarIVSteps(VPlan &Plan,
return Steps;
}
-/// If any user of a VPWidenIntOrFpInductionRecipe needs scalar values,
-/// provide them by building scalar steps off of the canonical scalar IV and
+/// Legalize VPWidenPointer induction, by replacing it with a PtrAdd (IndStart,
+/// ScalarIVSteps (0, Step)) if only its scalar values are used. Als optimize
+/// VPWidenIntOrFpInductionRecipe, if any of its users needs scalar values, by
+/// providing them by building scalar steps off of the canonical scalar IV and
/// update the original IV's users. This is an optional optimization to reduce
/// the needs of vector extracts.
-/// If all users of VPWidenPointerInductionRecipe only use its scalar values,
-/// replace it with a PtrAdd (IndStart, ScalarIVSteps (0, Step)).
-static void optimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
+static void legalizeAndOptimizeInductions(VPlan &Plan, ScalarEvolution &SE) {
SmallVector<VPRecipeBase *> ToRemove;
VPBasicBlock *HeaderVPBB = Plan.getVectorLoopRegion()->getEntryBasicBlock();
bool HasOnlyVectorVFs = !Plan.hasVF(ElementCount::getFixed(1));
@@ -1096,7 +1096,7 @@ void VPlanTransforms::optimize(VPlan &Plan, ScalarEvolution &SE) {
removeRedundantInductionCasts(Plan);
simplifyRecipes(Plan, SE.getContext());
- optimizeInductions(Plan, SE);
+ legalizeAndOptimizeInductions(Plan, SE);
removeDeadRecipes(Plan);
createAndOptimizeReplicateRegions(Plan);
More information about the llvm-commits
mailing list