[llvm] [LAA] Rework getStrideFromPointer, speculate more (PR #212816)
Ramkumar Ramachandra via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 30 06:16:25 PDT 2026
https://github.com/artagnon updated https://github.com/llvm/llvm-project/pull/212816
>From 8e5368f900705b1428a1c9549e979734f5cbbd4a Mon Sep 17 00:00:00 2001
From: Ramkumar Ramachandra <artagnon at tenstorrent.com>
Date: Wed, 29 Jul 2026 17:11:59 +0100
Subject: [PATCH 1/4] [LAA] Rework getStrideFromPointer, speculate more
getLoopVaryingGEPOperand can be replaced by the more powerful
ScalarEvolution::removePointerBase. Fix this and rework the surrounding
code in getStrideFromPointer to speculate more strides.
---
llvm/lib/Analysis/LoopAccessAnalysis.cpp | 76 +---
.../no-dep-via-loop-guards.ll | 33 +-
...ter-dependence-analysis-forked-pointers.ll | 4 +-
.../reverse-memcheck-bounds.ll | 9 +-
.../partial-reduce-replicate-extends.ll | 20 +-
.../AArch64/pr60831-sve-inv-store-crash.ll | 142 +++----
.../RISCV/blocks-with-dead-instructions.ll | 26 +-
.../truncate-to-minimal-bitwidth-cost.ll | 54 ++-
.../SystemZ/load-scalarization-cost-0.ll | 6 +-
.../LoopVectorize/pointer-induction.ll | 38 +-
.../pr59319-loop-access-info-invalidation.ll | 37 +-
.../runtime-check-needed-but-empty.ll | 35 +-
.../LoopVectorize/runtime-checks-hoist.ll | 8 +-
.../LoopVectorize/vector-no-scevcheck.ll | 73 +++-
.../LoopVectorize/version-mem-access.ll | 70 +++-
.../LoopVectorize/vplan-based-stride-mv.ll | 379 +++++++++++-------
16 files changed, 587 insertions(+), 423 deletions(-)
diff --git a/llvm/lib/Analysis/LoopAccessAnalysis.cpp b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
index 1f13106214ea9..e53fc9df90227 100644
--- a/llvm/lib/Analysis/LoopAccessAnalysis.cpp
+++ b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
@@ -3033,66 +3033,24 @@ bool LoopAccessInfo::isInvariant(Value *V) const {
return SE->isLoopInvariant(S, TheLoop);
}
-/// If \p Ptr is a GEP, which has a loop-variant operand, return that operand.
-/// Otherwise, return \p Ptr.
-static Value *getLoopVariantGEPOperand(Value *Ptr, ScalarEvolution *SE,
- Loop *Lp) {
- auto *GEP = dyn_cast<GetElementPtrInst>(Ptr);
- if (!GEP)
- return Ptr;
-
- Value *V = Ptr;
- for (const Use &U : GEP->operands()) {
- if (!SE->isLoopInvariant(SE->getSCEV(U), Lp)) {
- if (V == Ptr)
- V = U;
- else
- // There must be exactly one loop-variant operand.
- return Ptr;
- }
- }
- return V;
-}
-
/// Get the stride of a pointer access in a loop. Looks for symbolic
/// strides "a[i*stride]". Returns the symbolic stride, or null otherwise.
-static const SCEV *getStrideFromPointer(Value *Ptr, ScalarEvolution *SE, Loop *Lp) {
- auto *PtrTy = dyn_cast<PointerType>(Ptr->getType());
- if (!PtrTy)
- return nullptr;
-
- // Try to remove a gep instruction to make the pointer (actually index at this
- // point) easier analyzable. If OrigPtr is equal to Ptr we are analyzing the
- // pointer, otherwise, we are analyzing the index.
- Value *OrigPtr = Ptr;
+static const SCEVUnknown *getStrideFromPointer(Value *Ptr, ScalarEvolution *SE,
+ Loop *Lp) {
+ assert(Ptr->getType()->isPointerTy() && "Pointer type expected");
+ const SCEV *V = SE->removePointerBase(SE->getSCEV(Ptr));
- Ptr = getLoopVariantGEPOperand(Ptr, SE, Lp);
- const SCEV *V = SE->getSCEV(Ptr);
+ while (auto *C = dyn_cast<SCEVIntegralCastExpr>(V))
+ V = C->getOperand();
- if (Ptr != OrigPtr)
- // Strip off casts.
- while (auto *C = dyn_cast<SCEVIntegralCastExpr>(V))
- V = C->getOperand();
-
- if (!match(V, m_scev_AffineAddRec(m_SCEV(), m_SCEV(V), m_SpecificLoop(Lp))))
- return nullptr;
-
- // Note that the restriction after this loop invariant check are only
- // profitability restrictions.
- if (!SE->isLoopInvariant(V, Lp))
- return nullptr;
-
- // Look for the loop invariant symbolic value.
- if (isa<SCEVUnknown>(V))
- return V;
-
- // Look through multiplies that scale a stride by a constant.
+ match(V, m_scev_AffineAddRec(m_SCEV(), m_SCEV(V), m_SpecificLoop(Lp)));
match(V, m_scev_Mul(m_SCEVConstant(), m_SCEV(V)));
- if (auto *C = dyn_cast<SCEVIntegralCastExpr>(V))
- if (isa<SCEVUnknown>(C->getOperand()))
- return V;
- return nullptr;
+ while (auto *C = dyn_cast<SCEVIntegralCastExpr>(V))
+ V = C->getOperand();
+
+ // It wouldn't make sense to speculate on a loop-varying stride.
+ return SE->isLoopInvariant(V, Lp) ? dyn_cast<SCEVUnknown>(V) : nullptr;
}
void LoopAccessInfo::collectStridedAccess(Value *MemAccess) {
@@ -3106,7 +3064,8 @@ void LoopAccessInfo::collectStridedAccess(Value *MemAccess) {
// computation of an interesting IV - but we chose not to as we
// don't have a cost model here, and broadening the scope exposes
// far too many unprofitable cases.
- const SCEV *StrideExpr = getStrideFromPointer(Ptr, PSE->getSE(), TheLoop);
+ const SCEVUnknown *StrideExpr =
+ getStrideFromPointer(Ptr, PSE->getSE(), TheLoop);
if (!StrideExpr)
return;
@@ -3163,12 +3122,7 @@ void LoopAccessInfo::collectStridedAccess(Value *MemAccess) {
}
LLVM_DEBUG(dbgs() << "LAA: Found a strided access that we can version.\n");
- // Strip back off the integer cast, and check that our result is a
- // SCEVUnknown as we expect.
- const SCEV *StrideBase = StrideExpr;
- if (const auto *C = dyn_cast<SCEVIntegralCastExpr>(StrideBase))
- StrideBase = C->getOperand();
- SymbolicStrides[Ptr] = cast<SCEVUnknown>(StrideBase);
+ SymbolicStrides[Ptr] = StrideExpr;
}
LoopAccessInfo::LoopAccessInfo(Loop *L, ScalarEvolution *SE,
diff --git a/llvm/test/Analysis/LoopAccessAnalysis/no-dep-via-loop-guards.ll b/llvm/test/Analysis/LoopAccessAnalysis/no-dep-via-loop-guards.ll
index e00f91d9b2e8b..943c9452de3b0 100644
--- a/llvm/test/Analysis/LoopAccessAnalysis/no-dep-via-loop-guards.ll
+++ b/llvm/test/Analysis/LoopAccessAnalysis/no-dep-via-loop-guards.ll
@@ -5,13 +5,19 @@
define void @access_after_via_loop_guard(ptr %a, i64 %off) {
; CHECK-LABEL: 'access_after_via_loop_guard'
; CHECK-NEXT: loop:
-; CHECK-NEXT: Memory dependences are safe
+; CHECK-NEXT: Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
+; CHECK-NEXT: Unknown data dependence.
; CHECK-NEXT: Dependences:
+; CHECK-NEXT: Unknown:
+; CHECK-NEXT: %l = load i32, ptr %gep.after, align 4 ->
+; CHECK-NEXT: store i32 %add, ptr %gep, align 4
+; CHECK-EMPTY:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Grouped accesses:
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
+; CHECK-NEXT: Equal predicate: %off == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
@@ -53,6 +59,7 @@ define void @access_after_via_loop_guard_sge(ptr %a, i64 %off) {
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
+; CHECK-NEXT: Equal predicate: %off == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
@@ -92,6 +99,7 @@ define void @access_after_via_loop_guard_99(ptr %a, i64 %off) {
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
+; CHECK-NEXT: Equal predicate: %off == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
@@ -133,6 +141,7 @@ define void @access_after_via_loop_guard_sge_99(ptr %a, i64 %off) {
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
+; CHECK-NEXT: Equal predicate: %off == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
@@ -172,6 +181,7 @@ define void @access_after_via_loop_guard_uge(ptr %a, i64 %off) {
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
+; CHECK-NEXT: Equal predicate: %off == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
@@ -354,20 +364,22 @@ exit:
define void @nodep_via_logical_and_1(ptr %A, i32 %index, i32 %n) {
; CHECK-LABEL: 'nodep_via_logical_and_1'
; CHECK-NEXT: loop:
-; CHECK-NEXT: Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
-; CHECK-NEXT: Unknown data dependence.
+; CHECK-NEXT: Memory dependences are safe
; CHECK-NEXT: Dependences:
-; CHECK-NEXT: Unknown:
-; CHECK-NEXT: %0 = load double, ptr %gep.load, align 8 ->
-; CHECK-NEXT: store double %0, ptr %gep.store, align 8
-; CHECK-EMPTY:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Grouped accesses:
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
+; CHECK-NEXT: Equal predicate: %index == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
+; CHECK-NEXT: [PSE] %gep.load = getelementptr double, ptr %A, i64 %iv:
+; CHECK-NEXT: {((8 * (zext i32 (1 + %index) to i64))<nuw><nsw> + %A),+,8}<nw><%loop>
+; CHECK-NEXT: --> {(16 + %A),+,8}<nw><%loop>
+; CHECK-NEXT: [PSE] %gep.store = getelementptr double, ptr %A, i64 %index.ext:
+; CHECK-NEXT: ((8 * (zext i32 %index to i64))<nuw><nsw> + %A)
+; CHECK-NEXT: --> (8 + %A)
;
entry:
%pre.0 = icmp sgt i32 %index, 0
@@ -407,8 +419,15 @@ define void @nodep_via_logical_and_2(ptr %A, i32 %index, i32 %n) {
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
+; CHECK-NEXT: Equal predicate: %index == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
+; CHECK-NEXT: [PSE] %gep.load = getelementptr double, ptr %A, i64 %iv:
+; CHECK-NEXT: {((8 * (zext i32 (1 + %index) to i64))<nuw><nsw> + %A),+,8}<nw><%loop>
+; CHECK-NEXT: --> {(16 + %A),+,8}<nw><%loop>
+; CHECK-NEXT: [PSE] %gep.store = getelementptr double, ptr %A, i64 %index.ext:
+; CHECK-NEXT: ((8 * (zext i32 %index to i64))<nuw><nsw> + %A)
+; CHECK-NEXT: --> (8 + %A)
;
entry:
%pre.0 = icmp sgt i32 %index, 0
diff --git a/llvm/test/Analysis/LoopAccessAnalysis/retry-runtime-checks-after-dependence-analysis-forked-pointers.ll b/llvm/test/Analysis/LoopAccessAnalysis/retry-runtime-checks-after-dependence-analysis-forked-pointers.ll
index 3a5730eb471ed..be91a172132a9 100644
--- a/llvm/test/Analysis/LoopAccessAnalysis/retry-runtime-checks-after-dependence-analysis-forked-pointers.ll
+++ b/llvm/test/Analysis/LoopAccessAnalysis/retry-runtime-checks-after-dependence-analysis-forked-pointers.ll
@@ -112,10 +112,10 @@ define void @dependency_check_and_runtime_checks_needed_select_of_ptr_add_recs(p
; CHECK-NEXT: Member: {%a,+,4}<nuw><%loop>
; CHECK-NEXT: Group GRP1:
; CHECK-NEXT: (Low: %b High: ((4 * %n) + %b))
-; CHECK-NEXT: Member: {%b,+,4}<%loop>
+; CHECK-NEXT: Member: {%b,+,4}<nw><%loop>
; CHECK-NEXT: Group GRP2:
; CHECK-NEXT: (Low: %c High: ((4 * %n) + %c))
-; CHECK-NEXT: Member: {%c,+,4}<%loop>
+; CHECK-NEXT: Member: {%c,+,4}<nw><%loop>
; CHECK-NEXT: Group GRP3:
; CHECK-NEXT: (Low: ((4 * %offset) + %a) High: ((4 * %offset) + (4 * %n) + %a))
; CHECK-NEXT: Member: {((4 * %offset) + %a),+,4}<%loop>
diff --git a/llvm/test/Analysis/LoopAccessAnalysis/reverse-memcheck-bounds.ll b/llvm/test/Analysis/LoopAccessAnalysis/reverse-memcheck-bounds.ll
index 1496e1b0be82b..c86d145d3d701 100644
--- a/llvm/test/Analysis/LoopAccessAnalysis/reverse-memcheck-bounds.ll
+++ b/llvm/test/Analysis/LoopAccessAnalysis/reverse-memcheck-bounds.ll
@@ -48,17 +48,14 @@ for.end: ; preds = %for.body
}
; CHECK: function 'g':
-; When the stride is not constant, we are forced to do umin/umax to get
-; the interval limits.
+; When the stride is not constant, we can speculate the stride.
; for (i = 0; i < 10000; i++) {
; B[i] = A[i] * 3;
; }
-; Here it is not obvious what the limits are, since 'step' could be negative.
-
-; CHECK: Low: ((60000 + %a) umin (60000 + (-40000 * %step) + %a))
-; CHECK: High: (4 + ((60000 + %a) umax (60000 + (-40000 * %step) + %a)))
+; CHECK: Low: (20000 + %a) High: (60004 + %a)
+; CHECK: Equal predicate: %step == 1
define void @g(i64 %step) {
entry:
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-replicate-extends.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-replicate-extends.ll
index 2684e33cf051c..a9619383cefe7 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-replicate-extends.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-replicate-extends.ll
@@ -7,20 +7,20 @@ target triple = "aarch64"
define double @test(i32 %0) {
; CHECK-LABEL: define double @test(
; CHECK-SAME: i32 [[TMP0:%.*]]) {
-; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: br label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[REDUX:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[BINOP:%.*]], %[[VECTOR_PH]] ]
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[NEXT_IV:%.*]], %[[VECTOR_PH]] ]
; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
-; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[TMP1]], i64 0
-; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
-; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
-; CHECK: [[VECTOR_BODY]]:
-; CHECK-NEXT: [[TMP2:%.*]] = or <2 x i64> zeroinitializer, [[BROADCAST_SPLAT]]
-; CHECK-NEXT: br label %[[MIDDLE_BLOCK:.*]]
-; CHECK: [[MIDDLE_BLOCK]]:
-; CHECK-NEXT: [[TMP3:%.*]] = call i64 @llvm.vector.reduce.or.v2i64(<2 x i64> [[TMP2]])
-; CHECK-NEXT: br label %[[EXIT:.*]]
+; CHECK-NEXT: [[BINOP]] = or i64 [[REDUX]], [[TMP1]]
+; CHECK-NEXT: [[GEP:%.*]] = getelementptr [8 x i8], ptr null, i64 [[TMP1]]
+; CHECK-NEXT: [[LOAD:%.*]] = load double, ptr [[GEP]], align 8
+; CHECK-NEXT: [[NEXT_IV]] = add i64 [[IV]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV]], 1
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[VECTOR_PH]]
; CHECK: [[EXIT]]:
+; CHECK-NEXT: [[TMP3:%.*]] = phi i64 [ [[BINOP]], %[[VECTOR_PH]] ]
; CHECK-NEXT: [[RET:%.*]] = uitofp i64 [[TMP3]] to double
; CHECK-NEXT: ret double [[RET]]
;
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/pr60831-sve-inv-store-crash.ll b/llvm/test/Transforms/LoopVectorize/AArch64/pr60831-sve-inv-store-crash.ll
index 6bc2c098576c7..af61e38a5908e 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/pr60831-sve-inv-store-crash.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/pr60831-sve-inv-store-crash.ll
@@ -343,6 +343,9 @@ define void @test_loop2(i64 %n, ptr %dst) {
; CHECK-LABEL: @test_loop2(
; CHECK-NEXT: iter.check:
; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
+; CHECK: vector.scevcheck:
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[N:%.*]], 1
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK1:%.*]]
; CHECK: vector.main.loop.iter.check:
; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
@@ -364,22 +367,22 @@ define void @test_loop2(i64 %n, ptr %dst) {
; CHECK-NEXT: [[TMP13:%.*]] = add i64 [[INDEX]], 13
; CHECK-NEXT: [[TMP14:%.*]] = add i64 [[INDEX]], 14
; CHECK-NEXT: [[TMP15:%.*]] = add i64 [[INDEX]], 15
-; CHECK-NEXT: [[TMP16:%.*]] = sub nsw i64 [[N:%.*]], [[INDEX]]
-; CHECK-NEXT: [[TMP17:%.*]] = sub nsw i64 [[N]], [[TMP1]]
-; CHECK-NEXT: [[TMP18:%.*]] = sub nsw i64 [[N]], [[TMP2]]
-; CHECK-NEXT: [[TMP19:%.*]] = sub nsw i64 [[N]], [[TMP3]]
-; CHECK-NEXT: [[TMP20:%.*]] = sub nsw i64 [[N]], [[TMP4]]
-; CHECK-NEXT: [[TMP21:%.*]] = sub nsw i64 [[N]], [[TMP5]]
-; CHECK-NEXT: [[TMP22:%.*]] = sub nsw i64 [[N]], [[TMP6]]
-; CHECK-NEXT: [[TMP23:%.*]] = sub nsw i64 [[N]], [[TMP7]]
-; CHECK-NEXT: [[TMP24:%.*]] = sub nsw i64 [[N]], [[TMP8]]
-; CHECK-NEXT: [[TMP25:%.*]] = sub nsw i64 [[N]], [[TMP9]]
-; CHECK-NEXT: [[TMP26:%.*]] = sub nsw i64 [[N]], [[TMP10]]
-; CHECK-NEXT: [[TMP27:%.*]] = sub nsw i64 [[N]], [[TMP11]]
-; CHECK-NEXT: [[TMP28:%.*]] = sub nsw i64 [[N]], [[TMP12]]
-; CHECK-NEXT: [[TMP29:%.*]] = sub nsw i64 [[N]], [[TMP13]]
-; CHECK-NEXT: [[TMP30:%.*]] = sub nsw i64 [[N]], [[TMP14]]
-; CHECK-NEXT: [[TMP31:%.*]] = sub nsw i64 [[N]], [[TMP15]]
+; CHECK-NEXT: [[TMP16:%.*]] = sub nsw i64 1, [[INDEX]]
+; CHECK-NEXT: [[TMP17:%.*]] = sub nsw i64 1, [[TMP1]]
+; CHECK-NEXT: [[TMP18:%.*]] = sub nsw i64 1, [[TMP2]]
+; CHECK-NEXT: [[TMP19:%.*]] = sub nsw i64 1, [[TMP3]]
+; CHECK-NEXT: [[TMP20:%.*]] = sub nsw i64 1, [[TMP4]]
+; CHECK-NEXT: [[TMP21:%.*]] = sub nsw i64 1, [[TMP5]]
+; CHECK-NEXT: [[TMP22:%.*]] = sub nsw i64 1, [[TMP6]]
+; CHECK-NEXT: [[TMP23:%.*]] = sub nsw i64 1, [[TMP7]]
+; CHECK-NEXT: [[TMP24:%.*]] = sub nsw i64 1, [[TMP8]]
+; CHECK-NEXT: [[TMP25:%.*]] = sub nsw i64 1, [[TMP9]]
+; CHECK-NEXT: [[TMP26:%.*]] = sub nsw i64 1, [[TMP10]]
+; CHECK-NEXT: [[TMP27:%.*]] = sub nsw i64 1, [[TMP11]]
+; CHECK-NEXT: [[TMP28:%.*]] = sub nsw i64 1, [[TMP12]]
+; CHECK-NEXT: [[TMP29:%.*]] = sub nsw i64 1, [[TMP13]]
+; CHECK-NEXT: [[TMP30:%.*]] = sub nsw i64 1, [[TMP14]]
+; CHECK-NEXT: [[TMP31:%.*]] = sub nsw i64 1, [[TMP15]]
; CHECK-NEXT: [[TMP32:%.*]] = insertelement <16 x i64> poison, i64 [[TMP16]], i32 0
; CHECK-NEXT: [[TMP33:%.*]] = insertelement <16 x i64> [[TMP32]], i64 [[TMP17]], i32 1
; CHECK-NEXT: [[TMP34:%.*]] = insertelement <16 x i64> [[TMP33]], i64 [[TMP18]], i32 2
@@ -409,7 +412,7 @@ define void @test_loop2(i64 %n, ptr %dst) {
; CHECK: vec.epilog.iter.check:
; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF10:![0-9]+]]
; CHECK: vec.epilog.ph:
-; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
+; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK1]] ]
; 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]] ]
@@ -420,14 +423,14 @@ define void @test_loop2(i64 %n, ptr %dst) {
; CHECK-NEXT: [[TMP58:%.*]] = add i64 [[INDEX2]], 5
; CHECK-NEXT: [[TMP59:%.*]] = add i64 [[INDEX2]], 6
; CHECK-NEXT: [[TMP60:%.*]] = add i64 [[INDEX2]], 7
-; CHECK-NEXT: [[TMP61:%.*]] = sub nsw i64 [[N]], [[INDEX2]]
-; CHECK-NEXT: [[TMP62:%.*]] = sub nsw i64 [[N]], [[TMP54]]
-; CHECK-NEXT: [[TMP63:%.*]] = sub nsw i64 [[N]], [[TMP55]]
-; CHECK-NEXT: [[TMP64:%.*]] = sub nsw i64 [[N]], [[TMP56]]
-; CHECK-NEXT: [[TMP65:%.*]] = sub nsw i64 [[N]], [[TMP57]]
-; CHECK-NEXT: [[TMP66:%.*]] = sub nsw i64 [[N]], [[TMP58]]
-; CHECK-NEXT: [[TMP67:%.*]] = sub nsw i64 [[N]], [[TMP59]]
-; CHECK-NEXT: [[TMP68:%.*]] = sub nsw i64 [[N]], [[TMP60]]
+; CHECK-NEXT: [[TMP61:%.*]] = sub nsw i64 1, [[INDEX2]]
+; CHECK-NEXT: [[TMP62:%.*]] = sub nsw i64 1, [[TMP54]]
+; CHECK-NEXT: [[TMP63:%.*]] = sub nsw i64 1, [[TMP55]]
+; CHECK-NEXT: [[TMP64:%.*]] = sub nsw i64 1, [[TMP56]]
+; CHECK-NEXT: [[TMP65:%.*]] = sub nsw i64 1, [[TMP57]]
+; CHECK-NEXT: [[TMP66:%.*]] = sub nsw i64 1, [[TMP58]]
+; CHECK-NEXT: [[TMP67:%.*]] = sub nsw i64 1, [[TMP59]]
+; CHECK-NEXT: [[TMP68:%.*]] = sub nsw i64 1, [[TMP60]]
; CHECK-NEXT: [[TMP69:%.*]] = insertelement <8 x i64> poison, i64 [[TMP61]], i32 0
; CHECK-NEXT: [[TMP70:%.*]] = insertelement <8 x i64> [[TMP69]], i64 [[TMP62]], i32 1
; CHECK-NEXT: [[TMP71:%.*]] = insertelement <8 x i64> [[TMP70]], i64 [[TMP63]], i32 2
@@ -447,7 +450,7 @@ define void @test_loop2(i64 %n, ptr %dst) {
; 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_VAL:%.*]] = phi i64 [ 1000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
@@ -465,6 +468,9 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2-LABEL: @test_loop2(
; IC2-NEXT: iter.check:
; IC2-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
+; IC2: vector.scevcheck:
+; IC2-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[N:%.*]], 1
+; IC2-NEXT: br i1 [[IDENT_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK1:%.*]]
; IC2: vector.main.loop.iter.check:
; IC2-NEXT: br i1 false, label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
; IC2: vector.ph:
@@ -502,22 +508,22 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2-NEXT: [[TMP29:%.*]] = add i64 [[INDEX]], 29
; IC2-NEXT: [[TMP30:%.*]] = add i64 [[INDEX]], 30
; IC2-NEXT: [[TMP31:%.*]] = add i64 [[INDEX]], 31
-; IC2-NEXT: [[TMP32:%.*]] = sub nsw i64 [[N:%.*]], [[INDEX]]
-; IC2-NEXT: [[TMP33:%.*]] = sub nsw i64 [[N]], [[TMP1]]
-; IC2-NEXT: [[TMP34:%.*]] = sub nsw i64 [[N]], [[TMP2]]
-; IC2-NEXT: [[TMP35:%.*]] = sub nsw i64 [[N]], [[TMP3]]
-; IC2-NEXT: [[TMP36:%.*]] = sub nsw i64 [[N]], [[TMP4]]
-; IC2-NEXT: [[TMP37:%.*]] = sub nsw i64 [[N]], [[TMP5]]
-; IC2-NEXT: [[TMP38:%.*]] = sub nsw i64 [[N]], [[TMP6]]
-; IC2-NEXT: [[TMP39:%.*]] = sub nsw i64 [[N]], [[TMP7]]
-; IC2-NEXT: [[TMP40:%.*]] = sub nsw i64 [[N]], [[TMP8]]
-; IC2-NEXT: [[TMP41:%.*]] = sub nsw i64 [[N]], [[TMP9]]
-; IC2-NEXT: [[TMP42:%.*]] = sub nsw i64 [[N]], [[TMP10]]
-; IC2-NEXT: [[TMP43:%.*]] = sub nsw i64 [[N]], [[TMP11]]
-; IC2-NEXT: [[TMP44:%.*]] = sub nsw i64 [[N]], [[TMP12]]
-; IC2-NEXT: [[TMP45:%.*]] = sub nsw i64 [[N]], [[TMP13]]
-; IC2-NEXT: [[TMP46:%.*]] = sub nsw i64 [[N]], [[TMP14]]
-; IC2-NEXT: [[TMP47:%.*]] = sub nsw i64 [[N]], [[TMP15]]
+; IC2-NEXT: [[TMP32:%.*]] = sub nsw i64 1, [[INDEX]]
+; IC2-NEXT: [[TMP33:%.*]] = sub nsw i64 1, [[TMP1]]
+; IC2-NEXT: [[TMP34:%.*]] = sub nsw i64 1, [[TMP2]]
+; IC2-NEXT: [[TMP35:%.*]] = sub nsw i64 1, [[TMP3]]
+; IC2-NEXT: [[TMP36:%.*]] = sub nsw i64 1, [[TMP4]]
+; IC2-NEXT: [[TMP37:%.*]] = sub nsw i64 1, [[TMP5]]
+; IC2-NEXT: [[TMP38:%.*]] = sub nsw i64 1, [[TMP6]]
+; IC2-NEXT: [[TMP39:%.*]] = sub nsw i64 1, [[TMP7]]
+; IC2-NEXT: [[TMP40:%.*]] = sub nsw i64 1, [[TMP8]]
+; IC2-NEXT: [[TMP41:%.*]] = sub nsw i64 1, [[TMP9]]
+; IC2-NEXT: [[TMP42:%.*]] = sub nsw i64 1, [[TMP10]]
+; IC2-NEXT: [[TMP43:%.*]] = sub nsw i64 1, [[TMP11]]
+; IC2-NEXT: [[TMP44:%.*]] = sub nsw i64 1, [[TMP12]]
+; IC2-NEXT: [[TMP45:%.*]] = sub nsw i64 1, [[TMP13]]
+; IC2-NEXT: [[TMP46:%.*]] = sub nsw i64 1, [[TMP14]]
+; IC2-NEXT: [[TMP47:%.*]] = sub nsw i64 1, [[TMP15]]
; IC2-NEXT: [[TMP48:%.*]] = insertelement <16 x i64> poison, i64 [[TMP32]], i32 0
; IC2-NEXT: [[TMP49:%.*]] = insertelement <16 x i64> [[TMP48]], i64 [[TMP33]], i32 1
; IC2-NEXT: [[TMP50:%.*]] = insertelement <16 x i64> [[TMP49]], i64 [[TMP34]], i32 2
@@ -534,22 +540,22 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2-NEXT: [[TMP61:%.*]] = insertelement <16 x i64> [[TMP60]], i64 [[TMP45]], i32 13
; IC2-NEXT: [[TMP62:%.*]] = insertelement <16 x i64> [[TMP61]], i64 [[TMP46]], i32 14
; IC2-NEXT: [[TMP63:%.*]] = insertelement <16 x i64> [[TMP62]], i64 [[TMP47]], i32 15
-; IC2-NEXT: [[TMP64:%.*]] = sub nsw i64 [[N]], [[TMP16]]
-; IC2-NEXT: [[TMP65:%.*]] = sub nsw i64 [[N]], [[TMP17]]
-; IC2-NEXT: [[TMP66:%.*]] = sub nsw i64 [[N]], [[TMP18]]
-; IC2-NEXT: [[TMP67:%.*]] = sub nsw i64 [[N]], [[TMP19]]
-; IC2-NEXT: [[TMP68:%.*]] = sub nsw i64 [[N]], [[TMP20]]
-; IC2-NEXT: [[TMP69:%.*]] = sub nsw i64 [[N]], [[TMP21]]
-; IC2-NEXT: [[TMP70:%.*]] = sub nsw i64 [[N]], [[TMP22]]
-; IC2-NEXT: [[TMP71:%.*]] = sub nsw i64 [[N]], [[TMP23]]
-; IC2-NEXT: [[TMP72:%.*]] = sub nsw i64 [[N]], [[TMP24]]
-; IC2-NEXT: [[TMP73:%.*]] = sub nsw i64 [[N]], [[TMP25]]
-; IC2-NEXT: [[TMP74:%.*]] = sub nsw i64 [[N]], [[TMP26]]
-; IC2-NEXT: [[TMP75:%.*]] = sub nsw i64 [[N]], [[TMP27]]
-; IC2-NEXT: [[TMP76:%.*]] = sub nsw i64 [[N]], [[TMP28]]
-; IC2-NEXT: [[TMP77:%.*]] = sub nsw i64 [[N]], [[TMP29]]
-; IC2-NEXT: [[TMP78:%.*]] = sub nsw i64 [[N]], [[TMP30]]
-; IC2-NEXT: [[TMP79:%.*]] = sub nsw i64 [[N]], [[TMP31]]
+; IC2-NEXT: [[TMP64:%.*]] = sub nsw i64 1, [[TMP16]]
+; IC2-NEXT: [[TMP65:%.*]] = sub nsw i64 1, [[TMP17]]
+; IC2-NEXT: [[TMP66:%.*]] = sub nsw i64 1, [[TMP18]]
+; IC2-NEXT: [[TMP67:%.*]] = sub nsw i64 1, [[TMP19]]
+; IC2-NEXT: [[TMP68:%.*]] = sub nsw i64 1, [[TMP20]]
+; IC2-NEXT: [[TMP69:%.*]] = sub nsw i64 1, [[TMP21]]
+; IC2-NEXT: [[TMP70:%.*]] = sub nsw i64 1, [[TMP22]]
+; IC2-NEXT: [[TMP71:%.*]] = sub nsw i64 1, [[TMP23]]
+; IC2-NEXT: [[TMP72:%.*]] = sub nsw i64 1, [[TMP24]]
+; IC2-NEXT: [[TMP73:%.*]] = sub nsw i64 1, [[TMP25]]
+; IC2-NEXT: [[TMP74:%.*]] = sub nsw i64 1, [[TMP26]]
+; IC2-NEXT: [[TMP75:%.*]] = sub nsw i64 1, [[TMP27]]
+; IC2-NEXT: [[TMP76:%.*]] = sub nsw i64 1, [[TMP28]]
+; IC2-NEXT: [[TMP77:%.*]] = sub nsw i64 1, [[TMP29]]
+; IC2-NEXT: [[TMP78:%.*]] = sub nsw i64 1, [[TMP30]]
+; IC2-NEXT: [[TMP79:%.*]] = sub nsw i64 1, [[TMP31]]
; IC2-NEXT: [[TMP80:%.*]] = insertelement <16 x i64> poison, i64 [[TMP64]], i32 0
; IC2-NEXT: [[TMP81:%.*]] = insertelement <16 x i64> [[TMP80]], i64 [[TMP65]], i32 1
; IC2-NEXT: [[TMP82:%.*]] = insertelement <16 x i64> [[TMP81]], i64 [[TMP66]], i32 2
@@ -584,7 +590,7 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2: vec.epilog.iter.check:
; IC2-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF10:![0-9]+]]
; IC2: vec.epilog.ph:
-; IC2-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
+; IC2-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK1]] ]
; IC2-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; IC2: vec.epilog.vector.body:
; IC2-NEXT: [[INDEX1:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT2:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
@@ -595,14 +601,14 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2-NEXT: [[TMP110:%.*]] = add i64 [[INDEX1]], 5
; IC2-NEXT: [[TMP111:%.*]] = add i64 [[INDEX1]], 6
; IC2-NEXT: [[TMP112:%.*]] = add i64 [[INDEX1]], 7
-; IC2-NEXT: [[TMP113:%.*]] = sub nsw i64 [[N]], [[INDEX1]]
-; IC2-NEXT: [[TMP114:%.*]] = sub nsw i64 [[N]], [[TMP106]]
-; IC2-NEXT: [[TMP115:%.*]] = sub nsw i64 [[N]], [[TMP107]]
-; IC2-NEXT: [[TMP116:%.*]] = sub nsw i64 [[N]], [[TMP108]]
-; IC2-NEXT: [[TMP117:%.*]] = sub nsw i64 [[N]], [[TMP109]]
-; IC2-NEXT: [[TMP118:%.*]] = sub nsw i64 [[N]], [[TMP110]]
-; IC2-NEXT: [[TMP119:%.*]] = sub nsw i64 [[N]], [[TMP111]]
-; IC2-NEXT: [[TMP120:%.*]] = sub nsw i64 [[N]], [[TMP112]]
+; IC2-NEXT: [[TMP113:%.*]] = sub nsw i64 1, [[INDEX1]]
+; IC2-NEXT: [[TMP114:%.*]] = sub nsw i64 1, [[TMP106]]
+; IC2-NEXT: [[TMP115:%.*]] = sub nsw i64 1, [[TMP107]]
+; IC2-NEXT: [[TMP116:%.*]] = sub nsw i64 1, [[TMP108]]
+; IC2-NEXT: [[TMP117:%.*]] = sub nsw i64 1, [[TMP109]]
+; IC2-NEXT: [[TMP118:%.*]] = sub nsw i64 1, [[TMP110]]
+; IC2-NEXT: [[TMP119:%.*]] = sub nsw i64 1, [[TMP111]]
+; IC2-NEXT: [[TMP120:%.*]] = sub nsw i64 1, [[TMP112]]
; IC2-NEXT: [[TMP121:%.*]] = insertelement <8 x i64> poison, i64 [[TMP113]], i32 0
; IC2-NEXT: [[TMP122:%.*]] = insertelement <8 x i64> [[TMP121]], i64 [[TMP114]], i32 1
; IC2-NEXT: [[TMP123:%.*]] = insertelement <8 x i64> [[TMP122]], i64 [[TMP115]], i32 2
@@ -622,7 +628,7 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2: vec.epilog.middle.block:
; IC2-NEXT: br i1 false, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
; IC2: vec.epilog.scalar.ph:
-; IC2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
+; IC2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
; IC2-NEXT: br label [[LOOP:%.*]]
; IC2: loop:
; IC2-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll b/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll
index 28b418b5c0cdf..37d6a9414382c 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll
@@ -485,6 +485,9 @@ define void @dead_load_in_block(ptr %dst, ptr %src, i8 %N, i64 %x) #0 {
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[TMP2]], 1
; CHECK-NEXT: br label %[[VECTOR_MEMCHECK:.*]]
; CHECK: [[VECTOR_MEMCHECK]]:
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[X]], 1
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK1:.*]]
+; CHECK: [[VECTOR_MEMCHECK1]]:
; CHECK-NEXT: [[UMIN:%.*]] = call i64 @llvm.umin.i64(i64 [[N_EXT]], i64 1)
; CHECK-NEXT: [[TMP7:%.*]] = sub i64 [[N_EXT]], [[UMIN]]
; CHECK-NEXT: [[TMP8:%.*]] = udiv i64 [[TMP7]], 3
@@ -492,19 +495,11 @@ define void @dead_load_in_block(ptr %dst, ptr %src, i8 %N, i64 %x) #0 {
; CHECK-NEXT: [[TMP10:%.*]] = mul i64 [[TMP9]], 12
; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[TMP10]], 4
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP11]]
-; CHECK-NEXT: [[TMP12:%.*]] = shl i64 [[X]], 2
-; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP12]]
-; CHECK-NEXT: [[TMP13:%.*]] = add i64 [[TMP12]], 4
-; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP13]]
-; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[SRC]], i64 4
-; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[DST]], [[SCEVGEP2]]
-; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP1]], [[SCEVGEP]]
-; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
+; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[SRC]], i64 8
; CHECK-NEXT: [[BOUND04:%.*]] = icmp ult ptr [[DST]], [[SCEVGEP3]]
; CHECK-NEXT: [[BOUND15:%.*]] = icmp ult ptr [[SRC]], [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT6:%.*]] = and i1 [[BOUND04]], [[BOUND15]]
-; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT6]]
-; CHECK-NEXT: br i1 [[CONFLICT_RDX]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT: br i1 [[FOUND_CONFLICT6]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP24:%.*]] = call <vscale x 4 x i64> @llvm.stepvector.nxv4i64()
; CHECK-NEXT: [[TMP25:%.*]] = mul <vscale x 4 x i64> [[TMP24]], splat (i64 3)
@@ -522,7 +517,7 @@ define void @dead_load_in_block(ptr %dst, ptr %src, i8 %N, i64 %x) #0 {
; CHECK-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP17]]
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
-; CHECK-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: br label %[[EXIT:.*]]
; CHECK: [[SCALAR_PH]]:
@@ -541,7 +536,7 @@ define void @dead_load_in_block(ptr %dst, ptr %src, i8 %N, i64 %x) #0 {
; CHECK-NEXT: store i32 0, ptr [[GEP_DST]], align 4
; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 3
; CHECK-NEXT: [[CMP:%.*]] = icmp ult i64 [[IV]], [[N_EXT]]
-; CHECK-NEXT: br i1 [[CMP]], label %[[LOOP_HEADER]], label %[[EXIT]], !llvm.loop [[LOOP17:![0-9]+]]
+; CHECK-NEXT: br i1 [[CMP]], label %[[LOOP_HEADER]], label %[[EXIT]], !llvm.loop [[LOOP16:![0-9]+]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -586,9 +581,8 @@ attributes #0 = { "target-features"="+64bit,+v" }
; CHECK: [[META10]] = !{[[META11:![0-9]+]]}
; CHECK: [[META11]] = distinct !{[[META11]], [[META12:![0-9]+]]}
; CHECK: [[META12]] = distinct !{[[META12]], !"LVerDomain"}
-; CHECK: [[META13]] = !{[[META14:![0-9]+]], [[META15:![0-9]+]]}
+; CHECK: [[META13]] = !{[[META14:![0-9]+]]}
; CHECK: [[META14]] = distinct !{[[META14]], [[META12]]}
-; CHECK: [[META15]] = distinct !{[[META15]], [[META12]]}
-; CHECK: [[LOOP16]] = distinct !{[[LOOP16]], [[META1]], [[META2]]}
-; CHECK: [[LOOP17]] = distinct !{[[LOOP17]], [[META1]]}
+; CHECK: [[LOOP15]] = distinct !{[[LOOP15]], [[META1]], [[META2]]}
+; CHECK: [[LOOP16]] = distinct !{[[LOOP16]], [[META1]]}
;.
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/truncate-to-minimal-bitwidth-cost.ll b/llvm/test/Transforms/LoopVectorize/RISCV/truncate-to-minimal-bitwidth-cost.ll
index 6b9e3d49951a7..ef28ef73f09fd 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/truncate-to-minimal-bitwidth-cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/truncate-to-minimal-bitwidth-cost.ll
@@ -193,21 +193,23 @@ define void @icmp_only_first_op_truncated(ptr noalias %dst, i32 %x, i64 %N, i64
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[V]], 1
; CHECK-NEXT: br label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[X]], 1
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH1:.*]]
+; CHECK: [[VECTOR_PH1]]:
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[N]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i32> poison, i32 [[T]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i32> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i32> poison, <vscale x 2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP7:%.*]] = trunc <vscale x 2 x i64> [[BROADCAST_SPLAT]] to <vscale x 2 x i32>
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq <vscale x 2 x i32> [[TMP7]], [[BROADCAST_SPLAT2]]
-; CHECK-NEXT: [[TMP9:%.*]] = zext i32 [[X]] to i64
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr double, ptr [[SRC]], i64 [[TMP9]]
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr double, ptr [[SRC]], i64 1
; CHECK-NEXT: [[BROADCAST_SPLATINSERT6:%.*]] = insertelement <vscale x 2 x ptr> poison, ptr [[TMP10]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT4:%.*]] = shufflevector <vscale x 2 x ptr> [[BROADCAST_SPLATINSERT6]], <vscale x 2 x ptr> poison, <vscale x 2 x i32> zeroinitializer
; CHECK-NEXT: [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <vscale x 2 x ptr> poison, ptr [[DST]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT6:%.*]] = shufflevector <vscale x 2 x ptr> [[BROADCAST_SPLATINSERT3]], <vscale x 2 x ptr> poison, <vscale x 2 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
-; CHECK-NEXT: [[AVL:%.*]] = phi i64 [ [[TMP0]], %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[AVL:%.*]] = phi i64 [ [[TMP0]], %[[VECTOR_PH1]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP14:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 2, i1 true)
; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 2 x double> @llvm.vp.gather.nxv2f64.nxv2p0(<vscale x 2 x ptr> align 8 [[BROADCAST_SPLAT4]], <vscale x 2 x i1> [[TMP8]], i32 [[TMP14]])
; CHECK-NEXT: call void @llvm.vp.scatter.nxv2f64.nxv2p0(<vscale x 2 x double> [[WIDE_MASKED_GATHER]], <vscale x 2 x ptr> align 8 [[BROADCAST_SPLAT6]], <vscale x 2 x i1> [[TMP8]], i32 [[TMP14]])
@@ -217,6 +219,23 @@ define void @icmp_only_first_op_truncated(ptr noalias %dst, i32 %x, i64 %N, i64
; CHECK-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: br label %[[EXIT:.*]]
+; CHECK: [[SCALAR_PH]]:
+; CHECK-NEXT: br label %[[LOOP_HEADER:.*]]
+; CHECK: [[LOOP_HEADER]]:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ]
+; CHECK-NEXT: [[T1:%.*]] = trunc i64 [[N]] to i32
+; CHECK-NEXT: [[C:%.*]] = icmp eq i32 [[T1]], [[T]]
+; CHECK-NEXT: br i1 [[C]], label %[[THEN:.*]], label %[[LOOP_LATCH]]
+; CHECK: [[THEN]]:
+; CHECK-NEXT: [[IDXPROM:%.*]] = zext i32 [[X]] to i64
+; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr double, ptr [[SRC]], i64 [[IDXPROM]]
+; CHECK-NEXT: [[RETVAL:%.*]] = load double, ptr [[ARRAYIDX]], align 8
+; CHECK-NEXT: store double [[RETVAL]], ptr [[DST]], align 8
+; CHECK-NEXT: br label %[[LOOP_LATCH]]
+; CHECK: [[LOOP_LATCH]]:
+; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
+; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV]], [[V]]
+; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP_HEADER]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -303,17 +322,17 @@ define void @test_minbws_for_trunc(i32 %n, ptr noalias %p1, ptr noalias %p2) {
; CHECK-NEXT: [[TMP15:%.*]] = call <vscale x 2 x i32> @llvm.experimental.vp.strided.load.nxv2i32.p0.i64(ptr align 4 [[TMP14]], i64 16, <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]])
; CHECK-NEXT: [[TMP16:%.*]] = trunc <vscale x 2 x i32> [[TMP15]] to <vscale x 2 x i16>
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr [1 x [1 x i16]], ptr [[P2]], <vscale x 2 x i64> [[TMP12]]
-; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i16.nxv2p0(<vscale x 2 x i16> [[TMP16]], <vscale x 2 x ptr> align 2 [[TMP17]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META6:![0-9]+]], !noalias [[META9:![0-9]+]]
+; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i16.nxv2p0(<vscale x 2 x i16> [[TMP16]], <vscale x 2 x ptr> align 2 [[TMP17]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META7:![0-9]+]], !noalias [[META10:![0-9]+]]
; CHECK-NEXT: [[TMP18:%.*]] = trunc <vscale x 2 x i32> [[TMP15]] to <vscale x 2 x i8>
; CHECK-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[P2]], <vscale x 2 x i64> [[TMP12]]
-; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i8.nxv2p0(<vscale x 2 x i8> [[TMP18]], <vscale x 2 x ptr> align 1 [[TMP19]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META12:![0-9]+]], !noalias [[META13:![0-9]+]]
+; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i8.nxv2p0(<vscale x 2 x i8> [[TMP18]], <vscale x 2 x ptr> align 1 [[TMP19]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META13:![0-9]+]], !noalias [[META14:![0-9]+]]
; CHECK-NEXT: [[TMP20:%.*]] = getelementptr [1 x i64], ptr [[P2]], <vscale x 2 x i64> [[TMP12]]
-; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i64.nxv2p0(<vscale x 2 x i64> zeroinitializer, <vscale x 2 x ptr> align 8 [[TMP20]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META13]]
+; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i64.nxv2p0(<vscale x 2 x i64> zeroinitializer, <vscale x 2 x ptr> align 8 [[TMP20]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META14]]
; CHECK-NEXT: [[CURRENT_ITERATION_NEXT]] = add nuw i32 [[TMP9]], [[INDEX]]
; CHECK-NEXT: [[AVL_NEXT]] = sub nuw i32 [[AVL]], [[TMP9]]
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i16> [[VEC_IND]], [[BROADCAST_SPLAT]]
; CHECK-NEXT: [[TMP21:%.*]] = icmp eq i32 [[AVL_NEXT]], 0
-; CHECK-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: br label %[[EXIT:.*]]
; CHECK: [[SCALAR_PH]]:
@@ -334,7 +353,7 @@ define void @test_minbws_for_trunc(i32 %n, ptr noalias %p1, ptr noalias %p2) {
; CHECK-NEXT: [[IV_NEXT]] = add i16 [[IV]], 4
; CHECK-NEXT: [[IV_NEXT_EXT:%.*]] = sext i16 [[IV_NEXT]] to i32
; CHECK-NEXT: [[CMP:%.*]] = icmp ne i32 [[IV_NEXT_EXT]], 1024
-; CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1]], label %[[EXIT]], !llvm.loop [[LOOP15:![0-9]+]]
+; CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1]], label %[[EXIT]], !llvm.loop [[LOOP16:![0-9]+]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -373,14 +392,15 @@ attributes #1 = { "target-features"="+64bit,+v" }
; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]], [[META2]]}
; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]], [[META2]]}
-; CHECK: [[META6]] = !{[[META7:![0-9]+]]}
-; CHECK: [[META7]] = distinct !{[[META7]], [[META8:![0-9]+]]}
-; CHECK: [[META8]] = distinct !{[[META8]], !"LVerDomain"}
-; CHECK: [[META9]] = !{[[META10:![0-9]+]], [[META11:![0-9]+]]}
-; CHECK: [[META10]] = distinct !{[[META10]], [[META8]]}
-; CHECK: [[META11]] = distinct !{[[META11]], [[META8]]}
-; CHECK: [[META12]] = !{[[META10]]}
+; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]]}
+; CHECK: [[META7]] = !{[[META8:![0-9]+]]}
+; CHECK: [[META8]] = distinct !{[[META8]], [[META9:![0-9]+]]}
+; CHECK: [[META9]] = distinct !{[[META9]], !"LVerDomain"}
+; CHECK: [[META10]] = !{[[META11:![0-9]+]], [[META12:![0-9]+]]}
+; CHECK: [[META11]] = distinct !{[[META11]], [[META9]]}
+; CHECK: [[META12]] = distinct !{[[META12]], [[META9]]}
; CHECK: [[META13]] = !{[[META11]]}
-; CHECK: [[LOOP14]] = distinct !{[[LOOP14]], [[META1]], [[META2]]}
-; CHECK: [[LOOP15]] = distinct !{[[LOOP15]], [[META1]]}
+; CHECK: [[META14]] = !{[[META12]]}
+; CHECK: [[LOOP15]] = distinct !{[[LOOP15]], [[META1]], [[META2]]}
+; CHECK: [[LOOP16]] = distinct !{[[LOOP16]], [[META1]]}
;.
diff --git a/llvm/test/Transforms/LoopVectorize/SystemZ/load-scalarization-cost-0.ll b/llvm/test/Transforms/LoopVectorize/SystemZ/load-scalarization-cost-0.ll
index b6e5862d351cd..dba1a7e767a32 100644
--- a/llvm/test/Transforms/LoopVectorize/SystemZ/load-scalarization-cost-0.ll
+++ b/llvm/test/Transforms/LoopVectorize/SystemZ/load-scalarization-cost-0.ll
@@ -2,8 +2,6 @@
; RUN: -force-vector-width=2 -debug-only=loop-vectorize \
; RUN: -disable-output < %s 2>&1 | FileCheck %s
; REQUIRES: asserts
-;
-; Check that a scalarized load does not get operands scalarization costs added.
define void @fun(ptr %data, i64 %n, i64 %s, ptr %Src) {
entry:
@@ -24,6 +22,6 @@ for.body:
for.end:
ret void
-; CHECK: Cost of 2 for VF 2: forced scalar %mul = mul nsw i64 %iv, %s
-; CHECK: Cost of 2 for VF 2: REPLICATE ir<%ld> = load ir<%bct>
+; CHECK: Cost of 0 for VF 2: forced scalar %mul = mul nsw i64 %iv, %s
+; CHECK: Cost of 1 for VF 2: WIDEN ir<%ld> = load vp<{{.*}}>
}
diff --git a/llvm/test/Transforms/LoopVectorize/pointer-induction.ll b/llvm/test/Transforms/LoopVectorize/pointer-induction.ll
index db10937efc12c..ba34a665ca0ea 100644
--- a/llvm/test/Transforms/LoopVectorize/pointer-induction.ll
+++ b/llvm/test/Transforms/LoopVectorize/pointer-induction.ll
@@ -519,47 +519,33 @@ define i64 @ivopt_widen_ptr_indvar_2(ptr noalias %a, i64 %stride, i64 %n) {
; STRIDED-NEXT: [[TMP0:%.*]] = add i64 [[N:%.*]], 1
; STRIDED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 4
; STRIDED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; STRIDED: vector.scevcheck:
+; STRIDED-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE]], 1
+; STRIDED-NEXT: br i1 [[IDENT_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH1:%.*]]
; STRIDED: vector.ph:
; STRIDED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 4
; STRIDED-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
-; STRIDED-NEXT: [[TMP2:%.*]] = mul i64 [[N_VEC]], [[TMP1]]
+; STRIDED-NEXT: [[TMP2:%.*]] = shl i64 [[N_VEC]], 3
; STRIDED-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP2]]
; STRIDED-NEXT: br label [[VECTOR_BODY:%.*]]
; STRIDED: vector.body:
-; STRIDED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
-; STRIDED-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], [[TMP1]]
-; STRIDED-NEXT: [[TMP10:%.*]] = mul i64 1, [[TMP1]]
-; STRIDED-NEXT: [[TMP11:%.*]] = add i64 [[OFFSET_IDX]], [[TMP10]]
-; STRIDED-NEXT: [[TMP12:%.*]] = mul i64 2, [[TMP1]]
-; STRIDED-NEXT: [[TMP13:%.*]] = add i64 [[OFFSET_IDX]], [[TMP12]]
-; STRIDED-NEXT: [[TMP14:%.*]] = mul i64 3, [[TMP1]]
-; STRIDED-NEXT: [[TMP15:%.*]] = add i64 [[OFFSET_IDX]], [[TMP14]]
-; STRIDED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr null, i64 [[OFFSET_IDX]]
-; STRIDED-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr null, i64 [[TMP11]]
-; STRIDED-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr null, i64 [[TMP13]]
+; STRIDED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; STRIDED-NEXT: [[TMP15:%.*]] = shl i64 [[INDEX]], 3
; STRIDED-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr null, i64 [[TMP15]]
; STRIDED-NEXT: [[TMP21:%.*]] = getelementptr i64, ptr [[A:%.*]], i64 [[INDEX]]
; STRIDED-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP21]], align 8
-; STRIDED-NEXT: [[TMP23:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i64 0
-; STRIDED-NEXT: store i64 [[TMP23]], ptr [[NEXT_GEP]], align 8
-; STRIDED-NEXT: [[TMP24:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i64 1
-; STRIDED-NEXT: store i64 [[TMP24]], ptr [[NEXT_GEP1]], align 8
-; STRIDED-NEXT: [[TMP16:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i64 2
-; STRIDED-NEXT: store i64 [[TMP16]], ptr [[NEXT_GEP2]], align 8
-; STRIDED-NEXT: [[TMP25:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i64 3
-; STRIDED-NEXT: store i64 [[TMP25]], ptr [[NEXT_GEP3]], align 8
+; STRIDED-NEXT: store <4 x i64> [[WIDE_LOAD]], ptr [[NEXT_GEP3]], align 8
; STRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; STRIDED-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; STRIDED-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
; STRIDED: middle.block:
-; STRIDED-NEXT: [[TMP19:%.*]] = sub i64 0, [[TMP1]]
-; STRIDED-NEXT: [[IND_ESCAPE:%.*]] = getelementptr i8, ptr [[TMP3]], i64 [[TMP19]]
+; STRIDED-NEXT: [[IND_ESCAPE:%.*]] = getelementptr i8, ptr [[TMP3]], i64 -8
; STRIDED-NEXT: [[TMP17:%.*]] = extractelement <4 x i64> [[WIDE_LOAD]], i64 3
; STRIDED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
; STRIDED-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
; STRIDED: scalar.ph:
-; STRIDED-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
-; STRIDED-NEXT: [[BC_RESUME_VAL4:%.*]] = phi ptr [ [[TMP3]], [[MIDDLE_BLOCK]] ], [ null, [[ENTRY]] ]
+; STRIDED-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ], [ 0, [[VECTOR_PH]] ]
+; STRIDED-NEXT: [[BC_RESUME_VAL4:%.*]] = phi ptr [ [[TMP3]], [[MIDDLE_BLOCK]] ], [ null, [[ENTRY]] ], [ null, [[VECTOR_PH]] ]
; STRIDED-NEXT: br label [[FOR_BODY:%.*]]
; STRIDED: for.body:
; STRIDED-NEXT: [[INDVAR:%.*]] = phi i64 [ [[INDVAR_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
@@ -713,8 +699,8 @@ define void @strided_ptr_iv_runtime_stride(ptr %pIn, ptr %pOut, i32 %nCols, i32
; STRIDED-NEXT: br i1 [[IDENT_CHECK]], label [[SCALAR_PH]], label [[VECTOR_MEMCHECK:%.*]]
; STRIDED: vector.memcheck:
; STRIDED-NEXT: [[TMP3:%.*]] = sub i64 [[POUT1]], [[PIN2]]
-; STRIDED-NEXT: [[TMP10:%.*]] = sub i64 [[TMP3]], 1
-; STRIDED-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP10]], 15
+; STRIDED-NEXT: [[TMP7:%.*]] = sub i64 [[TMP3]], 1
+; STRIDED-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP7]], 15
; STRIDED-NEXT: br i1 [[DIFF_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
; STRIDED: vector.ph:
; STRIDED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[UMAX]], 4
diff --git a/llvm/test/Transforms/LoopVectorize/pr59319-loop-access-info-invalidation.ll b/llvm/test/Transforms/LoopVectorize/pr59319-loop-access-info-invalidation.ll
index 9ffee47c8a31d..7a2719ee7d6ad 100644
--- a/llvm/test/Transforms/LoopVectorize/pr59319-loop-access-info-invalidation.ll
+++ b/llvm/test/Transforms/LoopVectorize/pr59319-loop-access-info-invalidation.ll
@@ -38,12 +38,13 @@ define void @reduced(ptr %0, ptr %1, i64 %iv, ptr %2, i64 %iv76, i64 %iv93) {
; CHECK-NEXT: [[IV761_LCSSA:%.*]] = phi i64 [ [[IV761]], [[LOOP_1]] ], [ [[IND_ESCAPE]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: [[MIN_ITERS_CHECK6:%.*]] = icmp ult i64 [[TMP3]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK6]], label [[SCALAR_PH5:%.*]], label [[VECTOR_MEMCHECK:%.*]]
+; CHECK: vector.scevcheck:
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[IV761_LCSSA]], 1
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], label [[SCALAR_PH5]], label [[VECTOR_MEMCHECK1:%.*]]
; CHECK: vector.memcheck:
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[TMP1:%.*]], i64 4
-; CHECK-NEXT: [[TMP9:%.*]] = shl i64 [[IV761_LCSSA]], 2
-; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[TMP0]], i64 [[TMP9]]
-; CHECK-NEXT: [[TMP10:%.*]] = add i64 [[TMP9]], 4
-; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[TMP0]], i64 [[TMP10]]
+; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[TMP0]], i64 4
+; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[TMP0]], i64 8
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[TMP1]], [[SCEVGEP3]]
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP2]], [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
@@ -56,43 +57,45 @@ define void @reduced(ptr %0, ptr %1, i64 %iv, ptr %2, i64 %iv76, i64 %iv93) {
; CHECK-NEXT: [[INDEX12:%.*]] = phi i64 [ 0, [[VECTOR_PH6]] ], [ [[INDEX_NEXT13:%.*]], [[VECTOR_BODY10]] ]
; CHECK-NEXT: [[INDEX_NEXT13]] = add nuw i64 [[INDEX12]], 4
; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT13]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK12:%.*]], label [[VECTOR_BODY10]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK11:%.*]], label [[VECTOR_BODY10]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: middle.block11:
; CHECK-NEXT: store i32 0, ptr [[TMP1]], align 4, !alias.scope [[META5:![0-9]+]], !noalias [[META8:![0-9]+]]
; CHECK-NEXT: [[CMP_N10:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N10]], label [[LOOP_3_LR_PH:%.*]], label [[SCALAR_PH5]]
; CHECK: scalar.ph3:
-; CHECK-NEXT: [[BC_RESUME_VAL13:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK12]] ], [ 0, [[LOOP_2_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ]
+; CHECK-NEXT: [[BC_RESUME_VAL13:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK11]] ], [ 0, [[LOOP_2_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[VECTOR_MEMCHECK1]] ]
; CHECK-NEXT: br label [[LOOP_2:%.*]]
; CHECK: loop.3.lr.ph:
; CHECK-NEXT: [[IDXPROM_I_I61:%.*]] = and i64 [[IV761_LCSSA]], 1
; CHECK-NEXT: [[ARRAYIDX_I_I62:%.*]] = getelementptr i32, ptr [[TMP0]], i64 [[IDXPROM_I_I61]]
; CHECK-NEXT: [[MIN_ITERS_CHECK22:%.*]] = icmp ult i64 [[TMP3]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK22]], label [[SCALAR_PH21:%.*]], label [[VECTOR_MEMCHECK15:%.*]]
-; CHECK: vector.memcheck14:
+; CHECK: vector.scevcheck14:
+; CHECK-NEXT: [[IDENT_CHECK15:%.*]] = icmp ne i64 [[IV761_LCSSA]], 1
+; CHECK-NEXT: br i1 [[IDENT_CHECK15]], label [[SCALAR_PH21]], label [[VECTOR_MEMCHECK16:%.*]]
+; CHECK: vector.memcheck16:
; CHECK-NEXT: [[SCEVGEP15:%.*]] = getelementptr i8, ptr [[TMP1]], i64 4
-; CHECK-NEXT: [[TMP12:%.*]] = shl nuw nsw i64 [[IDXPROM_I_I61]], 2
-; CHECK-NEXT: [[TMP13:%.*]] = add nuw nsw i64 [[TMP12]], 4
-; CHECK-NEXT: [[SCEVGEP16:%.*]] = getelementptr i8, ptr [[TMP0]], i64 [[TMP13]]
+; CHECK-NEXT: [[SCEVGEP18:%.*]] = getelementptr i8, ptr [[TMP0]], i64 4
+; CHECK-NEXT: [[SCEVGEP16:%.*]] = getelementptr i8, ptr [[TMP0]], i64 8
; CHECK-NEXT: [[BOUND017:%.*]] = icmp ult ptr [[TMP1]], [[SCEVGEP16]]
-; CHECK-NEXT: [[BOUND118:%.*]] = icmp ult ptr [[ARRAYIDX_I_I62]], [[SCEVGEP15]]
+; CHECK-NEXT: [[BOUND118:%.*]] = icmp ult ptr [[SCEVGEP18]], [[SCEVGEP15]]
; CHECK-NEXT: [[FOUND_CONFLICT19:%.*]] = and i1 [[BOUND017]], [[BOUND118]]
; CHECK-NEXT: br i1 [[FOUND_CONFLICT19]], label [[SCALAR_PH21]], label [[VECTOR_PH24:%.*]]
-; CHECK: vector.ph22:
+; CHECK: vector.ph25:
; CHECK-NEXT: [[N_MOD_VF24:%.*]] = urem i64 [[TMP3]], 4
; CHECK-NEXT: [[N_VEC25:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF24]]
; CHECK-NEXT: br label [[VECTOR_BODY27:%.*]]
-; CHECK: vector.body25:
+; CHECK: vector.body28:
; CHECK-NEXT: [[INDEX29:%.*]] = phi i64 [ 0, [[VECTOR_PH24]] ], [ [[INDEX_NEXT29:%.*]], [[VECTOR_BODY27]] ]
; CHECK-NEXT: [[INDEX_NEXT29]] = add nuw i64 [[INDEX29]], 4
; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT29]], [[N_VEC25]]
-; CHECK-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK29:%.*]], label [[VECTOR_BODY27]], !llvm.loop [[LOOP10:![0-9]+]]
-; CHECK: middle.block28:
+; CHECK-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK31:%.*]], label [[VECTOR_BODY27]], !llvm.loop [[LOOP10:![0-9]+]]
+; CHECK: middle.block31:
; CHECK-NEXT: store i32 0, ptr [[TMP1]], align 4, !alias.scope [[META11:![0-9]+]], !noalias [[META14:![0-9]+]]
; CHECK-NEXT: [[CMP_N27:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC25]]
; CHECK-NEXT: br i1 [[CMP_N27]], label [[LOOP_CLEANUP:%.*]], label [[SCALAR_PH21]]
-; CHECK: scalar.ph20:
-; CHECK-NEXT: [[BC_RESUME_VAL26:%.*]] = phi i64 [ [[N_VEC25]], [[MIDDLE_BLOCK29]] ], [ 0, [[LOOP_3_LR_PH]] ], [ 0, [[VECTOR_MEMCHECK15]] ]
+; CHECK: scalar.ph23:
+; CHECK-NEXT: [[BC_RESUME_VAL26:%.*]] = phi i64 [ [[N_VEC25]], [[MIDDLE_BLOCK31]] ], [ 0, [[LOOP_3_LR_PH]] ], [ 0, [[VECTOR_MEMCHECK15]] ], [ 0, [[VECTOR_MEMCHECK16]] ]
; CHECK-NEXT: br label [[LOOP_3:%.*]]
; CHECK: loop.2:
; CHECK-NEXT: [[IV846:%.*]] = phi i64 [ [[IV_NEXT85:%.*]], [[LOOP_2]] ], [ [[BC_RESUME_VAL13]], [[SCALAR_PH5]] ]
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll b/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
index 0cb40f354cb4c..06e9554937a0e 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
@@ -4,30 +4,7 @@
define void @test(ptr %A, i32 %x) {
; CHECK-LABEL: define void @test(
; CHECK-SAME: ptr [[A:%.*]], i32 [[X:%.*]]) {
-; CHECK-NEXT: [[ENTRY:.*:]]
-; CHECK-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
-; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[X]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
-; CHECK: [[VECTOR_PH]]:
-; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
-; CHECK: [[VECTOR_BODY]]:
-; CHECK-NEXT: [[OFFSET_IDX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[OFFSET_IDX]], 1
-; CHECK-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
-; CHECK-NEXT: [[TMP5:%.*]] = zext i32 [[TMP4]] to i64
-; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP5]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP6]], align 4
-; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[OFFSET_IDX]] to i32
-; CHECK-NEXT: [[TMP8:%.*]] = zext i32 [[TMP0]] to i64
-; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP8]]
-; CHECK-NEXT: store <4 x float> [[WIDE_LOAD]], ptr [[TMP9]], align 4
-; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[OFFSET_IDX]], 4
-; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000
-; CHECK-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
-; CHECK: [[MIDDLE_BLOCK]]:
-; CHECK-NEXT: br label %[[EXIT:.*]]
-; CHECK: [[SCALAR_PH]]:
+; CHECK-NEXT: [[SCALAR_PH:.*]]:
; CHECK-NEXT: br label %[[LOOP:.*]]
; CHECK: [[LOOP]]:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], %[[LOOP]] ], [ 0, %[[SCALAR_PH]] ]
@@ -43,7 +20,7 @@ define void @test(ptr %A, i32 %x) {
; CHECK-NEXT: [[ARRAYIDX1209:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IDX_2]]
; CHECK-NEXT: store float [[LV]], ptr [[ARRAYIDX1209]], align 4
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], 1000
-; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK-NEXT: br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -102,7 +79,7 @@ define void @diff_memcheck_known_false_for_vf_4(ptr %B, ptr %A, ptr %end) {
; CHECK-NEXT: store <4 x i64> zeroinitializer, ptr [[TMP8]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 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: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
@@ -118,7 +95,7 @@ define void @diff_memcheck_known_false_for_vf_4(ptr %B, ptr %A, ptr %end) {
; CHECK-NEXT: [[TMP10:%.*]] = load i64, ptr [[IV_2_NEXT]], align 8
; CHECK-NEXT: store i64 0, ptr [[IV_1]], align 8
; CHECK-NEXT: [[CMP_NOT_I_I_I_I:%.*]] = icmp eq ptr [[END]], [[IV_2]]
-; CHECK-NEXT: br i1 [[CMP_NOT_I_I_I_I]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK-NEXT: br i1 [[CMP_NOT_I_I_I_I]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -146,7 +123,5 @@ exit:
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
-; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]}
-; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
-; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META2]], [[META1]]}
+; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
;.
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll b/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll
index 23ebd63d07cd9..adbe6bcb91a79 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll
@@ -1503,10 +1503,13 @@ define void @stride_check_known_via_loop_guard(ptr %C, ptr %A, i32 %Acols) {
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr double, ptr [[A]], i32 [[MUL_US]]
; CHECK-NEXT: br label %[[VECTOR_MEMCHECK:.*]]
; CHECK: [[VECTOR_MEMCHECK]]:
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[ACOLS]], 1
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK1:.*]]
+; CHECK: [[VECTOR_MEMCHECK1]]:
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[C]], [[SCEVGEP1]]
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
-; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP0:%.*]] = load double, ptr [[ARRAYIDX_US]], align 8, !alias.scope [[META69:![0-9]+]]
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x double> poison, double [[TMP0]], i64 0
@@ -1522,9 +1525,10 @@ define void @stride_check_known_via_loop_guard(ptr %C, ptr %A, i32 %Acols) {
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: br label %[[OUTER_LATCH]]
; CHECK: [[SCALAR_PH]]:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 0, %[[VECTOR_MEMCHECK]] ], [ 0, %[[VECTOR_MEMCHECK1]] ]
; CHECK-NEXT: br label %[[INNER:.*]]
; CHECK: [[INNER]]:
-; CHECK-NEXT: [[INNER_IV:%.*]] = phi i32 [ 0, %[[SCALAR_PH]] ], [ [[INNER_IV_NEXT:%.*]], %[[INNER]] ]
+; CHECK-NEXT: [[INNER_IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INNER_IV_NEXT:%.*]], %[[INNER]] ]
; CHECK-NEXT: [[GEP_C:%.*]] = getelementptr inbounds double, ptr [[C]], i32 [[INNER_IV]]
; CHECK-NEXT: [[L:%.*]] = load double, ptr [[ARRAYIDX_US]], align 8
; CHECK-NEXT: store double [[L]], ptr [[GEP_C]], align 8
diff --git a/llvm/test/Transforms/LoopVectorize/vector-no-scevcheck.ll b/llvm/test/Transforms/LoopVectorize/vector-no-scevcheck.ll
index 56a8b5a93b851..d1b5cc947f54a 100644
--- a/llvm/test/Transforms/LoopVectorize/vector-no-scevcheck.ll
+++ b/llvm/test/Transforms/LoopVectorize/vector-no-scevcheck.ll
@@ -1,12 +1,40 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^vector.memcheck:" --version 6
; RUN: opt -passes=loop-vectorize -force-vector-width=2 -force-vector-interleave=1 -S < %s | FileCheck %s
-; This test is to ensure that SCEV checks (which are costly performancewise) are
-; not generated when appropriate aliasing checks are sufficient.
+; SCEV checks are cheap, due to stride being speculated.
define void @foo(ptr %pout, ptr %pin, i64 %val0, i64 %val1, i64 %val2) {
-; CHECK-LABEL: @foo(
-; CHECK-NOT: vector.scevcheck
-; CHECK: vector.body
+; CHECK-LABEL: define void @foo(
+; CHECK-SAME: ptr [[POUT:%.*]], ptr [[PIN:%.*]], i64 [[VAL0:%.*]], i64 [[VAL1:%.*]], i64 [[VAL2:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[TMP0:%.*]] = getelementptr double, ptr [[PIN]], i64 [[VAL0]]
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[VAL1]], -1
+; CHECK-NEXT: [[TMP2:%.*]] = mul i64 [[VAL1]], [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 3
+; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[VAL1]], i64 1)
+; CHECK-NEXT: [[TMP4:%.*]] = shl i64 [[SMAX]], 3
+; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[TMP3]], [[TMP4]]
+; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[POUT]], i64 [[TMP5]]
+; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[VAL1]], 3
+; CHECK-NEXT: [[TMP7:%.*]] = shl i64 [[VAL0]], 3
+; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[TMP7]], [[TMP4]]
+; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[PIN]], i64 [[TMP8]]
+; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[TMP6]], [[TMP7]]
+; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[PIN]], i64 [[TMP9]]
+; CHECK-NEXT: br label %[[LOOP1_HEADER:.*]]
+; CHECK: [[LOOP1_HEADER]]:
+; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[LOOP1_LATCH:%.*]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT: [[MUL0:%.*]] = mul nsw i64 [[I]], [[VAL2]]
+; CHECK-NEXT: [[ARRAYIDX0:%.*]] = getelementptr inbounds double, ptr [[TMP0]], i64 [[MUL0]]
+; CHECK-NEXT: [[MUL1:%.*]] = mul nsw i64 [[I]], [[VAL1]]
+; CHECK-NEXT: [[TMP10:%.*]] = call i64 @llvm.smax.i64(i64 [[VAL1]], i64 1)
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP10]], 2
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], [[SCALAR_PH:label %.*]], label %[[VECTOR_SCEVCHECK:.*]]
+; CHECK: [[VECTOR_SCEVCHECK]]:
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[VAL2]], 1
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], [[SCALAR_PH]], label %[[EXIT:.*]]
+; CHECK: [[EXIT]]:
+;
entry:
%0 = getelementptr double, ptr %pin, i64 %val0
br label %loop1.header
@@ -40,12 +68,37 @@ exit:
ret void
}
-; Similar test to the above but with the %arrayidx0 moved to the loop2.header
-
define void @bar(ptr %pout, ptr %pin, i64 %val0, i64 %val1, i64 %val2) {
-; CHECK-LABEL: @bar(
-; CHECK-NOT: vector.scevcheck
-; CHECK: vector.body
+; CHECK-LABEL: define void @bar(
+; CHECK-SAME: ptr [[POUT:%.*]], ptr [[PIN:%.*]], i64 [[VAL0:%.*]], i64 [[VAL1:%.*]], i64 [[VAL2:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[TMP0:%.*]] = getelementptr double, ptr [[PIN]], i64 [[VAL0]]
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[VAL1]], -1
+; CHECK-NEXT: [[TMP2:%.*]] = mul i64 [[VAL1]], [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 3
+; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[VAL1]], i64 1)
+; CHECK-NEXT: [[TMP4:%.*]] = shl i64 [[SMAX]], 3
+; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[TMP3]], [[TMP4]]
+; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[POUT]], i64 [[TMP5]]
+; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[VAL1]], 3
+; CHECK-NEXT: [[TMP7:%.*]] = shl i64 [[VAL0]], 3
+; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[TMP7]], [[TMP4]]
+; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[PIN]], i64 [[TMP8]]
+; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[TMP6]], [[TMP7]]
+; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[PIN]], i64 [[TMP9]]
+; CHECK-NEXT: br label %[[LOOP1_HEADER:.*]]
+; CHECK: [[LOOP1_HEADER]]:
+; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[LOOP1_LATCH:%.*]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT: [[MUL0:%.*]] = mul nsw i64 [[I]], [[VAL2]]
+; CHECK-NEXT: [[MUL1:%.*]] = mul nsw i64 [[I]], [[VAL1]]
+; CHECK-NEXT: [[TMP10:%.*]] = call i64 @llvm.smax.i64(i64 [[VAL1]], i64 1)
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP10]], 2
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], [[SCALAR_PH:label %.*]], label %[[VECTOR_SCEVCHECK:.*]]
+; CHECK: [[VECTOR_SCEVCHECK]]:
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[VAL2]], 1
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], [[SCALAR_PH]], label %[[EXIT:.*]]
+; CHECK: [[EXIT]]:
+;
entry:
%0 = getelementptr double, ptr %pin, i64 %val0
br label %loop1.header
diff --git a/llvm/test/Transforms/LoopVectorize/version-mem-access.ll b/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
index 3a98ae31b02b4..c533c95a444f1 100644
--- a/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
+++ b/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
@@ -13,26 +13,45 @@ define void @test(ptr noalias %A, i64 %AStride, ptr noalias %B, i32 %BStride, pt
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[IDENT_CHECK1:%.*]] = icmp ne i32 [[BSTRIDE]], 1
+; CHECK-NEXT: [[TMP0:%.*]] = sub i32 0, [[BSTRIDE]]
+; CHECK-NEXT: [[TMP1:%.*]] = icmp slt i32 [[BSTRIDE]], 0
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i32 [[TMP0]], i32 [[BSTRIDE]]
+; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP2]], i32 99)
+; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
+; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
+; CHECK-NEXT: [[TMP3:%.*]] = sub i32 0, [[MUL_RESULT]]
+; CHECK-NEXT: [[TMP4:%.*]] = icmp ugt i32 [[TMP3]], 0
+; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP1]], i1 [[TMP4]], i1 false
+; CHECK-NEXT: [[IDENT_CHECK1:%.*]] = or i1 [[TMP11]], [[MUL_OVERFLOW]]
; CHECK-NEXT: [[IDENT_CHECK2:%.*]] = icmp ne i64 [[CSTRIDE]], 1
; CHECK-NEXT: [[DIFF_CHECK6:%.*]] = icmp ne i64 [[ASTRIDE]], 1
; CHECK-NEXT: [[DIFF_CHECK:%.*]] = or i1 [[IDENT_CHECK1]], [[IDENT_CHECK2]]
; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[DIFF_CHECK]], [[DIFF_CHECK6]]
; CHECK-NEXT: br i1 [[CONFLICT_RDX]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[BSTRIDE]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP4:%.*]] = trunc i64 [[INDEX]] to i32
-; CHECK-NEXT: [[TMP5:%.*]] = zext i32 [[TMP4]] to i64
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP12:%.*]] = mul <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP18:%.*]] = zext <2 x i32> [[TMP12]] to <2 x i64>
+; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i64> [[TMP18]], i64 0
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP5]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP6]], align 4
+; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i64> [[TMP18]], i64 1
+; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP13]]
+; CHECK-NEXT: [[TMP15:%.*]] = load i32, ptr [[TMP6]], align 4
+; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP14]], align 4
+; CHECK-NEXT: [[TMP17:%.*]] = insertelement <2 x i32> poison, i32 [[TMP15]], i32 0
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = insertelement <2 x i32> [[TMP17]], i32 [[TMP16]], i32 1
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[INDEX]]
; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <2 x i32>, ptr [[TMP7]], align 4
; CHECK-NEXT: [[TMP8:%.*]] = mul nsw <2 x i32> [[WIDE_LOAD7]], [[WIDE_LOAD]]
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT: store <2 x i32> [[TMP8]], ptr [[TMP9]], align 4
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2)
; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
; CHECK-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
@@ -78,25 +97,50 @@ define void @fn1(ptr noalias %x, ptr noalias %c, double %a) {
; CHECK-NEXT: br i1 [[CMP8]], label %[[LOOP_PREHEADER:.*]], [[EXIT:label %.*]]
; CHECK: [[LOOP_PREHEADER]]:
; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[CONV2]] to i64
-; CHECK-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 2
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[CONV]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[CONV]], 3
+; CHECK-NEXT: [[TMP9:%.*]] = sub i32 0, [[CONV]]
+; CHECK-NEXT: [[TMP12:%.*]] = icmp slt i32 [[CONV]], 0
+; CHECK-NEXT: [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP9]], i32 [[CONV]]
+; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP13]], i32 [[TMP1]])
+; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
+; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
+; CHECK-NEXT: [[TMP5:%.*]] = sub i32 0, [[MUL_RESULT]]
+; CHECK-NEXT: [[TMP6:%.*]] = icmp slt i32 [[MUL_RESULT]], 0
+; CHECK-NEXT: [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], 0
+; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP12]], i1 [[TMP7]], i1 [[TMP6]]
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = or i1 [[TMP8]], [[MUL_OVERFLOW]]
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 2
+; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[CONV]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDEX]] to i32
-; CHECK-NEXT: [[TMP2:%.*]] = sext i32 [[TMP1]] to i64
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP10:%.*]] = mul nsw <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP11:%.*]] = sext <2 x i32> [[TMP10]] to <2 x i64>
+; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i64> [[TMP11]], i64 0
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds double, ptr [[X]], i64 [[TMP2]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x double>, ptr [[TMP3]], align 8
+; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i64> [[TMP11]], i64 1
+; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds double, ptr [[X]], i64 [[TMP14]]
+; CHECK-NEXT: [[TMP16:%.*]] = load double, ptr [[TMP3]], align 8
+; CHECK-NEXT: [[TMP17:%.*]] = load double, ptr [[TMP15]], align 8
+; CHECK-NEXT: [[TMP18:%.*]] = insertelement <2 x double> poison, double [[TMP16]], i32 0
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = insertelement <2 x double> [[TMP18]], double [[TMP17]], i32 1
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds double, ptr [[C]], i64 [[INDEX]]
; CHECK-NEXT: store <2 x double> [[WIDE_LOAD]], ptr [[TMP4]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4
-; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2)
+; CHECK-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
-; CHECK-NEXT: br label %[[SCALAR_PH]]
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], [[EXIT_LOOPEXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
;
entry:
diff --git a/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll b/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
index 4429b0e731616..5a7588d04793b 100644
--- a/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
+++ b/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
@@ -360,38 +360,35 @@ define void @byte_gep_scaled_stride(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[STRIDE]], i64 0
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
-; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = shl <4 x i64> [[BROADCAST_SPLAT]], splat (i64 3)
+; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE]], 1
+; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH1:.*]]
+; COMPARE-LAA-MV: [[VECTOR_PH1]]:
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = mul <4 x i64> [[VEC_IND]], [[TMP0]]
-; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = extractelement <4 x i64> [[TMP1]], i64 0
-; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP2]]
-; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = extractelement <4 x i64> [[TMP1]], i64 1
-; COMPARE-LAA-MV-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP3]]
-; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = extractelement <4 x i64> [[TMP1]], i64 2
-; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP4]]
-; COMPARE-LAA-MV-NEXT: [[TMP5:%.*]] = extractelement <4 x i64> [[TMP1]], i64 3
+; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[TMP5:%.*]] = shl i64 [[INDEX]], 3
; COMPARE-LAA-MV-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP5]]
-; COMPARE-LAA-MV-NEXT: [[TMP10:%.*]] = load i64, ptr [[TMP6]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP11:%.*]] = load i64, ptr [[TMP7]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP12:%.*]] = load i64, ptr [[TMP8]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP13:%.*]] = load i64, ptr [[TMP9]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP14:%.*]] = insertelement <4 x i64> poison, i64 [[TMP10]], i32 0
-; COMPARE-LAA-MV-NEXT: [[TMP15:%.*]] = insertelement <4 x i64> [[TMP14]], i64 [[TMP11]], i32 1
-; COMPARE-LAA-MV-NEXT: [[TMP16:%.*]] = insertelement <4 x i64> [[TMP15]], i64 [[TMP12]], i32 2
-; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = insertelement <4 x i64> [[TMP16]], i64 [[TMP13]], i32 3
+; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = load <4 x i64>, ptr [[TMP9]], align 8
; COMPARE-LAA-MV-NEXT: [[TMP18:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[INDEX]]
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[TMP17]], ptr [[TMP18]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
-; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
; COMPARE-LAA-MV-NEXT: br i1 [[TMP19]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
+; COMPARE-LAA-MV: [[SCALAR_PH]]:
+; COMPARE-LAA-MV-NEXT: br label %[[HEADER:.*]]
+; COMPARE-LAA-MV: [[HEADER]]:
+; COMPARE-LAA-MV-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[HEADER]] ]
+; COMPARE-LAA-MV-NEXT: [[IV_NEXT]] = add nsw i64 [[IV]], 1
+; COMPARE-LAA-MV-NEXT: [[STRIDE_X8:%.*]] = mul i64 [[STRIDE]], 8
+; COMPARE-LAA-MV-NEXT: [[IDX:%.*]] = mul i64 [[IV]], [[STRIDE_X8]]
+; COMPARE-LAA-MV-NEXT: [[GEP_LD:%.*]] = getelementptr i8, ptr [[P]], i64 [[IDX]]
+; COMPARE-LAA-MV-NEXT: [[LD:%.*]] = load i64, ptr [[GEP_LD]], align 8
+; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
+; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
+; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP7:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -466,14 +463,14 @@ define void @byte_gep_under_scaled_stride(ptr noalias %p.out, ptr %p, i64 %strid
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[STRIDE]], i64 0
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
-; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = shl <4 x i64> [[BROADCAST_SPLAT]], splat (i64 2)
+; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE]], 1
+; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH1:.*]]
+; COMPARE-LAA-MV: [[VECTOR_PH1]]:
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = mul <4 x i64> [[VEC_IND]], [[TMP0]]
+; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH1]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = shl <4 x i64> [[VEC_IND]], splat (i64 2)
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = extractelement <4 x i64> [[TMP1]], i64 0
; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP2]]
; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = extractelement <4 x i64> [[TMP1]], i64 1
@@ -495,9 +492,22 @@ define void @byte_gep_under_scaled_stride(ptr noalias %p.out, ptr %p, i64 %strid
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP19]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP19]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
+; COMPARE-LAA-MV: [[SCALAR_PH]]:
+; COMPARE-LAA-MV-NEXT: br label %[[HEADER:.*]]
+; COMPARE-LAA-MV: [[HEADER]]:
+; COMPARE-LAA-MV-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[HEADER]] ]
+; COMPARE-LAA-MV-NEXT: [[IV_NEXT]] = add nsw i64 [[IV]], 1
+; COMPARE-LAA-MV-NEXT: [[STRIDE_X8:%.*]] = mul i64 [[STRIDE]], 4
+; COMPARE-LAA-MV-NEXT: [[IDX:%.*]] = mul i64 [[IV]], [[STRIDE_X8]]
+; COMPARE-LAA-MV-NEXT: [[GEP_LD:%.*]] = getelementptr i8, ptr [[P]], i64 [[IDX]]
+; COMPARE-LAA-MV-NEXT: [[LD:%.*]] = load i64, ptr [[GEP_LD]], align 8
+; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
+; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
+; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP9:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -573,14 +583,14 @@ define void @byte_gep_over_scaled_stride(ptr noalias %p.out, ptr %p, i64 %stride
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[STRIDE]], i64 0
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
-; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = shl <4 x i64> [[BROADCAST_SPLAT]], splat (i64 4)
+; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE]], 1
+; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH1:.*]]
+; COMPARE-LAA-MV: [[VECTOR_PH1]]:
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = mul <4 x i64> [[VEC_IND]], [[TMP0]]
+; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH1]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = shl <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = extractelement <4 x i64> [[TMP1]], i64 0
; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP2]]
; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = extractelement <4 x i64> [[TMP1]], i64 1
@@ -602,9 +612,22 @@ define void @byte_gep_over_scaled_stride(ptr noalias %p.out, ptr %p, i64 %stride
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP19]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP19]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
+; COMPARE-LAA-MV: [[SCALAR_PH]]:
+; COMPARE-LAA-MV-NEXT: br label %[[HEADER:.*]]
+; COMPARE-LAA-MV: [[HEADER]]:
+; COMPARE-LAA-MV-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[HEADER]] ]
+; COMPARE-LAA-MV-NEXT: [[IV_NEXT]] = add nsw i64 [[IV]], 1
+; COMPARE-LAA-MV-NEXT: [[STRIDE_X8:%.*]] = mul i64 [[STRIDE]], 16
+; COMPARE-LAA-MV-NEXT: [[IDX:%.*]] = mul i64 [[IV]], [[STRIDE_X8]]
+; COMPARE-LAA-MV-NEXT: [[GEP_LD:%.*]] = getelementptr i8, ptr [[P]], i64 [[IDX]]
+; COMPARE-LAA-MV-NEXT: [[LD:%.*]] = load i64, ptr [[GEP_LD]], align 8
+; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
+; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
+; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP11:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -679,14 +702,14 @@ define void @byte_gep_non_power_of_two_scaled_stride(ptr noalias %p.out, ptr %p,
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
-; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = mul i64 [[STRIDE]], 11
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TMP0]], i64 0
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
+; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE]], 1
+; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH1:.*]]
+; COMPARE-LAA-MV: [[VECTOR_PH1]]:
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = mul <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH1]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = mul <4 x i64> [[VEC_IND]], splat (i64 11)
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = extractelement <4 x i64> [[TMP1]], i64 0
; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP2]]
; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = extractelement <4 x i64> [[TMP1]], i64 1
@@ -708,9 +731,22 @@ define void @byte_gep_non_power_of_two_scaled_stride(ptr noalias %p.out, ptr %p,
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP19]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP19]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
+; COMPARE-LAA-MV: [[SCALAR_PH]]:
+; COMPARE-LAA-MV-NEXT: br label %[[HEADER:.*]]
+; COMPARE-LAA-MV: [[HEADER]]:
+; COMPARE-LAA-MV-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[HEADER]] ]
+; COMPARE-LAA-MV-NEXT: [[IV_NEXT]] = add nsw i64 [[IV]], 1
+; COMPARE-LAA-MV-NEXT: [[STRIDE_X8:%.*]] = mul i64 [[STRIDE]], 11
+; COMPARE-LAA-MV-NEXT: [[IDX:%.*]] = mul i64 [[IV]], [[STRIDE_X8]]
+; COMPARE-LAA-MV-NEXT: [[GEP_LD:%.*]] = getelementptr i8, ptr [[P]], i64 [[IDX]]
+; COMPARE-LAA-MV-NEXT: [[LD:%.*]] = load i64, ptr [[GEP_LD]], align 8
+; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
+; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
+; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP13:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -808,7 +844,7 @@ define void @byte_gep_nonscaled_stride(ptr noalias %p.out, ptr %p, i64 %stride)
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[TMP15]], ptr [[TMP16]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP17]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP17]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -822,7 +858,7 @@ define void @byte_gep_nonscaled_stride(ptr noalias %p.out, ptr %p, i64 %stride)
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP11:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP15:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -897,14 +933,14 @@ define void @byte_gep_negated_stride(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
-; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = sub i64 0, [[STRIDE]]
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[TMP0]], i64 0
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
+; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE]], 1
+; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH1:.*]]
+; COMPARE-LAA-MV: [[VECTOR_PH1]]:
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = mul <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH1]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = sub <4 x i64> zeroinitializer, [[VEC_IND]]
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = extractelement <4 x i64> [[TMP1]], i64 0
; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP2]]
; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = extractelement <4 x i64> [[TMP1]], i64 1
@@ -926,9 +962,22 @@ define void @byte_gep_negated_stride(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP19]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP19]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
+; COMPARE-LAA-MV: [[SCALAR_PH]]:
+; COMPARE-LAA-MV-NEXT: br label %[[HEADER:.*]]
+; COMPARE-LAA-MV: [[HEADER]]:
+; COMPARE-LAA-MV-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[HEADER]] ]
+; COMPARE-LAA-MV-NEXT: [[IV_NEXT]] = add nsw i64 [[IV]], 1
+; COMPARE-LAA-MV-NEXT: [[STRIDE_NEG:%.*]] = sub i64 0, [[STRIDE]]
+; COMPARE-LAA-MV-NEXT: [[IDX:%.*]] = mul i64 [[IV]], [[STRIDE_NEG]]
+; COMPARE-LAA-MV-NEXT: [[GEP_LD:%.*]] = getelementptr i8, ptr [[P]], i64 [[IDX]]
+; COMPARE-LAA-MV-NEXT: [[LD:%.*]] = load i64, ptr [[GEP_LD]], align 8
+; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
+; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
+; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP17:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -1029,7 +1078,7 @@ define void @shared_stride(ptr noalias %p.out, ptr %p0, ptr %p1, i64 %stride) {
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[TMP2]], ptr [[TMP3]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -1046,7 +1095,7 @@ define void @shared_stride(ptr noalias %p.out, ptr %p0, ptr %p1, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP14:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP19:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -1158,7 +1207,7 @@ define void @independent_strides(ptr noalias %p.out, ptr %p0, ptr %p1, i64 %stri
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[TMP3]], ptr [[TMP4]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -1176,7 +1225,7 @@ define void @independent_strides(ptr noalias %p.out, ptr %p0, ptr %p1, i64 %stri
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP16:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP21:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -1303,7 +1352,7 @@ define void @dependent_strides(ptr noalias %p.out, ptr %p0, ptr %p1, i64 %stride
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP17:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -1322,7 +1371,7 @@ define void @dependent_strides(ptr noalias %p.out, ptr %p0, ptr %p1, i64 %stride
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP18:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP23:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -1452,7 +1501,7 @@ define void @dependent_strides_reverse_order(ptr noalias %p.out, ptr %p0, ptr %p
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -1471,7 +1520,7 @@ define void @dependent_strides_reverse_order(ptr noalias %p.out, ptr %p0, ptr %p
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP20:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP25:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -1602,7 +1651,7 @@ define void @byte_dependent_byte_geps(ptr noalias %p.out, ptr %p0, ptr %p1, i64
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[TMP29]], ptr [[TMP30]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP31:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP31]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP21:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP31]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP26:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -1620,7 +1669,7 @@ define void @byte_dependent_byte_geps(ptr noalias %p.out, ptr %p0, ptr %p1, i64
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP22:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP27:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -1751,7 +1800,7 @@ define void @byte_dependent_byte_geps_reverse_order(ptr noalias %p.out, ptr %p0,
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[TMP29]], ptr [[TMP30]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP31:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP31]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP23:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP31]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -1769,7 +1818,7 @@ define void @byte_dependent_byte_geps_reverse_order(ptr noalias %p.out, ptr %p0,
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP24:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP29:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -1877,7 +1926,7 @@ define void @strided_interleave(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[TMP2]], ptr [[TMP3]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP25:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP30:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -1894,7 +1943,7 @@ define void @strided_interleave(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP26:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP31:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -1986,7 +2035,7 @@ define void @in_loop_base(ptr noalias %p.out, ptr %p, i64 %stride, i64 %offset)
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[WIDE_LOAD]], ptr [[TMP2]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP27:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP32:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -2002,7 +2051,7 @@ define void @in_loop_base(ptr noalias %p.out, ptr %p, i64 %stride, i64 %offset)
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP28:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP33:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -2093,7 +2142,7 @@ define void @base_not_in_ir(ptr noalias %p.out, ptr %p, i64 %stride, i64 %offset
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[WIDE_LOAD]], ptr [[TMP2]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP29:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP34:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -2108,7 +2157,7 @@ define void @base_not_in_ir(ptr noalias %p.out, ptr %p, i64 %stride, i64 %offset
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP30:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP35:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -2214,7 +2263,7 @@ define void @non_invariant_uniform_base(ptr noalias %p.out, ptr %p, i64 %stride)
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP31:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP36:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[EXIT]]:
@@ -2324,7 +2373,7 @@ define void @non_invariant_uniform_stride(ptr noalias %p.out, ptr %p, ptr %p.uni
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP32:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP37:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[EXIT]]:
@@ -2440,7 +2489,7 @@ define void @non_constant_btc(ptr noalias %p.out, ptr %p, i64 %stride, i64 %n) {
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[WIDE_LOAD]], ptr [[TMP1]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP33:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP38:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
; COMPARE-LAA-MV-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
@@ -2456,7 +2505,7 @@ define void @non_constant_btc(ptr noalias %p.out, ptr %p, i64 %stride, i64 %n) {
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], [[N]]
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP34:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP39:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -2768,7 +2817,7 @@ define void @stride_btc_checks_order(ptr noalias %p.out, ptr %p, i64 %stride, i6
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[WIDE_LOAD]], ptr [[TMP1]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP35:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP40:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX1]], [[N_VEC]]
; COMPARE-LAA-MV-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
@@ -2784,7 +2833,7 @@ define void @stride_btc_checks_order(ptr noalias %p.out, ptr %p, i64 %stride, i6
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], [[N]]
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP36:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP41:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -2899,7 +2948,7 @@ define void @stride_dependent_btc_non_preventive(ptr noalias %p.out, ptr %p, i64
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], [[N]]
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP37:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP42:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -2987,7 +3036,7 @@ define void @stride_btc_memdep_triple_check(ptr %p, i64 %stride, i64 %out.offset
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], [[N]]
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP38:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP43:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -3125,7 +3174,7 @@ define void @stride_btc_independent_memdep_triple_check(ptr %p, ptr noalias %p2,
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], [[N]]
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP39:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP44:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -3207,20 +3256,18 @@ define void @actual_stride_not_in_ir(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[STRIDE]], i64 0
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
+; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE]], 1
+; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH1:.*]]
+; COMPARE-LAA-MV: [[VECTOR_PH1]]:
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = mul <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = extractelement <4 x i64> [[TMP0]], i64 0
+; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = phi i64 [ 0, %[[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], 1
+; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = add i64 [[TMP1]], 2
+; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = add i64 [[TMP1]], 3
; COMPARE-LAA-MV-NEXT: [[TMP5:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP1]]
-; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = extractelement <4 x i64> [[TMP0]], i64 1
; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP2]]
-; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = extractelement <4 x i64> [[TMP0]], i64 2
; COMPARE-LAA-MV-NEXT: [[TMP7:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP3]]
-; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = extractelement <4 x i64> [[TMP0]], i64 3
; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = getelementptr i64, ptr [[P]], i64 [[TMP4]]
; COMPARE-LAA-MV-NEXT: [[TMP9:%.*]] = getelementptr i64, ptr [[TMP5]], i64 [[TMP1]]
; COMPARE-LAA-MV-NEXT: [[TMP10:%.*]] = getelementptr i64, ptr [[TMP6]], i64 [[TMP2]]
@@ -3234,14 +3281,26 @@ define void @actual_stride_not_in_ir(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[TMP18:%.*]] = insertelement <4 x i64> [[TMP17]], i64 [[TMP14]], i32 1
; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = insertelement <4 x i64> [[TMP18]], i64 [[TMP15]], i32 2
; COMPARE-LAA-MV-NEXT: [[TMP20:%.*]] = insertelement <4 x i64> [[TMP19]], i64 [[TMP16]], i32 3
-; COMPARE-LAA-MV-NEXT: [[TMP21:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[INDEX]]
+; COMPARE-LAA-MV-NEXT: [[TMP21:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[TMP1]]
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[TMP20]], ptr [[TMP21]], align 8
-; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
-; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
+; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP1]], 4
; COMPARE-LAA-MV-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP22]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP40:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP22]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP45:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
+; COMPARE-LAA-MV: [[SCALAR_PH]]:
+; COMPARE-LAA-MV-NEXT: br label %[[HEADER:.*]]
+; COMPARE-LAA-MV: [[HEADER]]:
+; COMPARE-LAA-MV-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[HEADER]] ]
+; COMPARE-LAA-MV-NEXT: [[IV_NEXT]] = add nsw i64 [[IV]], 1
+; COMPARE-LAA-MV-NEXT: [[IDX:%.*]] = mul i64 [[IV]], [[STRIDE]]
+; COMPARE-LAA-MV-NEXT: [[BASE:%.*]] = getelementptr i64, ptr [[P]], i64 [[IDX]]
+; COMPARE-LAA-MV-NEXT: [[GEP_LD:%.*]] = getelementptr i64, ptr [[BASE]], i64 [[IDX]]
+; COMPARE-LAA-MV-NEXT: [[LD:%.*]] = load i64, ptr [[GEP_LD]], align 8
+; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
+; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
+; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP46:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -3328,7 +3387,7 @@ define void @nd_array_last_idx(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[WIDE_LOAD]], ptr [[TMP1]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP41:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP47:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -3342,7 +3401,7 @@ define void @nd_array_last_idx(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP42:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP48:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -3441,7 +3500,7 @@ define void @nd_array_non_last_idx(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[TMP15]], ptr [[TMP16]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP17]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP43:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP17]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP49:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -3455,7 +3514,7 @@ define void @nd_array_non_last_idx(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP44:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP50:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -3527,20 +3586,18 @@ define void @nd_array_multiple_idxs(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[STRIDE]], i64 0
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
+; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE]], 1
+; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH1:.*]]
+; COMPARE-LAA-MV: [[VECTOR_PH1]]:
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = mul <4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = extractelement <4 x i64> [[TMP0]], i64 0
+; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = phi i64 [ 0, %[[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], 1
+; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = add i64 [[TMP1]], 2
+; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = add i64 [[TMP1]], 3
; COMPARE-LAA-MV-NEXT: [[TMP5:%.*]] = getelementptr [256 x [256 x i64]], ptr [[P]], i64 1, i64 [[TMP1]], i64 [[TMP1]]
-; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = extractelement <4 x i64> [[TMP0]], i64 1
; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = getelementptr [256 x [256 x i64]], ptr [[P]], i64 1, i64 [[TMP2]], i64 [[TMP2]]
-; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = extractelement <4 x i64> [[TMP0]], i64 2
; COMPARE-LAA-MV-NEXT: [[TMP7:%.*]] = getelementptr [256 x [256 x i64]], ptr [[P]], i64 1, i64 [[TMP3]], i64 [[TMP3]]
-; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = extractelement <4 x i64> [[TMP0]], i64 3
; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = getelementptr [256 x [256 x i64]], ptr [[P]], i64 1, i64 [[TMP4]], i64 [[TMP4]]
; COMPARE-LAA-MV-NEXT: [[TMP9:%.*]] = load i64, ptr [[TMP5]], align 8
; COMPARE-LAA-MV-NEXT: [[TMP10:%.*]] = load i64, ptr [[TMP6]], align 8
@@ -3550,14 +3607,25 @@ define void @nd_array_multiple_idxs(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[TMP14:%.*]] = insertelement <4 x i64> [[TMP13]], i64 [[TMP10]], i32 1
; COMPARE-LAA-MV-NEXT: [[TMP15:%.*]] = insertelement <4 x i64> [[TMP14]], i64 [[TMP11]], i32 2
; COMPARE-LAA-MV-NEXT: [[TMP16:%.*]] = insertelement <4 x i64> [[TMP15]], i64 [[TMP12]], i32 3
-; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[INDEX]]
+; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[TMP1]]
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[TMP16]], ptr [[TMP17]], align 8
-; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
-; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
+; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP1]], 4
; COMPARE-LAA-MV-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP18]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP45:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP18]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP51:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
+; COMPARE-LAA-MV: [[SCALAR_PH]]:
+; COMPARE-LAA-MV-NEXT: br label %[[HEADER:.*]]
+; COMPARE-LAA-MV: [[HEADER]]:
+; COMPARE-LAA-MV-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[HEADER]] ]
+; COMPARE-LAA-MV-NEXT: [[IV_NEXT]] = add nsw i64 [[IV]], 1
+; COMPARE-LAA-MV-NEXT: [[IDX:%.*]] = mul i64 [[IV]], [[STRIDE]]
+; COMPARE-LAA-MV-NEXT: [[GEP_LD:%.*]] = getelementptr [256 x [256 x i64]], ptr [[P]], i64 1, i64 [[IDX]], i64 [[IDX]]
+; COMPARE-LAA-MV-NEXT: [[LD:%.*]] = load i64, ptr [[GEP_LD]], align 8
+; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
+; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
+; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP52:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -3642,7 +3710,7 @@ define void @sext_stride(ptr noalias %p.out, ptr %p, i32 %stride.i32) {
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[WIDE_LOAD]], ptr [[TMP1]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP46:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP53:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -3657,7 +3725,7 @@ define void @sext_stride(ptr noalias %p.out, ptr %p, i32 %stride.i32) {
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP47:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP54:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -3758,19 +3826,50 @@ define void @trunc_stride(ptr noalias %p.out, ptr %p, i64 %stride.i64) {
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; COMPARE-LAA-MV: [[VECTOR_SCEVCHECK]]:
-; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE_I64]], 1
+; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = trunc i64 [[STRIDE_I64]] to i32
+; COMPARE-LAA-MV-NEXT: [[TMP11:%.*]] = sub i32 0, [[TMP8]]
+; COMPARE-LAA-MV-NEXT: [[TMP12:%.*]] = icmp slt i32 [[TMP8]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = select i1 [[TMP12]], i32 [[TMP11]], i32 [[TMP8]]
+; COMPARE-LAA-MV-NEXT: [[MUL:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP3]], i32 127)
+; COMPARE-LAA-MV-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL]], 0
+; COMPARE-LAA-MV-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL]], 1
+; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = sub i32 0, [[MUL_RESULT]]
+; COMPARE-LAA-MV-NEXT: [[TMP5:%.*]] = icmp slt i32 [[MUL_RESULT]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = icmp sgt i32 [[TMP4]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP7:%.*]] = select i1 [[TMP12]], i1 [[TMP6]], i1 [[TMP5]]
+; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW]]
; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
+; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[STRIDE_I64]], i64 0
+; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
+; COMPARE-LAA-MV-NEXT: [[TMP9:%.*]] = trunc <4 x i64> [[BROADCAST_SPLAT]] to <4 x i32>
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[INDEX1:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[TMP10:%.*]] = mul <4 x i32> [[VEC_IND]], [[TMP9]]
+; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = extractelement <4 x i32> [[TMP10]], i64 0
; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = getelementptr i32, ptr [[P]], i32 [[INDEX]]
-; COMPARE-LAA-MV-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP0]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[P_OUT]], i32 [[INDEX]]
+; COMPARE-LAA-MV-NEXT: [[TMP13:%.*]] = extractelement <4 x i32> [[TMP10]], i64 1
+; COMPARE-LAA-MV-NEXT: [[TMP14:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP13]]
+; COMPARE-LAA-MV-NEXT: [[TMP15:%.*]] = extractelement <4 x i32> [[TMP10]], i64 2
+; COMPARE-LAA-MV-NEXT: [[TMP16:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP15]]
+; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = extractelement <4 x i32> [[TMP10]], i64 3
+; COMPARE-LAA-MV-NEXT: [[TMP18:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP17]]
+; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP0]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP20:%.*]] = load i32, ptr [[TMP14]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP16]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP18]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP23:%.*]] = insertelement <4 x i32> poison, i32 [[TMP19]], i32 0
+; COMPARE-LAA-MV-NEXT: [[TMP24:%.*]] = insertelement <4 x i32> [[TMP23]], i32 [[TMP20]], i32 1
+; COMPARE-LAA-MV-NEXT: [[TMP25:%.*]] = insertelement <4 x i32> [[TMP24]], i32 [[TMP21]], i32 2
+; COMPARE-LAA-MV-NEXT: [[WIDE_LOAD:%.*]] = insertelement <4 x i32> [[TMP25]], i32 [[TMP22]], i32 3
+; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[P_OUT]], i32 [[INDEX1]]
; COMPARE-LAA-MV-NEXT: store <4 x i32> [[WIDE_LOAD]], ptr [[TMP1]], align 8
-; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
+; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX1]], 4
+; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i32> [[VEC_IND]], splat (i32 4)
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP48:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP55:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -3785,7 +3884,7 @@ define void @trunc_stride(ptr noalias %p.out, ptr %p, i64 %stride.i64) {
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i32, ptr [[P_OUT]], i32 [[IV]]
; COMPARE-LAA-MV-NEXT: store i32 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i32 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP49:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP56:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -3916,8 +4015,20 @@ define void @trunc_ext_stride(ptr noalias %p.out, ptr %p0, ptr %p1, i32 %stride)
; COMPARE-LAA-MV-NEXT: [[STRIDE_EXT:%.*]] = sext i32 [[STRIDE]] to i64
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; COMPARE-LAA-MV: [[VECTOR_SCEVCHECK]]:
+; COMPARE-LAA-MV-NEXT: [[TMP9:%.*]] = sub i16 0, [[STRIDE_TRUNC]]
+; COMPARE-LAA-MV-NEXT: [[TMP10:%.*]] = icmp slt i16 [[STRIDE_TRUNC]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP11:%.*]] = select i1 [[TMP10]], i16 [[TMP9]], i16 [[STRIDE_TRUNC]]
+; COMPARE-LAA-MV-NEXT: [[MUL:%.*]] = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 [[TMP11]], i16 127)
+; COMPARE-LAA-MV-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i16, i1 } [[MUL]], 0
+; COMPARE-LAA-MV-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i16, i1 } [[MUL]], 1
+; COMPARE-LAA-MV-NEXT: [[TMP12:%.*]] = sub i16 0, [[MUL_RESULT]]
+; COMPARE-LAA-MV-NEXT: [[TMP13:%.*]] = icmp slt i16 [[MUL_RESULT]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP14:%.*]] = icmp sgt i16 [[TMP12]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP15:%.*]] = select i1 [[TMP10]], i1 [[TMP14]], i1 [[TMP13]]
+; COMPARE-LAA-MV-NEXT: [[TMP16:%.*]] = or i1 [[TMP15]], [[MUL_OVERFLOW]]
; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[STRIDE]], 1
-; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = or i1 [[TMP16]], [[IDENT_CHECK]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP8]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
@@ -3934,7 +4045,7 @@ define void @trunc_ext_stride(ptr noalias %p.out, ptr %p0, ptr %p1, i32 %stride)
; COMPARE-LAA-MV-NEXT: store <4 x i32> [[TMP5]], ptr [[TMP6]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP7:%.*]] = icmp eq i32 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP50:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP57:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -3954,7 +4065,7 @@ define void @trunc_ext_stride(ptr noalias %p.out, ptr %p0, ptr %p1, i32 %stride)
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i32, ptr [[P_OUT]], i32 [[IV]]
; COMPARE-LAA-MV-NEXT: store i32 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i32 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP51:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP58:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -4118,7 +4229,7 @@ define void @basic_masked(ptr noalias %p.out, ptr %p, i64 %stride, i64 %x) {
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP2]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP52:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP59:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -4137,7 +4248,7 @@ define void @basic_masked(ptr noalias %p.out, ptr %p, i64 %stride, i64 %x) {
; COMPARE-LAA-MV-NEXT: br label %[[LATCH]]
; COMPARE-LAA-MV: [[LATCH]]:
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP53:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP60:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -4238,7 +4349,7 @@ define void @stride_poison(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP18]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP54:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP18]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP61:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[EXIT]]:
@@ -4324,7 +4435,7 @@ define void @basic_strided_store(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: store <4 x i64> [[WIDE_LOAD]], ptr [[TMP1]], align 8
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP55:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP62:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -4338,7 +4449,7 @@ define void @basic_strided_store(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IDX]]
; COMPARE-LAA-MV-NEXT: store i64 [[LD]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP56:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP63:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -4427,7 +4538,7 @@ define void @ptr_vec_use(ptr noalias %p.out, ptr noalias %p.ptr.out, ptr %p, i64
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP57:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP64:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -4443,7 +4554,7 @@ define void @ptr_vec_use(ptr noalias %p.out, ptr noalias %p.ptr.out, ptr %p, i64
; COMPARE-LAA-MV-NEXT: [[GEP_PTR_ST:%.*]] = getelementptr ptr, ptr [[P_PTR_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store ptr [[GEP_LD]], ptr [[GEP_PTR_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP58:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP65:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -4535,7 +4646,7 @@ define void @stride_idx_vec_use(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP59:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP66:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -4550,7 +4661,7 @@ define void @stride_idx_vec_use(ptr noalias %p.out, ptr %p, i64 %stride) {
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP60:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP67:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -4643,7 +4754,7 @@ define void @offset_stride_idx_vec_use(ptr noalias %p.out, ptr %p, i64 %stride)
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i64> [[VEC_IND]], splat (i64 4)
; COMPARE-LAA-MV-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP61:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP68:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[EXIT:.*]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -4659,7 +4770,7 @@ define void @offset_stride_idx_vec_use(ptr noalias %p.out, ptr %p, i64 %stride)
; COMPARE-LAA-MV-NEXT: [[GEP_ST:%.*]] = getelementptr i64, ptr [[P_OUT]], i64 [[IV]]
; COMPARE-LAA-MV-NEXT: store i64 [[VAL]], ptr [[GEP_ST]], align 8
; COMPARE-LAA-MV-NEXT: [[EXITCOND:%.*]] = icmp slt i64 [[IV_NEXT]], 128
-; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP62:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EXITCOND]], label %[[HEADER]], label %[[EXIT]], !llvm.loop [[LOOP69:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
@@ -4822,7 +4933,7 @@ define void @test_rewrite_iv_scevs(i32 %start, ptr %dst) {
; COMPARE-LAA-MV-NEXT: store <4 x float> zeroinitializer, ptr [[TMP1]], align 4
; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = icmp eq i64 [[INDEX_NEXT]], 96
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP63:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP70:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
; COMPARE-LAA-MV-NEXT: br label %[[SCALAR_PH]]
; COMPARE-LAA-MV: [[SCALAR_PH]]:
@@ -4837,7 +4948,7 @@ define void @test_rewrite_iv_scevs(i32 %start, ptr %dst) {
; COMPARE-LAA-MV-NEXT: [[IV_1_NEXT]] = add i64 [[IV_1]], [[START_EXT]]
; COMPARE-LAA-MV-NEXT: [[IV_0_NEXT]] = add i64 [[IV_0]], 1
; COMPARE-LAA-MV-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_0_NEXT]], 100
-; COMPARE-LAA-MV-NEXT: br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP64:![0-9]+]]
+; COMPARE-LAA-MV-NEXT: br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP71:![0-9]+]]
; COMPARE-LAA-MV: [[EXIT]]:
; COMPARE-LAA-MV-NEXT: ret void
;
>From 19784dcc2a7355d0b0da00fb79284612ecf6579e Mon Sep 17 00:00:00 2001
From: Ramkumar Ramachandra <artagnon at tenstorrent.com>
Date: Wed, 29 Jul 2026 18:38:16 +0100
Subject: [PATCH 2/4] [LAA] Squelch regression; more improvements
---
.../Analysis/ScalarEvolutionPatternMatch.h | 6 ++
llvm/lib/Analysis/LoopAccessAnalysis.cpp | 15 ++--
.../LoopAccessAnalysis/symbolic-stride.ll | 18 ++---
llvm/test/Transforms/LoopVectorize/pr34681.ll | 65 +++++------------
.../runtime-check-needed-but-empty.ll | 35 ++++++++--
.../LoopVectorize/version-mem-access.ll | 70 ++++---------------
.../LoopVectorize/vplan-based-stride-mv.ll | 55 ++-------------
7 files changed, 88 insertions(+), 176 deletions(-)
diff --git a/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h b/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h
index aeeb88d722cf0..4202a4fd1d745 100644
--- a/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h
+++ b/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h
@@ -174,6 +174,12 @@ inline SCEVUnaryExpr_match<SCEVTy, Op0_t> m_scev_Unary(const Op0_t &Op0) {
return SCEVUnaryExpr_match<SCEVTy, Op0_t>(Op0);
}
+template <typename Op0_t>
+inline SCEVUnaryExpr_match<SCEVIntegralCastExpr, Op0_t>
+m_scev_IntegralCast(const Op0_t &Op0) {
+ return m_scev_Unary<SCEVIntegralCastExpr>(Op0);
+}
+
template <typename Op0_t>
inline SCEVUnaryExpr_match<SCEVSignExtendExpr, Op0_t>
m_scev_SExt(const Op0_t &Op0) {
diff --git a/llvm/lib/Analysis/LoopAccessAnalysis.cpp b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
index e53fc9df90227..b4ab97e390111 100644
--- a/llvm/lib/Analysis/LoopAccessAnalysis.cpp
+++ b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
@@ -3040,14 +3040,13 @@ static const SCEVUnknown *getStrideFromPointer(Value *Ptr, ScalarEvolution *SE,
assert(Ptr->getType()->isPointerTy() && "Pointer type expected");
const SCEV *V = SE->removePointerBase(SE->getSCEV(Ptr));
- while (auto *C = dyn_cast<SCEVIntegralCastExpr>(V))
- V = C->getOperand();
-
- match(V, m_scev_AffineAddRec(m_SCEV(), m_SCEV(V), m_SpecificLoop(Lp)));
- match(V, m_scev_Mul(m_SCEVConstant(), m_SCEV(V)));
-
- while (auto *C = dyn_cast<SCEVIntegralCastExpr>(V))
- V = C->getOperand();
+ bool Changed = true;
+ while (Changed) {
+ Changed = match(V, m_scev_AffineAddRec(m_SCEV(), m_SCEV(V),
+ m_SpecificLoop(Lp))) ||
+ match(V, m_scev_Mul(m_SCEVConstant(), m_SCEV(V))) ||
+ match(V, m_scev_IntegralCast(m_SCEV(V)));
+ }
// It wouldn't make sense to speculate on a loop-varying stride.
return SE->isLoopInvariant(V, Lp) ? dyn_cast<SCEVUnknown>(V) : nullptr;
diff --git a/llvm/test/Analysis/LoopAccessAnalysis/symbolic-stride.ll b/llvm/test/Analysis/LoopAccessAnalysis/symbolic-stride.ll
index 88cf23e412183..4b407c10820b4 100644
--- a/llvm/test/Analysis/LoopAccessAnalysis/symbolic-stride.ll
+++ b/llvm/test/Analysis/LoopAccessAnalysis/symbolic-stride.ll
@@ -521,17 +521,18 @@ define void @unknown_stride_equalto_zext_tc(i16 zeroext %N, ptr %A, ptr %B, i32
; CHECK-NEXT: (Low: %A High: (4 + %A))
; CHECK-NEXT: Member: %A
; CHECK-NEXT: Group GRP1:
-; CHECK-NEXT: (Low: (((2 * (sext i32 %j to i64))<nsw> + %B) umin ((2 * (sext i32 %j to i64))<nsw> + (2 * (zext i32 (-1 + (zext i16 %N to i32))<nsw> to i64) * (zext i16 %N to i64)) + %B)) High: (2 + (((2 * (sext i32 %j to i64))<nsw> + %B) umax ((2 * (sext i32 %j to i64))<nsw> + (2 * (zext i32 (-1 + (zext i16 %N to i32))<nsw> to i64) * (zext i16 %N to i64)) + %B))))
-; CHECK-NEXT: Member: {((2 * (sext i32 %j to i64))<nsw> + %B),+,(2 * (zext i16 %N to i64))<nuw><nsw>}<nw><%loop>
+; CHECK-NEXT: (Low: ((2 * (sext i32 %j to i64))<nsw> + %B) High: (2 + (2 * (zext i32 (-1 + (zext i16 %N to i32))<nsw> to i64))<nuw><nsw> + (2 * (sext i32 %j to i64))<nsw> + %B))
+; CHECK-NEXT: Member: {((2 * (sext i32 %j to i64))<nsw> + %B),+,2}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: {%j,+,(zext i16 %N to i32)}<nw><%loop> Added Flags: <nssw>
+; CHECK-NEXT: Equal predicate: %N == 1
+; CHECK-NEXT: {%j,+,1}<nw><%loop> Added Flags: <nssw>
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
; CHECK-NEXT: [PSE] %arrayidx = getelementptr inbounds i16, ptr %B, i32 %add:
; CHECK-NEXT: ((2 * (sext i32 {%j,+,(zext i16 %N to i32)}<nw><%loop> to i64))<nsw> + %B)
-; CHECK-NEXT: --> {((2 * (sext i32 %j to i64))<nsw> + %B),+,(2 * (zext i16 %N to i64))<nuw><nsw>}<nw><%loop>
+; CHECK-NEXT: --> {((2 * (sext i32 %j to i64))<nsw> + %B),+,2}<nw><%loop>
;
entry:
%N.ext = zext i16 %N to i32
@@ -623,17 +624,18 @@ define void @unknown_stride_equalto_trunc_tc(i64 %N, ptr %A, ptr %B, i32 %j) {
; CHECK-NEXT: (Low: %A High: (4 + %A))
; CHECK-NEXT: Member: %A
; CHECK-NEXT: Group GRP1:
-; CHECK-NEXT: (Low: (((2 * (sext i32 %j to i64))<nsw> + %B) umin ((2 * (sext i32 %j to i64))<nsw> + (2 * (zext i32 (-1 + (trunc i64 %N to i32)) to i64) * (sext i32 (trunc i64 %N to i32) to i64)) + %B)) High: (2 + (((2 * (sext i32 %j to i64))<nsw> + %B) umax ((2 * (sext i32 %j to i64))<nsw> + (2 * (zext i32 (-1 + (trunc i64 %N to i32)) to i64) * (sext i32 (trunc i64 %N to i32) to i64)) + %B))))
-; CHECK-NEXT: Member: {((2 * (sext i32 %j to i64))<nsw> + %B),+,(2 * (sext i32 (trunc i64 %N to i32) to i64))<nsw>}<%loop>
+; CHECK-NEXT: (Low: ((2 * (sext i32 %j to i64))<nsw> + %B) High: (2 + (2 * (zext i32 (-1 + (trunc i64 %N to i32)) to i64))<nuw><nsw> + (2 * (sext i32 %j to i64))<nsw> + %B))
+; CHECK-NEXT: Member: {((2 * (sext i32 %j to i64))<nsw> + %B),+,2}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: {%j,+,(trunc i64 %N to i32)}<nw><%loop> Added Flags: <nssw>
+; CHECK-NEXT: Equal predicate: %N == 1
+; CHECK-NEXT: {%j,+,1}<nw><%loop> Added Flags: <nssw>
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
; CHECK-NEXT: [PSE] %arrayidx = getelementptr inbounds i16, ptr %B, i32 %add:
; CHECK-NEXT: ((2 * (sext i32 {%j,+,(trunc i64 %N to i32)}<nw><%loop> to i64))<nsw> + %B)
-; CHECK-NEXT: --> {((2 * (sext i32 %j to i64))<nsw> + %B),+,(2 * (sext i32 (trunc i64 %N to i32) to i64))<nsw>}<%loop>
+; CHECK-NEXT: --> {((2 * (sext i32 %j to i64))<nsw> + %B),+,2}<nw><%loop>
;
entry:
%N.trunc = trunc i64 %N to i32
diff --git a/llvm/test/Transforms/LoopVectorize/pr34681.ll b/llvm/test/Transforms/LoopVectorize/pr34681.ll
index ce942fefa6e78..3ae164dd0fc83 100644
--- a/llvm/test/Transforms/LoopVectorize/pr34681.ll
+++ b/llvm/test/Transforms/LoopVectorize/pr34681.ll
@@ -136,63 +136,30 @@ for.end:
define i32 @foo2(i16 zeroext %N, ptr nocapture readnone %A, ptr nocapture readonly %B, i32 %i, i32 %j) {
; CHECK-LABEL: define i32 @foo2(
; CHECK-SAME: i16 zeroext [[N:%.*]], ptr readnone captures(none) [[A:%.*]], ptr readonly captures(none) [[B:%.*]], i32 [[I:%.*]], i32 [[J:%.*]]) {
-; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[ENTRY:.*]]:
; CHECK-NEXT: [[CONV:%.*]] = zext i16 [[N]] to i32
; CHECK-NEXT: [[CMP11:%.*]] = icmp eq i16 [[N]], 0
-; CHECK-NEXT: br i1 [[CMP11]], [[FOR_END:label %.*]], label %[[FOR_BODY_LR_PH:.*]]
-; CHECK: [[FOR_BODY_LR_PH]]:
-; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[CONV]], 4
-; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
-; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[TMP0:%.*]] = add nsw i32 [[CONV]], -1
-; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[CONV]], i32 [[TMP0]])
-; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
-; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
-; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[J]], [[MUL_RESULT]]
-; CHECK-NEXT: [[TMP3:%.*]] = icmp slt i32 [[TMP1]], [[J]]
-; CHECK-NEXT: [[TMP4:%.*]] = or i1 [[TMP3]], [[MUL_OVERFLOW]]
-; CHECK-NEXT: br i1 [[TMP4]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT: br i1 [[CMP11]], label %[[FOR_END:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
-; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[CONV]], 4
-; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[CONV]], [[N_MOD_VF]]
-; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[CONV]], i64 0
-; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
-; CHECK-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <4 x i32> poison, i32 [[J]], i64 0
-; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT2]], <4 x i32> poison, <4 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP26:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP7:%.*]] = mul nuw <4 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; CHECK-NEXT: [[TMP8:%.*]] = add <4 x i32> [[TMP7]], [[BROADCAST_SPLAT3]]
-; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i32> [[TMP8]], i64 0
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP9]]
-; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x i32> [[TMP8]], i64 1
-; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP11]]
-; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x i32> [[TMP8]], i64 2
-; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP13]]
-; CHECK-NEXT: [[TMP15:%.*]] = extractelement <4 x i32> [[TMP8]], i64 3
+; CHECK-NEXT: [[TMP_013:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[ADD4:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[K_012:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INC:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[MUL:%.*]] = mul nuw i32 [[K_012]], [[CONV]]
+; CHECK-NEXT: [[TMP15:%.*]] = add i32 [[MUL]], [[J]]
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP15]]
-; CHECK-NEXT: [[TMP17:%.*]] = load i16, ptr [[TMP10]], align 2
-; CHECK-NEXT: [[TMP18:%.*]] = load i16, ptr [[TMP12]], align 2
-; CHECK-NEXT: [[TMP19:%.*]] = load i16, ptr [[TMP14]], align 2
; CHECK-NEXT: [[TMP20:%.*]] = load i16, ptr [[TMP16]], align 2
-; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i16> poison, i16 [[TMP17]], i32 0
-; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x i16> [[TMP21]], i16 [[TMP18]], i32 1
-; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x i16> [[TMP22]], i16 [[TMP19]], i32 2
-; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x i16> [[TMP23]], i16 [[TMP20]], i32 3
-; CHECK-NEXT: [[TMP25:%.*]] = sext <4 x i16> [[TMP24]] to <4 x i32>
-; CHECK-NEXT: [[TMP26]] = add <4 x i32> [[VEC_PHI]], [[TMP25]]
-; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
-; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i32> [[VEC_IND]], splat (i32 4)
-; CHECK-NEXT: [[TMP27:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP27]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
-; CHECK: [[MIDDLE_BLOCK]]:
-; CHECK-NEXT: [[TMP28:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP26]])
-; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[CONV]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[CMP_N]], [[FOR_END_LOOPEXIT:label %.*]], label %[[SCALAR_PH]]
+; CHECK-NEXT: [[CONV3:%.*]] = sext i16 [[TMP20]] to i32
+; CHECK-NEXT: [[ADD4]] = add nsw i32 [[TMP_013]], [[CONV3]]
+; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[K_012]], 1
+; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INC]], [[CONV]]
+; CHECK-NEXT: br i1 [[EXITCOND]], label %[[SCALAR_PH:.*]], label %[[VECTOR_BODY]]
; CHECK: [[SCALAR_PH]]:
+; CHECK-NEXT: [[ADD4_LCSSA:%.*]] = phi i32 [ [[ADD4]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: br label %[[FOR_END]]
+; CHECK: [[FOR_END]]:
+; CHECK-NEXT: [[TMP_0_LCSSA:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[ADD4_LCSSA]], %[[SCALAR_PH]] ]
+; CHECK-NEXT: ret i32 [[TMP_0_LCSSA]]
;
entry:
%conv = zext i16 %N to i32
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll b/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
index 06e9554937a0e..0cb40f354cb4c 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
@@ -4,7 +4,30 @@
define void @test(ptr %A, i32 %x) {
; CHECK-LABEL: define void @test(
; CHECK-SAME: ptr [[A:%.*]], i32 [[X:%.*]]) {
-; CHECK-NEXT: [[SCALAR_PH:.*]]:
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
+; CHECK: [[VECTOR_SCEVCHECK]]:
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[X]], 1
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK: [[VECTOR_BODY]]:
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[OFFSET_IDX]], 1
+; CHECK-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
+; CHECK-NEXT: [[TMP5:%.*]] = zext i32 [[TMP4]] to i64
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP5]]
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP6]], align 4
+; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[OFFSET_IDX]] to i32
+; CHECK-NEXT: [[TMP8:%.*]] = zext i32 [[TMP0]] to i64
+; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP8]]
+; CHECK-NEXT: store <4 x float> [[WIDE_LOAD]], ptr [[TMP9]], align 4
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[OFFSET_IDX]], 4
+; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000
+; CHECK-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: br label %[[EXIT:.*]]
+; CHECK: [[SCALAR_PH]]:
; CHECK-NEXT: br label %[[LOOP:.*]]
; CHECK: [[LOOP]]:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], %[[LOOP]] ], [ 0, %[[SCALAR_PH]] ]
@@ -20,7 +43,7 @@ define void @test(ptr %A, i32 %x) {
; CHECK-NEXT: [[ARRAYIDX1209:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IDX_2]]
; CHECK-NEXT: store float [[LV]], ptr [[ARRAYIDX1209]], align 4
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], 1000
-; CHECK-NEXT: br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]]
+; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -79,7 +102,7 @@ define void @diff_memcheck_known_false_for_vf_4(ptr %B, ptr %A, ptr %end) {
; CHECK-NEXT: store <4 x i64> zeroinitializer, ptr [[TMP8]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
@@ -95,7 +118,7 @@ define void @diff_memcheck_known_false_for_vf_4(ptr %B, ptr %A, ptr %end) {
; CHECK-NEXT: [[TMP10:%.*]] = load i64, ptr [[IV_2_NEXT]], align 8
; CHECK-NEXT: store i64 0, ptr [[IV_1]], align 8
; CHECK-NEXT: [[CMP_NOT_I_I_I_I:%.*]] = icmp eq ptr [[END]], [[IV_2]]
-; CHECK-NEXT: br i1 [[CMP_NOT_I_I_I_I]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK-NEXT: br i1 [[CMP_NOT_I_I_I_I]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -123,5 +146,7 @@ exit:
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
-; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
+; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]}
+; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
+; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META2]], [[META1]]}
;.
diff --git a/llvm/test/Transforms/LoopVectorize/version-mem-access.ll b/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
index c533c95a444f1..1d7d99d691e26 100644
--- a/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
+++ b/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
@@ -13,45 +13,26 @@ define void @test(ptr noalias %A, i64 %AStride, ptr noalias %B, i32 %BStride, pt
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[TMP0:%.*]] = sub i32 0, [[BSTRIDE]]
-; CHECK-NEXT: [[TMP1:%.*]] = icmp slt i32 [[BSTRIDE]], 0
-; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i32 [[TMP0]], i32 [[BSTRIDE]]
-; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP2]], i32 99)
-; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
-; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
-; CHECK-NEXT: [[TMP3:%.*]] = sub i32 0, [[MUL_RESULT]]
-; CHECK-NEXT: [[TMP4:%.*]] = icmp ugt i32 [[TMP3]], 0
-; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP1]], i1 [[TMP4]], i1 false
-; CHECK-NEXT: [[IDENT_CHECK1:%.*]] = or i1 [[TMP11]], [[MUL_OVERFLOW]]
+; CHECK-NEXT: [[IDENT_CHECK1:%.*]] = icmp ne i32 [[BSTRIDE]], 1
; CHECK-NEXT: [[IDENT_CHECK2:%.*]] = icmp ne i64 [[CSTRIDE]], 1
; CHECK-NEXT: [[DIFF_CHECK6:%.*]] = icmp ne i64 [[ASTRIDE]], 1
; CHECK-NEXT: [[DIFF_CHECK:%.*]] = or i1 [[IDENT_CHECK1]], [[IDENT_CHECK2]]
; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[DIFF_CHECK]], [[DIFF_CHECK6]]
; CHECK-NEXT: br i1 [[CONFLICT_RDX]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
-; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[BSTRIDE]], i64 0
-; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP12:%.*]] = mul <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; CHECK-NEXT: [[TMP18:%.*]] = zext <2 x i32> [[TMP12]] to <2 x i64>
-; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i64> [[TMP18]], i64 0
-; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP5]]
-; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i64> [[TMP18]], i64 1
+; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[INDEX]] to i32
+; CHECK-NEXT: [[TMP13:%.*]] = zext i32 [[TMP2]] to i64
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP13]]
-; CHECK-NEXT: [[TMP15:%.*]] = load i32, ptr [[TMP6]], align 4
-; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP14]], align 4
-; CHECK-NEXT: [[TMP17:%.*]] = insertelement <2 x i32> poison, i32 [[TMP15]], i32 0
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = insertelement <2 x i32> [[TMP17]], i32 [[TMP16]], i32 1
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP14]], align 4
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[INDEX]]
; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <2 x i32>, ptr [[TMP7]], align 4
; CHECK-NEXT: [[TMP8:%.*]] = mul nsw <2 x i32> [[WIDE_LOAD7]], [[WIDE_LOAD]]
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT: store <2 x i32> [[TMP8]], ptr [[TMP9]], align 4
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
-; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2)
; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
; CHECK-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
@@ -97,50 +78,25 @@ define void @fn1(ptr noalias %x, ptr noalias %c, double %a) {
; CHECK-NEXT: br i1 [[CMP8]], label %[[LOOP_PREHEADER:.*]], [[EXIT:label %.*]]
; CHECK: [[LOOP_PREHEADER]]:
; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[CONV2]] to i64
-; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 2
-; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
+; CHECK-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[CONV]], 3
-; CHECK-NEXT: [[TMP9:%.*]] = sub i32 0, [[CONV]]
-; CHECK-NEXT: [[TMP12:%.*]] = icmp slt i32 [[CONV]], 0
-; CHECK-NEXT: [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP9]], i32 [[CONV]]
-; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP13]], i32 [[TMP1]])
-; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
-; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
-; CHECK-NEXT: [[TMP5:%.*]] = sub i32 0, [[MUL_RESULT]]
-; CHECK-NEXT: [[TMP6:%.*]] = icmp slt i32 [[MUL_RESULT]], 0
-; CHECK-NEXT: [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], 0
-; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP12]], i1 [[TMP7]], i1 [[TMP6]]
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = or i1 [[TMP8]], [[MUL_OVERFLOW]]
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[CONV]], 1
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
-; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 2
-; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
-; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[CONV]], i64 0
-; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP10:%.*]] = mul nsw <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; CHECK-NEXT: [[TMP11:%.*]] = sext <2 x i32> [[TMP10]] to <2 x i64>
-; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i64> [[TMP11]], i64 0
-; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds double, ptr [[X]], i64 [[TMP2]]
-; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i64> [[TMP11]], i64 1
+; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDEX]] to i32
+; CHECK-NEXT: [[TMP14:%.*]] = sext i32 [[TMP1]] to i64
; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds double, ptr [[X]], i64 [[TMP14]]
-; CHECK-NEXT: [[TMP16:%.*]] = load double, ptr [[TMP3]], align 8
-; CHECK-NEXT: [[TMP17:%.*]] = load double, ptr [[TMP15]], align 8
-; CHECK-NEXT: [[TMP18:%.*]] = insertelement <2 x double> poison, double [[TMP16]], i32 0
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = insertelement <2 x double> [[TMP18]], double [[TMP17]], i32 1
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x double>, ptr [[TMP15]], align 8
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds double, ptr [[C]], i64 [[INDEX]]
; CHECK-NEXT: store <2 x double> [[WIDE_LOAD]], ptr [[TMP4]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
-; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2)
-; CHECK-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4
+; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
-; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[CMP_N]], [[EXIT_LOOPEXIT:label %.*]], label %[[SCALAR_PH]]
+; CHECK-NEXT: br label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
;
entry:
diff --git a/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll b/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
index 5a7588d04793b..d38b56bc62f42 100644
--- a/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
+++ b/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
@@ -3826,48 +3826,17 @@ define void @trunc_stride(ptr noalias %p.out, ptr %p, i64 %stride.i64) {
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; COMPARE-LAA-MV: [[VECTOR_SCEVCHECK]]:
-; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = trunc i64 [[STRIDE_I64]] to i32
-; COMPARE-LAA-MV-NEXT: [[TMP11:%.*]] = sub i32 0, [[TMP8]]
-; COMPARE-LAA-MV-NEXT: [[TMP12:%.*]] = icmp slt i32 [[TMP8]], 0
-; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = select i1 [[TMP12]], i32 [[TMP11]], i32 [[TMP8]]
-; COMPARE-LAA-MV-NEXT: [[MUL:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP3]], i32 127)
-; COMPARE-LAA-MV-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL]], 0
-; COMPARE-LAA-MV-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL]], 1
-; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = sub i32 0, [[MUL_RESULT]]
-; COMPARE-LAA-MV-NEXT: [[TMP5:%.*]] = icmp slt i32 [[MUL_RESULT]], 0
-; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = icmp sgt i32 [[TMP4]], 0
-; COMPARE-LAA-MV-NEXT: [[TMP7:%.*]] = select i1 [[TMP12]], i1 [[TMP6]], i1 [[TMP5]]
-; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW]]
+; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE_I64]], 1
; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[STRIDE_I64]], i64 0
-; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
-; COMPARE-LAA-MV-NEXT: [[TMP9:%.*]] = trunc <4 x i64> [[BROADCAST_SPLAT]] to <4 x i32>
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[INDEX1:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; COMPARE-LAA-MV-NEXT: [[TMP10:%.*]] = mul <4 x i32> [[VEC_IND]], [[TMP9]]
-; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = extractelement <4 x i32> [[TMP10]], i64 0
-; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = getelementptr i32, ptr [[P]], i32 [[INDEX]]
-; COMPARE-LAA-MV-NEXT: [[TMP13:%.*]] = extractelement <4 x i32> [[TMP10]], i64 1
-; COMPARE-LAA-MV-NEXT: [[TMP14:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP13]]
-; COMPARE-LAA-MV-NEXT: [[TMP15:%.*]] = extractelement <4 x i32> [[TMP10]], i64 2
-; COMPARE-LAA-MV-NEXT: [[TMP16:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP15]]
-; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = extractelement <4 x i32> [[TMP10]], i64 3
+; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; COMPARE-LAA-MV-NEXT: [[TMP18:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP17]]
-; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP0]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP20:%.*]] = load i32, ptr [[TMP14]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP16]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP18]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP23:%.*]] = insertelement <4 x i32> poison, i32 [[TMP19]], i32 0
-; COMPARE-LAA-MV-NEXT: [[TMP24:%.*]] = insertelement <4 x i32> [[TMP23]], i32 [[TMP20]], i32 1
-; COMPARE-LAA-MV-NEXT: [[TMP25:%.*]] = insertelement <4 x i32> [[TMP24]], i32 [[TMP21]], i32 2
-; COMPARE-LAA-MV-NEXT: [[WIDE_LOAD:%.*]] = insertelement <4 x i32> [[TMP25]], i32 [[TMP22]], i32 3
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[P_OUT]], i32 [[INDEX1]]
+; COMPARE-LAA-MV-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP18]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[P_OUT]], i32 [[TMP17]]
; COMPARE-LAA-MV-NEXT: store <4 x i32> [[WIDE_LOAD]], ptr [[TMP1]], align 8
-; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX1]], 4
-; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i32> [[VEC_IND]], splat (i32 4)
+; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i32 [[TMP17]], 4
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 128
; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP55:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
@@ -4015,20 +3984,8 @@ define void @trunc_ext_stride(ptr noalias %p.out, ptr %p0, ptr %p1, i32 %stride)
; COMPARE-LAA-MV-NEXT: [[STRIDE_EXT:%.*]] = sext i32 [[STRIDE]] to i64
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; COMPARE-LAA-MV: [[VECTOR_SCEVCHECK]]:
-; COMPARE-LAA-MV-NEXT: [[TMP9:%.*]] = sub i16 0, [[STRIDE_TRUNC]]
-; COMPARE-LAA-MV-NEXT: [[TMP10:%.*]] = icmp slt i16 [[STRIDE_TRUNC]], 0
-; COMPARE-LAA-MV-NEXT: [[TMP11:%.*]] = select i1 [[TMP10]], i16 [[TMP9]], i16 [[STRIDE_TRUNC]]
-; COMPARE-LAA-MV-NEXT: [[MUL:%.*]] = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 [[TMP11]], i16 127)
-; COMPARE-LAA-MV-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i16, i1 } [[MUL]], 0
-; COMPARE-LAA-MV-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i16, i1 } [[MUL]], 1
-; COMPARE-LAA-MV-NEXT: [[TMP12:%.*]] = sub i16 0, [[MUL_RESULT]]
-; COMPARE-LAA-MV-NEXT: [[TMP13:%.*]] = icmp slt i16 [[MUL_RESULT]], 0
-; COMPARE-LAA-MV-NEXT: [[TMP14:%.*]] = icmp sgt i16 [[TMP12]], 0
-; COMPARE-LAA-MV-NEXT: [[TMP15:%.*]] = select i1 [[TMP10]], i1 [[TMP14]], i1 [[TMP13]]
-; COMPARE-LAA-MV-NEXT: [[TMP16:%.*]] = or i1 [[TMP15]], [[MUL_OVERFLOW]]
; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[STRIDE]], 1
-; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = or i1 [[TMP16]], [[IDENT_CHECK]]
-; COMPARE-LAA-MV-NEXT: br i1 [[TMP8]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
>From 3d50e686988856f2afb5f55371bc976e95555fe5 Mon Sep 17 00:00:00 2001
From: Ramkumar Ramachandra <artagnon at tenstorrent.com>
Date: Thu, 30 Jul 2026 12:14:53 +0100
Subject: [PATCH 3/4] [LAA] Fix thinko; remove stray predicates
---
llvm/lib/Analysis/LoopAccessAnalysis.cpp | 9 +--
.../no-dep-via-loop-guards.ll | 33 ++-------
.../LoopAccessAnalysis/symbolic-stride.ll | 18 ++---
.../partial-reduce-replicate-extends.ll | 20 ++---
.../RISCV/blocks-with-dead-instructions.ll | 26 ++++---
.../truncate-to-minimal-bitwidth-cost.ll | 54 +++++---------
llvm/test/Transforms/LoopVectorize/pr34681.ll | 65 +++++++++++++----
.../pr59319-loop-access-info-invalidation.ll | 37 +++++-----
.../runtime-check-needed-but-empty.ll | 35 ++-------
.../LoopVectorize/runtime-checks-hoist.ll | 8 +-
.../LoopVectorize/vector-no-scevcheck.ll | 73 +++----------------
.../LoopVectorize/version-mem-access.ll | 70 ++++++++++++++----
.../LoopVectorize/vplan-based-stride-mv.ll | 55 ++++++++++++--
13 files changed, 250 insertions(+), 253 deletions(-)
diff --git a/llvm/lib/Analysis/LoopAccessAnalysis.cpp b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
index b4ab97e390111..ce0b5b3e63c20 100644
--- a/llvm/lib/Analysis/LoopAccessAnalysis.cpp
+++ b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
@@ -3044,8 +3044,9 @@ static const SCEVUnknown *getStrideFromPointer(Value *Ptr, ScalarEvolution *SE,
while (Changed) {
Changed = match(V, m_scev_AffineAddRec(m_SCEV(), m_SCEV(V),
m_SpecificLoop(Lp))) ||
- match(V, m_scev_Mul(m_SCEVConstant(), m_SCEV(V))) ||
match(V, m_scev_IntegralCast(m_SCEV(V)));
+ if (Changed)
+ match(V, m_scev_Mul(m_SCEVConstant(), m_SCEV(V)));
}
// It wouldn't make sense to speculate on a loop-varying stride.
@@ -3065,10 +3066,7 @@ void LoopAccessInfo::collectStridedAccess(Value *MemAccess) {
// far too many unprofitable cases.
const SCEVUnknown *StrideExpr =
getStrideFromPointer(Ptr, PSE->getSE(), TheLoop);
- if (!StrideExpr)
- return;
-
- if (match(StrideExpr, m_scev_UndefOrPoison()))
+ if (!StrideExpr || match(StrideExpr, m_scev_UndefOrPoison()))
return;
LLVM_DEBUG(dbgs() << "LAA: Found a strided access that is a candidate for "
@@ -3120,7 +3118,6 @@ void LoopAccessInfo::collectStridedAccess(Value *MemAccess) {
return;
}
LLVM_DEBUG(dbgs() << "LAA: Found a strided access that we can version.\n");
-
SymbolicStrides[Ptr] = StrideExpr;
}
diff --git a/llvm/test/Analysis/LoopAccessAnalysis/no-dep-via-loop-guards.ll b/llvm/test/Analysis/LoopAccessAnalysis/no-dep-via-loop-guards.ll
index 943c9452de3b0..e00f91d9b2e8b 100644
--- a/llvm/test/Analysis/LoopAccessAnalysis/no-dep-via-loop-guards.ll
+++ b/llvm/test/Analysis/LoopAccessAnalysis/no-dep-via-loop-guards.ll
@@ -5,19 +5,13 @@
define void @access_after_via_loop_guard(ptr %a, i64 %off) {
; CHECK-LABEL: 'access_after_via_loop_guard'
; CHECK-NEXT: loop:
-; CHECK-NEXT: Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
-; CHECK-NEXT: Unknown data dependence.
+; CHECK-NEXT: Memory dependences are safe
; CHECK-NEXT: Dependences:
-; CHECK-NEXT: Unknown:
-; CHECK-NEXT: %l = load i32, ptr %gep.after, align 4 ->
-; CHECK-NEXT: store i32 %add, ptr %gep, align 4
-; CHECK-EMPTY:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Grouped accesses:
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: Equal predicate: %off == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
@@ -59,7 +53,6 @@ define void @access_after_via_loop_guard_sge(ptr %a, i64 %off) {
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: Equal predicate: %off == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
@@ -99,7 +92,6 @@ define void @access_after_via_loop_guard_99(ptr %a, i64 %off) {
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: Equal predicate: %off == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
@@ -141,7 +133,6 @@ define void @access_after_via_loop_guard_sge_99(ptr %a, i64 %off) {
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: Equal predicate: %off == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
@@ -181,7 +172,6 @@ define void @access_after_via_loop_guard_uge(ptr %a, i64 %off) {
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: Equal predicate: %off == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
;
@@ -364,22 +354,20 @@ exit:
define void @nodep_via_logical_and_1(ptr %A, i32 %index, i32 %n) {
; CHECK-LABEL: 'nodep_via_logical_and_1'
; CHECK-NEXT: loop:
-; CHECK-NEXT: Memory dependences are safe
+; CHECK-NEXT: Report: unsafe dependent memory operations in loop. Use #pragma clang loop distribute(enable) to allow loop distribution to attempt to isolate the offending operations into a separate loop
+; CHECK-NEXT: Unknown data dependence.
; CHECK-NEXT: Dependences:
+; CHECK-NEXT: Unknown:
+; CHECK-NEXT: %0 = load double, ptr %gep.load, align 8 ->
+; CHECK-NEXT: store double %0, ptr %gep.store, align 8
+; CHECK-EMPTY:
; CHECK-NEXT: Run-time memory checks:
; CHECK-NEXT: Grouped accesses:
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: Equal predicate: %index == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
-; CHECK-NEXT: [PSE] %gep.load = getelementptr double, ptr %A, i64 %iv:
-; CHECK-NEXT: {((8 * (zext i32 (1 + %index) to i64))<nuw><nsw> + %A),+,8}<nw><%loop>
-; CHECK-NEXT: --> {(16 + %A),+,8}<nw><%loop>
-; CHECK-NEXT: [PSE] %gep.store = getelementptr double, ptr %A, i64 %index.ext:
-; CHECK-NEXT: ((8 * (zext i32 %index to i64))<nuw><nsw> + %A)
-; CHECK-NEXT: --> (8 + %A)
;
entry:
%pre.0 = icmp sgt i32 %index, 0
@@ -419,15 +407,8 @@ define void @nodep_via_logical_and_2(ptr %A, i32 %index, i32 %n) {
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: Equal predicate: %index == 1
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
-; CHECK-NEXT: [PSE] %gep.load = getelementptr double, ptr %A, i64 %iv:
-; CHECK-NEXT: {((8 * (zext i32 (1 + %index) to i64))<nuw><nsw> + %A),+,8}<nw><%loop>
-; CHECK-NEXT: --> {(16 + %A),+,8}<nw><%loop>
-; CHECK-NEXT: [PSE] %gep.store = getelementptr double, ptr %A, i64 %index.ext:
-; CHECK-NEXT: ((8 * (zext i32 %index to i64))<nuw><nsw> + %A)
-; CHECK-NEXT: --> (8 + %A)
;
entry:
%pre.0 = icmp sgt i32 %index, 0
diff --git a/llvm/test/Analysis/LoopAccessAnalysis/symbolic-stride.ll b/llvm/test/Analysis/LoopAccessAnalysis/symbolic-stride.ll
index 4b407c10820b4..88cf23e412183 100644
--- a/llvm/test/Analysis/LoopAccessAnalysis/symbolic-stride.ll
+++ b/llvm/test/Analysis/LoopAccessAnalysis/symbolic-stride.ll
@@ -521,18 +521,17 @@ define void @unknown_stride_equalto_zext_tc(i16 zeroext %N, ptr %A, ptr %B, i32
; CHECK-NEXT: (Low: %A High: (4 + %A))
; CHECK-NEXT: Member: %A
; CHECK-NEXT: Group GRP1:
-; CHECK-NEXT: (Low: ((2 * (sext i32 %j to i64))<nsw> + %B) High: (2 + (2 * (zext i32 (-1 + (zext i16 %N to i32))<nsw> to i64))<nuw><nsw> + (2 * (sext i32 %j to i64))<nsw> + %B))
-; CHECK-NEXT: Member: {((2 * (sext i32 %j to i64))<nsw> + %B),+,2}<nw><%loop>
+; CHECK-NEXT: (Low: (((2 * (sext i32 %j to i64))<nsw> + %B) umin ((2 * (sext i32 %j to i64))<nsw> + (2 * (zext i32 (-1 + (zext i16 %N to i32))<nsw> to i64) * (zext i16 %N to i64)) + %B)) High: (2 + (((2 * (sext i32 %j to i64))<nsw> + %B) umax ((2 * (sext i32 %j to i64))<nsw> + (2 * (zext i32 (-1 + (zext i16 %N to i32))<nsw> to i64) * (zext i16 %N to i64)) + %B))))
+; CHECK-NEXT: Member: {((2 * (sext i32 %j to i64))<nsw> + %B),+,(2 * (zext i16 %N to i64))<nuw><nsw>}<nw><%loop>
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: Equal predicate: %N == 1
-; CHECK-NEXT: {%j,+,1}<nw><%loop> Added Flags: <nssw>
+; CHECK-NEXT: {%j,+,(zext i16 %N to i32)}<nw><%loop> Added Flags: <nssw>
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
; CHECK-NEXT: [PSE] %arrayidx = getelementptr inbounds i16, ptr %B, i32 %add:
; CHECK-NEXT: ((2 * (sext i32 {%j,+,(zext i16 %N to i32)}<nw><%loop> to i64))<nsw> + %B)
-; CHECK-NEXT: --> {((2 * (sext i32 %j to i64))<nsw> + %B),+,2}<nw><%loop>
+; CHECK-NEXT: --> {((2 * (sext i32 %j to i64))<nsw> + %B),+,(2 * (zext i16 %N to i64))<nuw><nsw>}<nw><%loop>
;
entry:
%N.ext = zext i16 %N to i32
@@ -624,18 +623,17 @@ define void @unknown_stride_equalto_trunc_tc(i64 %N, ptr %A, ptr %B, i32 %j) {
; CHECK-NEXT: (Low: %A High: (4 + %A))
; CHECK-NEXT: Member: %A
; CHECK-NEXT: Group GRP1:
-; CHECK-NEXT: (Low: ((2 * (sext i32 %j to i64))<nsw> + %B) High: (2 + (2 * (zext i32 (-1 + (trunc i64 %N to i32)) to i64))<nuw><nsw> + (2 * (sext i32 %j to i64))<nsw> + %B))
-; CHECK-NEXT: Member: {((2 * (sext i32 %j to i64))<nsw> + %B),+,2}<nw><%loop>
+; CHECK-NEXT: (Low: (((2 * (sext i32 %j to i64))<nsw> + %B) umin ((2 * (sext i32 %j to i64))<nsw> + (2 * (zext i32 (-1 + (trunc i64 %N to i32)) to i64) * (sext i32 (trunc i64 %N to i32) to i64)) + %B)) High: (2 + (((2 * (sext i32 %j to i64))<nsw> + %B) umax ((2 * (sext i32 %j to i64))<nsw> + (2 * (zext i32 (-1 + (trunc i64 %N to i32)) to i64) * (sext i32 (trunc i64 %N to i32) to i64)) + %B))))
+; CHECK-NEXT: Member: {((2 * (sext i32 %j to i64))<nsw> + %B),+,(2 * (sext i32 (trunc i64 %N to i32) to i64))<nsw>}<%loop>
; CHECK-EMPTY:
; CHECK-NEXT: Non vectorizable stores to invariant address were not found in loop.
; CHECK-NEXT: SCEV assumptions:
-; CHECK-NEXT: Equal predicate: %N == 1
-; CHECK-NEXT: {%j,+,1}<nw><%loop> Added Flags: <nssw>
+; CHECK-NEXT: {%j,+,(trunc i64 %N to i32)}<nw><%loop> Added Flags: <nssw>
; CHECK-EMPTY:
; CHECK-NEXT: Expressions re-written:
; CHECK-NEXT: [PSE] %arrayidx = getelementptr inbounds i16, ptr %B, i32 %add:
; CHECK-NEXT: ((2 * (sext i32 {%j,+,(trunc i64 %N to i32)}<nw><%loop> to i64))<nsw> + %B)
-; CHECK-NEXT: --> {((2 * (sext i32 %j to i64))<nsw> + %B),+,2}<nw><%loop>
+; CHECK-NEXT: --> {((2 * (sext i32 %j to i64))<nsw> + %B),+,(2 * (sext i32 (trunc i64 %N to i32) to i64))<nsw>}<%loop>
;
entry:
%N.trunc = trunc i64 %N to i32
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-replicate-extends.ll b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-replicate-extends.ll
index a9619383cefe7..2684e33cf051c 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-replicate-extends.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-replicate-extends.ll
@@ -7,20 +7,20 @@ target triple = "aarch64"
define double @test(i32 %0) {
; CHECK-LABEL: define double @test(
; CHECK-SAME: i32 [[TMP0:%.*]]) {
-; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: br label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
-; CHECK-NEXT: [[REDUX:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[BINOP:%.*]], %[[VECTOR_PH]] ]
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[NEXT_IV:%.*]], %[[VECTOR_PH]] ]
; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
-; CHECK-NEXT: [[BINOP]] = or i64 [[REDUX]], [[TMP1]]
-; CHECK-NEXT: [[GEP:%.*]] = getelementptr [8 x i8], ptr null, i64 [[TMP1]]
-; CHECK-NEXT: [[LOAD:%.*]] = load double, ptr [[GEP]], align 8
-; CHECK-NEXT: [[NEXT_IV]] = add i64 [[IV]], 1
-; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV]], 1
-; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT:.*]], label %[[VECTOR_PH]]
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[TMP1]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK: [[VECTOR_BODY]]:
+; CHECK-NEXT: [[TMP2:%.*]] = or <2 x i64> zeroinitializer, [[BROADCAST_SPLAT]]
+; CHECK-NEXT: br label %[[MIDDLE_BLOCK:.*]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: [[TMP3:%.*]] = call i64 @llvm.vector.reduce.or.v2i64(<2 x i64> [[TMP2]])
+; CHECK-NEXT: br label %[[EXIT:.*]]
; CHECK: [[EXIT]]:
-; CHECK-NEXT: [[TMP3:%.*]] = phi i64 [ [[BINOP]], %[[VECTOR_PH]] ]
; CHECK-NEXT: [[RET:%.*]] = uitofp i64 [[TMP3]] to double
; CHECK-NEXT: ret double [[RET]]
;
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll b/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll
index 37d6a9414382c..28b418b5c0cdf 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/blocks-with-dead-instructions.ll
@@ -485,9 +485,6 @@ define void @dead_load_in_block(ptr %dst, ptr %src, i8 %N, i64 %x) #0 {
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[TMP2]], 1
; CHECK-NEXT: br label %[[VECTOR_MEMCHECK:.*]]
; CHECK: [[VECTOR_MEMCHECK]]:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[X]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK1:.*]]
-; CHECK: [[VECTOR_MEMCHECK1]]:
; CHECK-NEXT: [[UMIN:%.*]] = call i64 @llvm.umin.i64(i64 [[N_EXT]], i64 1)
; CHECK-NEXT: [[TMP7:%.*]] = sub i64 [[N_EXT]], [[UMIN]]
; CHECK-NEXT: [[TMP8:%.*]] = udiv i64 [[TMP7]], 3
@@ -495,11 +492,19 @@ define void @dead_load_in_block(ptr %dst, ptr %src, i8 %N, i64 %x) #0 {
; CHECK-NEXT: [[TMP10:%.*]] = mul i64 [[TMP9]], 12
; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[TMP10]], 4
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP11]]
-; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[SRC]], i64 8
+; CHECK-NEXT: [[TMP12:%.*]] = shl i64 [[X]], 2
+; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP12]]
+; CHECK-NEXT: [[TMP13:%.*]] = add i64 [[TMP12]], 4
+; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP13]]
+; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[SRC]], i64 4
+; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[DST]], [[SCEVGEP2]]
+; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP1]], [[SCEVGEP]]
+; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: [[BOUND04:%.*]] = icmp ult ptr [[DST]], [[SCEVGEP3]]
; CHECK-NEXT: [[BOUND15:%.*]] = icmp ult ptr [[SRC]], [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT6:%.*]] = and i1 [[BOUND04]], [[BOUND15]]
-; CHECK-NEXT: br i1 [[FOUND_CONFLICT6]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT6]]
+; CHECK-NEXT: br i1 [[CONFLICT_RDX]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP24:%.*]] = call <vscale x 4 x i64> @llvm.stepvector.nxv4i64()
; CHECK-NEXT: [[TMP25:%.*]] = mul <vscale x 4 x i64> [[TMP24]], splat (i64 3)
@@ -517,7 +522,7 @@ define void @dead_load_in_block(ptr %dst, ptr %src, i8 %N, i64 %x) #0 {
; CHECK-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP17]]
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
-; CHECK-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: br label %[[EXIT:.*]]
; CHECK: [[SCALAR_PH]]:
@@ -536,7 +541,7 @@ define void @dead_load_in_block(ptr %dst, ptr %src, i8 %N, i64 %x) #0 {
; CHECK-NEXT: store i32 0, ptr [[GEP_DST]], align 4
; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 3
; CHECK-NEXT: [[CMP:%.*]] = icmp ult i64 [[IV]], [[N_EXT]]
-; CHECK-NEXT: br i1 [[CMP]], label %[[LOOP_HEADER]], label %[[EXIT]], !llvm.loop [[LOOP16:![0-9]+]]
+; CHECK-NEXT: br i1 [[CMP]], label %[[LOOP_HEADER]], label %[[EXIT]], !llvm.loop [[LOOP17:![0-9]+]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -581,8 +586,9 @@ attributes #0 = { "target-features"="+64bit,+v" }
; CHECK: [[META10]] = !{[[META11:![0-9]+]]}
; CHECK: [[META11]] = distinct !{[[META11]], [[META12:![0-9]+]]}
; CHECK: [[META12]] = distinct !{[[META12]], !"LVerDomain"}
-; CHECK: [[META13]] = !{[[META14:![0-9]+]]}
+; CHECK: [[META13]] = !{[[META14:![0-9]+]], [[META15:![0-9]+]]}
; CHECK: [[META14]] = distinct !{[[META14]], [[META12]]}
-; CHECK: [[LOOP15]] = distinct !{[[LOOP15]], [[META1]], [[META2]]}
-; CHECK: [[LOOP16]] = distinct !{[[LOOP16]], [[META1]]}
+; CHECK: [[META15]] = distinct !{[[META15]], [[META12]]}
+; CHECK: [[LOOP16]] = distinct !{[[LOOP16]], [[META1]], [[META2]]}
+; CHECK: [[LOOP17]] = distinct !{[[LOOP17]], [[META1]]}
;.
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/truncate-to-minimal-bitwidth-cost.ll b/llvm/test/Transforms/LoopVectorize/RISCV/truncate-to-minimal-bitwidth-cost.ll
index ef28ef73f09fd..6b9e3d49951a7 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/truncate-to-minimal-bitwidth-cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/truncate-to-minimal-bitwidth-cost.ll
@@ -193,23 +193,21 @@ define void @icmp_only_first_op_truncated(ptr noalias %dst, i32 %x, i64 %N, i64
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[V]], 1
; CHECK-NEXT: br label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[X]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH1:.*]]
-; CHECK: [[VECTOR_PH1]]:
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[N]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <vscale x 2 x i32> poison, i32 [[T]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <vscale x 2 x i32> [[BROADCAST_SPLATINSERT1]], <vscale x 2 x i32> poison, <vscale x 2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP7:%.*]] = trunc <vscale x 2 x i64> [[BROADCAST_SPLAT]] to <vscale x 2 x i32>
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq <vscale x 2 x i32> [[TMP7]], [[BROADCAST_SPLAT2]]
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr double, ptr [[SRC]], i64 1
+; CHECK-NEXT: [[TMP9:%.*]] = zext i32 [[X]] to i64
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr double, ptr [[SRC]], i64 [[TMP9]]
; CHECK-NEXT: [[BROADCAST_SPLATINSERT6:%.*]] = insertelement <vscale x 2 x ptr> poison, ptr [[TMP10]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT4:%.*]] = shufflevector <vscale x 2 x ptr> [[BROADCAST_SPLATINSERT6]], <vscale x 2 x ptr> poison, <vscale x 2 x i32> zeroinitializer
; CHECK-NEXT: [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <vscale x 2 x ptr> poison, ptr [[DST]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT6:%.*]] = shufflevector <vscale x 2 x ptr> [[BROADCAST_SPLATINSERT3]], <vscale x 2 x ptr> poison, <vscale x 2 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
-; CHECK-NEXT: [[AVL:%.*]] = phi i64 [ [[TMP0]], %[[VECTOR_PH1]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[AVL:%.*]] = phi i64 [ [[TMP0]], %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP14:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 2, i1 true)
; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 2 x double> @llvm.vp.gather.nxv2f64.nxv2p0(<vscale x 2 x ptr> align 8 [[BROADCAST_SPLAT4]], <vscale x 2 x i1> [[TMP8]], i32 [[TMP14]])
; CHECK-NEXT: call void @llvm.vp.scatter.nxv2f64.nxv2p0(<vscale x 2 x double> [[WIDE_MASKED_GATHER]], <vscale x 2 x ptr> align 8 [[BROADCAST_SPLAT6]], <vscale x 2 x i1> [[TMP8]], i32 [[TMP14]])
@@ -219,23 +217,6 @@ define void @icmp_only_first_op_truncated(ptr noalias %dst, i32 %x, i64 %N, i64
; CHECK-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: br label %[[EXIT:.*]]
-; CHECK: [[SCALAR_PH]]:
-; CHECK-NEXT: br label %[[LOOP_HEADER:.*]]
-; CHECK: [[LOOP_HEADER]]:
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ 0, %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ]
-; CHECK-NEXT: [[T1:%.*]] = trunc i64 [[N]] to i32
-; CHECK-NEXT: [[C:%.*]] = icmp eq i32 [[T1]], [[T]]
-; CHECK-NEXT: br i1 [[C]], label %[[THEN:.*]], label %[[LOOP_LATCH]]
-; CHECK: [[THEN]]:
-; CHECK-NEXT: [[IDXPROM:%.*]] = zext i32 [[X]] to i64
-; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr double, ptr [[SRC]], i64 [[IDXPROM]]
-; CHECK-NEXT: [[RETVAL:%.*]] = load double, ptr [[ARRAYIDX]], align 8
-; CHECK-NEXT: store double [[RETVAL]], ptr [[DST]], align 8
-; CHECK-NEXT: br label %[[LOOP_LATCH]]
-; CHECK: [[LOOP_LATCH]]:
-; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
-; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV]], [[V]]
-; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP_HEADER]], !llvm.loop [[LOOP6:![0-9]+]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -322,17 +303,17 @@ define void @test_minbws_for_trunc(i32 %n, ptr noalias %p1, ptr noalias %p2) {
; CHECK-NEXT: [[TMP15:%.*]] = call <vscale x 2 x i32> @llvm.experimental.vp.strided.load.nxv2i32.p0.i64(ptr align 4 [[TMP14]], i64 16, <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]])
; CHECK-NEXT: [[TMP16:%.*]] = trunc <vscale x 2 x i32> [[TMP15]] to <vscale x 2 x i16>
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr [1 x [1 x i16]], ptr [[P2]], <vscale x 2 x i64> [[TMP12]]
-; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i16.nxv2p0(<vscale x 2 x i16> [[TMP16]], <vscale x 2 x ptr> align 2 [[TMP17]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META7:![0-9]+]], !noalias [[META10:![0-9]+]]
+; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i16.nxv2p0(<vscale x 2 x i16> [[TMP16]], <vscale x 2 x ptr> align 2 [[TMP17]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META6:![0-9]+]], !noalias [[META9:![0-9]+]]
; CHECK-NEXT: [[TMP18:%.*]] = trunc <vscale x 2 x i32> [[TMP15]] to <vscale x 2 x i8>
; CHECK-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[P2]], <vscale x 2 x i64> [[TMP12]]
-; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i8.nxv2p0(<vscale x 2 x i8> [[TMP18]], <vscale x 2 x ptr> align 1 [[TMP19]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META13:![0-9]+]], !noalias [[META14:![0-9]+]]
+; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i8.nxv2p0(<vscale x 2 x i8> [[TMP18]], <vscale x 2 x ptr> align 1 [[TMP19]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META12:![0-9]+]], !noalias [[META13:![0-9]+]]
; CHECK-NEXT: [[TMP20:%.*]] = getelementptr [1 x i64], ptr [[P2]], <vscale x 2 x i64> [[TMP12]]
-; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i64.nxv2p0(<vscale x 2 x i64> zeroinitializer, <vscale x 2 x ptr> align 8 [[TMP20]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META14]]
+; CHECK-NEXT: call void @llvm.vp.scatter.nxv2i64.nxv2p0(<vscale x 2 x i64> zeroinitializer, <vscale x 2 x ptr> align 8 [[TMP20]], <vscale x 2 x i1> splat (i1 true), i32 [[TMP9]]), !alias.scope [[META13]]
; CHECK-NEXT: [[CURRENT_ITERATION_NEXT]] = add nuw i32 [[TMP9]], [[INDEX]]
; CHECK-NEXT: [[AVL_NEXT]] = sub nuw i32 [[AVL]], [[TMP9]]
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i16> [[VEC_IND]], [[BROADCAST_SPLAT]]
; CHECK-NEXT: [[TMP21:%.*]] = icmp eq i32 [[AVL_NEXT]], 0
-; CHECK-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP15:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: br label %[[EXIT:.*]]
; CHECK: [[SCALAR_PH]]:
@@ -353,7 +334,7 @@ define void @test_minbws_for_trunc(i32 %n, ptr noalias %p1, ptr noalias %p2) {
; CHECK-NEXT: [[IV_NEXT]] = add i16 [[IV]], 4
; CHECK-NEXT: [[IV_NEXT_EXT:%.*]] = sext i16 [[IV_NEXT]] to i32
; CHECK-NEXT: [[CMP:%.*]] = icmp ne i32 [[IV_NEXT_EXT]], 1024
-; CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1]], label %[[EXIT]], !llvm.loop [[LOOP16:![0-9]+]]
+; CHECK-NEXT: br i1 [[CMP]], label %[[LOOP1]], label %[[EXIT]], !llvm.loop [[LOOP15:![0-9]+]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -392,15 +373,14 @@ attributes #1 = { "target-features"="+64bit,+v" }
; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]], [[META2]]}
; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]], [[META2]]}
-; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]]}
-; CHECK: [[META7]] = !{[[META8:![0-9]+]]}
-; CHECK: [[META8]] = distinct !{[[META8]], [[META9:![0-9]+]]}
-; CHECK: [[META9]] = distinct !{[[META9]], !"LVerDomain"}
-; CHECK: [[META10]] = !{[[META11:![0-9]+]], [[META12:![0-9]+]]}
-; CHECK: [[META11]] = distinct !{[[META11]], [[META9]]}
-; CHECK: [[META12]] = distinct !{[[META12]], [[META9]]}
+; CHECK: [[META6]] = !{[[META7:![0-9]+]]}
+; CHECK: [[META7]] = distinct !{[[META7]], [[META8:![0-9]+]]}
+; CHECK: [[META8]] = distinct !{[[META8]], !"LVerDomain"}
+; CHECK: [[META9]] = !{[[META10:![0-9]+]], [[META11:![0-9]+]]}
+; CHECK: [[META10]] = distinct !{[[META10]], [[META8]]}
+; CHECK: [[META11]] = distinct !{[[META11]], [[META8]]}
+; CHECK: [[META12]] = !{[[META10]]}
; CHECK: [[META13]] = !{[[META11]]}
-; CHECK: [[META14]] = !{[[META12]]}
-; CHECK: [[LOOP15]] = distinct !{[[LOOP15]], [[META1]], [[META2]]}
-; CHECK: [[LOOP16]] = distinct !{[[LOOP16]], [[META1]]}
+; CHECK: [[LOOP14]] = distinct !{[[LOOP14]], [[META1]], [[META2]]}
+; CHECK: [[LOOP15]] = distinct !{[[LOOP15]], [[META1]]}
;.
diff --git a/llvm/test/Transforms/LoopVectorize/pr34681.ll b/llvm/test/Transforms/LoopVectorize/pr34681.ll
index 3ae164dd0fc83..ce942fefa6e78 100644
--- a/llvm/test/Transforms/LoopVectorize/pr34681.ll
+++ b/llvm/test/Transforms/LoopVectorize/pr34681.ll
@@ -136,30 +136,63 @@ for.end:
define i32 @foo2(i16 zeroext %N, ptr nocapture readnone %A, ptr nocapture readonly %B, i32 %i, i32 %j) {
; CHECK-LABEL: define i32 @foo2(
; CHECK-SAME: i16 zeroext [[N:%.*]], ptr readnone captures(none) [[A:%.*]], ptr readonly captures(none) [[B:%.*]], i32 [[I:%.*]], i32 [[J:%.*]]) {
-; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[CONV:%.*]] = zext i16 [[N]] to i32
; CHECK-NEXT: [[CMP11:%.*]] = icmp eq i16 [[N]], 0
-; CHECK-NEXT: br i1 [[CMP11]], label %[[FOR_END:.*]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT: br i1 [[CMP11]], [[FOR_END:label %.*]], label %[[FOR_BODY_LR_PH:.*]]
+; CHECK: [[FOR_BODY_LR_PH]]:
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[CONV]], 4
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
+; CHECK: [[VECTOR_SCEVCHECK]]:
+; CHECK-NEXT: [[TMP0:%.*]] = add nsw i32 [[CONV]], -1
+; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[CONV]], i32 [[TMP0]])
+; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
+; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
+; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[J]], [[MUL_RESULT]]
+; CHECK-NEXT: [[TMP3:%.*]] = icmp slt i32 [[TMP1]], [[J]]
+; CHECK-NEXT: [[TMP4:%.*]] = or i1 [[TMP3]], [[MUL_OVERFLOW]]
+; CHECK-NEXT: br i1 [[TMP4]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[CONV]], 4
+; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[CONV]], [[N_MOD_VF]]
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[CONV]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT]], <4 x i32> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT2:%.*]] = insertelement <4 x i32> poison, i32 [[J]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT2]], <4 x i32> poison, <4 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
-; CHECK-NEXT: [[TMP_013:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[ADD4:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[K_012:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INC:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[MUL:%.*]] = mul nuw i32 [[K_012]], [[CONV]]
-; CHECK-NEXT: [[TMP15:%.*]] = add i32 [[MUL]], [[J]]
+; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP26:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP7:%.*]] = mul nuw <4 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP8:%.*]] = add <4 x i32> [[TMP7]], [[BROADCAST_SPLAT3]]
+; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i32> [[TMP8]], i64 0
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP9]]
+; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x i32> [[TMP8]], i64 1
+; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP11]]
+; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x i32> [[TMP8]], i64 2
+; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP13]]
+; CHECK-NEXT: [[TMP15:%.*]] = extractelement <4 x i32> [[TMP8]], i64 3
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP15]]
+; CHECK-NEXT: [[TMP17:%.*]] = load i16, ptr [[TMP10]], align 2
+; CHECK-NEXT: [[TMP18:%.*]] = load i16, ptr [[TMP12]], align 2
+; CHECK-NEXT: [[TMP19:%.*]] = load i16, ptr [[TMP14]], align 2
; CHECK-NEXT: [[TMP20:%.*]] = load i16, ptr [[TMP16]], align 2
-; CHECK-NEXT: [[CONV3:%.*]] = sext i16 [[TMP20]] to i32
-; CHECK-NEXT: [[ADD4]] = add nsw i32 [[TMP_013]], [[CONV3]]
-; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[K_012]], 1
-; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INC]], [[CONV]]
-; CHECK-NEXT: br i1 [[EXITCOND]], label %[[SCALAR_PH:.*]], label %[[VECTOR_BODY]]
+; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i16> poison, i16 [[TMP17]], i32 0
+; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x i16> [[TMP21]], i16 [[TMP18]], i32 1
+; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x i16> [[TMP22]], i16 [[TMP19]], i32 2
+; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x i16> [[TMP23]], i16 [[TMP20]], i32 3
+; CHECK-NEXT: [[TMP25:%.*]] = sext <4 x i16> [[TMP24]] to <4 x i32>
+; CHECK-NEXT: [[TMP26]] = add <4 x i32> [[VEC_PHI]], [[TMP25]]
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i32> [[VEC_IND]], splat (i32 4)
+; CHECK-NEXT: [[TMP27:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP27]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: [[TMP28:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP26]])
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[CONV]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], [[FOR_END_LOOPEXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
-; CHECK-NEXT: [[ADD4_LCSSA:%.*]] = phi i32 [ [[ADD4]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: br label %[[FOR_END]]
-; CHECK: [[FOR_END]]:
-; CHECK-NEXT: [[TMP_0_LCSSA:%.*]] = phi i32 [ 0, %[[ENTRY]] ], [ [[ADD4_LCSSA]], %[[SCALAR_PH]] ]
-; CHECK-NEXT: ret i32 [[TMP_0_LCSSA]]
;
entry:
%conv = zext i16 %N to i32
diff --git a/llvm/test/Transforms/LoopVectorize/pr59319-loop-access-info-invalidation.ll b/llvm/test/Transforms/LoopVectorize/pr59319-loop-access-info-invalidation.ll
index 7a2719ee7d6ad..9ffee47c8a31d 100644
--- a/llvm/test/Transforms/LoopVectorize/pr59319-loop-access-info-invalidation.ll
+++ b/llvm/test/Transforms/LoopVectorize/pr59319-loop-access-info-invalidation.ll
@@ -38,13 +38,12 @@ define void @reduced(ptr %0, ptr %1, i64 %iv, ptr %2, i64 %iv76, i64 %iv93) {
; CHECK-NEXT: [[IV761_LCSSA:%.*]] = phi i64 [ [[IV761]], [[LOOP_1]] ], [ [[IND_ESCAPE]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: [[MIN_ITERS_CHECK6:%.*]] = icmp ult i64 [[TMP3]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK6]], label [[SCALAR_PH5:%.*]], label [[VECTOR_MEMCHECK:%.*]]
-; CHECK: vector.scevcheck:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[IV761_LCSSA]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label [[SCALAR_PH5]], label [[VECTOR_MEMCHECK1:%.*]]
; CHECK: vector.memcheck:
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[TMP1:%.*]], i64 4
-; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[TMP0]], i64 4
-; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[TMP0]], i64 8
+; CHECK-NEXT: [[TMP9:%.*]] = shl i64 [[IV761_LCSSA]], 2
+; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[TMP0]], i64 [[TMP9]]
+; CHECK-NEXT: [[TMP10:%.*]] = add i64 [[TMP9]], 4
+; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[TMP0]], i64 [[TMP10]]
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[TMP1]], [[SCEVGEP3]]
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP2]], [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
@@ -57,45 +56,43 @@ define void @reduced(ptr %0, ptr %1, i64 %iv, ptr %2, i64 %iv76, i64 %iv93) {
; CHECK-NEXT: [[INDEX12:%.*]] = phi i64 [ 0, [[VECTOR_PH6]] ], [ [[INDEX_NEXT13:%.*]], [[VECTOR_BODY10]] ]
; CHECK-NEXT: [[INDEX_NEXT13]] = add nuw i64 [[INDEX12]], 4
; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT13]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK11:%.*]], label [[VECTOR_BODY10]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK12:%.*]], label [[VECTOR_BODY10]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: middle.block11:
; CHECK-NEXT: store i32 0, ptr [[TMP1]], align 4, !alias.scope [[META5:![0-9]+]], !noalias [[META8:![0-9]+]]
; CHECK-NEXT: [[CMP_N10:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N10]], label [[LOOP_3_LR_PH:%.*]], label [[SCALAR_PH5]]
; CHECK: scalar.ph3:
-; CHECK-NEXT: [[BC_RESUME_VAL13:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK11]] ], [ 0, [[LOOP_2_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[VECTOR_MEMCHECK1]] ]
+; CHECK-NEXT: [[BC_RESUME_VAL13:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK12]] ], [ 0, [[LOOP_2_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ]
; CHECK-NEXT: br label [[LOOP_2:%.*]]
; CHECK: loop.3.lr.ph:
; CHECK-NEXT: [[IDXPROM_I_I61:%.*]] = and i64 [[IV761_LCSSA]], 1
; CHECK-NEXT: [[ARRAYIDX_I_I62:%.*]] = getelementptr i32, ptr [[TMP0]], i64 [[IDXPROM_I_I61]]
; CHECK-NEXT: [[MIN_ITERS_CHECK22:%.*]] = icmp ult i64 [[TMP3]], 4
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK22]], label [[SCALAR_PH21:%.*]], label [[VECTOR_MEMCHECK15:%.*]]
-; CHECK: vector.scevcheck14:
-; CHECK-NEXT: [[IDENT_CHECK15:%.*]] = icmp ne i64 [[IV761_LCSSA]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK15]], label [[SCALAR_PH21]], label [[VECTOR_MEMCHECK16:%.*]]
-; CHECK: vector.memcheck16:
+; CHECK: vector.memcheck14:
; CHECK-NEXT: [[SCEVGEP15:%.*]] = getelementptr i8, ptr [[TMP1]], i64 4
-; CHECK-NEXT: [[SCEVGEP18:%.*]] = getelementptr i8, ptr [[TMP0]], i64 4
-; CHECK-NEXT: [[SCEVGEP16:%.*]] = getelementptr i8, ptr [[TMP0]], i64 8
+; CHECK-NEXT: [[TMP12:%.*]] = shl nuw nsw i64 [[IDXPROM_I_I61]], 2
+; CHECK-NEXT: [[TMP13:%.*]] = add nuw nsw i64 [[TMP12]], 4
+; CHECK-NEXT: [[SCEVGEP16:%.*]] = getelementptr i8, ptr [[TMP0]], i64 [[TMP13]]
; CHECK-NEXT: [[BOUND017:%.*]] = icmp ult ptr [[TMP1]], [[SCEVGEP16]]
-; CHECK-NEXT: [[BOUND118:%.*]] = icmp ult ptr [[SCEVGEP18]], [[SCEVGEP15]]
+; CHECK-NEXT: [[BOUND118:%.*]] = icmp ult ptr [[ARRAYIDX_I_I62]], [[SCEVGEP15]]
; CHECK-NEXT: [[FOUND_CONFLICT19:%.*]] = and i1 [[BOUND017]], [[BOUND118]]
; CHECK-NEXT: br i1 [[FOUND_CONFLICT19]], label [[SCALAR_PH21]], label [[VECTOR_PH24:%.*]]
-; CHECK: vector.ph25:
+; CHECK: vector.ph22:
; CHECK-NEXT: [[N_MOD_VF24:%.*]] = urem i64 [[TMP3]], 4
; CHECK-NEXT: [[N_VEC25:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF24]]
; CHECK-NEXT: br label [[VECTOR_BODY27:%.*]]
-; CHECK: vector.body28:
+; CHECK: vector.body25:
; CHECK-NEXT: [[INDEX29:%.*]] = phi i64 [ 0, [[VECTOR_PH24]] ], [ [[INDEX_NEXT29:%.*]], [[VECTOR_BODY27]] ]
; CHECK-NEXT: [[INDEX_NEXT29]] = add nuw i64 [[INDEX29]], 4
; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT29]], [[N_VEC25]]
-; CHECK-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK31:%.*]], label [[VECTOR_BODY27]], !llvm.loop [[LOOP10:![0-9]+]]
-; CHECK: middle.block31:
+; CHECK-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK29:%.*]], label [[VECTOR_BODY27]], !llvm.loop [[LOOP10:![0-9]+]]
+; CHECK: middle.block28:
; CHECK-NEXT: store i32 0, ptr [[TMP1]], align 4, !alias.scope [[META11:![0-9]+]], !noalias [[META14:![0-9]+]]
; CHECK-NEXT: [[CMP_N27:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC25]]
; CHECK-NEXT: br i1 [[CMP_N27]], label [[LOOP_CLEANUP:%.*]], label [[SCALAR_PH21]]
-; CHECK: scalar.ph23:
-; CHECK-NEXT: [[BC_RESUME_VAL26:%.*]] = phi i64 [ [[N_VEC25]], [[MIDDLE_BLOCK31]] ], [ 0, [[LOOP_3_LR_PH]] ], [ 0, [[VECTOR_MEMCHECK15]] ], [ 0, [[VECTOR_MEMCHECK16]] ]
+; CHECK: scalar.ph20:
+; CHECK-NEXT: [[BC_RESUME_VAL26:%.*]] = phi i64 [ [[N_VEC25]], [[MIDDLE_BLOCK29]] ], [ 0, [[LOOP_3_LR_PH]] ], [ 0, [[VECTOR_MEMCHECK15]] ]
; CHECK-NEXT: br label [[LOOP_3:%.*]]
; CHECK: loop.2:
; CHECK-NEXT: [[IV846:%.*]] = phi i64 [ [[IV_NEXT85:%.*]], [[LOOP_2]] ], [ [[BC_RESUME_VAL13]], [[SCALAR_PH5]] ]
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll b/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
index 0cb40f354cb4c..06e9554937a0e 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
@@ -4,30 +4,7 @@
define void @test(ptr %A, i32 %x) {
; CHECK-LABEL: define void @test(
; CHECK-SAME: ptr [[A:%.*]], i32 [[X:%.*]]) {
-; CHECK-NEXT: [[ENTRY:.*:]]
-; CHECK-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
-; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[X]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
-; CHECK: [[VECTOR_PH]]:
-; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
-; CHECK: [[VECTOR_BODY]]:
-; CHECK-NEXT: [[OFFSET_IDX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[OFFSET_IDX]], 1
-; CHECK-NEXT: [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
-; CHECK-NEXT: [[TMP5:%.*]] = zext i32 [[TMP4]] to i64
-; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP5]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP6]], align 4
-; CHECK-NEXT: [[TMP0:%.*]] = trunc i64 [[OFFSET_IDX]] to i32
-; CHECK-NEXT: [[TMP8:%.*]] = zext i32 [[TMP0]] to i64
-; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP8]]
-; CHECK-NEXT: store <4 x float> [[WIDE_LOAD]], ptr [[TMP9]], align 4
-; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[OFFSET_IDX]], 4
-; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000
-; CHECK-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
-; CHECK: [[MIDDLE_BLOCK]]:
-; CHECK-NEXT: br label %[[EXIT:.*]]
-; CHECK: [[SCALAR_PH]]:
+; CHECK-NEXT: [[SCALAR_PH:.*]]:
; CHECK-NEXT: br label %[[LOOP:.*]]
; CHECK: [[LOOP]]:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], %[[LOOP]] ], [ 0, %[[SCALAR_PH]] ]
@@ -43,7 +20,7 @@ define void @test(ptr %A, i32 %x) {
; CHECK-NEXT: [[ARRAYIDX1209:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IDX_2]]
; CHECK-NEXT: store float [[LV]], ptr [[ARRAYIDX1209]], align 4
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], 1000
-; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK-NEXT: br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -102,7 +79,7 @@ define void @diff_memcheck_known_false_for_vf_4(ptr %B, ptr %A, ptr %end) {
; CHECK-NEXT: store <4 x i64> zeroinitializer, ptr [[TMP8]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 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: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
@@ -118,7 +95,7 @@ define void @diff_memcheck_known_false_for_vf_4(ptr %B, ptr %A, ptr %end) {
; CHECK-NEXT: [[TMP10:%.*]] = load i64, ptr [[IV_2_NEXT]], align 8
; CHECK-NEXT: store i64 0, ptr [[IV_1]], align 8
; CHECK-NEXT: [[CMP_NOT_I_I_I_I:%.*]] = icmp eq ptr [[END]], [[IV_2]]
-; CHECK-NEXT: br i1 [[CMP_NOT_I_I_I_I]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
+; CHECK-NEXT: br i1 [[CMP_NOT_I_I_I_I]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
@@ -146,7 +123,5 @@ exit:
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
-; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]}
-; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
-; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META2]], [[META1]]}
+; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
;.
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll b/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll
index adbe6bcb91a79..23ebd63d07cd9 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-checks-hoist.ll
@@ -1503,13 +1503,10 @@ define void @stride_check_known_via_loop_guard(ptr %C, ptr %A, i32 %Acols) {
; CHECK-NEXT: [[ARRAYIDX_US:%.*]] = getelementptr double, ptr [[A]], i32 [[MUL_US]]
; CHECK-NEXT: br label %[[VECTOR_MEMCHECK:.*]]
; CHECK: [[VECTOR_MEMCHECK]]:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[ACOLS]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK1:.*]]
-; CHECK: [[VECTOR_MEMCHECK1]]:
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[C]], [[SCEVGEP1]]
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[A]], [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
-; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[TMP0:%.*]] = load double, ptr [[ARRAYIDX_US]], align 8, !alias.scope [[META69:![0-9]+]]
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x double> poison, double [[TMP0]], i64 0
@@ -1525,10 +1522,9 @@ define void @stride_check_known_via_loop_guard(ptr %C, ptr %A, i32 %Acols) {
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: br label %[[OUTER_LATCH]]
; CHECK: [[SCALAR_PH]]:
-; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 0, %[[VECTOR_MEMCHECK]] ], [ 0, %[[VECTOR_MEMCHECK1]] ]
; CHECK-NEXT: br label %[[INNER:.*]]
; CHECK: [[INNER]]:
-; CHECK-NEXT: [[INNER_IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INNER_IV_NEXT:%.*]], %[[INNER]] ]
+; CHECK-NEXT: [[INNER_IV:%.*]] = phi i32 [ 0, %[[SCALAR_PH]] ], [ [[INNER_IV_NEXT:%.*]], %[[INNER]] ]
; CHECK-NEXT: [[GEP_C:%.*]] = getelementptr inbounds double, ptr [[C]], i32 [[INNER_IV]]
; CHECK-NEXT: [[L:%.*]] = load double, ptr [[ARRAYIDX_US]], align 8
; CHECK-NEXT: store double [[L]], ptr [[GEP_C]], align 8
diff --git a/llvm/test/Transforms/LoopVectorize/vector-no-scevcheck.ll b/llvm/test/Transforms/LoopVectorize/vector-no-scevcheck.ll
index d1b5cc947f54a..56a8b5a93b851 100644
--- a/llvm/test/Transforms/LoopVectorize/vector-no-scevcheck.ll
+++ b/llvm/test/Transforms/LoopVectorize/vector-no-scevcheck.ll
@@ -1,40 +1,12 @@
-; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^vector.memcheck:" --version 6
; RUN: opt -passes=loop-vectorize -force-vector-width=2 -force-vector-interleave=1 -S < %s | FileCheck %s
-; SCEV checks are cheap, due to stride being speculated.
+; This test is to ensure that SCEV checks (which are costly performancewise) are
+; not generated when appropriate aliasing checks are sufficient.
define void @foo(ptr %pout, ptr %pin, i64 %val0, i64 %val1, i64 %val2) {
-; CHECK-LABEL: define void @foo(
-; CHECK-SAME: ptr [[POUT:%.*]], ptr [[PIN:%.*]], i64 [[VAL0:%.*]], i64 [[VAL1:%.*]], i64 [[VAL2:%.*]]) {
-; CHECK-NEXT: [[ENTRY:.*]]:
-; CHECK-NEXT: [[TMP0:%.*]] = getelementptr double, ptr [[PIN]], i64 [[VAL0]]
-; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[VAL1]], -1
-; CHECK-NEXT: [[TMP2:%.*]] = mul i64 [[VAL1]], [[TMP1]]
-; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 3
-; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[VAL1]], i64 1)
-; CHECK-NEXT: [[TMP4:%.*]] = shl i64 [[SMAX]], 3
-; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[TMP3]], [[TMP4]]
-; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[POUT]], i64 [[TMP5]]
-; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[VAL1]], 3
-; CHECK-NEXT: [[TMP7:%.*]] = shl i64 [[VAL0]], 3
-; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[TMP7]], [[TMP4]]
-; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[PIN]], i64 [[TMP8]]
-; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[TMP6]], [[TMP7]]
-; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[PIN]], i64 [[TMP9]]
-; CHECK-NEXT: br label %[[LOOP1_HEADER:.*]]
-; CHECK: [[LOOP1_HEADER]]:
-; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[LOOP1_LATCH:%.*]] ], [ 0, %[[ENTRY]] ]
-; CHECK-NEXT: [[MUL0:%.*]] = mul nsw i64 [[I]], [[VAL2]]
-; CHECK-NEXT: [[ARRAYIDX0:%.*]] = getelementptr inbounds double, ptr [[TMP0]], i64 [[MUL0]]
-; CHECK-NEXT: [[MUL1:%.*]] = mul nsw i64 [[I]], [[VAL1]]
-; CHECK-NEXT: [[TMP10:%.*]] = call i64 @llvm.smax.i64(i64 [[VAL1]], i64 1)
-; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP10]], 2
-; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], [[SCALAR_PH:label %.*]], label %[[VECTOR_SCEVCHECK:.*]]
-; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[VAL2]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], [[SCALAR_PH]], label %[[EXIT:.*]]
-; CHECK: [[EXIT]]:
-;
+; CHECK-LABEL: @foo(
+; CHECK-NOT: vector.scevcheck
+; CHECK: vector.body
entry:
%0 = getelementptr double, ptr %pin, i64 %val0
br label %loop1.header
@@ -68,37 +40,12 @@ exit:
ret void
}
+; Similar test to the above but with the %arrayidx0 moved to the loop2.header
+
define void @bar(ptr %pout, ptr %pin, i64 %val0, i64 %val1, i64 %val2) {
-; CHECK-LABEL: define void @bar(
-; CHECK-SAME: ptr [[POUT:%.*]], ptr [[PIN:%.*]], i64 [[VAL0:%.*]], i64 [[VAL1:%.*]], i64 [[VAL2:%.*]]) {
-; CHECK-NEXT: [[ENTRY:.*]]:
-; CHECK-NEXT: [[TMP0:%.*]] = getelementptr double, ptr [[PIN]], i64 [[VAL0]]
-; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[VAL1]], -1
-; CHECK-NEXT: [[TMP2:%.*]] = mul i64 [[VAL1]], [[TMP1]]
-; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 3
-; CHECK-NEXT: [[SMAX:%.*]] = call i64 @llvm.smax.i64(i64 [[VAL1]], i64 1)
-; CHECK-NEXT: [[TMP4:%.*]] = shl i64 [[SMAX]], 3
-; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[TMP3]], [[TMP4]]
-; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[POUT]], i64 [[TMP5]]
-; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[VAL1]], 3
-; CHECK-NEXT: [[TMP7:%.*]] = shl i64 [[VAL0]], 3
-; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[TMP7]], [[TMP4]]
-; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[PIN]], i64 [[TMP8]]
-; CHECK-NEXT: [[TMP9:%.*]] = add i64 [[TMP6]], [[TMP7]]
-; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[PIN]], i64 [[TMP9]]
-; CHECK-NEXT: br label %[[LOOP1_HEADER:.*]]
-; CHECK: [[LOOP1_HEADER]]:
-; CHECK-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[LOOP1_LATCH:%.*]] ], [ 0, %[[ENTRY]] ]
-; CHECK-NEXT: [[MUL0:%.*]] = mul nsw i64 [[I]], [[VAL2]]
-; CHECK-NEXT: [[MUL1:%.*]] = mul nsw i64 [[I]], [[VAL1]]
-; CHECK-NEXT: [[TMP10:%.*]] = call i64 @llvm.smax.i64(i64 [[VAL1]], i64 1)
-; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP10]], 2
-; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], [[SCALAR_PH:label %.*]], label %[[VECTOR_SCEVCHECK:.*]]
-; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[VAL2]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], [[SCALAR_PH]], label %[[EXIT:.*]]
-; CHECK: [[EXIT]]:
-;
+; CHECK-LABEL: @bar(
+; CHECK-NOT: vector.scevcheck
+; CHECK: vector.body
entry:
%0 = getelementptr double, ptr %pin, i64 %val0
br label %loop1.header
diff --git a/llvm/test/Transforms/LoopVectorize/version-mem-access.ll b/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
index 1d7d99d691e26..c533c95a444f1 100644
--- a/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
+++ b/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
@@ -13,26 +13,45 @@ define void @test(ptr noalias %A, i64 %AStride, ptr noalias %B, i32 %BStride, pt
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[IDENT_CHECK1:%.*]] = icmp ne i32 [[BSTRIDE]], 1
+; CHECK-NEXT: [[TMP0:%.*]] = sub i32 0, [[BSTRIDE]]
+; CHECK-NEXT: [[TMP1:%.*]] = icmp slt i32 [[BSTRIDE]], 0
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i32 [[TMP0]], i32 [[BSTRIDE]]
+; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP2]], i32 99)
+; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
+; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
+; CHECK-NEXT: [[TMP3:%.*]] = sub i32 0, [[MUL_RESULT]]
+; CHECK-NEXT: [[TMP4:%.*]] = icmp ugt i32 [[TMP3]], 0
+; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP1]], i1 [[TMP4]], i1 false
+; CHECK-NEXT: [[IDENT_CHECK1:%.*]] = or i1 [[TMP11]], [[MUL_OVERFLOW]]
; CHECK-NEXT: [[IDENT_CHECK2:%.*]] = icmp ne i64 [[CSTRIDE]], 1
; CHECK-NEXT: [[DIFF_CHECK6:%.*]] = icmp ne i64 [[ASTRIDE]], 1
; CHECK-NEXT: [[DIFF_CHECK:%.*]] = or i1 [[IDENT_CHECK1]], [[IDENT_CHECK2]]
; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[DIFF_CHECK]], [[DIFF_CHECK6]]
; CHECK-NEXT: br i1 [[CONFLICT_RDX]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[BSTRIDE]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[INDEX]] to i32
-; CHECK-NEXT: [[TMP13:%.*]] = zext i32 [[TMP2]] to i64
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP12:%.*]] = mul <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP18:%.*]] = zext <2 x i32> [[TMP12]] to <2 x i64>
+; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i64> [[TMP18]], i64 0
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP5]]
+; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i64> [[TMP18]], i64 1
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP13]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP14]], align 4
+; CHECK-NEXT: [[TMP15:%.*]] = load i32, ptr [[TMP6]], align 4
+; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP14]], align 4
+; CHECK-NEXT: [[TMP17:%.*]] = insertelement <2 x i32> poison, i32 [[TMP15]], i32 0
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = insertelement <2 x i32> [[TMP17]], i32 [[TMP16]], i32 1
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[INDEX]]
; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <2 x i32>, ptr [[TMP7]], align 4
; CHECK-NEXT: [[TMP8:%.*]] = mul nsw <2 x i32> [[WIDE_LOAD7]], [[WIDE_LOAD]]
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX]]
; CHECK-NEXT: store <2 x i32> [[TMP8]], ptr [[TMP9]], align 4
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2)
; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
; CHECK-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
@@ -78,25 +97,50 @@ define void @fn1(ptr noalias %x, ptr noalias %c, double %a) {
; CHECK-NEXT: br i1 [[CMP8]], label %[[LOOP_PREHEADER:.*]], [[EXIT:label %.*]]
; CHECK: [[LOOP_PREHEADER]]:
; CHECK-NEXT: [[TMP0:%.*]] = zext i32 [[CONV2]] to i64
-; CHECK-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 2
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
; CHECK: [[VECTOR_SCEVCHECK]]:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[CONV]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[CONV]], 3
+; CHECK-NEXT: [[TMP9:%.*]] = sub i32 0, [[CONV]]
+; CHECK-NEXT: [[TMP12:%.*]] = icmp slt i32 [[CONV]], 0
+; CHECK-NEXT: [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP9]], i32 [[CONV]]
+; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP13]], i32 [[TMP1]])
+; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
+; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
+; CHECK-NEXT: [[TMP5:%.*]] = sub i32 0, [[MUL_RESULT]]
+; CHECK-NEXT: [[TMP6:%.*]] = icmp slt i32 [[MUL_RESULT]], 0
+; CHECK-NEXT: [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], 0
+; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP12]], i1 [[TMP7]], i1 [[TMP6]]
+; CHECK-NEXT: [[IDENT_CHECK:%.*]] = or i1 [[TMP8]], [[MUL_OVERFLOW]]
+; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 2
+; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]]
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[CONV]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP1:%.*]] = trunc i64 [[INDEX]] to i32
-; CHECK-NEXT: [[TMP14:%.*]] = sext i32 [[TMP1]] to i64
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP10:%.*]] = mul nsw <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP11:%.*]] = sext <2 x i32> [[TMP10]] to <2 x i64>
+; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i64> [[TMP11]], i64 0
+; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds double, ptr [[X]], i64 [[TMP2]]
+; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i64> [[TMP11]], i64 1
; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds double, ptr [[X]], i64 [[TMP14]]
-; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x double>, ptr [[TMP15]], align 8
+; CHECK-NEXT: [[TMP16:%.*]] = load double, ptr [[TMP3]], align 8
+; CHECK-NEXT: [[TMP17:%.*]] = load double, ptr [[TMP15]], align 8
+; CHECK-NEXT: [[TMP18:%.*]] = insertelement <2 x double> poison, double [[TMP16]], i32 0
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = insertelement <2 x double> [[TMP18]], double [[TMP17]], i32 1
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds double, ptr [[C]], i64 [[INDEX]]
; CHECK-NEXT: store <2 x double> [[WIDE_LOAD]], ptr [[TMP4]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
-; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4
-; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 2)
+; CHECK-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP21]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
-; CHECK-NEXT: br label %[[SCALAR_PH]]
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], [[EXIT_LOOPEXIT:label %.*]], label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
;
entry:
diff --git a/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll b/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
index d38b56bc62f42..5a7588d04793b 100644
--- a/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
+++ b/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
@@ -3826,17 +3826,48 @@ define void @trunc_stride(ptr noalias %p.out, ptr %p, i64 %stride.i64) {
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; COMPARE-LAA-MV: [[VECTOR_SCEVCHECK]]:
-; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[STRIDE_I64]], 1
+; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = trunc i64 [[STRIDE_I64]] to i32
+; COMPARE-LAA-MV-NEXT: [[TMP11:%.*]] = sub i32 0, [[TMP8]]
+; COMPARE-LAA-MV-NEXT: [[TMP12:%.*]] = icmp slt i32 [[TMP8]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = select i1 [[TMP12]], i32 [[TMP11]], i32 [[TMP8]]
+; COMPARE-LAA-MV-NEXT: [[MUL:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP3]], i32 127)
+; COMPARE-LAA-MV-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL]], 0
+; COMPARE-LAA-MV-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL]], 1
+; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = sub i32 0, [[MUL_RESULT]]
+; COMPARE-LAA-MV-NEXT: [[TMP5:%.*]] = icmp slt i32 [[MUL_RESULT]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = icmp sgt i32 [[TMP4]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP7:%.*]] = select i1 [[TMP12]], i1 [[TMP6]], i1 [[TMP5]]
+; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW]]
; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
+; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i64> poison, i64 [[STRIDE_I64]], i64 0
+; COMPARE-LAA-MV-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT]], <4 x i64> poison, <4 x i32> zeroinitializer
+; COMPARE-LAA-MV-NEXT: [[TMP9:%.*]] = trunc <4 x i64> [[BROADCAST_SPLAT]] to <4 x i32>
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[INDEX1:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[TMP10:%.*]] = mul <4 x i32> [[VEC_IND]], [[TMP9]]
+; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = extractelement <4 x i32> [[TMP10]], i64 0
+; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = getelementptr i32, ptr [[P]], i32 [[INDEX]]
+; COMPARE-LAA-MV-NEXT: [[TMP13:%.*]] = extractelement <4 x i32> [[TMP10]], i64 1
+; COMPARE-LAA-MV-NEXT: [[TMP14:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP13]]
+; COMPARE-LAA-MV-NEXT: [[TMP15:%.*]] = extractelement <4 x i32> [[TMP10]], i64 2
+; COMPARE-LAA-MV-NEXT: [[TMP16:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP15]]
+; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = extractelement <4 x i32> [[TMP10]], i64 3
; COMPARE-LAA-MV-NEXT: [[TMP18:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP17]]
-; COMPARE-LAA-MV-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP18]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[P_OUT]], i32 [[TMP17]]
+; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP0]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP20:%.*]] = load i32, ptr [[TMP14]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP16]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP18]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP23:%.*]] = insertelement <4 x i32> poison, i32 [[TMP19]], i32 0
+; COMPARE-LAA-MV-NEXT: [[TMP24:%.*]] = insertelement <4 x i32> [[TMP23]], i32 [[TMP20]], i32 1
+; COMPARE-LAA-MV-NEXT: [[TMP25:%.*]] = insertelement <4 x i32> [[TMP24]], i32 [[TMP21]], i32 2
+; COMPARE-LAA-MV-NEXT: [[WIDE_LOAD:%.*]] = insertelement <4 x i32> [[TMP25]], i32 [[TMP22]], i32 3
+; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[P_OUT]], i32 [[INDEX1]]
; COMPARE-LAA-MV-NEXT: store <4 x i32> [[WIDE_LOAD]], ptr [[TMP1]], align 8
-; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i32 [[TMP17]], 4
+; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX1]], 4
+; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i32> [[VEC_IND]], splat (i32 4)
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 128
; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP55:![0-9]+]]
; COMPARE-LAA-MV: [[MIDDLE_BLOCK]]:
@@ -3984,8 +4015,20 @@ define void @trunc_ext_stride(ptr noalias %p.out, ptr %p0, ptr %p1, i32 %stride)
; COMPARE-LAA-MV-NEXT: [[STRIDE_EXT:%.*]] = sext i32 [[STRIDE]] to i64
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; COMPARE-LAA-MV: [[VECTOR_SCEVCHECK]]:
+; COMPARE-LAA-MV-NEXT: [[TMP9:%.*]] = sub i16 0, [[STRIDE_TRUNC]]
+; COMPARE-LAA-MV-NEXT: [[TMP10:%.*]] = icmp slt i16 [[STRIDE_TRUNC]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP11:%.*]] = select i1 [[TMP10]], i16 [[TMP9]], i16 [[STRIDE_TRUNC]]
+; COMPARE-LAA-MV-NEXT: [[MUL:%.*]] = call { i16, i1 } @llvm.umul.with.overflow.i16(i16 [[TMP11]], i16 127)
+; COMPARE-LAA-MV-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i16, i1 } [[MUL]], 0
+; COMPARE-LAA-MV-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i16, i1 } [[MUL]], 1
+; COMPARE-LAA-MV-NEXT: [[TMP12:%.*]] = sub i16 0, [[MUL_RESULT]]
+; COMPARE-LAA-MV-NEXT: [[TMP13:%.*]] = icmp slt i16 [[MUL_RESULT]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP14:%.*]] = icmp sgt i16 [[TMP12]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP15:%.*]] = select i1 [[TMP10]], i1 [[TMP14]], i1 [[TMP13]]
+; COMPARE-LAA-MV-NEXT: [[TMP16:%.*]] = or i1 [[TMP15]], [[MUL_OVERFLOW]]
; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[STRIDE]], 1
-; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = or i1 [[TMP16]], [[IDENT_CHECK]]
+; COMPARE-LAA-MV-NEXT: br i1 [[TMP8]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
>From 72e66f3ec64accfc92ad64c76f18c0fc3e2c24d6 Mon Sep 17 00:00:00 2001
From: Ramkumar Ramachandra <artagnon at tenstorrent.com>
Date: Thu, 30 Jul 2026 13:35:55 +0100
Subject: [PATCH 4/4] [LAA] Ugh, another attempt
---
.../Analysis/ScalarEvolutionPatternMatch.h | 6 +
llvm/lib/Analysis/LoopAccessAnalysis.cpp | 44 ++++--
.../AArch64/pr60831-sve-inv-store-crash.ll | 142 +++++++++---------
llvm/test/Transforms/LoopVectorize/pr34681.ll | 68 ++++-----
.../runtime-check-needed-but-empty.ll | 20 +--
.../LoopVectorize/version-mem-access.ll | 38 ++---
.../version-stride-with-integer-casts.ll | 14 +-
.../LoopVectorize/vplan-based-stride-mv.ll | 28 ++--
8 files changed, 194 insertions(+), 166 deletions(-)
diff --git a/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h b/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h
index 4202a4fd1d745..558265045b8eb 100644
--- a/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h
+++ b/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h
@@ -79,6 +79,7 @@ inline cst_pred_ty<is_all_ones> m_scev_AllOnes() {
}
inline auto m_SCEV() { return m_Isa<const SCEV>(); }
+inline auto m_SCEVUnknown() { return m_Isa<const SCEVUnknown>(); }
inline auto m_SCEVConstant() { return m_Isa<const SCEVConstant>(); }
inline auto m_SCEVVScale() { return m_Isa<const SCEVVScale>(); }
@@ -180,6 +181,11 @@ m_scev_IntegralCast(const Op0_t &Op0) {
return m_scev_Unary<SCEVIntegralCastExpr>(Op0);
}
+template <typename Op0_t>
+inline auto m_scev_IntegralCastOrSelf(const Op0_t &Op0) {
+ return m_CombineOr(m_scev_IntegralCast(Op0), Op0);
+}
+
template <typename Op0_t>
inline SCEVUnaryExpr_match<SCEVSignExtendExpr, Op0_t>
m_scev_SExt(const Op0_t &Op0) {
diff --git a/llvm/lib/Analysis/LoopAccessAnalysis.cpp b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
index ce0b5b3e63c20..3b2e3b792c875 100644
--- a/llvm/lib/Analysis/LoopAccessAnalysis.cpp
+++ b/llvm/lib/Analysis/LoopAccessAnalysis.cpp
@@ -3035,22 +3035,30 @@ bool LoopAccessInfo::isInvariant(Value *V) const {
/// Get the stride of a pointer access in a loop. Looks for symbolic
/// strides "a[i*stride]". Returns the symbolic stride, or null otherwise.
-static const SCEVUnknown *getStrideFromPointer(Value *Ptr, ScalarEvolution *SE,
- Loop *Lp) {
+static const SCEV *getStrideFromPointer(Value *Ptr, ScalarEvolution *SE,
+ Loop *Lp) {
assert(Ptr->getType()->isPointerTy() && "Pointer type expected");
const SCEV *V = SE->removePointerBase(SE->getSCEV(Ptr));
- bool Changed = true;
- while (Changed) {
- Changed = match(V, m_scev_AffineAddRec(m_SCEV(), m_SCEV(V),
- m_SpecificLoop(Lp))) ||
- match(V, m_scev_IntegralCast(m_SCEV(V)));
- if (Changed)
- match(V, m_scev_Mul(m_SCEVConstant(), m_SCEV(V)));
+ // A stride to speculate is either an AddRec of a optional Mul with a
+ // constant, or a Mul of a constant with an AddRec, ignoring casts in AddRec
+ // steps. We strip casts off the Mul in the caller.
+ if (match(V,
+ m_scev_AffineAddRec(m_SCEV(), m_scev_IntegralCastOrSelf(m_SCEV(V)),
+ m_SpecificLoop(Lp)))) {
+ match(V, m_scev_Mul(m_SCEVConstant(), m_SCEV(V)));
+ } else if (!match(V, m_scev_Mul(m_SCEVConstant(), m_SCEV(V))) ||
+ !match(V, m_scev_AffineAddRec(m_SCEV(),
+ m_scev_IntegralCastOrSelf(m_SCEV(V)),
+ m_SpecificLoop(Lp)))) {
+ return nullptr;
}
// It wouldn't make sense to speculate on a loop-varying stride.
- return SE->isLoopInvariant(V, Lp) ? dyn_cast<SCEVUnknown>(V) : nullptr;
+ return match(V, m_scev_IntegralCastOrSelf(m_SCEVUnknown())) &&
+ SE->isLoopInvariant(V, Lp)
+ ? V
+ : nullptr;
}
void LoopAccessInfo::collectStridedAccess(Value *MemAccess) {
@@ -3064,9 +3072,11 @@ void LoopAccessInfo::collectStridedAccess(Value *MemAccess) {
// computation of an interesting IV - but we chose not to as we
// don't have a cost model here, and broadening the scope exposes
// far too many unprofitable cases.
- const SCEVUnknown *StrideExpr =
- getStrideFromPointer(Ptr, PSE->getSE(), TheLoop);
- if (!StrideExpr || match(StrideExpr, m_scev_UndefOrPoison()))
+ const SCEV *StrideExpr = getStrideFromPointer(Ptr, PSE->getSE(), TheLoop);
+ if (!StrideExpr)
+ return;
+
+ if (match(StrideExpr, m_scev_UndefOrPoison()))
return;
LLVM_DEBUG(dbgs() << "LAA: Found a strided access that is a candidate for "
@@ -3118,7 +3128,13 @@ void LoopAccessInfo::collectStridedAccess(Value *MemAccess) {
return;
}
LLVM_DEBUG(dbgs() << "LAA: Found a strided access that we can version.\n");
- SymbolicStrides[Ptr] = StrideExpr;
+
+ // Strip back off the integer cast, and check that our result is a
+ // SCEVUnknown as we expect.
+ const SCEV *StrideBase = StrideExpr;
+ if (const auto *C = dyn_cast<SCEVIntegralCastExpr>(StrideBase))
+ StrideBase = C->getOperand();
+ SymbolicStrides[Ptr] = cast<SCEVUnknown>(StrideBase);
}
LoopAccessInfo::LoopAccessInfo(Loop *L, ScalarEvolution *SE,
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/pr60831-sve-inv-store-crash.ll b/llvm/test/Transforms/LoopVectorize/AArch64/pr60831-sve-inv-store-crash.ll
index af61e38a5908e..6bc2c098576c7 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/pr60831-sve-inv-store-crash.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/pr60831-sve-inv-store-crash.ll
@@ -343,9 +343,6 @@ define void @test_loop2(i64 %n, ptr %dst) {
; CHECK-LABEL: @test_loop2(
; CHECK-NEXT: iter.check:
; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
-; CHECK: vector.scevcheck:
-; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[N:%.*]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK1:%.*]]
; CHECK: vector.main.loop.iter.check:
; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
@@ -367,22 +364,22 @@ define void @test_loop2(i64 %n, ptr %dst) {
; CHECK-NEXT: [[TMP13:%.*]] = add i64 [[INDEX]], 13
; CHECK-NEXT: [[TMP14:%.*]] = add i64 [[INDEX]], 14
; CHECK-NEXT: [[TMP15:%.*]] = add i64 [[INDEX]], 15
-; CHECK-NEXT: [[TMP16:%.*]] = sub nsw i64 1, [[INDEX]]
-; CHECK-NEXT: [[TMP17:%.*]] = sub nsw i64 1, [[TMP1]]
-; CHECK-NEXT: [[TMP18:%.*]] = sub nsw i64 1, [[TMP2]]
-; CHECK-NEXT: [[TMP19:%.*]] = sub nsw i64 1, [[TMP3]]
-; CHECK-NEXT: [[TMP20:%.*]] = sub nsw i64 1, [[TMP4]]
-; CHECK-NEXT: [[TMP21:%.*]] = sub nsw i64 1, [[TMP5]]
-; CHECK-NEXT: [[TMP22:%.*]] = sub nsw i64 1, [[TMP6]]
-; CHECK-NEXT: [[TMP23:%.*]] = sub nsw i64 1, [[TMP7]]
-; CHECK-NEXT: [[TMP24:%.*]] = sub nsw i64 1, [[TMP8]]
-; CHECK-NEXT: [[TMP25:%.*]] = sub nsw i64 1, [[TMP9]]
-; CHECK-NEXT: [[TMP26:%.*]] = sub nsw i64 1, [[TMP10]]
-; CHECK-NEXT: [[TMP27:%.*]] = sub nsw i64 1, [[TMP11]]
-; CHECK-NEXT: [[TMP28:%.*]] = sub nsw i64 1, [[TMP12]]
-; CHECK-NEXT: [[TMP29:%.*]] = sub nsw i64 1, [[TMP13]]
-; CHECK-NEXT: [[TMP30:%.*]] = sub nsw i64 1, [[TMP14]]
-; CHECK-NEXT: [[TMP31:%.*]] = sub nsw i64 1, [[TMP15]]
+; CHECK-NEXT: [[TMP16:%.*]] = sub nsw i64 [[N:%.*]], [[INDEX]]
+; CHECK-NEXT: [[TMP17:%.*]] = sub nsw i64 [[N]], [[TMP1]]
+; CHECK-NEXT: [[TMP18:%.*]] = sub nsw i64 [[N]], [[TMP2]]
+; CHECK-NEXT: [[TMP19:%.*]] = sub nsw i64 [[N]], [[TMP3]]
+; CHECK-NEXT: [[TMP20:%.*]] = sub nsw i64 [[N]], [[TMP4]]
+; CHECK-NEXT: [[TMP21:%.*]] = sub nsw i64 [[N]], [[TMP5]]
+; CHECK-NEXT: [[TMP22:%.*]] = sub nsw i64 [[N]], [[TMP6]]
+; CHECK-NEXT: [[TMP23:%.*]] = sub nsw i64 [[N]], [[TMP7]]
+; CHECK-NEXT: [[TMP24:%.*]] = sub nsw i64 [[N]], [[TMP8]]
+; CHECK-NEXT: [[TMP25:%.*]] = sub nsw i64 [[N]], [[TMP9]]
+; CHECK-NEXT: [[TMP26:%.*]] = sub nsw i64 [[N]], [[TMP10]]
+; CHECK-NEXT: [[TMP27:%.*]] = sub nsw i64 [[N]], [[TMP11]]
+; CHECK-NEXT: [[TMP28:%.*]] = sub nsw i64 [[N]], [[TMP12]]
+; CHECK-NEXT: [[TMP29:%.*]] = sub nsw i64 [[N]], [[TMP13]]
+; CHECK-NEXT: [[TMP30:%.*]] = sub nsw i64 [[N]], [[TMP14]]
+; CHECK-NEXT: [[TMP31:%.*]] = sub nsw i64 [[N]], [[TMP15]]
; CHECK-NEXT: [[TMP32:%.*]] = insertelement <16 x i64> poison, i64 [[TMP16]], i32 0
; CHECK-NEXT: [[TMP33:%.*]] = insertelement <16 x i64> [[TMP32]], i64 [[TMP17]], i32 1
; CHECK-NEXT: [[TMP34:%.*]] = insertelement <16 x i64> [[TMP33]], i64 [[TMP18]], i32 2
@@ -412,7 +409,7 @@ define void @test_loop2(i64 %n, ptr %dst) {
; CHECK: vec.epilog.iter.check:
; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF10:![0-9]+]]
; CHECK: vec.epilog.ph:
-; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK1]] ]
+; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 992, [[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]] ]
@@ -423,14 +420,14 @@ define void @test_loop2(i64 %n, ptr %dst) {
; CHECK-NEXT: [[TMP58:%.*]] = add i64 [[INDEX2]], 5
; CHECK-NEXT: [[TMP59:%.*]] = add i64 [[INDEX2]], 6
; CHECK-NEXT: [[TMP60:%.*]] = add i64 [[INDEX2]], 7
-; CHECK-NEXT: [[TMP61:%.*]] = sub nsw i64 1, [[INDEX2]]
-; CHECK-NEXT: [[TMP62:%.*]] = sub nsw i64 1, [[TMP54]]
-; CHECK-NEXT: [[TMP63:%.*]] = sub nsw i64 1, [[TMP55]]
-; CHECK-NEXT: [[TMP64:%.*]] = sub nsw i64 1, [[TMP56]]
-; CHECK-NEXT: [[TMP65:%.*]] = sub nsw i64 1, [[TMP57]]
-; CHECK-NEXT: [[TMP66:%.*]] = sub nsw i64 1, [[TMP58]]
-; CHECK-NEXT: [[TMP67:%.*]] = sub nsw i64 1, [[TMP59]]
-; CHECK-NEXT: [[TMP68:%.*]] = sub nsw i64 1, [[TMP60]]
+; CHECK-NEXT: [[TMP61:%.*]] = sub nsw i64 [[N]], [[INDEX2]]
+; CHECK-NEXT: [[TMP62:%.*]] = sub nsw i64 [[N]], [[TMP54]]
+; CHECK-NEXT: [[TMP63:%.*]] = sub nsw i64 [[N]], [[TMP55]]
+; CHECK-NEXT: [[TMP64:%.*]] = sub nsw i64 [[N]], [[TMP56]]
+; CHECK-NEXT: [[TMP65:%.*]] = sub nsw i64 [[N]], [[TMP57]]
+; CHECK-NEXT: [[TMP66:%.*]] = sub nsw i64 [[N]], [[TMP58]]
+; CHECK-NEXT: [[TMP67:%.*]] = sub nsw i64 [[N]], [[TMP59]]
+; CHECK-NEXT: [[TMP68:%.*]] = sub nsw i64 [[N]], [[TMP60]]
; CHECK-NEXT: [[TMP69:%.*]] = insertelement <8 x i64> poison, i64 [[TMP61]], i32 0
; CHECK-NEXT: [[TMP70:%.*]] = insertelement <8 x i64> [[TMP69]], i64 [[TMP62]], i32 1
; CHECK-NEXT: [[TMP71:%.*]] = insertelement <8 x i64> [[TMP70]], i64 [[TMP63]], i32 2
@@ -450,7 +447,7 @@ define void @test_loop2(i64 %n, ptr %dst) {
; 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_VAL:%.*]] = phi i64 [ 1000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
; CHECK-NEXT: br label [[LOOP:%.*]]
; CHECK: loop:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
@@ -468,9 +465,6 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2-LABEL: @test_loop2(
; IC2-NEXT: iter.check:
; IC2-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
-; IC2: vector.scevcheck:
-; IC2-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i64 [[N:%.*]], 1
-; IC2-NEXT: br i1 [[IDENT_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VECTOR_MAIN_LOOP_ITER_CHECK1:%.*]]
; IC2: vector.main.loop.iter.check:
; IC2-NEXT: br i1 false, label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
; IC2: vector.ph:
@@ -508,22 +502,22 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2-NEXT: [[TMP29:%.*]] = add i64 [[INDEX]], 29
; IC2-NEXT: [[TMP30:%.*]] = add i64 [[INDEX]], 30
; IC2-NEXT: [[TMP31:%.*]] = add i64 [[INDEX]], 31
-; IC2-NEXT: [[TMP32:%.*]] = sub nsw i64 1, [[INDEX]]
-; IC2-NEXT: [[TMP33:%.*]] = sub nsw i64 1, [[TMP1]]
-; IC2-NEXT: [[TMP34:%.*]] = sub nsw i64 1, [[TMP2]]
-; IC2-NEXT: [[TMP35:%.*]] = sub nsw i64 1, [[TMP3]]
-; IC2-NEXT: [[TMP36:%.*]] = sub nsw i64 1, [[TMP4]]
-; IC2-NEXT: [[TMP37:%.*]] = sub nsw i64 1, [[TMP5]]
-; IC2-NEXT: [[TMP38:%.*]] = sub nsw i64 1, [[TMP6]]
-; IC2-NEXT: [[TMP39:%.*]] = sub nsw i64 1, [[TMP7]]
-; IC2-NEXT: [[TMP40:%.*]] = sub nsw i64 1, [[TMP8]]
-; IC2-NEXT: [[TMP41:%.*]] = sub nsw i64 1, [[TMP9]]
-; IC2-NEXT: [[TMP42:%.*]] = sub nsw i64 1, [[TMP10]]
-; IC2-NEXT: [[TMP43:%.*]] = sub nsw i64 1, [[TMP11]]
-; IC2-NEXT: [[TMP44:%.*]] = sub nsw i64 1, [[TMP12]]
-; IC2-NEXT: [[TMP45:%.*]] = sub nsw i64 1, [[TMP13]]
-; IC2-NEXT: [[TMP46:%.*]] = sub nsw i64 1, [[TMP14]]
-; IC2-NEXT: [[TMP47:%.*]] = sub nsw i64 1, [[TMP15]]
+; IC2-NEXT: [[TMP32:%.*]] = sub nsw i64 [[N:%.*]], [[INDEX]]
+; IC2-NEXT: [[TMP33:%.*]] = sub nsw i64 [[N]], [[TMP1]]
+; IC2-NEXT: [[TMP34:%.*]] = sub nsw i64 [[N]], [[TMP2]]
+; IC2-NEXT: [[TMP35:%.*]] = sub nsw i64 [[N]], [[TMP3]]
+; IC2-NEXT: [[TMP36:%.*]] = sub nsw i64 [[N]], [[TMP4]]
+; IC2-NEXT: [[TMP37:%.*]] = sub nsw i64 [[N]], [[TMP5]]
+; IC2-NEXT: [[TMP38:%.*]] = sub nsw i64 [[N]], [[TMP6]]
+; IC2-NEXT: [[TMP39:%.*]] = sub nsw i64 [[N]], [[TMP7]]
+; IC2-NEXT: [[TMP40:%.*]] = sub nsw i64 [[N]], [[TMP8]]
+; IC2-NEXT: [[TMP41:%.*]] = sub nsw i64 [[N]], [[TMP9]]
+; IC2-NEXT: [[TMP42:%.*]] = sub nsw i64 [[N]], [[TMP10]]
+; IC2-NEXT: [[TMP43:%.*]] = sub nsw i64 [[N]], [[TMP11]]
+; IC2-NEXT: [[TMP44:%.*]] = sub nsw i64 [[N]], [[TMP12]]
+; IC2-NEXT: [[TMP45:%.*]] = sub nsw i64 [[N]], [[TMP13]]
+; IC2-NEXT: [[TMP46:%.*]] = sub nsw i64 [[N]], [[TMP14]]
+; IC2-NEXT: [[TMP47:%.*]] = sub nsw i64 [[N]], [[TMP15]]
; IC2-NEXT: [[TMP48:%.*]] = insertelement <16 x i64> poison, i64 [[TMP32]], i32 0
; IC2-NEXT: [[TMP49:%.*]] = insertelement <16 x i64> [[TMP48]], i64 [[TMP33]], i32 1
; IC2-NEXT: [[TMP50:%.*]] = insertelement <16 x i64> [[TMP49]], i64 [[TMP34]], i32 2
@@ -540,22 +534,22 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2-NEXT: [[TMP61:%.*]] = insertelement <16 x i64> [[TMP60]], i64 [[TMP45]], i32 13
; IC2-NEXT: [[TMP62:%.*]] = insertelement <16 x i64> [[TMP61]], i64 [[TMP46]], i32 14
; IC2-NEXT: [[TMP63:%.*]] = insertelement <16 x i64> [[TMP62]], i64 [[TMP47]], i32 15
-; IC2-NEXT: [[TMP64:%.*]] = sub nsw i64 1, [[TMP16]]
-; IC2-NEXT: [[TMP65:%.*]] = sub nsw i64 1, [[TMP17]]
-; IC2-NEXT: [[TMP66:%.*]] = sub nsw i64 1, [[TMP18]]
-; IC2-NEXT: [[TMP67:%.*]] = sub nsw i64 1, [[TMP19]]
-; IC2-NEXT: [[TMP68:%.*]] = sub nsw i64 1, [[TMP20]]
-; IC2-NEXT: [[TMP69:%.*]] = sub nsw i64 1, [[TMP21]]
-; IC2-NEXT: [[TMP70:%.*]] = sub nsw i64 1, [[TMP22]]
-; IC2-NEXT: [[TMP71:%.*]] = sub nsw i64 1, [[TMP23]]
-; IC2-NEXT: [[TMP72:%.*]] = sub nsw i64 1, [[TMP24]]
-; IC2-NEXT: [[TMP73:%.*]] = sub nsw i64 1, [[TMP25]]
-; IC2-NEXT: [[TMP74:%.*]] = sub nsw i64 1, [[TMP26]]
-; IC2-NEXT: [[TMP75:%.*]] = sub nsw i64 1, [[TMP27]]
-; IC2-NEXT: [[TMP76:%.*]] = sub nsw i64 1, [[TMP28]]
-; IC2-NEXT: [[TMP77:%.*]] = sub nsw i64 1, [[TMP29]]
-; IC2-NEXT: [[TMP78:%.*]] = sub nsw i64 1, [[TMP30]]
-; IC2-NEXT: [[TMP79:%.*]] = sub nsw i64 1, [[TMP31]]
+; IC2-NEXT: [[TMP64:%.*]] = sub nsw i64 [[N]], [[TMP16]]
+; IC2-NEXT: [[TMP65:%.*]] = sub nsw i64 [[N]], [[TMP17]]
+; IC2-NEXT: [[TMP66:%.*]] = sub nsw i64 [[N]], [[TMP18]]
+; IC2-NEXT: [[TMP67:%.*]] = sub nsw i64 [[N]], [[TMP19]]
+; IC2-NEXT: [[TMP68:%.*]] = sub nsw i64 [[N]], [[TMP20]]
+; IC2-NEXT: [[TMP69:%.*]] = sub nsw i64 [[N]], [[TMP21]]
+; IC2-NEXT: [[TMP70:%.*]] = sub nsw i64 [[N]], [[TMP22]]
+; IC2-NEXT: [[TMP71:%.*]] = sub nsw i64 [[N]], [[TMP23]]
+; IC2-NEXT: [[TMP72:%.*]] = sub nsw i64 [[N]], [[TMP24]]
+; IC2-NEXT: [[TMP73:%.*]] = sub nsw i64 [[N]], [[TMP25]]
+; IC2-NEXT: [[TMP74:%.*]] = sub nsw i64 [[N]], [[TMP26]]
+; IC2-NEXT: [[TMP75:%.*]] = sub nsw i64 [[N]], [[TMP27]]
+; IC2-NEXT: [[TMP76:%.*]] = sub nsw i64 [[N]], [[TMP28]]
+; IC2-NEXT: [[TMP77:%.*]] = sub nsw i64 [[N]], [[TMP29]]
+; IC2-NEXT: [[TMP78:%.*]] = sub nsw i64 [[N]], [[TMP30]]
+; IC2-NEXT: [[TMP79:%.*]] = sub nsw i64 [[N]], [[TMP31]]
; IC2-NEXT: [[TMP80:%.*]] = insertelement <16 x i64> poison, i64 [[TMP64]], i32 0
; IC2-NEXT: [[TMP81:%.*]] = insertelement <16 x i64> [[TMP80]], i64 [[TMP65]], i32 1
; IC2-NEXT: [[TMP82:%.*]] = insertelement <16 x i64> [[TMP81]], i64 [[TMP66]], i32 2
@@ -590,7 +584,7 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2: vec.epilog.iter.check:
; IC2-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF10:![0-9]+]]
; IC2: vec.epilog.ph:
-; IC2-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK1]] ]
+; IC2-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; IC2-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; IC2: vec.epilog.vector.body:
; IC2-NEXT: [[INDEX1:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT2:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
@@ -601,14 +595,14 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2-NEXT: [[TMP110:%.*]] = add i64 [[INDEX1]], 5
; IC2-NEXT: [[TMP111:%.*]] = add i64 [[INDEX1]], 6
; IC2-NEXT: [[TMP112:%.*]] = add i64 [[INDEX1]], 7
-; IC2-NEXT: [[TMP113:%.*]] = sub nsw i64 1, [[INDEX1]]
-; IC2-NEXT: [[TMP114:%.*]] = sub nsw i64 1, [[TMP106]]
-; IC2-NEXT: [[TMP115:%.*]] = sub nsw i64 1, [[TMP107]]
-; IC2-NEXT: [[TMP116:%.*]] = sub nsw i64 1, [[TMP108]]
-; IC2-NEXT: [[TMP117:%.*]] = sub nsw i64 1, [[TMP109]]
-; IC2-NEXT: [[TMP118:%.*]] = sub nsw i64 1, [[TMP110]]
-; IC2-NEXT: [[TMP119:%.*]] = sub nsw i64 1, [[TMP111]]
-; IC2-NEXT: [[TMP120:%.*]] = sub nsw i64 1, [[TMP112]]
+; IC2-NEXT: [[TMP113:%.*]] = sub nsw i64 [[N]], [[INDEX1]]
+; IC2-NEXT: [[TMP114:%.*]] = sub nsw i64 [[N]], [[TMP106]]
+; IC2-NEXT: [[TMP115:%.*]] = sub nsw i64 [[N]], [[TMP107]]
+; IC2-NEXT: [[TMP116:%.*]] = sub nsw i64 [[N]], [[TMP108]]
+; IC2-NEXT: [[TMP117:%.*]] = sub nsw i64 [[N]], [[TMP109]]
+; IC2-NEXT: [[TMP118:%.*]] = sub nsw i64 [[N]], [[TMP110]]
+; IC2-NEXT: [[TMP119:%.*]] = sub nsw i64 [[N]], [[TMP111]]
+; IC2-NEXT: [[TMP120:%.*]] = sub nsw i64 [[N]], [[TMP112]]
; IC2-NEXT: [[TMP121:%.*]] = insertelement <8 x i64> poison, i64 [[TMP113]], i32 0
; IC2-NEXT: [[TMP122:%.*]] = insertelement <8 x i64> [[TMP121]], i64 [[TMP114]], i32 1
; IC2-NEXT: [[TMP123:%.*]] = insertelement <8 x i64> [[TMP122]], i64 [[TMP115]], i32 2
@@ -628,7 +622,7 @@ define void @test_loop2(i64 %n, ptr %dst) {
; IC2: vec.epilog.middle.block:
; IC2-NEXT: br i1 false, label [[EXIT]], label [[VEC_EPILOG_SCALAR_PH]]
; IC2: vec.epilog.scalar.ph:
-; IC2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
+; IC2-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1000, [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 992, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK:%.*]] ]
; IC2-NEXT: br label [[LOOP:%.*]]
; IC2: loop:
; IC2-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
diff --git a/llvm/test/Transforms/LoopVectorize/pr34681.ll b/llvm/test/Transforms/LoopVectorize/pr34681.ll
index ce942fefa6e78..7fb239d0d013d 100644
--- a/llvm/test/Transforms/LoopVectorize/pr34681.ll
+++ b/llvm/test/Transforms/LoopVectorize/pr34681.ll
@@ -139,20 +139,20 @@ define i32 @foo2(i16 zeroext %N, ptr nocapture readnone %A, ptr nocapture readon
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[CONV:%.*]] = zext i16 [[N]] to i32
; CHECK-NEXT: [[CMP11:%.*]] = icmp eq i16 [[N]], 0
-; CHECK-NEXT: br i1 [[CMP11]], [[FOR_END:label %.*]], label %[[FOR_BODY_LR_PH:.*]]
-; CHECK: [[FOR_BODY_LR_PH]]:
+; CHECK-NEXT: br i1 [[CMP11]], [[FOR_END:label %.*]], label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[CONV]], 4
-; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH1:.*]], label %[[VECTOR_SCEVCHECK:.*]]
; CHECK: [[VECTOR_SCEVCHECK]]:
; CHECK-NEXT: [[TMP0:%.*]] = add nsw i32 [[CONV]], -1
; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[CONV]], i32 [[TMP0]])
; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[J]], [[MUL_RESULT]]
-; CHECK-NEXT: [[TMP3:%.*]] = icmp slt i32 [[TMP1]], [[J]]
-; CHECK-NEXT: [[TMP4:%.*]] = or i1 [[TMP3]], [[MUL_OVERFLOW]]
-; CHECK-NEXT: br i1 [[TMP4]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
-; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[TMP2:%.*]] = icmp slt i32 [[TMP1]], [[J]]
+; CHECK-NEXT: [[TMP3:%.*]] = or i1 [[TMP2]], [[MUL_OVERFLOW]]
+; CHECK-NEXT: br i1 [[TMP3]], label %[[SCALAR_PH1]], label %[[VECTOR_PH1:.*]]
+; CHECK: [[VECTOR_PH1]]:
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[CONV]], 4
; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[CONV]], [[N_MOD_VF]]
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i32> poison, i32 [[CONV]], i64 0
@@ -161,38 +161,38 @@ define i32 @foo2(i16 zeroext %N, ptr nocapture readnone %A, ptr nocapture readon
; CHECK-NEXT: [[BROADCAST_SPLAT3:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT2]], <4 x i32> poison, <4 x i32> zeroinitializer
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH]] ], [ [[TMP26:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP7:%.*]] = mul nuw <4 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; CHECK-NEXT: [[TMP8:%.*]] = add <4 x i32> [[TMP7]], [[BROADCAST_SPLAT3]]
-; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i32> [[TMP8]], i64 0
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP9]]
-; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x i32> [[TMP8]], i64 1
-; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP11]]
-; CHECK-NEXT: [[TMP13:%.*]] = extractelement <4 x i32> [[TMP8]], i64 2
-; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP13]]
-; CHECK-NEXT: [[TMP15:%.*]] = extractelement <4 x i32> [[TMP8]], i64 3
+; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH1]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %[[VECTOR_PH1]] ], [ [[TMP23:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH1]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP4:%.*]] = mul nuw <4 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP5:%.*]] = add <4 x i32> [[TMP4]], [[BROADCAST_SPLAT3]]
+; CHECK-NEXT: [[TMP15:%.*]] = extractelement <4 x i32> [[TMP5]], i64 0
; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP15]]
-; CHECK-NEXT: [[TMP17:%.*]] = load i16, ptr [[TMP10]], align 2
-; CHECK-NEXT: [[TMP18:%.*]] = load i16, ptr [[TMP12]], align 2
-; CHECK-NEXT: [[TMP19:%.*]] = load i16, ptr [[TMP14]], align 2
+; CHECK-NEXT: [[TMP8:%.*]] = extractelement <4 x i32> [[TMP5]], i64 1
+; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP8]]
+; CHECK-NEXT: [[TMP10:%.*]] = extractelement <4 x i32> [[TMP5]], i64 2
+; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP10]]
+; CHECK-NEXT: [[TMP12:%.*]] = extractelement <4 x i32> [[TMP5]], i64 3
+; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i16, ptr [[B]], i32 [[TMP12]]
; CHECK-NEXT: [[TMP20:%.*]] = load i16, ptr [[TMP16]], align 2
-; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i16> poison, i16 [[TMP17]], i32 0
-; CHECK-NEXT: [[TMP22:%.*]] = insertelement <4 x i16> [[TMP21]], i16 [[TMP18]], i32 1
-; CHECK-NEXT: [[TMP23:%.*]] = insertelement <4 x i16> [[TMP22]], i16 [[TMP19]], i32 2
-; CHECK-NEXT: [[TMP24:%.*]] = insertelement <4 x i16> [[TMP23]], i16 [[TMP20]], i32 3
-; CHECK-NEXT: [[TMP25:%.*]] = sext <4 x i16> [[TMP24]] to <4 x i32>
-; CHECK-NEXT: [[TMP26]] = add <4 x i32> [[VEC_PHI]], [[TMP25]]
+; CHECK-NEXT: [[TMP26:%.*]] = load i16, ptr [[TMP9]], align 2
+; CHECK-NEXT: [[TMP27:%.*]] = load i16, ptr [[TMP11]], align 2
+; CHECK-NEXT: [[TMP17:%.*]] = load i16, ptr [[TMP13]], align 2
+; CHECK-NEXT: [[TMP18:%.*]] = insertelement <4 x i16> poison, i16 [[TMP20]], i32 0
+; CHECK-NEXT: [[TMP19:%.*]] = insertelement <4 x i16> [[TMP18]], i16 [[TMP26]], i32 1
+; CHECK-NEXT: [[TMP28:%.*]] = insertelement <4 x i16> [[TMP19]], i16 [[TMP27]], i32 2
+; CHECK-NEXT: [[TMP21:%.*]] = insertelement <4 x i16> [[TMP28]], i16 [[TMP17]], i32 3
+; CHECK-NEXT: [[TMP22:%.*]] = sext <4 x i16> [[TMP21]] to <4 x i32>
+; CHECK-NEXT: [[TMP23]] = add <4 x i32> [[VEC_PHI]], [[TMP22]]
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <4 x i32> [[VEC_IND]], splat (i32 4)
-; CHECK-NEXT: [[TMP27:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[TMP27]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
-; CHECK: [[MIDDLE_BLOCK]]:
-; CHECK-NEXT: [[TMP28:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP26]])
-; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[CONV]], [[N_VEC]]
-; CHECK-NEXT: br i1 [[CMP_N]], [[FOR_END_LOOPEXIT:label %.*]], label %[[SCALAR_PH]]
+; CHECK-NEXT: [[TMP24:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP24]], label %[[SCALAR_PH:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: [[SCALAR_PH]]:
+; CHECK-NEXT: [[TMP25:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP23]])
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[CONV]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], [[FOR_END_LOOPEXIT:label %.*]], label %[[SCALAR_PH1]]
+; CHECK: [[SCALAR_PH1]]:
;
entry:
%conv = zext i16 %N to i32
diff --git a/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll b/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
index 06e9554937a0e..d3925e8f225f0 100644
--- a/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
+++ b/llvm/test/Transforms/LoopVectorize/runtime-check-needed-but-empty.ll
@@ -4,23 +4,23 @@
define void @test(ptr %A, i32 %x) {
; CHECK-LABEL: define void @test(
; CHECK-SAME: ptr [[A:%.*]], i32 [[X:%.*]]) {
-; CHECK-NEXT: [[SCALAR_PH:.*]]:
-; CHECK-NEXT: br label %[[LOOP:.*]]
-; CHECK: [[LOOP]]:
-; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[IV_NEXT:%.*]], %[[LOOP]] ], [ 0, %[[SCALAR_PH]] ]
-; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
-; CHECK-NEXT: [[T_IV_NEXT:%.*]] = trunc i64 [[IV_NEXT]] to i32
-; CHECK-NEXT: [[MUL_IV_NEXT:%.*]] = mul i32 [[T_IV_NEXT]], [[X]]
+; CHECK-NEXT: [[SCALAR_PH1:.*]]:
+; CHECK-NEXT: br label %[[LOOP1:.*]]
+; CHECK: [[LOOP1]]:
+; CHECK-NEXT: [[IV1:%.*]] = phi i64 [ [[IV_NEXT1:%.*]], %[[LOOP1]] ], [ 0, %[[SCALAR_PH1]] ]
+; CHECK-NEXT: [[IV_NEXT1]] = add nuw nsw i64 [[IV1]], 1
+; CHECK-NEXT: [[T_IV_NEXT1:%.*]] = trunc i64 [[IV_NEXT1]] to i32
+; CHECK-NEXT: [[MUL_IV_NEXT:%.*]] = mul i32 [[T_IV_NEXT1]], [[X]]
; CHECK-NEXT: [[IDX_1:%.*]] = zext i32 [[MUL_IV_NEXT]] to i64
; CHECK-NEXT: [[ARRAYIDX1215:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IDX_1]]
; CHECK-NEXT: [[LV:%.*]] = load float, ptr [[ARRAYIDX1215]], align 4
-; CHECK-NEXT: [[T_IV:%.*]] = trunc i64 [[IV]] to i32
+; CHECK-NEXT: [[T_IV:%.*]] = trunc i64 [[IV1]] to i32
; CHECK-NEXT: [[MUL_IV:%.*]] = mul i32 [[T_IV]], [[X]]
; CHECK-NEXT: [[IDX_2:%.*]] = zext i32 [[MUL_IV]] to i64
; CHECK-NEXT: [[ARRAYIDX1209:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IDX_2]]
; CHECK-NEXT: store float [[LV]], ptr [[ARRAYIDX1209]], align 4
-; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], 1000
-; CHECK-NEXT: br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP]]
+; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT1]], 1000
+; CHECK-NEXT: br i1 [[EC]], label %[[EXIT:.*]], label %[[LOOP1]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
diff --git a/llvm/test/Transforms/LoopVectorize/version-mem-access.ll b/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
index c533c95a444f1..a0c77a964cc86 100644
--- a/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
+++ b/llvm/test/Transforms/LoopVectorize/version-mem-access.ll
@@ -21,8 +21,8 @@ define void @test(ptr noalias %A, i64 %AStride, ptr noalias %B, i32 %BStride, pt
; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
; CHECK-NEXT: [[TMP3:%.*]] = sub i32 0, [[MUL_RESULT]]
; CHECK-NEXT: [[TMP4:%.*]] = icmp ugt i32 [[TMP3]], 0
-; CHECK-NEXT: [[TMP11:%.*]] = select i1 [[TMP1]], i1 [[TMP4]], i1 false
-; CHECK-NEXT: [[IDENT_CHECK1:%.*]] = or i1 [[TMP11]], [[MUL_OVERFLOW]]
+; CHECK-NEXT: [[TMP5:%.*]] = select i1 [[TMP1]], i1 [[TMP4]], i1 false
+; CHECK-NEXT: [[IDENT_CHECK1:%.*]] = or i1 [[TMP5]], [[MUL_OVERFLOW]]
; CHECK-NEXT: [[IDENT_CHECK2:%.*]] = icmp ne i64 [[CSTRIDE]], 1
; CHECK-NEXT: [[DIFF_CHECK6:%.*]] = icmp ne i64 [[ASTRIDE]], 1
; CHECK-NEXT: [[DIFF_CHECK:%.*]] = or i1 [[IDENT_CHECK1]], [[IDENT_CHECK2]]
@@ -35,14 +35,14 @@ define void @test(ptr noalias %A, i64 %AStride, ptr noalias %B, i32 %BStride, pt
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT: [[TMP12:%.*]] = mul <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
-; CHECK-NEXT: [[TMP18:%.*]] = zext <2 x i32> [[TMP12]] to <2 x i64>
-; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i64> [[TMP18]], i64 0
-; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP5]]
-; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i64> [[TMP18]], i64 1
+; CHECK-NEXT: [[TMP11:%.*]] = mul <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP12:%.*]] = zext <2 x i32> [[TMP11]] to <2 x i64>
+; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i64> [[TMP12]], i64 0
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP13]]
-; CHECK-NEXT: [[TMP15:%.*]] = load i32, ptr [[TMP6]], align 4
-; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP14]], align 4
+; CHECK-NEXT: [[TMP18:%.*]] = extractelement <2 x i64> [[TMP12]], i64 1
+; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP18]]
+; CHECK-NEXT: [[TMP15:%.*]] = load i32, ptr [[TMP14]], align 4
+; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[TMP19]], align 4
; CHECK-NEXT: [[TMP17:%.*]] = insertelement <2 x i32> poison, i32 [[TMP15]], i32 0
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = insertelement <2 x i32> [[TMP17]], i32 [[TMP16]], i32 1
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[C]], i64 [[INDEX]]
@@ -101,16 +101,16 @@ define void @fn1(ptr noalias %x, ptr noalias %c, double %a) {
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]]
; CHECK: [[VECTOR_SCEVCHECK]]:
; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[CONV]], 3
-; CHECK-NEXT: [[TMP9:%.*]] = sub i32 0, [[CONV]]
-; CHECK-NEXT: [[TMP12:%.*]] = icmp slt i32 [[CONV]], 0
-; CHECK-NEXT: [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP9]], i32 [[CONV]]
-; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP13]], i32 [[TMP1]])
+; CHECK-NEXT: [[TMP2:%.*]] = sub i32 0, [[CONV]]
+; CHECK-NEXT: [[TMP3:%.*]] = icmp slt i32 [[CONV]], 0
+; CHECK-NEXT: [[TMP9:%.*]] = select i1 [[TMP3]], i32 [[TMP2]], i32 [[CONV]]
+; CHECK-NEXT: [[MUL1:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP9]], i32 [[TMP1]])
; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL1]], 0
; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL1]], 1
; CHECK-NEXT: [[TMP5:%.*]] = sub i32 0, [[MUL_RESULT]]
; CHECK-NEXT: [[TMP6:%.*]] = icmp slt i32 [[MUL_RESULT]], 0
; CHECK-NEXT: [[TMP7:%.*]] = icmp sgt i32 [[TMP5]], 0
-; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP12]], i1 [[TMP7]], i1 [[TMP6]]
+; CHECK-NEXT: [[TMP8:%.*]] = select i1 [[TMP3]], i1 [[TMP7]], i1 [[TMP6]]
; CHECK-NEXT: [[IDENT_CHECK:%.*]] = or i1 [[TMP8]], [[MUL_OVERFLOW]]
; CHECK-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
@@ -124,12 +124,12 @@ define void @fn1(ptr noalias %x, ptr noalias %c, double %a) {
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP10:%.*]] = mul nsw <2 x i32> [[VEC_IND]], [[BROADCAST_SPLAT]]
; CHECK-NEXT: [[TMP11:%.*]] = sext <2 x i32> [[TMP10]] to <2 x i64>
-; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i64> [[TMP11]], i64 0
-; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds double, ptr [[X]], i64 [[TMP2]]
-; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i64> [[TMP11]], i64 1
+; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x i64> [[TMP11]], i64 0
; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds double, ptr [[X]], i64 [[TMP14]]
-; CHECK-NEXT: [[TMP16:%.*]] = load double, ptr [[TMP3]], align 8
-; CHECK-NEXT: [[TMP17:%.*]] = load double, ptr [[TMP15]], align 8
+; CHECK-NEXT: [[TMP19:%.*]] = extractelement <2 x i64> [[TMP11]], i64 1
+; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds double, ptr [[X]], i64 [[TMP19]]
+; CHECK-NEXT: [[TMP16:%.*]] = load double, ptr [[TMP15]], align 8
+; CHECK-NEXT: [[TMP17:%.*]] = load double, ptr [[TMP20]], align 8
; CHECK-NEXT: [[TMP18:%.*]] = insertelement <2 x double> poison, double [[TMP16]], i32 0
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = insertelement <2 x double> [[TMP18]], double [[TMP17]], i32 1
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds double, ptr [[C]], i64 [[INDEX]]
diff --git a/llvm/test/Transforms/LoopVectorize/version-stride-with-integer-casts.ll b/llvm/test/Transforms/LoopVectorize/version-stride-with-integer-casts.ll
index 92a2f97245e37..a4529a75a6687 100644
--- a/llvm/test/Transforms/LoopVectorize/version-stride-with-integer-casts.ll
+++ b/llvm/test/Transforms/LoopVectorize/version-stride-with-integer-casts.ll
@@ -214,6 +214,7 @@ define void @test_versioned_with_different_uses(i32 %offset, ptr noalias %dst.1,
; CHECK-SAME: i32 [[OFFSET:%.*]], ptr noalias [[DST_1:%.*]], ptr [[DST_2:%.*]]) {
; CHECK-NEXT: entry:
; CHECK-NEXT: [[OFFSET_EXT:%.*]] = zext i32 [[OFFSET]] to i64
+; CHECK-NEXT: [[TMP0:%.*]] = sub i32 0, [[OFFSET]]
; CHECK-NEXT: br label [[OUTER_HEADER:%.*]]
; CHECK: outer.header.loopexit:
; CHECK-NEXT: [[IV_2_NEXT_LCSSA:%.*]] = phi i64 [ [[IV_2_NEXT:%.*]], [[INNER_LOOP:%.*]] ]
@@ -225,8 +226,19 @@ define void @test_versioned_with_different_uses(i32 %offset, ptr noalias %dst.1,
; CHECK: inner.loop.preheader:
; CHECK-NEXT: br label [[VECTOR_SCEVCHECK:%.*]]
; CHECK: vector.scevcheck:
+; CHECK-NEXT: [[TMP1:%.*]] = icmp slt i32 [[OFFSET]], 0
+; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i32 [[TMP0]], i32 [[OFFSET]]
+; CHECK-NEXT: [[MUL:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP2]], i32 200)
+; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL]], 0
+; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL]], 1
+; CHECK-NEXT: [[TMP13:%.*]] = sub i32 0, [[MUL_RESULT]]
+; CHECK-NEXT: [[TMP4:%.*]] = icmp slt i32 [[MUL_RESULT]], 0
+; CHECK-NEXT: [[TMP14:%.*]] = icmp sgt i32 [[TMP13]], 0
+; CHECK-NEXT: [[TMP16:%.*]] = select i1 [[TMP1]], i1 [[TMP14]], i1 [[TMP4]]
+; CHECK-NEXT: [[TMP17:%.*]] = or i1 [[TMP16]], [[MUL_OVERFLOW]]
; CHECK-NEXT: [[IDENT_CHECK:%.*]] = icmp ne i32 [[OFFSET]], 1
-; CHECK-NEXT: br i1 [[IDENT_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK-NEXT: [[TMP18:%.*]] = or i1 [[TMP17]], [[IDENT_CHECK]]
+; CHECK-NEXT: br i1 [[TMP18]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[IND_END:%.*]] = add i64 [[IV_1]], 200
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
diff --git a/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll b/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
index 5a7588d04793b..fa69a07fdd28a 100644
--- a/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
+++ b/llvm/test/Transforms/LoopVectorize/vplan-based-stride-mv.ll
@@ -3826,17 +3826,17 @@ define void @trunc_stride(ptr noalias %p.out, ptr %p, i64 %stride.i64) {
; COMPARE-LAA-MV-NEXT: [[ENTRY:.*:]]
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_SCEVCHECK:.*]]
; COMPARE-LAA-MV: [[VECTOR_SCEVCHECK]]:
-; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = trunc i64 [[STRIDE_I64]] to i32
-; COMPARE-LAA-MV-NEXT: [[TMP11:%.*]] = sub i32 0, [[TMP8]]
-; COMPARE-LAA-MV-NEXT: [[TMP12:%.*]] = icmp slt i32 [[TMP8]], 0
-; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = select i1 [[TMP12]], i32 [[TMP11]], i32 [[TMP8]]
+; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = trunc i64 [[STRIDE_I64]] to i32
+; COMPARE-LAA-MV-NEXT: [[TMP8:%.*]] = sub i32 0, [[TMP0]]
+; COMPARE-LAA-MV-NEXT: [[TMP11:%.*]] = icmp slt i32 [[TMP0]], 0
+; COMPARE-LAA-MV-NEXT: [[TMP3:%.*]] = select i1 [[TMP11]], i32 [[TMP8]], i32 [[TMP0]]
; COMPARE-LAA-MV-NEXT: [[MUL:%.*]] = call { i32, i1 } @llvm.umul.with.overflow.i32(i32 [[TMP3]], i32 127)
; COMPARE-LAA-MV-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i32, i1 } [[MUL]], 0
; COMPARE-LAA-MV-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i32, i1 } [[MUL]], 1
; COMPARE-LAA-MV-NEXT: [[TMP4:%.*]] = sub i32 0, [[MUL_RESULT]]
; COMPARE-LAA-MV-NEXT: [[TMP5:%.*]] = icmp slt i32 [[MUL_RESULT]], 0
; COMPARE-LAA-MV-NEXT: [[TMP6:%.*]] = icmp sgt i32 [[TMP4]], 0
-; COMPARE-LAA-MV-NEXT: [[TMP7:%.*]] = select i1 [[TMP12]], i1 [[TMP6]], i1 [[TMP5]]
+; COMPARE-LAA-MV-NEXT: [[TMP7:%.*]] = select i1 [[TMP11]], i1 [[TMP6]], i1 [[TMP5]]
; COMPARE-LAA-MV-NEXT: [[IDENT_CHECK:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW]]
; COMPARE-LAA-MV-NEXT: br i1 [[IDENT_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
; COMPARE-LAA-MV: [[VECTOR_PH]]:
@@ -3845,28 +3845,28 @@ define void @trunc_stride(ptr noalias %p.out, ptr %p, i64 %stride.i64) {
; COMPARE-LAA-MV-NEXT: [[TMP9:%.*]] = trunc <4 x i64> [[BROADCAST_SPLAT]] to <4 x i32>
; COMPARE-LAA-MV-NEXT: br label %[[VECTOR_BODY:.*]]
; COMPARE-LAA-MV: [[VECTOR_BODY]]:
-; COMPARE-LAA-MV-NEXT: [[INDEX1:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; COMPARE-LAA-MV-NEXT: [[VEC_IND:%.*]] = phi <4 x i32> [ <i32 0, i32 1, i32 2, i32 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
; COMPARE-LAA-MV-NEXT: [[TMP10:%.*]] = mul <4 x i32> [[VEC_IND]], [[TMP9]]
-; COMPARE-LAA-MV-NEXT: [[INDEX:%.*]] = extractelement <4 x i32> [[TMP10]], i64 0
-; COMPARE-LAA-MV-NEXT: [[TMP0:%.*]] = getelementptr i32, ptr [[P]], i32 [[INDEX]]
+; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = extractelement <4 x i32> [[TMP10]], i64 0
+; COMPARE-LAA-MV-NEXT: [[TMP18:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP17]]
; COMPARE-LAA-MV-NEXT: [[TMP13:%.*]] = extractelement <4 x i32> [[TMP10]], i64 1
; COMPARE-LAA-MV-NEXT: [[TMP14:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP13]]
; COMPARE-LAA-MV-NEXT: [[TMP15:%.*]] = extractelement <4 x i32> [[TMP10]], i64 2
; COMPARE-LAA-MV-NEXT: [[TMP16:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP15]]
-; COMPARE-LAA-MV-NEXT: [[TMP17:%.*]] = extractelement <4 x i32> [[TMP10]], i64 3
-; COMPARE-LAA-MV-NEXT: [[TMP18:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP17]]
-; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP0]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP26:%.*]] = extractelement <4 x i32> [[TMP10]], i64 3
+; COMPARE-LAA-MV-NEXT: [[TMP27:%.*]] = getelementptr i32, ptr [[P]], i32 [[TMP26]]
+; COMPARE-LAA-MV-NEXT: [[TMP19:%.*]] = load i32, ptr [[TMP18]], align 8
; COMPARE-LAA-MV-NEXT: [[TMP20:%.*]] = load i32, ptr [[TMP14]], align 8
; COMPARE-LAA-MV-NEXT: [[TMP21:%.*]] = load i32, ptr [[TMP16]], align 8
-; COMPARE-LAA-MV-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP18]], align 8
+; COMPARE-LAA-MV-NEXT: [[TMP22:%.*]] = load i32, ptr [[TMP27]], align 8
; COMPARE-LAA-MV-NEXT: [[TMP23:%.*]] = insertelement <4 x i32> poison, i32 [[TMP19]], i32 0
; COMPARE-LAA-MV-NEXT: [[TMP24:%.*]] = insertelement <4 x i32> [[TMP23]], i32 [[TMP20]], i32 1
; COMPARE-LAA-MV-NEXT: [[TMP25:%.*]] = insertelement <4 x i32> [[TMP24]], i32 [[TMP21]], i32 2
; COMPARE-LAA-MV-NEXT: [[WIDE_LOAD:%.*]] = insertelement <4 x i32> [[TMP25]], i32 [[TMP22]], i32 3
-; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[P_OUT]], i32 [[INDEX1]]
+; COMPARE-LAA-MV-NEXT: [[TMP1:%.*]] = getelementptr i32, ptr [[P_OUT]], i32 [[INDEX]]
; COMPARE-LAA-MV-NEXT: store <4 x i32> [[WIDE_LOAD]], ptr [[TMP1]], align 8
-; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX1]], 4
+; COMPARE-LAA-MV-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4
; COMPARE-LAA-MV-NEXT: [[VEC_IND_NEXT]] = add nsw <4 x i32> [[VEC_IND]], splat (i32 4)
; COMPARE-LAA-MV-NEXT: [[TMP2:%.*]] = icmp eq i32 [[INDEX_NEXT]], 128
; COMPARE-LAA-MV-NEXT: br i1 [[TMP2]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP55:![0-9]+]]
More information about the llvm-commits
mailing list