[llvm] [LV] Only collect strides without predicates under OptForSize when interleaved access analysis (PR #205793)
Mel Chen via llvm-commits
llvm-commits at lists.llvm.org
Fri Jun 26 01:08:24 PDT 2026
https://github.com/Mel-Chen updated https://github.com/llvm/llvm-project/pull/205793
>From 5b5e6ae31962219b3a04831adb0c4595b34cecd6 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Thu, 25 Jun 2026 04:56:18 -0700
Subject: [PATCH] [LV] Only collect interleave strides without predicates under
OptForSize
---
llvm/include/llvm/Analysis/VectorUtils.h | 10 ++--
llvm/lib/Analysis/VectorUtils.cpp | 7 +--
.../Transforms/Vectorize/LoopVectorize.cpp | 2 +-
.../AArch64/discarded-interleave-group.ll | 49 ++++++++++++-------
4 files changed, 42 insertions(+), 26 deletions(-)
diff --git a/llvm/include/llvm/Analysis/VectorUtils.h b/llvm/include/llvm/Analysis/VectorUtils.h
index e088371be9464..f70ae3ff57f88 100644
--- a/llvm/include/llvm/Analysis/VectorUtils.h
+++ b/llvm/include/llvm/Analysis/VectorUtils.h
@@ -665,8 +665,9 @@ class InterleavedAccessInfo {
public:
InterleavedAccessInfo(PredicatedScalarEvolution &PSE, Loop *L,
DominatorTree *DT, LoopInfo *LI,
- const LoopAccessInfo *LAI)
- : PSE(PSE), TheLoop(L), DT(DT), LI(LI), LAI(LAI) {}
+ const LoopAccessInfo *LAI, bool OptForSize)
+ : PSE(PSE), TheLoop(L), DT(DT), LI(LI), LAI(LAI), OptForSize(OptForSize) {
+ }
~InterleavedAccessInfo() { invalidateGroups(); }
@@ -738,6 +739,7 @@ class InterleavedAccessInfo {
DominatorTree *DT;
LoopInfo *LI;
const LoopAccessInfo *LAI;
+ bool OptForSize;
/// True if the loop may contain non-reversed interleaved groups with
/// out-of-bounds accesses. We ensure we don't speculatively access memory
@@ -807,11 +809,11 @@ class InterleavedAccessInfo {
/// Collect all the accesses with a constant stride in program order. Any
/// SCEV predicates needed to compute the strides are added to \p
- /// Predicates.
+ /// Predicates if it is not nullptr.
void collectConstStrideAccesses(
MapVector<Instruction *, StrideDescriptor> &AccessStrideInfo,
const DenseMap<Value *, const SCEV *> &Strides,
- SmallVectorImpl<const SCEVPredicate *> &Predicates);
+ SmallVectorImpl<const SCEVPredicate *> *Predicates);
/// Returns true if \p Stride is allowed in an interleaved group.
LLVM_ABI static bool isStrided(int Stride);
diff --git a/llvm/lib/Analysis/VectorUtils.cpp b/llvm/lib/Analysis/VectorUtils.cpp
index 884229e09c6be..bcdb6df47d757 100644
--- a/llvm/lib/Analysis/VectorUtils.cpp
+++ b/llvm/lib/Analysis/VectorUtils.cpp
@@ -1310,7 +1310,7 @@ bool InterleavedAccessInfo::isStrided(int Stride) {
void InterleavedAccessInfo::collectConstStrideAccesses(
MapVector<Instruction *, StrideDescriptor> &AccessStrideInfo,
const DenseMap<Value *, const SCEV *> &Strides,
- SmallVectorImpl<const SCEVPredicate *> &Predicates) {
+ SmallVectorImpl<const SCEVPredicate *> *Predicates) {
auto &DL = TheLoop->getHeader()->getDataLayout();
// Since it's desired that the load/store instructions be maintained in
@@ -1342,7 +1342,7 @@ void InterleavedAccessInfo::collectConstStrideAccesses(
// even without the transformation. The wrapping checks are therefore
// deferred until after we've formed the interleaved groups.
int64_t Stride = getPtrStride(PSE, ElementTy, Ptr, TheLoop, *DT, Strides,
- /*ShouldCheckWrap=*/false, &Predicates)
+ /*ShouldCheckWrap=*/false, Predicates)
.value_or(0);
const SCEV *Scev = replaceSymbolicStrideSCEV(PSE, Strides, Ptr);
@@ -1395,7 +1395,8 @@ void InterleavedAccessInfo::analyzeInterleaving(
// Holds all accesses with a constant stride.
MapVector<Instruction *, StrideDescriptor> AccessStrideInfo;
SmallVector<const SCEVPredicate *> Predicates;
- collectConstStrideAccesses(AccessStrideInfo, Strides, Predicates);
+ collectConstStrideAccesses(AccessStrideInfo, Strides,
+ OptForSize ? nullptr : &Predicates);
if (AccessStrideInfo.empty())
return;
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 18e6890d13e13..41112574d76e3 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -7942,7 +7942,7 @@ bool LoopVectorizePass::processLoop(Loop *L) {
}
}
- InterleavedAccessInfo IAI(PSE, L, DT, LI, LVL.getLAI());
+ InterleavedAccessInfo IAI(PSE, L, DT, LI, LVL.getLAI(), OptForSize);
bool UseInterleaved =
IsInnerLoop && TTI->enableInterleavedAccessVectorization();
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/discarded-interleave-group.ll b/llvm/test/Transforms/LoopVectorize/AArch64/discarded-interleave-group.ll
index 1c79629ac1754..01a12b67dcf4b 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/discarded-interleave-group.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/discarded-interleave-group.ll
@@ -65,27 +65,40 @@ exit:
}
; The stride-2 interleave group on %B survives, no predicate should be needed.
-; FIXME: only predicates required by surviving group members should be added.
define void @over_conservative_merge(ptr noalias %A, ptr noalias %B, i64 %N) #0 {
; CHECK-LABEL: define void @over_conservative_merge(
; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
-; CHECK-NEXT: [[ENTRY:.*]]:
-; CHECK-NEXT: br label %[[LOOP:.*]]
-; CHECK: [[LOOP]]:
-; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[ENTRY]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
-; CHECK-NEXT: [[CLAMPED:%.*]] = urem i64 [[INDEX]], 4
-; CHECK-NEXT: [[GEP_A:%.*]] = getelementptr float, ptr [[A]], i64 [[CLAMPED]]
-; CHECK-NEXT: [[LA:%.*]] = load float, ptr [[GEP_A]], align 4
-; CHECK-NEXT: [[M7:%.*]] = fmul float [[LA]], [[LA]]
-; CHECK-NEXT: [[TMP12:%.*]] = shl nuw nsw i64 [[INDEX]], 1
-; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP12]]
-; CHECK-NEXT: store float [[M7]], ptr [[TMP13]], align 4
-; CHECK-NEXT: [[IDX1:%.*]] = add nuw nsw i64 [[TMP12]], 1
-; CHECK-NEXT: [[GEP_B1:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[IDX1]]
-; CHECK-NEXT: store float [[M7]], ptr [[GEP_B1]], align 4
-; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[INDEX]], 1
-; CHECK-NEXT: [[COND:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
-; CHECK-NEXT: br i1 [[COND]], label %[[EXIT:.*]], label %[[LOOP]]
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: br label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
+; CHECK-NEXT: [[TMP1:%.*]] = shl nuw i64 [[TMP0]], 2
+; CHECK-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 4 x i1> @llvm.get.active.lane.mask.nxv4i1.i64(i64 0, i64 [[N]])
+; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 4 x i64> @llvm.stepvector.nxv4i64()
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 4 x i64> poison, i64 [[TMP1]], i64 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 4 x i64> poison, <vscale x 4 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: [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 4 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], %[[VECTOR_PH]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i64> [ [[TMP2]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP3:%.*]] = urem <vscale x 4 x i64> [[VEC_IND]], splat (i64 4)
+; CHECK-NEXT: [[WIDE_GEP:%.*]] = getelementptr float, ptr [[A]], <vscale x 4 x i64> [[TMP3]]
+; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x float> @llvm.masked.gather.nxv4f32.nxv4p0(<vscale x 4 x ptr> align 4 [[WIDE_GEP]], <vscale x 4 x i1> [[ACTIVE_LANE_MASK]], <vscale x 4 x float> poison)
+; CHECK-NEXT: [[TMP4:%.*]] = fmul <vscale x 4 x float> [[WIDE_MASKED_GATHER]], [[WIDE_MASKED_GATHER]]
+; CHECK-NEXT: [[TMP5:%.*]] = shl nuw nsw i64 [[INDEX]], 1
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP5]]
+; CHECK-NEXT: [[INTERLEAVED_VEC:%.*]] = call <vscale x 8 x float> @llvm.vector.interleave2.nxv8f32(<vscale x 4 x float> [[TMP4]], <vscale x 4 x float> [[TMP4]])
+; CHECK-NEXT: [[INTERLEAVED_MASK:%.*]] = call <vscale x 8 x i1> @llvm.vector.interleave2.nxv8i1(<vscale x 4 x i1> [[ACTIVE_LANE_MASK]], <vscale x 4 x i1> [[ACTIVE_LANE_MASK]])
+; CHECK-NEXT: call void @llvm.masked.store.nxv8f32.p0(<vscale x 8 x float> [[INTERLEAVED_VEC]], ptr align 4 [[TMP6]], <vscale x 8 x i1> [[INTERLEAVED_MASK]])
+; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP1]]
+; CHECK-NEXT: [[ACTIVE_LANE_MASK_NEXT]] = call <vscale x 4 x i1> @llvm.get.active.lane.mask.nxv4i1.i64(i64 [[INDEX_NEXT]], i64 [[N]])
+; CHECK-NEXT: [[TMP7:%.*]] = extractelement <vscale x 4 x i1> [[ACTIVE_LANE_MASK_NEXT]], i64 0
+; CHECK-NEXT: [[TMP8:%.*]] = xor i1 [[TMP7]], true
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add nuw nsw <vscale x 4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: br label %[[EXIT:.*]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret void
;
More information about the llvm-commits
mailing list