[llvm] [LV] Use index type for base pointer computation in convertToStridedAccesses (PR #201070)
Mel Chen via llvm-commits
llvm-commits at lists.llvm.org
Thu Jun 4 01:21:23 PDT 2026
https://github.com/Mel-Chen updated https://github.com/llvm/llvm-project/pull/201070
>From 60aca66ff06b3a0d1f1623c6a15d371357b1ef68 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Tue, 2 Jun 2026 01:10:45 -0700
Subject: [PATCH 1/4] use stride type as offset type
---
llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp | 10 ++++++----
1 file changed, 6 insertions(+), 4 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 62d56fa762874..04e4f68187151 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -7206,17 +7206,19 @@ void VPlanTransforms::convertToStridedAccesses(VPlan &Plan,
VPBuilder Builder(LoadR);
// Create the base pointer of strided access.
+ // TODO: reuse VPDerivedIVRecipe for base pointer computation when it
+ // supports a general VPValue as the start value.
VPValue *StartVPV = vputils::getOrCreateVPValueForSCEVExpr(Plan, Start);
VPValue *StrideInBytes = Plan.getOrAddLiveIn(Step->getValue());
Type *IndexTy = Plan.getDataLayout().getIndexType(Ptr->getScalarType());
assert(IndexTy == StrideInBytes->getScalarType() &&
"Stride type from SCEV must match the index type");
- VPValue *CanIVTyStride = Builder.createScalarSExtOrTrunc(
- StrideInBytes, VectorLoop->getCanonicalIVType(), IndexTy,
- DebugLoc::getUnknown());
+ VPValue *CanIV = Builder.createScalarSExtOrTrunc(
+ VectorLoop->getCanonicalIV(), IndexTy,
+ VectorLoop->getCanonicalIVType(), DebugLoc::getUnknown());
auto *AddRecPtr = cast<SCEVAddRecExpr>(PtrSCEV);
auto *Offset = Builder.createOverflowingOp(
- Instruction::Mul, {VectorLoop->getCanonicalIV(), CanIVTyStride},
+ Instruction::Mul, {CanIV, StrideInBytes},
{AddRecPtr->hasNoUnsignedWrap(), AddRecPtr->hasNoSignedWrap()});
auto *BasePtr = Builder.createNoWrapPtrAdd(
StartVPV, Offset,
>From 09a3e712cc431fd3cd0404056a0016a853a24528 Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Tue, 2 Jun 2026 01:25:03 -0700
Subject: [PATCH 2/4] update test
---
.../LoopVectorize/RISCV/dead-ops-cost.ll | 7 ++++---
.../RISCV/masked_gather_scatter.ll | 13 ++++++------
.../RISCV/strided-accesses-narrow-iv.ll | 20 +++++++++++--------
.../LoopVectorize/RISCV/strided-accesses.ll | 10 ++++++----
.../truncate-to-minimal-bitwidth-cost.ll | 5 +++--
5 files changed, 32 insertions(+), 23 deletions(-)
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/dead-ops-cost.ll b/llvm/test/Transforms/LoopVectorize/RISCV/dead-ops-cost.ll
index 96a8594f21ed6..a9db4458a34ff 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/dead-ops-cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/dead-ops-cost.ll
@@ -91,9 +91,10 @@ define i8 @dead_live_out_due_to_scalar_epilogue_required(ptr %src, ptr %dst) {
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 16 x i32> poison, i32 [[TMP3]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 16 x i32> [[BROADCAST_SPLATINSERT]], <vscale x 16 x i32> poison, <vscale x 16 x i32> zeroinitializer
; CHECK-NEXT: [[TMP9:%.*]] = sext <vscale x 16 x i32> [[VEC_IND]] to <vscale x 16 x i64>
-; CHECK-NEXT: [[TMP5:%.*]] = shl i32 [[INDEX]], 2
-; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[SRC]], i32 [[TMP5]]
-; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 16 x i8> @llvm.experimental.vp.strided.load.nxv16i8.p0.i64(ptr align 1 [[TMP6]], i64 4, <vscale x 16 x i1> splat (i1 true), i32 [[TMP2]]), !alias.scope [[META3:![0-9]+]]
+; CHECK-NEXT: [[TMP5:%.*]] = sext i32 [[INDEX]] to i64
+; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[TMP5]], 2
+; CHECK-NEXT: [[TMP12:%.*]] = getelementptr i8, ptr [[SRC]], i64 [[TMP6]]
+; CHECK-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 16 x i8> @llvm.experimental.vp.strided.load.nxv16i8.p0.i64(ptr align 1 [[TMP12]], i64 4, <vscale x 16 x i1> splat (i1 true), i32 [[TMP2]]), !alias.scope [[META3:![0-9]+]]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[DST]], <vscale x 16 x i64> [[TMP9]]
; CHECK-NEXT: call void @llvm.vp.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> zeroinitializer, <vscale x 16 x ptr> align 1 [[TMP7]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP2]]), !alias.scope [[META6:![0-9]+]], !noalias [[META3]]
; CHECK-NEXT: [[CURRENT_ITERATION_NEXT]] = add nuw i32 [[TMP2]], [[INDEX]]
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/masked_gather_scatter.ll b/llvm/test/Transforms/LoopVectorize/RISCV/masked_gather_scatter.ll
index 395a081b3df76..1c67b243429b5 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/masked_gather_scatter.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/masked_gather_scatter.ll
@@ -57,13 +57,14 @@ define void @foo4(ptr nocapture %A, ptr nocapture readonly %B, ptr nocapture rea
; RV32-NEXT: [[TMP11:%.*]] = shl nuw nsw i64 [[TMP8]], 4
; RV32-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i64> poison, i64 [[TMP11]], i64 0
; RV32-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i64> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i64> poison, <vscale x 2 x i32> zeroinitializer
-; RV32-NEXT: [[TMP12:%.*]] = shl i64 [[INDEX]], 6
-; RV32-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[TRIGGER]], i64 [[TMP12]]
-; RV32-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 2 x i32> @llvm.experimental.vp.strided.load.nxv2i32.p0.i32(ptr align 4 [[TMP13]], i32 64, <vscale x 2 x i1> splat (i1 true), i32 [[TMP10]]), !alias.scope [[META0:![0-9]+]], !noalias [[META3:![0-9]+]]
+; RV32-NEXT: [[TMP12:%.*]] = trunc i64 [[INDEX]] to i32
+; RV32-NEXT: [[TMP13:%.*]] = shl i32 [[TMP12]], 6
+; RV32-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr [[TRIGGER]], i32 [[TMP13]]
+; RV32-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 2 x i32> @llvm.experimental.vp.strided.load.nxv2i32.p0.i32(ptr align 4 [[TMP15]], i32 64, <vscale x 2 x i1> splat (i1 true), i32 [[TMP10]]), !alias.scope [[META0:![0-9]+]], !noalias [[META3:![0-9]+]]
; RV32-NEXT: [[TMP14:%.*]] = icmp slt <vscale x 2 x i32> [[WIDE_MASKED_GATHER]], splat (i32 100)
-; RV32-NEXT: [[TMP16:%.*]] = shl i64 [[INDEX]], 8
-; RV32-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP16]]
-; RV32-NEXT: [[WIDE_MASKED_GATHER6:%.*]] = call <vscale x 2 x double> @llvm.experimental.vp.strided.load.nxv2f64.p0.i32(ptr align 8 [[TMP20]], i32 256, <vscale x 2 x i1> [[TMP14]], i32 [[TMP10]]), !alias.scope [[META5:![0-9]+]]
+; RV32-NEXT: [[TMP20:%.*]] = shl i32 [[TMP12]], 8
+; RV32-NEXT: [[TMP25:%.*]] = getelementptr i8, ptr [[B]], i32 [[TMP20]]
+; RV32-NEXT: [[WIDE_MASKED_GATHER6:%.*]] = call <vscale x 2 x double> @llvm.experimental.vp.strided.load.nxv2f64.p0.i32(ptr align 8 [[TMP25]], i32 256, <vscale x 2 x i1> [[TMP14]], i32 [[TMP10]]), !alias.scope [[META5:![0-9]+]]
; RV32-NEXT: [[TMP17:%.*]] = sitofp <vscale x 2 x i32> [[WIDE_MASKED_GATHER]] to <vscale x 2 x double>
; RV32-NEXT: [[TMP18:%.*]] = fadd <vscale x 2 x double> [[WIDE_MASKED_GATHER6]], [[TMP17]]
; RV32-NEXT: [[TMP19:%.*]] = getelementptr inbounds double, ptr [[A]], <vscale x 2 x i64> [[VEC_IND]]
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses-narrow-iv.ll b/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses-narrow-iv.ll
index c0de36196379e..cee73c92460b5 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses-narrow-iv.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses-narrow-iv.ll
@@ -15,8 +15,9 @@ define void @narrow_iv_i8_sext_i64(ptr noalias %arr, ptr noalias %out) {
; RV64-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[CURRENT_ITERATION_NEXT:%.*]], %[[VECTOR_BODY]] ]
; RV64-NEXT: [[AVL:%.*]] = phi i32 [ 16, %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; RV64-NEXT: [[TMP0:%.*]] = call i32 @llvm.experimental.get.vector.length.i32(i32 [[AVL]], i32 8, i1 true)
-; RV64-NEXT: [[TMP1:%.*]] = shl i32 [[INDEX]], 12
-; RV64-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[ARR]], i32 [[TMP1]]
+; RV64-NEXT: [[TMP1:%.*]] = sext i32 [[INDEX]] to i64
+; RV64-NEXT: [[TMP5:%.*]] = shl i64 [[TMP1]], 12
+; RV64-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[ARR]], i64 [[TMP5]]
; RV64-NEXT: [[TMP3:%.*]] = call <vscale x 8 x i8> @llvm.experimental.vp.strided.load.nxv8i8.p0.i64(ptr align 1 [[TMP2]], i64 4096, <vscale x 8 x i1> splat (i1 true), i32 [[TMP0]])
; RV64-NEXT: call void @llvm.vp.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP3]], <vscale x 8 x ptr> align 1 [[BROADCAST_SPLAT]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP0]])
; RV64-NEXT: [[CURRENT_ITERATION_NEXT]] = add nuw i32 [[TMP0]], [[INDEX]]
@@ -83,8 +84,9 @@ define void @narrow_iv_i8_sext_i16(ptr noalias %arr, ptr noalias %out) {
; RV64-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[CURRENT_ITERATION_NEXT:%.*]], %[[VECTOR_BODY]] ]
; RV64-NEXT: [[AVL:%.*]] = phi i32 [ 16, %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; RV64-NEXT: [[TMP0:%.*]] = call i32 @llvm.experimental.get.vector.length.i32(i32 [[AVL]], i32 8, i1 true)
-; RV64-NEXT: [[TMP1:%.*]] = shl i32 [[INDEX]], 12
-; RV64-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[ARR]], i32 [[TMP1]]
+; RV64-NEXT: [[TMP1:%.*]] = sext i32 [[INDEX]] to i64
+; RV64-NEXT: [[TMP5:%.*]] = shl i64 [[TMP1]], 12
+; RV64-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[ARR]], i64 [[TMP5]]
; RV64-NEXT: [[TMP3:%.*]] = call <vscale x 8 x i8> @llvm.experimental.vp.strided.load.nxv8i8.p0.i64(ptr align 1 [[TMP2]], i64 4096, <vscale x 8 x i1> splat (i1 true), i32 [[TMP0]])
; RV64-NEXT: call void @llvm.vp.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP3]], <vscale x 8 x ptr> align 1 [[BROADCAST_SPLAT]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP0]])
; RV64-NEXT: [[CURRENT_ITERATION_NEXT]] = add nuw i32 [[TMP0]], [[INDEX]]
@@ -151,8 +153,9 @@ define void @narrow_iv_i8_zext_i64(ptr noalias %arr, ptr noalias %out) {
; RV64-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[CURRENT_ITERATION_NEXT:%.*]], %[[VECTOR_BODY]] ]
; RV64-NEXT: [[AVL:%.*]] = phi i32 [ 16, %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; RV64-NEXT: [[TMP0:%.*]] = call i32 @llvm.experimental.get.vector.length.i32(i32 [[AVL]], i32 8, i1 true)
-; RV64-NEXT: [[TMP1:%.*]] = shl i32 [[INDEX]], 12
-; RV64-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[ARR]], i32 [[TMP1]]
+; RV64-NEXT: [[TMP1:%.*]] = sext i32 [[INDEX]] to i64
+; RV64-NEXT: [[TMP5:%.*]] = shl i64 [[TMP1]], 12
+; RV64-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[ARR]], i64 [[TMP5]]
; RV64-NEXT: [[TMP3:%.*]] = call <vscale x 8 x i8> @llvm.experimental.vp.strided.load.nxv8i8.p0.i64(ptr align 1 [[TMP2]], i64 4096, <vscale x 8 x i1> splat (i1 true), i32 [[TMP0]])
; RV64-NEXT: call void @llvm.vp.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP3]], <vscale x 8 x ptr> align 1 [[BROADCAST_SPLAT]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP0]])
; RV64-NEXT: [[CURRENT_ITERATION_NEXT]] = add nuw i32 [[TMP0]], [[INDEX]]
@@ -219,8 +222,9 @@ define void @narrow_iv_i8_zext_i16(ptr noalias %arr, ptr noalias %out) {
; RV64-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[CURRENT_ITERATION_NEXT:%.*]], %[[VECTOR_BODY]] ]
; RV64-NEXT: [[AVL:%.*]] = phi i32 [ 16, %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; RV64-NEXT: [[TMP0:%.*]] = call i32 @llvm.experimental.get.vector.length.i32(i32 [[AVL]], i32 8, i1 true)
-; RV64-NEXT: [[TMP1:%.*]] = shl i32 [[INDEX]], 12
-; RV64-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[ARR]], i32 [[TMP1]]
+; RV64-NEXT: [[TMP1:%.*]] = sext i32 [[INDEX]] to i64
+; RV64-NEXT: [[TMP5:%.*]] = shl i64 [[TMP1]], 12
+; RV64-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[ARR]], i64 [[TMP5]]
; RV64-NEXT: [[TMP3:%.*]] = call <vscale x 8 x i8> @llvm.experimental.vp.strided.load.nxv8i8.p0.i64(ptr align 1 [[TMP2]], i64 4096, <vscale x 8 x i1> splat (i1 true), i32 [[TMP0]])
; RV64-NEXT: call void @llvm.vp.scatter.nxv8i8.nxv8p0(<vscale x 8 x i8> [[TMP3]], <vscale x 8 x ptr> align 1 [[BROADCAST_SPLAT]], <vscale x 8 x i1> splat (i1 true), i32 [[TMP0]])
; RV64-NEXT: [[CURRENT_ITERATION_NEXT]] = add nuw i32 [[TMP0]], [[INDEX]]
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll b/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll
index 9c56e77a4c883..ae49a5bd4c83d 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll
@@ -1559,8 +1559,9 @@ define void @type_cache_crash(ptr noalias %a, ptr noalias %b, ptr noalias %c, i3
; CHECK-NEXT: [[TMP8:%.*]] = load double, ptr [[TMP6]], align 8
; CHECK-NEXT: [[BROADCAST_SPLATINSERT5:%.*]] = insertelement <vscale x 2 x double> poison, double [[TMP8]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT6:%.*]] = shufflevector <vscale x 2 x double> [[BROADCAST_SPLATINSERT5]], <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer
-; CHECK-NEXT: [[TMP9:%.*]] = shl i32 [[INDEX]], 5
-; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[B]], i32 [[TMP9]]
+; CHECK-NEXT: [[TMP9:%.*]] = sext i32 [[INDEX]] to i64
+; CHECK-NEXT: [[TMP16:%.*]] = shl i64 [[TMP9]], 5
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP16]]
; CHECK-NEXT: [[TMP11:%.*]] = call <vscale x 2 x double> @llvm.experimental.vp.strided.load.nxv2f64.p0.i64(ptr align 8 [[TMP10]], i64 32, <vscale x 2 x i1> splat (i1 true), i32 [[TMP7]])
; CHECK-NEXT: [[TMP12:%.*]] = fsub <vscale x 2 x double> [[BROADCAST_SPLAT2]], [[TMP11]]
; CHECK-NEXT: [[TMP13:%.*]] = fmul <vscale x 2 x double> [[BROADCAST_SPLAT6]], [[TMP12]]
@@ -1611,8 +1612,9 @@ define void @type_cache_crash(ptr noalias %a, ptr noalias %b, ptr noalias %c, i3
; CHECK-UF2-NEXT: [[TMP14:%.*]] = load double, ptr [[TMP13]], align 8
; CHECK-UF2-NEXT: [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <vscale x 2 x double> poison, double [[TMP14]], i64 0
; CHECK-UF2-NEXT: [[BROADCAST_SPLAT4:%.*]] = shufflevector <vscale x 2 x double> [[BROADCAST_SPLATINSERT3]], <vscale x 2 x double> poison, <vscale x 2 x i32> zeroinitializer
-; CHECK-UF2-NEXT: [[TMP15:%.*]] = shl i32 [[INDEX]], 5
-; CHECK-UF2-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[B]], i32 [[TMP15]]
+; CHECK-UF2-NEXT: [[TMP15:%.*]] = sext i32 [[INDEX]] to i64
+; CHECK-UF2-NEXT: [[TMP29:%.*]] = shl i64 [[TMP15]], 5
+; CHECK-UF2-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP29]]
; CHECK-UF2-NEXT: [[TMP17:%.*]] = zext i32 [[TMP5]] to i64
; CHECK-UF2-NEXT: [[TMP18:%.*]] = mul i64 [[TMP17]], 32
; CHECK-UF2-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[TMP16]], i64 [[TMP18]]
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 7b5a74d89cce1..30ea84b642489 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
@@ -297,8 +297,9 @@ define void @test_minbws_for_trunc(i32 %n, ptr noalias %p1, ptr noalias %p2) {
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <vscale x 2 x i16> poison, i16 [[TMP11]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 2 x i16> [[BROADCAST_SPLATINSERT]], <vscale x 2 x i16> poison, <vscale x 2 x i32> zeroinitializer
; CHECK-NEXT: [[TMP12:%.*]] = sext <vscale x 2 x i16> [[VEC_IND]] to <vscale x 2 x i64>
-; CHECK-NEXT: [[TMP13:%.*]] = shl i32 [[INDEX]], 4
-; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[P1]], i32 [[TMP13]]
+; CHECK-NEXT: [[TMP13:%.*]] = sext i32 [[INDEX]] to i64
+; CHECK-NEXT: [[TMP22:%.*]] = shl i64 [[TMP13]], 4
+; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[P1]], i64 [[TMP22]]
; 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]]
>From ecb7e94c7e254b89a8a006dcf95d007db44897ff Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Tue, 2 Jun 2026 06:48:26 -0700
Subject: [PATCH 3/4] f
---
.../LoopVectorize/RISCV/strided-access-wide-stride.ll | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/strided-access-wide-stride.ll b/llvm/test/Transforms/LoopVectorize/RISCV/strided-access-wide-stride.ll
index 25a88f630f08c..246ef7c561f61 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/strided-access-wide-stride.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/strided-access-wide-stride.ll
@@ -27,8 +27,9 @@ define void @stride_exceeds_i32_max(ptr noalias readonly %src, ptr noalias %dst,
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[CURRENT_ITERATION_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[AVL:%.*]] = phi i32 [ [[TMP4]], %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.experimental.get.vector.length.i32(i32 [[AVL]], i32 16, i1 true)
-; CHECK-NEXT: [[TMP7:%.*]] = mul i32 [[INDEX]], -1294967296
-; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i32 [[TMP7]]
+; CHECK-NEXT: [[TMP7:%.*]] = sext i32 [[INDEX]] to i64
+; CHECK-NEXT: [[TMP11:%.*]] = mul i64 [[TMP7]], 3000000000
+; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[TMP11]]
; CHECK-NEXT: [[TMP9:%.*]] = call <vscale x 16 x i8> @llvm.experimental.vp.strided.load.nxv16i8.p0.i64(ptr align 1 [[TMP8]], i64 3000000000, <vscale x 16 x i1> splat (i1 true), i32 [[TMP6]])
; CHECK-NEXT: call void @llvm.vp.scatter.nxv16i8.nxv16p0(<vscale x 16 x i8> [[TMP9]], <vscale x 16 x ptr> align 1 [[BROADCAST_SPLAT]], <vscale x 16 x i1> splat (i1 true), i32 [[TMP6]])
; CHECK-NEXT: [[CURRENT_ITERATION_NEXT]] = add nuw i32 [[TMP6]], [[INDEX]]
>From 07a40881aab3a75e217f54f0dac4899479c70c9a Mon Sep 17 00:00:00 2001
From: Mel Chen <mel.chen at sifive.com>
Date: Tue, 2 Jun 2026 06:49:13 -0700
Subject: [PATCH 4/4] add negative test for wrapping index
---
.../RISCV/strided-accesses-narrow-iv.ll | 51 +++++++++++++++++++
1 file changed, 51 insertions(+)
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses-narrow-iv.ll b/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses-narrow-iv.ll
index cee73c92460b5..fa5c758d6138e 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses-narrow-iv.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses-narrow-iv.ll
@@ -277,3 +277,54 @@ loop:
exit:
ret void
}
+
+; Negative test
+define void @narrow_iv_i8_sext_i64_wrapping(ptr noalias %arr, ptr noalias %out) {
+; RV64-LABEL: define void @narrow_iv_i8_sext_i64_wrapping(
+; RV64-SAME: ptr noalias [[ARR:%.*]], ptr noalias [[OUT:%.*]]) #[[ATTR0]] {
+; RV64-NEXT: [[ENTRY:.*]]:
+; RV64-NEXT: br label %[[LOOP:.*]]
+; RV64: [[LOOP]]:
+; RV64-NEXT: [[I:%.*]] = phi i8 [ 0, %[[ENTRY]] ], [ [[I_NEXT:%.*]], %[[LOOP]] ]
+; RV64-NEXT: [[IDX:%.*]] = sext i8 [[I]] to i64
+; RV64-NEXT: [[PTR:%.*]] = getelementptr [1024 x i8], ptr [[ARR]], i64 [[IDX]]
+; RV64-NEXT: [[VAL:%.*]] = load i8, ptr [[PTR]], align 1
+; RV64-NEXT: store i8 [[VAL]], ptr [[OUT]], align 1
+; RV64-NEXT: [[I_NEXT]] = add i8 [[I]], 4
+; RV64-NEXT: [[CMP:%.*]] = icmp ne i8 [[I_NEXT]], 0
+; RV64-NEXT: br i1 [[CMP]], label %[[LOOP]], label %[[EXIT:.*]]
+; RV64: [[EXIT]]:
+; RV64-NEXT: ret void
+;
+; RV32-LABEL: define void @narrow_iv_i8_sext_i64_wrapping(
+; RV32-SAME: ptr noalias [[ARR:%.*]], ptr noalias [[OUT:%.*]]) #[[ATTR0]] {
+; RV32-NEXT: [[ENTRY:.*]]:
+; RV32-NEXT: br label %[[LOOP:.*]]
+; RV32: [[LOOP]]:
+; RV32-NEXT: [[I:%.*]] = phi i8 [ 0, %[[ENTRY]] ], [ [[I_NEXT:%.*]], %[[LOOP]] ]
+; RV32-NEXT: [[IDX:%.*]] = sext i8 [[I]] to i64
+; RV32-NEXT: [[PTR:%.*]] = getelementptr [1024 x i8], ptr [[ARR]], i64 [[IDX]]
+; RV32-NEXT: [[VAL:%.*]] = load i8, ptr [[PTR]], align 1
+; RV32-NEXT: store i8 [[VAL]], ptr [[OUT]], align 1
+; RV32-NEXT: [[I_NEXT]] = add i8 [[I]], 4
+; RV32-NEXT: [[CMP:%.*]] = icmp ne i8 [[I_NEXT]], 0
+; RV32-NEXT: br i1 [[CMP]], label %[[LOOP]], label %[[EXIT:.*]]
+; RV32: [[EXIT]]:
+; RV32-NEXT: ret void
+;
+entry:
+ br label %loop
+
+loop:
+ %i = phi i8 [ 0, %entry ], [ %i.next, %loop ]
+ %idx = sext i8 %i to i64
+ %ptr = getelementptr [1024 x i8], ptr %arr, i64 %idx
+ %val = load i8, ptr %ptr, align 1
+ store i8 %val, ptr %out, align 1
+ %i.next = add i8 %i, 4
+ %cmp = icmp ne i8 %i.next, 0
+ br i1 %cmp, label %loop, label %exit
+
+exit:
+ ret void
+}
More information about the llvm-commits
mailing list