[llvm] [VPlan] Preserve nusw in createInBoundsPtrAdd (PR #151549)
Ramkumar Ramachandra via llvm-commits
llvm-commits at lists.llvm.org
Mon Aug 11 05:10:19 PDT 2025
https://github.com/artagnon updated https://github.com/llvm/llvm-project/pull/151549
>From d69f681c118ca2cf813cdbdd9c7b39b869c79106 Mon Sep 17 00:00:00 2001
From: Ramkumar Ramachandra <ramkumar.ramachandra at codasip.com>
Date: Thu, 31 Jul 2025 16:49:43 +0100
Subject: [PATCH] [VPlan] Preserve nusw in createInBoundsPtrAdd
---
.../Vectorize/LoopVectorizationPlanner.h | 14 +-
.../Transforms/Vectorize/VPlanTransforms.cpp | 4 +-
...aved-accesses-different-insert-position.ll | 2 +-
.../interleaved-accesses-gep-nowrap-flags.ll | 148 ++++++++++++++++++
4 files changed, 158 insertions(+), 10 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h b/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h
index 912c893123095..3adb4de1a5345 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorizationPlanner.h
@@ -256,13 +256,15 @@ class VPBuilder {
new VPInstruction(VPInstruction::PtrAdd, {Ptr, Offset},
GEPNoWrapFlags::none(), DL, Name));
}
- VPInstruction *createInBoundsPtrAdd(VPValue *Ptr, VPValue *Offset,
- DebugLoc DL = DebugLoc::getUnknown(),
- const Twine &Name = "") {
- return tryInsertInstruction(
- new VPInstruction(VPInstruction::PtrAdd, {Ptr, Offset},
- GEPNoWrapFlags::inBounds(), DL, Name));
+
+ VPInstruction *createNoWrapPtrAdd(VPValue *Ptr, VPValue *Offset,
+ GEPNoWrapFlags GEPFlags,
+ DebugLoc DL = DebugLoc::getUnknown(),
+ const Twine &Name = "") {
+ return tryInsertInstruction(new VPInstruction(
+ VPInstruction::PtrAdd, {Ptr, Offset}, GEPFlags, DL, Name));
}
+
VPInstruction *createWidePtrAdd(VPValue *Ptr, VPValue *Offset,
DebugLoc DL = DebugLoc::getUnknown(),
const Twine &Name = "") {
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index 7979392f7dda8..2ccee2014e37a 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -2608,9 +2608,7 @@ void VPlanTransforms::createInterleaveGroups(
VPValue *OffsetVPV =
Plan.getOrAddLiveIn(ConstantInt::get(Plan.getContext(), -Offset));
VPBuilder B(InsertPos);
- Addr = NW.isInBounds()
- ? B.createInBoundsPtrAdd(InsertPos->getAddr(), OffsetVPV)
- : B.createPtrAdd(InsertPos->getAddr(), OffsetVPV);
+ Addr = B.createNoWrapPtrAdd(InsertPos->getAddr(), OffsetVPV, NW);
}
// If the group is reverse, adjust the index to refer to the last vector
// lane instead of the first. We adjust the index from the first vector
diff --git a/llvm/test/Transforms/LoopVectorize/interleaved-accesses-different-insert-position.ll b/llvm/test/Transforms/LoopVectorize/interleaved-accesses-different-insert-position.ll
index 651210df823dd..0c5d15b7ce953 100644
--- a/llvm/test/Transforms/LoopVectorize/interleaved-accesses-different-insert-position.ll
+++ b/llvm/test/Transforms/LoopVectorize/interleaved-accesses-different-insert-position.ll
@@ -88,7 +88,7 @@ define void @test_ig_insert_pos_at_end_of_vpbb(ptr noalias %dst, ptr noalias %sr
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[TMP3:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr nusw { i16, i16, i16, i16 }, ptr [[SRC]], i64 [[TMP3]], i32 2
-; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP4]], i32 -4
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr nusw i8, ptr [[TMP4]], i32 -4
; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <16 x i16>, ptr [[TMP5]], align 2
; CHECK-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <16 x i16> [[WIDE_VEC]], <16 x i16> poison, <4 x i32> <i32 0, i32 4, i32 8, i32 12>
; CHECK-NEXT: [[STRIDED_VEC1:%.*]] = shufflevector <16 x i16> [[WIDE_VEC]], <16 x i16> poison, <4 x i32> <i32 2, i32 6, i32 10, i32 14>
diff --git a/llvm/test/Transforms/LoopVectorize/interleaved-accesses-gep-nowrap-flags.ll b/llvm/test/Transforms/LoopVectorize/interleaved-accesses-gep-nowrap-flags.ll
index 552f6a4ec62d9..a6ba29ed1ca0e 100644
--- a/llvm/test/Transforms/LoopVectorize/interleaved-accesses-gep-nowrap-flags.ll
+++ b/llvm/test/Transforms/LoopVectorize/interleaved-accesses-gep-nowrap-flags.ll
@@ -185,3 +185,151 @@ loop:
exit:
ret void
}
+
+define void @nusw_preservation_2(ptr %src, ptr noalias %dst) {
+; CHECK-LABEL: define void @nusw_preservation_2(
+; CHECK-SAME: ptr [[SRC:%.*]], ptr noalias [[DST:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK: [[VECTOR_BODY]]:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 2
+; CHECK-NEXT: [[TMP0:%.*]] = or disjoint i64 [[OFFSET_IDX]], 1
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr nusw i8, ptr [[SRC]], i64 [[TMP0]]
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr nusw i8, ptr [[TMP1]], i32 -1
+; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <8 x i8> [[WIDE_VEC]], <8 x i8> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; CHECK-NEXT: [[STRIDED_VEC1:%.*]] = shufflevector <8 x i8> [[WIDE_VEC]], <8 x i8> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; CHECK-NEXT: [[TMP3:%.*]] = add <4 x i8> [[STRIDED_VEC1]], [[STRIDED_VEC]]
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr nusw i8, ptr [[DST]], i64 [[INDEX]]
+; CHECK-NEXT: store <4 x i8> [[TMP3]], ptr [[TMP4]], align 1
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
+; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: br [[EXIT:label %.*]]
+; CHECK: [[SCALAR_PH]]:
+;
+entry:
+ br label %loop
+
+loop: ; preds = %loop, %entry
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+ %iv2 = phi i64 [ 0, %entry ], [ %iv2.next, %loop ]
+ %or.1 = or disjoint i64 %iv2, 1
+ %gep.src.or.1 = getelementptr nusw i8, ptr %src, i64 %or.1
+ %load.src.1 = load i8, ptr %gep.src.or.1, align 1
+ %gep.src.iv2 = getelementptr nusw i8, ptr %src, i64 %iv2
+ %load.src.2 = load i8, ptr %gep.src.iv2, align 1
+ %add = add i8 %load.src.1, %load.src.2
+ %gep.dst.iv = getelementptr nusw i8, ptr %dst, i64 %iv
+ store i8 %add, ptr %gep.dst.iv, align 1
+ %iv2.next = add i64 %iv2, 2
+ %iv.next = add i64 %iv, 1
+ %exit.cond = icmp eq i64 %iv.next, 100
+ br i1 %exit.cond, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @inbounds_preservation_2(ptr %src, ptr noalias %dst) {
+; CHECK-LABEL: define void @inbounds_preservation_2(
+; CHECK-SAME: ptr [[SRC:%.*]], ptr noalias [[DST:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK: [[VECTOR_BODY]]:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 2
+; CHECK-NEXT: [[TMP0:%.*]] = or disjoint i64 [[OFFSET_IDX]], 1
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[SRC]], i64 [[TMP0]]
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i32 -1
+; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <8 x i8> [[WIDE_VEC]], <8 x i8> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; CHECK-NEXT: [[STRIDED_VEC1:%.*]] = shufflevector <8 x i8> [[WIDE_VEC]], <8 x i8> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; CHECK-NEXT: [[TMP3:%.*]] = add <4 x i8> [[STRIDED_VEC1]], [[STRIDED_VEC]]
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[DST]], i64 [[INDEX]]
+; CHECK-NEXT: store <4 x i8> [[TMP3]], ptr [[TMP4]], align 1
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
+; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: br [[EXIT:label %.*]]
+; CHECK: [[SCALAR_PH]]:
+;
+entry:
+ br label %loop
+
+loop: ; preds = %loop, %entry
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+ %iv2 = phi i64 [ 0, %entry ], [ %iv2.next, %loop ]
+ %or.1 = or disjoint i64 %iv2, 1
+ %gep.src.or.1 = getelementptr inbounds i8, ptr %src, i64 %or.1
+ %load.src.1 = load i8, ptr %gep.src.or.1, align 1
+ %gep.src.iv2 = getelementptr inbounds i8, ptr %src, i64 %iv2
+ %load.src.2 = load i8, ptr %gep.src.iv2, align 1
+ %add = add i8 %load.src.1, %load.src.2
+ %gep.dst.iv = getelementptr inbounds i8, ptr %dst, i64 %iv
+ store i8 %add, ptr %gep.dst.iv, align 1
+ %iv2.next = add i64 %iv2, 2
+ %iv.next = add i64 %iv, 1
+ %exit.cond = icmp eq i64 %iv.next, 100
+ br i1 %exit.cond, label %exit, label %loop
+
+exit:
+ ret void
+}
+
+define void @nuw_drop_2(ptr %src, ptr noalias %dst) {
+; CHECK-LABEL: define void @nuw_drop_2(
+; CHECK-SAME: ptr [[SRC:%.*]], ptr noalias [[DST:%.*]]) {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK: [[VECTOR_PH]]:
+; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
+; CHECK: [[VECTOR_BODY]]:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 2
+; CHECK-NEXT: [[TMP0:%.*]] = or disjoint i64 [[OFFSET_IDX]], 1
+; CHECK-NEXT: [[TMP1:%.*]] = getelementptr nuw i8, ptr [[SRC]], i64 [[TMP0]]
+; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[TMP1]], i32 -1
+; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <8 x i8>, ptr [[TMP2]], align 1
+; CHECK-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <8 x i8> [[WIDE_VEC]], <8 x i8> poison, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
+; CHECK-NEXT: [[STRIDED_VEC1:%.*]] = shufflevector <8 x i8> [[WIDE_VEC]], <8 x i8> poison, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
+; CHECK-NEXT: [[TMP3:%.*]] = add <4 x i8> [[STRIDED_VEC1]], [[STRIDED_VEC]]
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr nuw i8, ptr [[DST]], i64 [[INDEX]]
+; CHECK-NEXT: store <4 x i8> [[TMP3]], ptr [[TMP4]], align 1
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; CHECK-NEXT: [[TMP5:%.*]] = icmp eq i64 [[INDEX_NEXT]], 100
+; CHECK-NEXT: br i1 [[TMP5]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
+; CHECK: [[MIDDLE_BLOCK]]:
+; CHECK-NEXT: br [[EXIT:label %.*]]
+; CHECK: [[SCALAR_PH]]:
+;
+entry:
+ br label %loop
+
+loop: ; preds = %loop, %entry
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+ %iv2 = phi i64 [ 0, %entry ], [ %iv2.next, %loop ]
+ %or.1 = or disjoint i64 %iv2, 1
+ %gep.src.or.1 = getelementptr nuw i8, ptr %src, i64 %or.1
+ %load.src.1 = load i8, ptr %gep.src.or.1, align 1
+ %gep.src.iv2 = getelementptr nuw i8, ptr %src, i64 %iv2
+ %load.src.2 = load i8, ptr %gep.src.iv2, align 1
+ %add = add i8 %load.src.1, %load.src.2
+ %gep.dst.iv = getelementptr nuw i8, ptr %dst, i64 %iv
+ store i8 %add, ptr %gep.dst.iv, align 1
+ %iv2.next = add i64 %iv2, 2
+ %iv.next = add i64 %iv, 1
+ %exit.cond = icmp eq i64 %iv.next, 100
+ br i1 %exit.cond, label %exit, label %loop
+
+exit:
+ ret void
+}
+
More information about the llvm-commits
mailing list