[llvm] [VectorCombine] Fix iterator invalidation in foldSelectShuffle (PR #197914)

Julian Nagele via llvm-commits llvm-commits at lists.llvm.org
Fri May 15 06:51:51 PDT 2026


https://github.com/juliannagele updated https://github.com/llvm/llvm-project/pull/197914

>From 92c04cd028a8a9983d458a496ae5847bd07d86c9 Mon Sep 17 00:00:00 2001
From: Julian Nagele <j_nagele at apple.com>
Date: Fri, 15 May 2026 12:21:07 +0100
Subject: [PATCH 1/2] [VectorCombine] Fix iterator invalidation in
 foldSelectShuffle

A loop in foldSelectShuffle walked shuffles with a range-for while
push_back-ing into the same SmallVector. Once the inner pushes grew
the vector past its current heap capacity the buffer was reallocated,
leaving the range-for iterator dangling.

Iterate the original entries by index instead.

Fixes #197910.
---
 .../Transforms/Vectorize/VectorCombine.cpp    |  4 +-
 .../test/Transforms/VectorCombine/pr197910.ll | 57 +++++++++++++++++++
 2 files changed, 59 insertions(+), 2 deletions(-)
 create mode 100644 llvm/test/Transforms/VectorCombine/pr197910.ll

diff --git a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
index 5ba344ea9a808..e91d27a4fb178 100644
--- a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
+++ b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
@@ -4980,8 +4980,8 @@ bool VectorCombine::foldSelectShuffle(Instruction &I, bool FromReduction) {
   // Add any shuffle uses for the shuffles we have found, to include them in our
   // cost calculations.
   if (!FromReduction) {
-    for (ShuffleVectorInst *SV : Shuffles) {
-      for (auto *U : SV->users()) {
+    for (size_t Idx = 0, E = Shuffles.size(); Idx != E; ++Idx) {
+      for (auto *U : Shuffles[Idx]->users()) {
         ShuffleVectorInst *SSV = dyn_cast<ShuffleVectorInst>(U);
         if (SSV && isa<UndefValue>(SSV->getOperand(1)) && SSV->getType() == VT)
           Shuffles.push_back(SSV);
diff --git a/llvm/test/Transforms/VectorCombine/pr197910.ll b/llvm/test/Transforms/VectorCombine/pr197910.ll
new file mode 100644
index 0000000000000..d48e472921600
--- /dev/null
+++ b/llvm/test/Transforms/VectorCombine/pr197910.ll
@@ -0,0 +1,57 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -passes=vector-combine -S %s | FileCheck %s
+
+define <8 x i32> @PR197910() {
+; CHECK-LABEL: define <8 x i32> @PR197910() {
+; CHECK-NEXT:    [[PA:%.*]] = freeze <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[A:%.*]] = add <8 x i32> [[PA]], [[PA]]
+; CHECK-NEXT:    [[B:%.*]] = or <8 x i32> [[PA]], [[PA]]
+; CHECK-NEXT:    [[S2:%.*]] = shufflevector <8 x i32> [[B]], <8 x i32> [[A]], <8 x i32> <i32 8, i32 9, i32 2, i32 3, i32 12, i32 13, i32 6, i32 7>
+; CHECK-NEXT:    [[S3:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+; CHECK-NEXT:    [[S4:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
+; CHECK-NEXT:    [[S5:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    [[S6:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[S7:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 0, i32 9, i32 10, i32 3, i32 4, i32 13, i32 14, i32 7>
+; CHECK-NEXT:    [[S8:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> [[B]], <8 x i32> <i32 8, i32 1, i32 2, i32 11, i32 12, i32 5, i32 6, i32 15>
+; CHECK-NEXT:    [[U1:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[U6:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[R12:%.*]] = add <8 x i32> [[S2]], [[S3]]
+; CHECK-NEXT:    [[R34:%.*]] = add <8 x i32> [[S4]], [[S5]]
+; CHECK-NEXT:    [[R56:%.*]] = add <8 x i32> [[S6]], [[S7]]
+; CHECK-NEXT:    [[R1234:%.*]] = add <8 x i32> [[R12]], [[R34]]
+; CHECK-NEXT:    [[R123456:%.*]] = add <8 x i32> [[R1234]], [[R56]]
+; CHECK-NEXT:    [[R1234567:%.*]] = add <8 x i32> [[R123456]], [[S8]]
+; CHECK-NEXT:    [[RU:%.*]] = add <8 x i32> [[U1]], [[U6]]
+; CHECK-NEXT:    [[RET:%.*]] = add <8 x i32> [[R1234567]], [[RU]]
+; CHECK-NEXT:    ret <8 x i32> [[RET]]
+;
+  %pa = freeze <8 x i32> zeroinitializer
+  %a = add <8 x i32> %pa, %pa
+  %b = or <8 x i32> %pa, %pa
+
+  %s2 = shufflevector <8 x i32> %b, <8 x i32> %a, <8 x i32> <i32 8, i32 9, i32 2, i32 3, i32 12, i32 13, i32 6, i32 7>
+  %s3 = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %s4 = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
+  %s5 = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 12, i32 13, i32 14, i32 15>
+  %s6 = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7>
+  %s7 = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 9, i32 10, i32 3, i32 4, i32 13, i32 14, i32 7>
+  %s8 = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 8, i32 1, i32 2, i32 11, i32 12, i32 5, i32 6, i32 15>
+  %s1 = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 1, i32 10, i32 11, i32 4, i32 5, i32 14, i32 15>
+
+  %u1 = shufflevector <8 x i32> %s1, <8 x i32> poison, <8 x i32> zeroinitializer
+  %u2 = shufflevector <8 x i32> %s1, <8 x i32> poison, <8 x i32> zeroinitializer
+  %u3 = shufflevector <8 x i32> %s1, <8 x i32> poison, <8 x i32> zeroinitializer
+  %u4 = shufflevector <8 x i32> %s1, <8 x i32> poison, <8 x i32> zeroinitializer
+  %u5 = shufflevector <8 x i32> %s1, <8 x i32> poison, <8 x i32> zeroinitializer
+  %u6 = shufflevector <8 x i32> %s1, <8 x i32> poison, <8 x i32> zeroinitializer
+
+  %r12 = add <8 x i32> %s2, %s3
+  %r34 = add <8 x i32> %s4, %s5
+  %r56 = add <8 x i32> %s6, %s7
+  %r1234 = add <8 x i32> %r12, %r34
+  %r123456 = add <8 x i32> %r1234, %r56
+  %r1234567 = add <8 x i32> %r123456, %s8
+  %ru = add <8 x i32> %u1, %u6
+  %ret = add <8 x i32> %r1234567, %ru
+  ret <8 x i32> %ret
+}

>From 238e68a938ee2c12ef0ddfd2e90883d15689f12f Mon Sep 17 00:00:00 2001
From: Julian Nagele <j_nagele at apple.com>
Date: Fri, 15 May 2026 14:51:20 +0100
Subject: [PATCH 2/2] fixup! [VectorCombine] Fix iterator invalidation in
 foldSelectShuffle

---
 llvm/test/Transforms/VectorCombine/pr197910.ll | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/llvm/test/Transforms/VectorCombine/pr197910.ll b/llvm/test/Transforms/VectorCombine/pr197910.ll
index d48e472921600..8a3d50faf9260 100644
--- a/llvm/test/Transforms/VectorCombine/pr197910.ll
+++ b/llvm/test/Transforms/VectorCombine/pr197910.ll
@@ -1,9 +1,10 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
 ; RUN: opt -passes=vector-combine -S %s | FileCheck %s
 
-define <8 x i32> @PR197910() {
-; CHECK-LABEL: define <8 x i32> @PR197910() {
-; CHECK-NEXT:    [[PA:%.*]] = freeze <8 x i32> zeroinitializer
+define <8 x i32> @PR197910(<8 x i32> %arg) {
+; CHECK-LABEL: define <8 x i32> @PR197910(
+; CHECK-SAME: <8 x i32> [[ARG:%.*]]) {
+; CHECK-NEXT:    [[PA:%.*]] = freeze <8 x i32> [[ARG]]
 ; CHECK-NEXT:    [[A:%.*]] = add <8 x i32> [[PA]], [[PA]]
 ; CHECK-NEXT:    [[B:%.*]] = or <8 x i32> [[PA]], [[PA]]
 ; CHECK-NEXT:    [[S2:%.*]] = shufflevector <8 x i32> [[B]], <8 x i32> [[A]], <8 x i32> <i32 8, i32 9, i32 2, i32 3, i32 12, i32 13, i32 6, i32 7>
@@ -25,7 +26,7 @@ define <8 x i32> @PR197910() {
 ; CHECK-NEXT:    [[RET:%.*]] = add <8 x i32> [[R1234567]], [[RU]]
 ; CHECK-NEXT:    ret <8 x i32> [[RET]]
 ;
-  %pa = freeze <8 x i32> zeroinitializer
+  %pa = freeze <8 x i32> %arg
   %a = add <8 x i32> %pa, %pa
   %b = or <8 x i32> %pa, %pa
 



More information about the llvm-commits mailing list