[llvm] aa452b6 - [SLP]Restore insertion points after gathers vectorization

[Alexey Bataev via llvm-commits]

Author: Alexey Bataev
Date: 2025-05-24T07:25:20-07:00
New Revision: aa452b65fc7ebfee6f7e5b9d08aa418d532c7b88

URL: https://github.com/llvm/llvm-project/commit/aa452b65fc7ebfee6f7e5b9d08aa418d532c7b88
DIFF: https://github.com/llvm/llvm-project/commit/aa452b65fc7ebfee6f7e5b9d08aa418d532c7b88.diff

LOG: [SLP]Restore insertion points after gathers vectorization

Restore insertion points after gathers vectorization to avoid a crash in
a root node vectorization.

Fixes #141265

Added: 
    llvm/test/Transforms/SLPVectorizer/RISCV/gather-insert-point-restore.ll

Modified: 
    llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 0f86c572639ca..831703b375d9b 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -18535,6 +18535,7 @@ Value *BoUpSLP::vectorizeTree(
     }
   }
   for (auto &Entry : GatherEntries) {
+    IRBuilderBase::InsertPointGuard Guard(Builder);
     Builder.SetInsertPoint(Entry.second);
     Builder.SetCurrentDebugLocation(Entry.second->getDebugLoc());
     (void)vectorizeTree(Entry.first);

diff  --git a/llvm/test/Transforms/SLPVectorizer/RISCV/gather-insert-point-restore.ll b/llvm/test/Transforms/SLPVectorizer/RISCV/gather-insert-point-restore.ll
new file mode 100644
index 0000000000000..cc88718484172
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/RISCV/gather-insert-point-restore.ll
@@ -0,0 +1,55 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -mtriple=riscv64-unknown-linux-gnu -mattr=+v < %s | FileCheck %s
+
+define i16 @test(ptr %i) {
+; CHECK-LABEL: define i16 @test(
+; CHECK-SAME: ptr [[I:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x ptr> poison, ptr [[I]], i32 0
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <2 x ptr> [[TMP0]], <2 x ptr> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, <2 x ptr> [[TMP1]], <2 x i64> <i64 132860, i64 137774>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <2 x ptr> [[TMP2]], <2 x ptr> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
+; CHECK-NEXT:    [[GEP_US154_2:%.*]] = getelementptr i8, ptr [[I]], i64 142688
+; CHECK-NEXT:    br label %[[FOR_COND5_US:.*]]
+; CHECK:       [[FOR_COND5_US]]:
+; CHECK-NEXT:    [[TMP4:%.*]] = call <4 x i16> @llvm.experimental.vp.strided.load.v4i16.p0.i64(ptr align 2 [[GEP_US154_2]], i64 4914, <4 x i1> splat (i1 true), i32 4)
+; CHECK-NEXT:    [[TMP5:%.*]] = call <4 x i16> @llvm.masked.gather.v4i16.v4p0(<4 x ptr> [[TMP3]], i32 2, <4 x i1> splat (i1 true), <4 x i16> poison)
+; CHECK-NEXT:    [[TMP6:%.*]] = call <8 x i16> @llvm.vector.insert.v8i16.v4i16(<8 x i16> poison, <4 x i16> [[TMP4]], i64 0)
+; CHECK-NEXT:    [[TMP7:%.*]] = call <8 x i16> @llvm.vector.insert.v8i16.v4i16(<8 x i16> [[TMP6]], <4 x i16> [[TMP5]], i64 4)
+; CHECK-NEXT:    [[TMP8:%.*]] = call i16 @llvm.vector.reduce.umax.v8i16(<8 x i16> [[TMP7]])
+; CHECK-NEXT:    [[TMP9:%.*]] = call i16 @llvm.umax.i16(i16 [[TMP8]], i16 0)
+; CHECK-NEXT:    ret i16 [[TMP9]]
+;
+entry:
+  %gep.us154 = getelementptr i8, ptr %i, i64 132860
+  %gep.us154.1 = getelementptr i8, ptr %i, i64 137774
+  %gep.us154.2 = getelementptr i8, ptr %i, i64 142688
+  %gep.us154.3 = getelementptr i8, ptr %i, i64 147602
+  %gep.us154.4 = getelementptr i8, ptr %i, i64 152516
+  %gep.us154.5 = getelementptr i8, ptr %i, i64 157430
+  br label %for.cond5.us
+
+for.cond5.us:
+  %0 = load i16, ptr %gep.us154, align 2
+  %1 = load i16, ptr %gep.us154.1, align 2
+  %2 = load i16, ptr %gep.us154.2, align 2
+  %3 = load i16, ptr %gep.us154.3, align 2
+  %4 = load i16, ptr %gep.us154.4, align 2
+  %5 = load i16, ptr %gep.us154.5, align 2
+  %6 = call i16 @llvm.umax.i16(i16 %5, i16 0)
+  %7 = call i16 @llvm.umax.i16(i16 %0, i16 %6)
+  %8 = call i16 @llvm.umax.i16(i16 %1, i16 %7)
+  %9 = call i16 @llvm.umax.i16(i16 %2, i16 %8)
+  %10 = call i16 @llvm.umax.i16(i16 %3, i16 %9)
+  %11 = call i16 @llvm.umax.i16(i16 %2, i16 %10)
+  %12 = call i16 @llvm.umax.i16(i16 %3, i16 %11)
+  %13 = call i16 @llvm.umax.i16(i16 %4, i16 %12)
+  %14 = load i16, ptr %gep.us154, align 2
+  %15 = call i16 @llvm.umax.i16(i16 %14, i16 %13)
+  %16 = load i16, ptr %gep.us154.1, align 2
+  %17 = call i16 @llvm.umax.i16(i16 %16, i16 %15)
+  %18 = call i16 @llvm.umax.i16(i16 %4, i16 %17)
+  ret i16 %18
+}
+
+declare i16 @llvm.umax.i16(i16, i16) #1