[SLP]Fix a crash when trying to shuffle multiple nodes.

Need to transform mask after applying shuffle using the mask itself as
a base to correctly mark with identity those indices, actually used in
previous shuffle. Allows to fix a crash, if different sized vectors are
shuffled.

diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 1d5da9a..c763f67 100644 (file)
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -9142,11 +9142,11 @@ public:
     Value *Vec = InVectors.front();
     if (InVectors.size() == 2) {
       Vec = createShuffle(Vec, InVectors.back(), CommonMask);
-      transformMaskAfterShuffle(CommonMask, Mask);
+      transformMaskAfterShuffle(CommonMask, CommonMask);
     } else if (cast<FixedVectorType>(Vec->getType())->getNumElements() !=
                Mask.size()) {
       Vec = createShuffle(Vec, nullptr, CommonMask);
-      transformMaskAfterShuffle(CommonMask, Mask);
+      transformMaskAfterShuffle(CommonMask, CommonMask);
     }
     V1 = createShuffle(V1, V2, Mask);
     for (unsigned Idx = 0, Sz = CommonMask.size(); Idx < Sz; ++Idx)

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/shuffle-multiple-nodes.ll b/llvm/test/Transforms/SLPVectorizer/X86/shuffle-multiple-nodes.ll
new file mode 100644 (file)
index 0000000..370dc46
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/shuffle-multiple-nodes.ll
@@ -0,0 +1,63 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
+; RUN: opt -S --passes=slp-vectorizer -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f < %s | FileCheck %s
+
+define i32 @main(<16 x i32> %bc47.i, <16 x i32> %bc) {
+; CHECK-LABEL: define i32 @main
+; CHECK-SAME: (<16 x i32> [[BC47_I:%.*]], <16 x i32> [[BC:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = shufflevector <16 x i32> [[BC]], <16 x i32> [[BC47_I]], <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 undef, i32 16, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i32> [[TMP0]], <8 x i32> <i32 poison, i32 0, i32 0, i32 0, i32 0, i32 poison, i32 0, i32 0>, <8 x i32> <i32 0, i32 9, i32 10, i32 11, i32 12, i32 5, i32 14, i32 15>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <8 x i32> [[TMP0]], <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 -1, i32 0, i32 0>, <8 x i32> <i32 undef, i32 0, i32 undef, i32 5, i32 13, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <8 x i32> <i32 1, i32 undef, i32 0, i32 undef, i32 undef, i32 0, i32 0, i32 0>, <8 x i32> [[TMP2]], <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 12, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP4:%.*]] = mul <8 x i32> [[TMP1]], [[TMP3]]
+; CHECK-NEXT:    [[TMP5:%.*]] = sub <8 x i32> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP6:%.*]] = add <8 x i32> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <8 x i32> [[TMP5]], <8 x i32> [[TMP6]], <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+; CHECK-NEXT:    store <8 x i32> [[TMP7]], ptr null, align 16
+; CHECK-NEXT:    ret i32 0
+;
+entry:
+  %0 = extractelement <16 x i32> %bc47.i, i64 0
+  %1 = extractelement <16 x i32> %bc, i64 0
+  %2 = extractelement <16 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>, i64 0
+  %3 = mul i32 %1, %2
+  %4 = mul i32 0, 0
+  %5 = sub i32 %3, %4
+  %6 = mul i32 0, 0
+  %7 = mul i32 0, %1
+  %8 = add i32 %6, %7
+  store i32 %5, ptr null, align 16
+  store i32 %8, ptr getelementptr inbounds ([9 x i32], ptr null, i64 0, i64 1), align 2
+  %9 = extractelement <16 x i32> zeroinitializer, i64 0
+  %10 = mul i32 %9, 0
+  %11 = extractelement <16 x i32> zeroinitializer, i64 0
+  %12 = mul i32 0, 0
+  %13 = sub i32 %10, %12
+  %14 = mul i32 0, %0
+  %15 = mul i32 0, 0
+  %16 = add i32 %14, %15
+  store i32 %13, ptr getelementptr inbounds ([9 x i32], ptr null, i64 0, i64 2), align 4
+  store i32 %16, ptr getelementptr inbounds ([9 x i32], ptr null, i64 0, i64 3), align 2
+  %17 = extractelement <16 x i32> <i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1, i32 -1>, i64 0
+  %18 = mul i32 %17, 0
+  %19 = mul i32 0, 0
+  %20 = sub i32 %18, %19
+  %21 = extractelement <16 x i32> zeroinitializer, i64 1
+  %22 = mul i32 %21, %0
+  %23 = mul i32 0, %17
+  %24 = add i32 %22, %23
+  store i32 %20, ptr getelementptr inbounds ([9 x i32], ptr null, i64 0, i64 4), align 8
+  store i32 %24, ptr getelementptr inbounds ([9 x i32], ptr null, i64 0, i64 5), align 2
+  %25 = mul i32 0, %11
+  %26 = extractelement <16 x i32> zeroinitializer, i64 0
+  %27 = mul i32 0, 0
+  %28 = sub i32 %25, %27
+  %29 = mul i32 0, 0
+  %30 = mul i32 %26, 0
+  %31 = add i32 %29, %30
+  store i32 %28, ptr getelementptr inbounds ([9 x i32], ptr null, i64 0, i64 6), align 4
+  store i32 %31, ptr getelementptr inbounds ([9 x i32], ptr null, i64 0, i64 7), align 2
+  ret i32 0
+}
+
+