[llvm] 9d54668 - [SLP][NFC] Precommit test that exposes a bug in ShuffleBuilder.

[Vasileios Porpodas via llvm-commits]

Author: Vasileios Porpodas
Date: 2023-06-13T16:52:36-07:00
New Revision: 9d5466849a770eeab222d5a5890376d3596e8ad6

URL: https://github.com/llvm/llvm-project/commit/9d5466849a770eeab222d5a5890376d3596e8ad6
DIFF: https://github.com/llvm/llvm-project/commit/9d5466849a770eeab222d5a5890376d3596e8ad6.diff

LOG: [SLP][NFC] Precommit test that exposes a bug in ShuffleBuilder.

ShuffleBuilder generates a zero mask here:
`[[TMP6:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> poison, <4 x i32> zeroinitializer`
But the correct mask is `0,0,1,1`, or we should have reused `TMP4`.

Differential Revision: https://reviews.llvm.org/D152868

Added: 
    llvm/test/Transforms/SLPVectorizer/X86/shufflebuilder-bug.ll

Modified: 
    

Removed: 
    


################################################################################
diff  --git a/llvm/test/Transforms/SLPVectorizer/X86/shufflebuilder-bug.ll b/llvm/test/Transforms/SLPVectorizer/X86/shufflebuilder-bug.ll
new file mode 100644
index 0000000000000..9e25a709b5060
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/X86/shufflebuilder-bug.ll
@@ -0,0 +1,45 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
+; RUN: opt -S -p slp-vectorizer -mtriple=x86_64-- %s | FileCheck %s
+
+define void @foo(<4 x float> %vec, float %val, ptr %ptr) {
+; CHECK-LABEL: define void @foo
+; CHECK-SAME: (<4 x float> [[VEC:%.*]], float [[VAL:%.*]], ptr [[PTR:%.*]]) {
+; CHECK-NEXT:    [[GEP0:%.*]] = getelementptr inbounds float, ptr [[PTR]], i64 0
+; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, ptr [[GEP0]], align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x float> [[VEC]], <4 x float> poison, <2 x i32> <i32 3, i32 poison>
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x float> [[TMP2]], float [[VAL]], i32 1
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> poison, <4 x i32> <i32 0, i32 0, i32 1, i32 1>
+; CHECK-NEXT:    [[TMP5:%.*]] = fadd <4 x float> [[TMP1]], [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> poison, <4 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = fmul <4 x float> [[TMP5]], [[TMP6]]
+; CHECK-NEXT:    store <4 x float> [[TMP7]], ptr [[GEP0]], align 4
+; CHECK-NEXT:    ret void
+;
+  %vec_3 = extractelement <4 x float> %vec, i32 3
+
+  %gep0 = getelementptr inbounds float, ptr %ptr, i64 0
+  %gep1 = getelementptr inbounds float, ptr %ptr, i64 1
+  %gep2 = getelementptr inbounds float, ptr %ptr, i64 2
+  %gep3 = getelementptr inbounds float, ptr %ptr, i64 3
+
+  %l0 = load float, ptr %gep0, align 8
+  %l1 = load float, ptr %gep1, align 8
+  %l2 = load float, ptr %gep2, align 8
+  %l3 = load float, ptr %gep3, align 8
+
+  %fadd0 = fadd float %l0, %vec_3
+  %fadd1 = fadd float %l1, %vec_3
+  %fadd2 = fadd float %l2, %val
+  %fadd3 = fadd float %l3, %val
+
+  %fmul0 = fmul float %fadd0, %vec_3
+  %fmul1 = fmul float %fadd1, %vec_3
+  %fmul2 = fmul float %fadd2, %val
+  %fmul3 = fmul float %fadd3, %val
+
+  store float %fmul0, ptr %gep0, align 4
+  store float %fmul1, ptr %gep1, align 4
+  store float %fmul2, ptr %gep2, align 4
+  store float %fmul3, ptr %gep3, align 4
+  ret void
+}