[llvm] r322106 - [COST]Fix PR35865: Fix cost model evaluation for shuffle on X86.

[Alexey Bataev via llvm-commits]

Author: abataev
Date: Tue Jan  9 11:08:22 2018
New Revision: 322106

URL: http://llvm.org/viewvc/llvm-project?rev=322106&view=rev
Log:
[COST]Fix PR35865: Fix cost model evaluation for shuffle on X86.

Summary:
If the vector type is transformed to non-vector single type, the compile
may crash trying to get vector information about non-vector type.

Reviewers: RKSimon, spatel, mkuper, hfinkel

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D41862

Added:
    llvm/trunk/test/Transforms/SLPVectorizer/X86/PR35865.ll
Modified:
    llvm/trunk/lib/Target/X86/X86TargetTransformInfo.cpp

Modified: llvm/trunk/lib/Target/X86/X86TargetTransformInfo.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86TargetTransformInfo.cpp?rev=322106&r1=322105&r2=322106&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86TargetTransformInfo.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86TargetTransformInfo.cpp Tue Jan  9 11:08:22 2018
@@ -754,7 +754,8 @@ int X86TTIImpl::getShuffleCost(TTI::Shuf
   // type remains the same.
   if (Kind == TTI::SK_PermuteSingleSrc && LT.first != 1) {
     MVT LegalVT = LT.second;
-    if (LegalVT.getVectorElementType().getSizeInBits() ==
+    if (LegalVT.isVector() &&
+        LegalVT.getVectorElementType().getSizeInBits() ==
             Tp->getVectorElementType()->getPrimitiveSizeInBits() &&
         LegalVT.getVectorNumElements() < Tp->getVectorNumElements()) {
 

Added: llvm/trunk/test/Transforms/SLPVectorizer/X86/PR35865.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SLPVectorizer/X86/PR35865.ll?rev=322106&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/SLPVectorizer/X86/PR35865.ll (added)
+++ llvm/trunk/test/Transforms/SLPVectorizer/X86/PR35865.ll Tue Jan  9 11:08:22 2018
@@ -0,0 +1,27 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -slp-vectorizer < %s -S -o - -mtriple=x86_64-apple-macosx10.10.0 -mcpu=core2 | FileCheck %s
+
+define void @_Z10fooConvertPDv4_xS0_S0_PKS_() {
+; CHECK-LABEL: @_Z10fooConvertPDv4_xS0_S0_PKS_(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = extractelement <16 x half> undef, i32 4
+; CHECK-NEXT:    [[CONV_I_4_I:%.*]] = fpext half [[TMP0]] to float
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast float [[CONV_I_4_I]] to i32
+; CHECK-NEXT:    [[VECINS_I_4_I:%.*]] = insertelement <8 x i32> undef, i32 [[TMP1]], i32 4
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <16 x half> undef, i32 5
+; CHECK-NEXT:    [[CONV_I_5_I:%.*]] = fpext half [[TMP2]] to float
+; CHECK-NEXT:    [[TMP3:%.*]] = bitcast float [[CONV_I_5_I]] to i32
+; CHECK-NEXT:    [[VECINS_I_5_I:%.*]] = insertelement <8 x i32> [[VECINS_I_4_I]], i32 [[TMP3]], i32 5
+; CHECK-NEXT:    ret void
+;
+entry:
+  %0 = extractelement <16 x half> undef, i32 4
+  %conv.i.4.i = fpext half %0 to float
+  %1 = bitcast float %conv.i.4.i to i32
+  %vecins.i.4.i = insertelement <8 x i32> undef, i32 %1, i32 4
+  %2 = extractelement <16 x half> undef, i32 5
+  %conv.i.5.i = fpext half %2 to float
+  %3 = bitcast float %conv.i.5.i to i32
+  %vecins.i.5.i = insertelement <8 x i32> %vecins.i.4.i, i32 %3, i32 5
+  ret void
+}