[llvm] 6d86409 - [VectorCombine] fix crash while transforming constants

[Sanjay Patel via llvm-commits]

Author: Sanjay Patel
Date: 2020-06-19T12:30:32-04:00
New Revision: 6d864097a2b5e5ec6404700ac240b47932492a7f

URL: https://github.com/llvm/llvm-project/commit/6d864097a2b5e5ec6404700ac240b47932492a7f
DIFF: https://github.com/llvm/llvm-project/commit/6d864097a2b5e5ec6404700ac240b47932492a7f.diff

LOG: [VectorCombine] fix crash while transforming constants

This is a variation of the proposal in D82049 with an extra test.

Added: 
    

Modified: 
    llvm/lib/Transforms/Vectorize/VectorCombine.cpp
    llvm/test/Transforms/VectorCombine/X86/extract-binop.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
index 5cbe3d85fb7a..2ea19493a69e 100644
--- a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
+++ b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
@@ -233,6 +233,11 @@ static bool foldExtractExtract(Instruction &I, const TargetTransformInfo &TTI) {
     return false;
 
   if (ConvertToShuffle) {
+    // If the extract can be constant-folded, this code is unsimplified. Defer
+    // to other passes to handle that.
+    if (isa<Constant>(ConvertToShuffle->getOperand(0)))
+      return false;
+
     // The shuffle mask is undefined except for 1 lane that is being translated
     // to the cheap extraction lane. Example:
     // ShufMask = { 2, undef, undef, undef }

diff  --git a/llvm/test/Transforms/VectorCombine/X86/extract-binop.ll b/llvm/test/Transforms/VectorCombine/X86/extract-binop.ll
index 17fab6b5c3c1..e4f18f6083ac 100644
--- a/llvm/test/Transforms/VectorCombine/X86/extract-binop.ll
+++ b/llvm/test/Transforms/VectorCombine/X86/extract-binop.ll
@@ -547,3 +547,29 @@ define i32 @ext_ext_partial_add_reduction_and_extra_add_v4i32(<4 x i32> %x, <4 x
   %x2y210 = add i32 %x2, %y210
   ret i32 %x2y210
 }
+
+define i32 @constant_fold_crash(<4 x i32> %x) {
+; CHECK-LABEL: @constant_fold_crash(
+; CHECK-NEXT:    [[A:%.*]] = extractelement <4 x i32> <i32 16, i32 17, i32 18, i32 19>, i32 1
+; CHECK-NEXT:    [[B:%.*]] = extractelement <4 x i32> [[X:%.*]], i32 0
+; CHECK-NEXT:    [[C:%.*]] = add i32 [[A]], [[B]]
+; CHECK-NEXT:    ret i32 [[C]]
+;
+  %a = extractelement <4 x i32> <i32 16, i32 17, i32 18, i32 19>, i32 1
+  %b = extractelement <4 x i32> %x, i32 0
+  %c = add i32 %a, %b
+  ret i32 %c
+}
+
+define float @constant_fold_crash_commute(<4 x float> %x) {
+; CHECK-LABEL: @constant_fold_crash_commute(
+; CHECK-NEXT:    [[A:%.*]] = extractelement <4 x float> <float 1.600000e+01, float 1.700000e+01, float 1.800000e+01, float 1.900000e+01>, i32 3
+; CHECK-NEXT:    [[B:%.*]] = extractelement <4 x float> [[X:%.*]], i32 1
+; CHECK-NEXT:    [[C:%.*]] = fadd float [[B]], [[A]]
+; CHECK-NEXT:    ret float [[C]]
+;
+  %a = extractelement <4 x float> <float 16.0, float 17.0, float 18.0, float 19.0>, i32 3
+  %b = extractelement <4 x float> %x, i32 1
+  %c = fadd float %b, %a
+  ret float %c
+}