BBVectorize: Correctly merge SubclassOptionalData

When two instructions are combined into a vector instruction,
the resulting instruction must have the most-conservative flags.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@168765 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Vectorize/BBVectorize.cpp b/lib/Transforms/Vectorize/BBVectorize.cpp
index dacbc7f24257e95f48df3991bb64c8ac9a55d75d..f7be3e31240770e72eeb64c96a0e1cc1ec0ac29b 100644 (file)
--- a/lib/Transforms/Vectorize/BBVectorize.cpp
+++ b/lib/Transforms/Vectorize/BBVectorize.cpp
@@ -2903,6 +2903,7 @@ namespace {
         K->mutateType(getVecTypeForPair(L->getType(), H->getType()));
 
       combineMetadata(K, H);
+      K->intersectOptionalDataWith(H);
 
       for (unsigned o = 0; o < NumOperands; ++o)
         K->setOperand(o, ReplacedOperands[o]);

diff --git a/test/Transforms/BBVectorize/simple.ll b/test/Transforms/BBVectorize/simple.ll
index d9a12eebed85534cc9f8295afa62909d349538b4..3527ae75b457d40bdf295ee9be143187cdab3046 100644 (file)
--- a/test/Transforms/BBVectorize/simple.ll
+++ b/test/Transforms/BBVectorize/simple.ll
@@ -173,3 +173,27 @@ define double @test7(double %A1, double %A2, double %B1, double %B2) {
 ; CHECK: ret double %R
 }
 
+; Basic depth-3 chain (subclass data)
+define i64 @test8(i64 %A1, i64 %A2, i64 %B1, i64 %B2) {
+; CHECK: @test8
+; CHECK: %X1.v.i1.1 = insertelement <2 x i64> undef, i64 %B1, i32 0
+; CHECK: %X1.v.i1.2 = insertelement <2 x i64> %X1.v.i1.1, i64 %B2, i32 1
+; CHECK: %X1.v.i0.1 = insertelement <2 x i64> undef, i64 %A1, i32 0
+; CHECK: %X1.v.i0.2 = insertelement <2 x i64> %X1.v.i0.1, i64 %A2, i32 1
+       %X1 = sub nsw i64 %A1, %B1
+       %X2 = sub i64 %A2, %B2
+; CHECK: %X1 = sub <2 x i64> %X1.v.i0.2, %X1.v.i1.2
+       %Y1 = mul i64 %X1, %A1
+       %Y2 = mul i64 %X2, %A2
+; CHECK: %Y1 = mul <2 x i64> %X1, %X1.v.i0.2
+       %Z1 = add i64 %Y1, %B1
+       %Z2 = add i64 %Y2, %B2
+; CHECK: %Z1 = add <2 x i64> %Y1, %X1.v.i1.2
+       %R  = mul i64 %Z1, %Z2
+; CHECK: %Z1.v.r1 = extractelement <2 x i64> %Z1, i32 0
+; CHECK: %Z1.v.r2 = extractelement <2 x i64> %Z1, i32 1
+; CHECK: %R = mul i64 %Z1.v.r1, %Z1.v.r2
+       ret i64 %R
+; CHECK: ret i64 %R
+}
+