Fix to handle all non-power-of-2 vector sizes in the mask form of _builtin_shuffle_ve...

Previously a 2-bit mask was used to mask each element of a vec6 mask before doing the extracts instead of 3-bit mask necessary to cover 0-5. vec3 was the only non-power-of-2 that worked correctly because a +1 conditionally added before calculating floor(log2(elements)).

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@187560 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/CGExprScalar.cpp b/lib/CodeGen/CGExprScalar.cpp
index d3250ea646383fa743db25fdd0472da9cb7aa2d2..c0355f690de11fbc09a6faeadb6c3e1e5ab6da24 100644 (file)
--- a/lib/CodeGen/CGExprScalar.cpp
+++ b/lib/CodeGen/CGExprScalar.cpp
@@ -922,16 +922,8 @@ Value *ScalarExprEmitter::VisitShuffleVectorExpr(ShuffleVectorExpr *E) {
     llvm::VectorType *MTy = cast<llvm::VectorType>(Mask->getType());
     llvm::Constant* EltMask;
 
-    // Treat vec3 like vec4.
-    if ((LHSElts == 6) && (E->getNumSubExprs() == 3))
-      EltMask = llvm::ConstantInt::get(MTy->getElementType(),
-                                       (1 << llvm::Log2_32(LHSElts+2))-1);
-    else if ((LHSElts == 3) && (E->getNumSubExprs() == 2))
-      EltMask = llvm::ConstantInt::get(MTy->getElementType(),
-                                       (1 << llvm::Log2_32(LHSElts+1))-1);
-    else
-      EltMask = llvm::ConstantInt::get(MTy->getElementType(),
-                                       (1 << llvm::Log2_32(LHSElts))-1);
+    EltMask = llvm::ConstantInt::get(MTy->getElementType(),
+                                     llvm::NextPowerOf2(LHSElts-1)-1);
 
     // Mask off the high bits of each shuffle index.
     Value *MaskBits = llvm::ConstantVector::getSplat(MTy->getNumElements(),