[ScalarizeMaskedMemIntrin] Add constant mask support to expandload and compressstore scalarization
This adds support for generating all the loads or stores for a constant mask into a single basic block with no conditionals.
Differential Revision: https://reviews.llvm.org/D65613
llvm-svn: 367715
diff --git a/llvm/lib/CodeGen/ScalarizeMaskedMemIntrin.cpp b/llvm/lib/CodeGen/ScalarizeMaskedMemIntrin.cpp
index 456c483..71bd0fe 100644
--- a/llvm/lib/CodeGen/ScalarizeMaskedMemIntrin.cpp
+++ b/llvm/lib/CodeGen/ScalarizeMaskedMemIntrin.cpp
@@ -616,6 +616,24 @@
// The result vector
Value *VResult = PassThru;
+ // Shorten the way if the mask is a vector of constants.
+ if (isConstantIntVector(Mask)) {
+ unsigned MemIndex = 0;
+ for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
+ if (cast<Constant>(Mask)->getAggregateElement(Idx)->isNullValue())
+ continue;
+ Value *NewPtr = Builder.CreateConstInBoundsGEP1_32(EltTy, Ptr, MemIndex);
+ LoadInst *Load =
+ Builder.CreateAlignedLoad(EltTy, NewPtr, 1, "Load" + Twine(Idx));
+ VResult =
+ Builder.CreateInsertElement(VResult, Load, Idx, "Res" + Twine(Idx));
+ ++MemIndex;
+ }
+ CI->replaceAllUsesWith(VResult);
+ CI->eraseFromParent();
+ return;
+ }
+
for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
// Fill the "else" block, created in the previous iteration
//
@@ -694,6 +712,22 @@
unsigned VectorWidth = VecType->getNumElements();
+ // Shorten the way if the mask is a vector of constants.
+ if (isConstantIntVector(Mask)) {
+ unsigned MemIndex = 0;
+ for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
+ if (cast<Constant>(Mask)->getAggregateElement(Idx)->isNullValue())
+ continue;
+ Value *OneElt =
+ Builder.CreateExtractElement(Src, Idx, "Elt" + Twine(Idx));
+ Value *NewPtr = Builder.CreateConstInBoundsGEP1_32(EltTy, Ptr, MemIndex);
+ Builder.CreateAlignedStore(OneElt, NewPtr, 1);
+ ++MemIndex;
+ }
+ CI->eraseFromParent();
+ return;
+ }
+
for (unsigned Idx = 0; Idx < VectorWidth; ++Idx) {
// Fill the "else" block, created in the previous iteration
//