maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

decouple MMX check from regular splat checks. Some refactoring is coming, and MMX should be left alone to be easily removed after moving to intrinsics 

llvm-svn: 113247
This commit is contained in:
Bruno Cardoso Lopes 2010-09-07 18:41:45 +00:00
parent 8499f47ef4
commit 5a45db3e6c
1 changed files with 8 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
llvm/lib/Target/X86/X86ISelLowering.cpp
View File

@ -5271,12 +5271,16 @@ X86TargetLowering::LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const {
if (isZeroShuffle(SVOp))
return getZeroVector(VT, Subtarget->hasSSE2(), DAG, dl);
// Promote splats to v4f32.
if (SVOp->isSplat()) {
if (isMMX)
// FIXME: this is somehow handled during isel by MMX pattern fragments. Remove
// the check or come up with another solution when all MMX move to intrinsics,
// but don't allow this to be considered legal, we don't want vector_shuffle
// operations to be matched during isel anymore.
if (isMMX && SVOp->isSplat())
return Op;
// Promote splats to v4f32.
if (SVOp->isSplat())
return PromoteSplat(SVOp, DAG);
}
// If the shuffle can be profitably rewritten as a narrower shuffle, then
// do it!