replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Allow v16i16 and v32i8 shuffles to be rewritten as narrower shuffles. 

llvm-svn: 156156
This commit is contained in:
Craig Topper 2012-05-04 04:44:49 +00:00
parent fa0ebcd1b0
commit 42f2182366
2 changed files with 16 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
llvm/lib/Target/X86/X86ISelLowering.cpp
View File

@ -5920,10 +5920,12 @@ SDValue RewriteAsNarrowerShuffle(ShuffleVectorSDNode *SVOp,
unsigned Scale;
switch (VT.SimpleTy) {
default: llvm_unreachable("Unexpected!");
case MVT::v4f32: NewVT = MVT::v2f64; Scale = 2; break;
case MVT::v4i32: NewVT = MVT::v2i64; Scale = 2; break;
case MVT::v8i16: NewVT = MVT::v4i32; Scale = 2; break;
case MVT::v16i8: NewVT = MVT::v4i32; Scale = 4; break;
case MVT::v4f32: NewVT = MVT::v2f64; Scale = 2; break;
case MVT::v4i32: NewVT = MVT::v2i64; Scale = 2; break;
case MVT::v8i16: NewVT = MVT::v4i32; Scale = 2; break;
case MVT::v16i8: NewVT = MVT::v4i32; Scale = 4; break;
case MVT::v16i16: NewVT = MVT::v8i32; Scale = 2; break;
case MVT::v32i8: NewVT = MVT::v8i32; Scale = 4; break;
}
SmallVector<int, 8> MaskVec;
@ -6370,7 +6372,8 @@ X86TargetLowering::NormalizeVectorShuffle(SDValue Op, SelectionDAG &DAG) const {
// If the shuffle can be profitably rewritten as a narrower shuffle, then
// do it!
if (VT == MVT::v8i16 || VT == MVT::v16i8) {
if (VT == MVT::v8i16 || VT == MVT::v16i8 ||
VT == MVT::v16i16 || VT == MVT::v32i8) {
SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG, dl);
if (NewOp.getNode())
return DAG.getNode(ISD::BITCAST, dl, VT, NewOp);

8
llvm/test/CodeGen/X86/avx-shuffle.ll
View File

@ -202,3 +202,11 @@ define <4 x i64> @blend4(<4 x i64> %a, <4 x i64> %b) nounwind alwaysinline {
%t = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 0, i32 1, i32 2, i32 7>
ret <4 x i64> %t
}
; CHECK: narrow
; CHECK: vpermilps
; CHECK: ret
define <16 x i16> @narrow(<16 x i16> %a) nounwind alwaysinline {
%t = shufflevector <16 x i16> %a, <16 x i16> undef, <16 x i32> <i32 2, i32 3, i32 undef, i32 1, i32 6, i32 7, i32 4, i32 5, i32 10, i32 11, i32 8, i32 undef, i32 14, i32 15, i32 undef, i32 undef>
ret <16 x i16> %t
}