6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-07-21 18:29:45 +00:00
Code Issues Projects Releases Wiki Activity

[x86] Teach the vector shuffle yet another step of canonicalization.

Browse Source 
No functionality changed yet, but this will prevent subsequent patches
from having to handle permutations of various interleaved shuffle
patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@222614 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Chandler Carruth 2014-11-22 09:18:53 +00:00
parent 8ff39c5c44
commit e915b4b7c8
1 changed files with 13 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
lib/Target/X86/X86ISelLowering.cpp
View File

@ -10833,7 +10833,8 @@ static SDValue lowerVectorShuffle(SDValue Op, const X86Subtarget *Subtarget,
// When the number of V1 and V2 elements are the same, try to minimize the
// number of uses of V2 in the low half of the vector. When that is tied,
// ensure that the sum of indices for V1 is equal to or lower than the sum
// indices for V2.
// indices for V2. When those are equal, try to ensure that the number of odd
// indices for V1 is lower than the number of odd indices for V2.
if (NumV1Elements == NumV2Elements) {
int LowV1Elements = 0, LowV2Elements = 0;
for (int M : SVOp->getMask().slice(0, NumElements / 2))
@ -10850,8 +10851,18 @@ static SDValue lowerVectorShuffle(SDValue Op, const X86Subtarget *Subtarget,
SumV2Indices += i;
else if (SVOp->getMask()[i] >= 0)
SumV1Indices += i;
if (SumV2Indices < SumV1Indices)
if (SumV2Indices < SumV1Indices) {
return DAG.getCommutedVectorShuffle(*SVOp);
} else if (SumV2Indices == SumV1Indices) {
int NumV1OddIndices = 0, NumV2OddIndices = 0;
for (int i = 0, Size = SVOp->getMask().size(); i < Size; ++i)
if (SVOp->getMask()[i] >= NumElements)
NumV2OddIndices += i % 2;
else if (SVOp->getMask()[i] >= 0)
NumV1OddIndices += i % 2;
if (NumV2OddIndices < NumV1OddIndices)
return DAG.getCommutedVectorShuffle(*SVOp);
}
}
}