[PATCH] D19661: [X86] Also try to zero elts when lowering v32i8 shuffle with PSHUFB.
Sanjay Patel via llvm-commits
llvm-commits at lists.llvm.org
Thu Apr 28 08:04:25 PDT 2016
spatel added inline comments.
================
Comment at: lib/Target/X86/X86ISelLowering.cpp:11427-11436
@@ -11426,7 +11426,12 @@
- if (isSingleInputShuffleMask(Mask)) {
- // There are no generalized cross-lane shuffle operations available on i8
- // element types.
- if (is128BitLaneCrossingShuffleMask(MVT::v32i8, Mask))
- return lowerVectorShuffleAsLanePermuteAndBlend(DL, MVT::v32i8, V1, V2,
- Mask, DAG);
+ SmallBitVector Zeroable = computeZeroableShuffleElements(Mask, V1, V2);
+ bool SingleInputMask = true;
+ bool SingleInputAndZeroesMask = true;
+ for (int i = 0, Size = Mask.size(); i < Size; ++i) {
+ if (Mask[i] >= Size) {
+ SingleInputMask = false;
+ if (!Zeroable[i])
+ SingleInputAndZeroesMask = false;
+ }
+ }
+ // There are no generalized cross-lane shuffle operations available on i8
----------------
Would it make sense to enhance isSingleInputShuffleMask() like:
bool isSingleInputShuffleMask(ArrayRef<int> Mask, bool OrZeroes = false) {}
and then optionally check for zeroes there?
It seems like vperm2* and maybe some other instructions would then be able to use the same helper.
================
Comment at: lib/Target/X86/X86ISelLowering.cpp:11445
@@ -11434,7 +11444,3 @@
SDValue PSHUFBMask[32];
- for (int i = 0; i < 32; ++i)
- PSHUFBMask[i] =
- Mask[i] < 0
- ? DAG.getUNDEF(MVT::i8)
- : DAG.getConstant(Mask[i] < 16 ? Mask[i] : Mask[i] - 16, DL,
- MVT::i8);
+ for (int i = 0; i < 32; ++i) {
+ if (Mask[i] < 0)
----------------
Please add a comment here to describe the pshufb encoding. Something like:
If the most-significant-bit of any byte of the PSHUFB mask is set, the destination byte is zeroed.
I hate having to look back at the manual. :)
================
Comment at: lib/Target/X86/X86ISelLowering.cpp:11446
@@ +11445,3 @@
+ for (int i = 0; i < 32; ++i) {
+ if (Mask[i] < 0)
+ PSHUFBMask[i] = DAG.getUNDEF(MVT::i8);
----------------
Use SM_SentinelUndef?
http://reviews.llvm.org/D19661
More information about the llvm-commits
mailing list