[PATCH][X86] Split VALIGN out of X86ISD::PALINGR. Add masking VALINGD/Q

[Adam Nemet]

On Aug 4, 2014, at 9:57 PM, Craig Topper <craig.topper at gmail.com> wrote:

> Just curious how much does 6 operands increase the size of the static tables for the disassembler and the asm matcher?

Sure.

The 	x86OperandSets disassembler table is small (thanks!) so even a 20% increase (5->6) only amounts to 1376 bytes.

The constant is not used directly in the asm matcher but the ConversionTable does become wider after the addition of valign.  There is an increase of 1034 bytes there.  The MatchTables do not become wider as a consequence of this change.

Adam


> On Mon, Aug 4, 2014 at 3:53 PM, Adam Nemet <anemet at apple.com> wrote:
> Apologies but the wrong version was sent out for the final patch in the series.  This should be the correct set now.
> 
> On Aug 4, 2014, at 12:12 PM, Adam Nemet <anemet at apple.com> wrote:
> 
> > Hi,
> >
> > I marked this patch X86 rather AVX512 because while its main purpose is to enable masking with valignd/q it also changes a bit how PALIGNR is handled.  I also had to increase X86_MAX_OPERANDS from 5 to 6 which is also an X86-wide change.
> >
> > The flow of the patchset is essentially to do some refactoring on the related TD part to make the rest of the patches simpler, then split the DAG nodes PALINGR and VALIGN, then add the masking support along with the intrinsics.
> >
> > See the individual patch files for more detailed explanation.
> >
> > Please let me know if it looks good.  The clang counterpart to add the intrinsic header piece is ready to go to the CFE list.
> >
> > Adam
> >
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
> 
> 
> 
> 
> -- 
> ~Craig

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140804/15b889a0/attachment.html>