[PATCH] R600: Match sign_extend_inreg to BFE instructions

Jan Vesely jan.vesely at rutgers.edu
Fri Mar 21 11:32:14 PDT 2014


On Thu, 2014-03-20 at 09:52 -0700, Matt Arsenault wrote:
> On 03/20/2014 09:30 AM, Jan Vesely wrote:
> > This patch regresses parts of the following piglit tests on my Turks
> > card:
> > builtin-char-abs-1.0.generated
> > builtin-char-abs_diff-1.0.generated
> > builtin-char-clamp-1.1.generated
> > builtin-char-hadd-1.0.generated
> > builtin-char-max-1.1.generated
> >
> > builtin-short-abs-1.0.generated
> > builtin-short-abs_diff-1.0.generated
> > builtin-short-clamp-1.1.generated
> > builtin-short-hadd-1.0.generated
> > builtin-short-max-1.1.generated
> >
> > Only the vector variants of byte/word size types are affected, so the
> > attached patch fixes the issue by expanding vectors.
> >
> > regards,
> > Jan
> 
> Can you add a test from one of these failing cases? I thought I had 
> added a bunch of tests for these vector cases, but I didn't check the 
> evergreen versions particularly carefully. As you can see, the EG checks 
> don't bother checking the operands.

I'll try to find some time this weekend or next week.

-- 
Jan Vesely <jan.vesely at rutgers.edu>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 836 bytes
Desc: This is a digitally signed message part
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140321/b2859caa/attachment.sig>


More information about the llvm-commits mailing list