[Liboil] x86 broken in CVS

[Eric Anholt]

On Tue, 2005-12-13 at 22:12 +0100, Stephane Fillod wrote:
> On Tue, Dec 13, 2005 at 02:26:33AM -0800, Eric Anholt wrote:
> > On Tue, 2005-12-13 at 10:54 +0100, Loïc Minier wrote:
> > > On Mon, Dec 12, 2005, Stephane Fillod wrote:
> > > > BTW, one test of the "make check" fails on fbCompositeSolid with 
> > > > the following reason:
> > > > OIL: ERROR ../../liboil/liboil/liboiltest.c 351: oil_test_check_impl(): function fbCompositeSolid_nx8888mmx wrote after area for parameter i_n
> > > > OIL: ERROR ../../liboil/liboil/liboiltest.c 368: oil_test_check_impl(): function fbCompositeSolid_nx8888mmx in class composite_over_argb_const_src failed check (0 > 91) outside=1
> > > 
> > >  This seems unrelated to your patch, as Debian's liboil produces this
> > >  errors but is not affected by the x86_64 rbx issue.
> > > 
> > >  This problem is the #1 reported installation issue with gst-plugins0.8
> > >  and liboil in Debian and AFAICT is present since 0.3.
> > 
> > Right, it's unrelated.  A fix might be attached, though it's late and I
> > don't fully trust myself at this point.
> [..]
> 
> I confirm, it was unrelated. 2 reports were put in the same mail, 
> the first one with a solution, the second without.
> The patch you submitted for the second issue seemed to have fixed 
> the problem on my system. FWIW:
> 
> composite_over_argb_const_src (uint32_t *i_n, uint32_t *s1_1, int n)
>   all implementations:
>     fbCompositeSolid_nx8888mmx
>       profile: 1950.5 ticks (std.dev. 1.26168)
>       sum abs difference: 0 (n=100)
>       currently chosen
>     composite_over_argb_const_src_ref
>       flags: REF
>       profile: 5551 ticks (std.dev. 1.125)

OK, I'll give it another look over and commit soon.

> However, the latest cvs update introduced a new regression:
> 
> OIL: ERROR ../../liboil/liboil/liboiltest.c 351: oil_test_check_impl(): function multiply_f32_sse_unroll2 wrote after area for parameter d
> OIL: ERROR ../../liboil/liboil/liboiltest.c 368: oil_test_check_impl(): function multiply_f32_sse_unroll2 in class multiply_f32 failed check (0 > 22) outside=1
> OIL: ERROR ../../liboil/liboil/liboiltest.c 351: oil_test_check_impl(): function multiply_f32_sse wrote after area for parameter d
> OIL: ERROR ../../liboil/liboil/liboiltest.c 368: oil_test_check_impl(): function multiply_f32_sse in class multiply_f32 failed check (0 > 22) outside=1

How latest?  I committed some fixes at:
date: 2005/12/13 21:17:26;  author: anholt;  state: Exp;  lines: +20 -0
according to CVS.

I'm working on some code so we can test alignment issues (which that
commit was about) automatically from the testsuite.  It's kinda working
already.

-- 
Eric Anholt                                     eta at lclark.edu
http://people.freebsd.org/~anholt/              anholt at FreeBSD.org
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: This is a digitally signed message part
Url : http://lists.freedesktop.org/archives/liboil/attachments/20051213/1c84fa19/attachment.pgp