[FFmpeg-devel] Fw: [foms] Paper submissions to LCA

[Måns Rullgård]

Carl Eugen Hoyos <cehoyos at ag.or.at> writes:

> M?ns Rullg?rd <mans <at> mansr.com> writes:
>
>> On that topic, does anyone know why so many FATE tests are failing
>> with icc?
>
> (It exclusively fails for 11.x 64bit and I wrote a wiki page about it.)
> Some files are mis-compiled at optimisations > 0:
> vc1.c
> h264.c
> h263.c
> snow.c
> wmv2dec.c

That much is more or less clear from the test results. ;-)

> Earlier this year, I looked into wmv2dec.c (it is the shortest of
> these files), and found that the problem is wmv2_pred_motion().
> (Adding printf's after the initialisation of *C shows that mot_val,
> A[], B[] and C[] often differ from compilation with -O0 while xy and
> wrap are unchanged.)
> Perhaps you could have a look what is so special about this (short)
> function?  I could upload the assembly of O0 and O1 if you want.
>
> I reported the problem to Intel, but it took them some time to reproduce the
> problem. It is reproduced (escalated) now (issue 540759):
> "I've narrowed down the problem to one of our internal modules"

Good to know they're working on it.

> I have no idea if the other files have the same problem (there is
> some indication that vc1.c has a different problem), but they are
> rather big making it difficult to find the failing function.

I just wanted to make sure it wasn't some silly aliasing violation in
our code causing it.  There are quite a few warnings of that kind from
h264.c.

-- 
M?ns Rullg?rd
mans at mansr.com