[MPlayer-users] Fresh svn mplayer crashes on many ac3 streams

[Ivan Kalvachev]

On 5/29/11, Ivan Kalvachev <ikalvachev at gmail.com> wrote:
> On 5/29/11, Vladimir Mosgalin <mosgalin at vm10124.spb.edu> wrote:
>> Hi Reimar Döffinger!
>>
>>  On 2011.05.28 at 22:14:58 +0200, Reimar Döffinger wrote next:
>>
>>> > The thing is, you see, this is default mode; on sandy bridge systems
>>> > "-march=native" with gcc 4.6 or higher
>>>
>>> Huh? 4.6.0 is the latest version.
>>> And honestly, the answer is: you never, never ever use a .0 version of
>>> gcc
>>> for stuff you actually use.
>>> To my knowledge there hasn't been a single one without serious issues
>>> and
>>> MPlayer most of the time is affected.
>>> Slightly more useful: -fno-tree-vectorize probably should always be used
>>> as for FFmpeg, the only thing it does is make the code slower and cause
>>> bugs.
>>
>> But it's latest stable version. Plus it's the only compiler in current
>> stable fedora release.
>>
>> Anyhow it's a bit unfair to blame it on ".0" here, I think, since this
>> issue is related to avx as well and 4.6.0 was first stable release to
>> include avx support. It's, like, not gccs fault that avx support
>> appeared in .0 version and not some other..
>> (plus, we don't yet known if it's gcc problem or actual bug in ac3dec.c
>> code..)
>
> This is phenomenon we have observed since 3.0.0 onward.
> (Actually gcc-2.96.0 was the worse compiler ever to compile linux.
> However it was not official release. It was just gcc3 ported back by
> RedHat in a very bad way. )
>
>
>> I can turn off avx or ftree-vectorize, it's np to me - though I expected
>> some serious speedups from gcc 4.6.0 and avx support, like shown in some
>> phoronix benchmarks, for example
>> http://www.phoronix.com/data/img/results/intel_avx_gcc/6.png
>> http://www.phoronix.com/data/img/results/intel_avx_gcc/7.png
>
> All speed critical parts of codecs are separated and assembly
> functions are written to use the SIMD available on the CPU. In that
> sense you are not going to get significant speedup from the automatic
> SIMD. That is if it produced optimal code.
>
> Of course, AVX optimizations are still in process of writing.
>
>> but I guess the real issue here is that ffmpeg's configure isn't
>> executed so ffmpeg's will to add -fno-tree-vectorize flag is ignored?
>
> The vectorize may break code and in other parts of MPlayer, so better
> be safe and use it everywhere.

Code to use -fno-tree-vectorize is already in the svn.