[FFmpeg-devel] armeb fix for mpegaudio

[matthieu castet]

Siarhei Siamashka wrote:
> On Friday 01 August 2008, matthieu castet wrote:
>> Hi,
>>
>> this patch fix mp2 regression test on arm big endian.
> 
> Does FFmpeg officially support big endian byte order on ARM?
If you look at the archive someone told that ffmpeg was working on 
armeb. There is also the debian armeb port.

> 
> Just if you look at 'armv4l' abbreviation, it can be decoded
> as ARMv4 or higher cores with little endian byte order.
May be. In that case we should disable them in case of armeb.

> 
>> PS : as Mans said dct is broken as you can see with the video regression
>> failure.
> 
> Most assembly optimizations for ARM (including IDCT) assume little endian byte
> order, some assembly optimized code even mentions it in comments as a
> requirement.
> 
Well I am for fixing easy stuff and disabling more complex thing.

> 
> BTW, if I'm not mistaken, regression test from FFmpeg uses just C
> implementation of (i)DCT at least on ARM unless you have specifically 
> modified something.
Ok, I will done more test on a codec that fails and try to analyse where 
the failure come from.


> 
>> PS2 : test were done with qemu-armeb
> 
> Do you know of any real systems (HW + OS) that could be used as ARM big endian
> target for FFmpeg?
Some Linksys NSLU2.
$grep BIG mach-*/Kconfig
mach-ixp2000/Kconfig:config ARCH_SUPPORTS_BIG_ENDIAN
mach-ixp23xx/Kconfig:config ARCH_SUPPORTS_BIG_ENDIAN
mach-ixp4xx/Kconfig:config ARCH_SUPPORTS_BIG_ENDIAN

samsung s3c HW also support big endian with minor Linux modification.


> Failed regression tests do not automatically mean problems
> in FFmpeg, it could be also some bugs in qemu.
Of course.


Matthieu