[FFmpeg-devel] [PATCH] Fix mm_flags, mm_support for ARM

Måns Rullgård mans
Fri Jun 27 21:49:08 CEST 2008


Siarhei Siamashka <siarhei.siamashka at gmail.com> writes:

> On Friday 27 June 2008, matthieu castet wrote:
>> Benoit Fouet wrote:
>> > Michael Niedermayer wrote:
>> >> On Thu, Jun 26, 2008 at 10:45:20PM +0200, matthieu castet wrote:
>> >
>> > Matthieu, could you please resend the complete patch ?
>>
>> I attach them as 2 separate patch.
>> fix_dct-test-arm.diff : allow to build dct-test on arm
>> arm-icdt-test.diff : add some arm idct (there are other version that
>> could be added, but I have no hardware to test).
>
>> +#ifdef ARCH_ARMV4L
>> + ?//{"IDCT_ARM", ? ? ? ?1, j_rev_dct_ARM, ? ? ?idct, MMX_PERM},
>> /* should > be MMX_PERM perm, but produce strange result */ 
>
> Yes, 'j_rev_dct_ARM' is broken, it produces quite noticeable
> artefacts on decoding video. I think I already reported this issue
> here long ago. But as it is armv4 only, I was not interested in
> investigating this problem further...

It is indeed in pretty poor shape.  It's pretty fast though, if that's
any excuse.

For completeness, here's the output of dct-test -i including all the
ARM IDCT variants on a 500MHz Cortex-A8.  It is interesting to note
that the ones targeting older ARM cores are significantly slower than
plain C on the Cortex.

I've also attached (part of) my dct-test.c patch.  It seems we all
have one...

-- 
M?ns Rullg?rd
mans at mansr.com
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: dct-test-arm-cortex-a8-500
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080627/c9432a8f/attachment.txt>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-add-ARM-IDCTs-to-dct-test.patch
Type: text/x-patch
Size: 1489 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20080627/c9432a8f/attachment.bin>



More information about the ffmpeg-devel mailing list