[FFmpeg-devel] [PATCH] avcodec: loongson3 optimized mpegvideo dct unquantize and denoise with mmi

[周晓勇]

> > > On Wed, May 27, 2015 at 06:10:31PM +0800, 周晓勇 wrote:
> > > > From eb476eba51fffd30989917606a9b239f7c67ad26 Mon Sep 17 00:00:00 2001
> > > > From: ZhouXiaoyong <zhouxiaoyong at loongson.cn>
> > > > Date: Thu, 21 May 2015 16:32:35 +0800
> > > > Subject: [PATCH] avcodec: loongson3 optimized mpegvideo dct unquantize and
> > > >  denoise with mmi
> > > > 
> > > > 
> > > > Signed-off-by: ZhouXiaoyong <zhouxiaoyong at loongson.cn>
> > > > ---
> > > >  libavcodec/mips/Makefile              |   2 +
> > > >  libavcodec/mips/mpegvideo_init_mips.c |  40 ++++
> > > >  libavcodec/mips/mpegvideo_mips.h      |  36 ++++
> > > >  libavcodec/mips/mpegvideo_mmi.c       | 365 ++++++++++++++++++++++++++++++++++
> > > >  libavcodec/mpegvideo.c                |   2 +
> > > >  libavcodec/mpegvideo.h                |   1 +
> > > >  6 files changed, 446 insertions(+)
> > > 
> > > this seems to break fate-mpeg4-resolution-change-up-down on loongson3
> > > 
> > > [...]
> > i have done the test, fate-mpeg4 passed all with no problem.
> > and also, all fate passed, not only the loongson3a but also loongson3b.
> > 
> > i wonder the break maybe caused by the os, the FC19 version on your machine is FC19-Build090-20150317.
> > here is the Build-010 download address: 
> > http://www.loongnix.org/dev/ftp/os/Fedora19/iso/Fedora19-Desktop-Loongson-Release-Build010-20150429.iso
> > 
> > i have the same problem, for example, make fate -j4 break down at vsynth_lena-h263p.
> > but do make fate-vsynth_lena-h263p again, i will pass.
> > i do the fate-test on loongson3b machine with same setting and evironment, it pass all correctly.
> > i doubt it's about the cpu's SIMD unit problem or the glib-math's, also maybe the multi-threads compiling.
> > so, you could remake the breakdown fate-mpeg4-resolution-change-up-down for a try.
> 
> trying
> make -j4 fate-vsynth_lena-mpeg4 fate-vsynth_lena-mpeg4-error fate-vsynth_lena-mpeg4-nsse fate-vsynth_lena-mpeg4-qpel fate-vsynth1-h263-obmc fate-vsynth1-mpeg4-adv fate-vsynth1-wmv2 fate-vsynth2-h263-obmc fate-vsynth2-rv10 fate-vsynth2-rv20 fate-lavf-ismv
> it succeeded 100% of the time before the patch while 50% of the
> time after the patch something random fails
> also -j1 fails after the patch occasionally
> 
> why do you think upgrading the OS would help ?
> i assume you have the latest and you said it happens for you too
> 
> [...]

why the test successed randomly? do you have encountered this situation?
why i suggest to update os is because i achieved more success probability on build-010.
by the way, all test successed 100% exactly on loongson3b machine.
so, if the random fail is not acceptable and you assure there's no relationship with the fate-test itself, i will abort this patch and rewrite.