Re: [FFmpeg-devel] [PATCH] avcodec/vp9: add tile threading support

2017-09-08 Thread Paul B Mahol
On 9/8/17, Ilia Valiakhmetov wrote: > Signed-off-by: Ilia Valiakhmetov > --- > Changelog | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/Changelog b/Changelog > index cae5254..8a4818a 100644 > --- a/Changelog > +++ b/Changelog > @@ -43,6 +43,7 @@

[FFmpeg-devel] [PATCH] avcodec/vp9: add tile threading support

2017-09-08 Thread Ilia Valiakhmetov
Signed-off-by: Ilia Valiakhmetov --- Changelog | 1 + 1 file changed, 1 insertion(+) diff --git a/Changelog b/Changelog index cae5254..8a4818a 100644 --- a/Changelog +++ b/Changelog @@ -43,6 +43,7 @@ version : - add --disable-autodetect build switch - drop deprecated qtkit

Re: [FFmpeg-devel] [PATCH] avcodec/vp9: Add tile threading support

2017-08-23 Thread Ronald S. Bultje
Hi, On Tue, Aug 22, 2017 at 7:59 PM, Ilia Valiakhmetov wrote: > +if (s->td) > +av_free(s->td); > + > +s->td = av_mallocz_array(s->s.h.tiling.tile_cols, > sizeof(VP9TileData)); The free() leaks s->td[*].b/block_base if they were already

Re: [FFmpeg-devel] [PATCH] avcodec/vp9: Add tile threading support

2017-08-23 Thread Ronald S. Bultje
Hi, now for review of the vp9-specific bits. On Tue, Aug 22, 2017 at 7:59 PM, Ilia Valiakhmetov wrote: > +lflvl_len*sizeof(*s->lflvl) + 16 * > sizeof(*s->above_mv_ctx))); > lflvl_len * sizeof (space). > -// these will be re-allocated a

Re: [FFmpeg-devel] [PATCH] avcodec/vp9: Add tile threading support

2017-08-23 Thread Ronald S. Bultje
Hi, I would also recommend to split out the changes to add the mainfunc-aware slice threading implementation (execute3) into pthread_slice.c into its own patch, that way others can review that without being obstructed by the vp9 bits. On Tue, Aug 22, 2017 at 7:59 PM, Ilia Valiakhmetov

Re: [FFmpeg-devel] [PATCH] avcodec/vp9: Add tile threading support

2017-08-22 Thread Ronald S. Bultje
Hi, overall pretty good. Nice performance improvement. We typically note performance improvement as a/b-1, not 1-b/a, so if runtime with 2 threads is 6 seconds and with 1 thread is 10 seconds, the improvement is not 1-6/10=0.4=40%, but rather 10/6-1=0.67=67%. On Tue, Aug 22, 2017 at 7:59 PM,