On Sat, Sep 12, 2015 at 3:35 AM, Ilia Mirkin <imir...@alum.mit.edu> wrote: > On Mon, Sep 7, 2015 at 8:15 AM, Julien Isorce <j.iso...@samsung.com> wrote: >> https://bugs.freedesktop.org/show_bug.cgi?id=89969 >> >> Signed-off-by: Julien Isorce <j.iso...@samsung.com> >> --- >> src/gallium/drivers/nouveau/nouveau_vp3_video.h | 3 +++ >> src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c | 8 ++++++++ >> 2 files changed, 11 insertions(+) >> >> diff --git a/src/gallium/drivers/nouveau/nouveau_vp3_video.h >> b/src/gallium/drivers/nouveau/nouveau_vp3_video.h >> index 43b9556..87ae012 100644 >> --- a/src/gallium/drivers/nouveau/nouveau_vp3_video.h >> +++ b/src/gallium/drivers/nouveau/nouveau_vp3_video.h >> @@ -121,6 +121,9 @@ struct nouveau_vp3_decoder { >> >> /* Total data appended so far after last begin frame. */ >> unsigned bsp_size; >> + >> + /* Number of slices so far since last begin frame */ >> + unsigned int nb_slices; >> }; >> >> struct comm { >> diff --git a/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c >> b/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c >> index c1d70ea..7776e4e 100644 >> --- a/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c >> +++ b/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c >> @@ -128,6 +128,8 @@ nouveau_vp3_fill_picparm_mpeg12_bsp(struct >> nouveau_vp3_decoder *dec, >> for (i = 0; i < 4; ++i) >> pic_bsp->f_code[i/2][i%2] = desc->f_code[i/2][i%2] + 1; // FU >> >> + desc->num_slices = dec->nb_slices; >> + >> return (desc->num_slices << 4) | (dec->base.profile != >> PIPE_VIDEO_PROFILE_MPEG1); >> } >> >> @@ -196,6 +198,8 @@ nouveau_vp3_fill_picparm_h264_bsp(struct >> nouveau_vp3_decoder *dec, >> struct h264_picparm_bsp stub_h = {}, *h = &stub_h; >> uint32_t caps = (d->slice_count << 4)&0xfff0; >> >> + d->slice_count = dec->nb_slices; > > But caps above is already set based on d->slice_count. With this diff, > vaapi works *much* better: > > diff --git a/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c > b/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c > index 7776e4e..7aa70e2 100644 > --- a/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c > +++ b/src/gallium/drivers/nouveau/nouveau_vp3_video_bsp.c > @@ -196,11 +196,12 @@ nouveau_vp3_fill_picparm_h264_bsp(struct > nouveau_vp3_decoder *dec, > char *map) > { > struct h264_picparm_bsp stub_h = {}, *h = &stub_h; > - uint32_t caps = (d->slice_count << 4)&0xfff0; > + uint32_t caps; > > - d->slice_count = dec->nb_slices; > + d->slice_count += dec->nb_slices; > > assert(!(d->slice_count & ~0xfff)); > + caps = (d->slice_count << 4)&0xfff0; > if (d->slice_count & 0x1000) > caps |= 1 << 20; > > > I'm not sure about the += -- can you investigate what the proper way > of handling this is? Also note that you appear to have forgotten to do > this for vc1. > > Very interesting by the way -- the vaapi decode does not have the same > problem as the vdpau decode on that simpsons trailer. (About 28s in, > when the sleigh is flying over the ravine.)
Errrr... nevermind. This in no way fixes the vaapi situation. I just had a stale makefile. But the issue I point out above still seems relevant. > >> + >> assert(!(d->slice_count & ~0xfff)); >> if (d->slice_count & 0x1000) >> caps |= 1 << 20; >> @@ -239,6 +243,7 @@ nouveau_vp3_bsp_begin(struct nouveau_vp3_decoder *dec) >> >> dec->bsp_ptr = bsp_bo->map; >> dec->bsp_size = NOUVEAU_VP3_BSP_RESERVED_SIZE; >> + dec->nb_slices = 0; >> >> dec->bsp_ptr += 0x100; >> >> @@ -266,6 +271,8 @@ nouveau_vp3_bsp_next(struct nouveau_vp3_decoder *dec, >> unsigned num_buffers, >> struct strparm_bsp *str_bsp = NULL; >> int i = 0; >> >> + ++dec->nb_slices; >> + >> bsp_origin += 0x100; >> str_bsp = (struct strparm_bsp *)bsp_origin; >> >> @@ -331,6 +338,7 @@ nouveau_vp3_bsp_end(struct nouveau_vp3_decoder *dec, >> union pipe_desc desc) >> >> dec->bsp_ptr = NULL; >> dec->bsp_size = 0; >> + dec->nb_slices = 0; >> >> return caps; >> } >> -- >> 1.9.1 >> >> _______________________________________________ >> mesa-dev mailing list >> mesa-dev@lists.freedesktop.org >> http://lists.freedesktop.org/mailman/listinfo/mesa-dev _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
