Hi, On Mon, Nov 28, 2016 at 7:26 PM, Mark Thompson <s...@jkqxz.net> wrote:
> On 28/11/16 21:22, Mathieu Velten wrote: > > --- > > libavcodec/vaapi_vp9.c | 1 + > > libavcodec/vp9.c | 32 +++++++++++++++++--------------- > > libavcodec/vp9.h | 1 + > > 3 files changed, 19 insertions(+), 15 deletions(-) > > Nice :) > > Tested on Kaby Lake, works for me (woo 180fps 4K 10-bit decode). > > This should probably be split into two patches, though - one for the > generic vp9 hwaccel support, a second then enabling it for VAAPI. > > > diff --git a/libavcodec/vaapi_vp9.c b/libavcodec/vaapi_vp9.c > > index b360dcb..9b3e81a 100644 > > --- a/libavcodec/vaapi_vp9.c > > +++ b/libavcodec/vaapi_vp9.c > > @@ -38,6 +38,7 @@ static void fill_picture_parameters(AVCodecContext > *avctx, > > pp->first_partition_size = h->h.compressed_header_size; > > > > pp->profile = h->h.profile; > > + pp->bit_depth = h->h.bpp; > > > > pp->filter_level = h->h.filter.level; > > pp->sharpness_level = h->h.filter.sharpness; > > diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c > > index 0ec895a..ff526da 100644 > > --- a/libavcodec/vp9.c > > +++ b/libavcodec/vp9.c > > @@ -68,7 +68,7 @@ typedef struct VP9Context { > > ptrdiff_t y_stride, uv_stride; > > > > uint8_t ss_h, ss_v; > > - uint8_t last_bpp, bpp, bpp_index, bytesperpixel; > > + uint8_t last_bpp, bpp_index, bytesperpixel; > > uint8_t last_keyframe; > > // sb_cols/rows, rows/cols and last_fmt are used for allocating all > internal > > // arrays, and are thus per-thread. w/h and gf_fmt are synced > between threads > > @@ -258,7 +258,9 @@ static int update_size(AVCodecContext *ctx, int w, > int h) > > if ((res = ff_set_dimensions(ctx, w, h)) < 0) > > return res; > > > > - if (s->pix_fmt == AV_PIX_FMT_YUV420P) { > > + if (s->pix_fmt == AV_PIX_FMT_YUV420P || > > + s->pix_fmt == AV_PIX_FMT_YUV420P10 || > > + s->pix_fmt == AV_PIX_FMT_YUV420P12) { > > #if CONFIG_VP9_DXVA2_HWACCEL > > *fmtp++ = AV_PIX_FMT_DXVA2_VLD; > > #endif > > This is enabling it for DXVA2 and D3D11VA as well? I'm guessing you > probably didn't want to do that - I think it would be better with something > more like <http://git.videolan.org/?p=ffmpeg.git;a=blob;f= > libavcodec/hevc.c;hb=HEAD#l350>. I'll let you guys figure out the details for this, but generic vp9.[ch] changes are OK with me. Thanks! Ronald _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel