[FFmpeg-devel] [PATCH v3 7/8] lavc/vvc_dec: Add hardware decode API
Wang, Fei W
fei.w.wang at intel.com
Mon Oct 28 10:54:27 EET 2024
On Mon, 2024-10-28 at 01:18 +0000, Xiang, Haihao wrote:
> On Wo, 2024-09-25 at 10:15 +0800,
> fei.w.wang-at-intel.com at ffmpeg.org wrote:
> > From: Fei Wang <fei.w.wang at intel.com>
> >
> > Signed-off-by: Fei Wang <fei.w.wang at intel.com>
> > ---
> > libavcodec/vvc/dec.c | 90 ++++++++++++++++++++++++++++++++++++---
> > ----
> > libavcodec/vvc/dec.h | 4 ++
> > libavcodec/vvc/refs.c | 6 +++
> > 3 files changed, 87 insertions(+), 13 deletions(-)
> >
> > diff --git a/libavcodec/vvc/dec.c b/libavcodec/vvc/dec.c
> > index edf2607f50..19770de3fa 100644
> > --- a/libavcodec/vvc/dec.c
> > +++ b/libavcodec/vvc/dec.c
> > @@ -22,6 +22,8 @@
> > */
> > #include "libavcodec/codec_internal.h"
> > #include "libavcodec/decode.h"
> > +#include "libavcodec/hwaccel_internal.h"
> > +#include "libavcodec/hwconfig.h"
> > #include "libavcodec/profiles.h"
> > #include "libavcodec/refstruct.h"
> > #include "libavutil/cpu.h"
> > @@ -610,6 +612,8 @@ static int ref_frame(VVCFrame *dst, const
> > VVCFrame *src)
> >
> > ff_refstruct_replace(&dst->rpl_tab, src->rpl_tab);
> > ff_refstruct_replace(&dst->rpl, src->rpl);
> > + ff_refstruct_replace(&dst->hwaccel_picture_private,
> > + src->hwaccel_picture_private);
> > dst->nb_rpl_elems = src->nb_rpl_elems;
> >
> > dst->poc = src->poc;
> > @@ -770,18 +774,50 @@ static int slice_start(SliceContext *sc,
> > VVCContext *s,
> > VVCFrameContext *fc,
> > return 0;
> > }
> >
> > -static void export_frame_params(VVCContext *s, const
> > VVCFrameContext *fc)
> > +static enum AVPixelFormat get_format(AVCodecContext *avctx, const
> > VVCSPS
> > *sps)
> > +{
> > +#define HWACCEL_MAX 0
> > +
> > + enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
> > +
> > + switch (sps->pix_fmt) {
> > + case AV_PIX_FMT_YUV420P:
> > + break;
> > + case AV_PIX_FMT_YUV420P10:
> > + break;
> > + }
> > +
> > + *fmt++ = sps->pix_fmt;
> > + *fmt = AV_PIX_FMT_NONE;
> > +
> > + return ff_get_format(avctx, pix_fmts);
> > +}
> > +
> > +static int export_frame_params(VVCContext *s, const
> > VVCFrameContext *fc)
> > {
> > AVCodecContext *c = s->avctx;
> > const VVCSPS *sps = fc->ps.sps;
> > const VVCPPS *pps = fc->ps.pps;
> > + int ret;
> >
> > - c->pix_fmt = sps->pix_fmt;
> > - c->coded_width = pps->width;
> > - c->coded_height = pps->height;
> > - c->width = pps->width - ((pps->r-
> > >pps_conf_win_left_offset + pps-
> > > r->pps_conf_win_right_offset) << sps->hshift[CHROMA]);
> > - c->height = pps->height - ((pps->r-
> > >pps_conf_win_top_offset + pps-
> > > r->pps_conf_win_bottom_offset) << sps->vshift[CHROMA]);
> > + // Reset HW config if pix_fmt/w/h change.
> > + if (s->pix_fmt != sps->pix_fmt || c->coded_width != pps->width
> > || c-
> > > coded_height != pps->height) {
> > + c->coded_width = pps->width;
> > + c->coded_height = pps->height;
> > + c->pix_fmt = get_format(c, sps);
> > + ret = get_format(c, sps);
>
> Needn't get_format() twice.
Fixed in V4.
Thanks
Fei
>
> Thanks
> Haihao
>
> > + if (ret < 0)
> > + return ret;
> > +
> > + c->pix_fmt = ret;
> > + s->pix_fmt = sps->pix_fmt;
> > + }
> > +
> > + c->width = pps->width - ((pps->r->pps_conf_win_left_offset +
> > pps->r-
> > > pps_conf_win_right_offset) << sps->hshift[CHROMA]);
> > + c->height = pps->height - ((pps->r->pps_conf_win_top_offset +
> > pps->r-
> > > pps_conf_win_bottom_offset) << sps->vshift[CHROMA]);
> > c->has_b_frames = sps->r-
> > >sps_dpb_params.dpb_max_num_reorder_pics[sps->r-
> > > sps_max_sublayers_minus1];
> > +
> > + return 0;
> > }
> >
> > static int frame_setup(VVCFrameContext *fc, VVCContext *s)
> > @@ -794,8 +830,11 @@ static int frame_setup(VVCFrameContext *fc,
> > VVCContext
> > *s)
> > if (ret < 0)
> > return ret;
> >
> > - export_frame_params(s, fc);
> > - return ret;
> > + ret = export_frame_params(s, fc);
> > + if (ret < 0)
> > + return ret;
> > +
> > + return 0;
> > }
> >
> > static int decode_slice(VVCContext *s, VVCFrameContext *fc, const
> > H2645NAL
> > *nal, const CodedBitstreamUnit *unit)
> > @@ -824,6 +863,20 @@ static int decode_slice(VVCContext *s,
> > VVCFrameContext
> > *fc, const H2645NAL *nal,
> > ret = slice_init_entry_points(sc, fc, nal, unit);
> > if (ret < 0)
> > return ret;
> > +
> > + if (s->avctx->hwaccel) {
> > + if (is_first_slice) {
> > + ret = FF_HW_CALL(s->avctx, start_frame, NULL, 0);
> > + if (ret < 0)
> > + return ret;
> > + }
> > +
> > + ret = FF_HW_CALL(s->avctx, decode_slice,
> > + nal->raw_data, nal->raw_size);
> > + if (ret < 0)
> > + return ret;
> > + }
> > +
> > fc->nb_slices++;
> >
> > return 0;
> > @@ -939,17 +992,26 @@ static int wait_delayed_frame(VVCContext *s,
> > AVFrame
> > *output, int *got_output)
> >
> > static int submit_frame(VVCContext *s, VVCFrameContext *fc,
> > AVFrame *output,
> > int *got_output)
> > {
> > - int ret = ff_vvc_frame_submit(s, fc);
> > + int ret;
> >
> > - if (ret < 0) {
> > - ff_vvc_report_frame_finished(fc->ref);
> > - return ret;
> > + if (s->avctx->hwaccel) {
> > + if (ret = FF_HW_SIMPLE_CALL(s->avctx, end_frame) < 0) {
> > + av_log(s->avctx, AV_LOG_ERROR,
> > + "Hardware accelerator failed to decode
> > picture\n");
> > + ff_vvc_unref_frame(fc, fc->ref, ~0);
> > + return ret;
> > + }
> > + } else {
> > + if (ret = ff_vvc_frame_submit(s, fc) < 0) {
> > + ff_vvc_report_frame_finished(fc->ref);
> > + return ret;
> > + }
> > }
> >
> > s->nb_frames++;
> > s->nb_delayed++;
> >
> > - if (s->nb_delayed >= s->nb_fcs) {
> > + if (s->nb_delayed >= s->nb_fcs || s->avctx->hwaccel) {
> > if ((ret = wait_delayed_frame(s, output, got_output)) < 0)
> > return ret;
> > }
> > @@ -1095,6 +1157,8 @@ static av_cold int
> > vvc_decode_init(AVCodecContext
> > *avctx)
> > GDR_SET_RECOVERED(s);
> > ff_thread_once(&init_static_once, init_default_scale_m);
> >
> > + s->pix_fmt = AV_PIX_FMT_NONE;
> > +
> > return 0;
> > }
> >
> > diff --git a/libavcodec/vvc/dec.h b/libavcodec/vvc/dec.h
> > index d27cf52ca2..776b38b20f 100644
> > --- a/libavcodec/vvc/dec.h
> > +++ b/libavcodec/vvc/dec.h
> > @@ -101,6 +101,8 @@ typedef struct VVCFrame {
> > * A combination of VVC_FRAME_FLAG_*
> > */
> > uint8_t flags;
> > +
> > + void *hwaccel_picture_private; ///< hardware accelerator
> > private data
> > } VVCFrame;
> >
> > typedef struct SliceContext {
> > @@ -243,6 +245,8 @@ typedef struct VVCContext {
> >
> > uint64_t nb_frames; ///< processed frames
> > int nb_delayed; ///< delayed frames
> > +
> > + enum AVPixelFormat pix_fmt; ///< pix format of current frame
> > } VVCContext ;
> >
> > #endif /* AVCODEC_VVC_DEC_H */
> > diff --git a/libavcodec/vvc/refs.c b/libavcodec/vvc/refs.c
> > index 3ed9d6768d..281410091e 100644
> > --- a/libavcodec/vvc/refs.c
> > +++ b/libavcodec/vvc/refs.c
> > @@ -26,6 +26,7 @@
> > #include "libavutil/thread.h"
> > #include "libavcodec/refstruct.h"
> > #include "libavcodec/thread.h"
> > +#include "libavcodec/decode.h"
> >
> > #include "refs.h"
> >
> > @@ -59,6 +60,7 @@ void ff_vvc_unref_frame(VVCFrameContext *fc,
> > VVCFrame
> > *frame, int flags)
> > ff_refstruct_unref(&frame->rpl_tab);
> >
> > frame->collocated_ref = NULL;
> > + ff_refstruct_unref(&frame->hwaccel_picture_private);
> > }
> > }
> >
> > @@ -153,6 +155,10 @@ static VVCFrame *alloc_frame(VVCContext *s,
> > VVCFrameContext *fc)
> > if (!frame->progress)
> > goto fail;
> >
> > + ret = ff_hwaccel_frame_priv_alloc(s->avctx, &frame-
> > > hwaccel_picture_private);
> > + if (ret < 0)
> > + goto fail;
> > +
> > return frame;
> > fail:
> > ff_vvc_unref_frame(fc, frame, ~0);
>
More information about the ffmpeg-devel
mailing list