[FFmpeg-devel] [PATCH 38/39] lavc/hevcdec: simplify output logic
Anton Khirnov
anton at khirnov.net
Tue Jul 16 20:11:53 EEST 2024
Current code is written around the "simple" decode API's limitation that
a single input packet (AU/coded frame) triggers the output of at most
one output frame. However the spec contains two cases where a coded
frame may cause multiple frames to be output (cf. C.5.2.2.2):
* start of a new sequence
* overflowing sps_max_dec_pic_buffering
The decoder currently contains rather convoluted logic to handle these
cases:
* decode/output/per-frame sequence counters,
* HEVC_FRAME_FLAG_BUMPING
* ff_hevc_bump_frame()
* special clauses in ff_hevc_output_frame()
However, with the receive_frame() API none of that is necessary, as we
can just output multiple frames at once. Previously added ContainerFifo
allows that to be done in a straightforward and efficient manner.
---
libavcodec/hevc/hevcdec.c | 33 +++++-------
libavcodec/hevc/hevcdec.h | 32 ++++--------
libavcodec/hevc/refs.c | 104 +++++++-------------------------------
3 files changed, 42 insertions(+), 127 deletions(-)
diff --git a/libavcodec/hevc/hevcdec.c b/libavcodec/hevc/hevcdec.c
index 5e28bfe54e..28835f48d6 100644
--- a/libavcodec/hevc/hevcdec.c
+++ b/libavcodec/hevc/hevcdec.c
@@ -2906,6 +2906,7 @@ static int hevc_frame_start(HEVCContext *s)
const HEVCSPS *const sps = pps->sps;
int pic_size_in_ctb = ((sps->width >> sps->log2_min_cb_size) + 1) *
((sps->height >> sps->log2_min_cb_size) + 1);
+ int new_sequence = IS_IDR(s) || IS_BLA(s) || s->last_eos;
int ret;
ff_refstruct_replace(&s->pps, pps);
@@ -2925,7 +2926,7 @@ static int hevc_frame_start(HEVCContext *s)
return pix_fmt;
s->avctx->pix_fmt = pix_fmt;
- s->seq_decode = (s->seq_decode + 1) & HEVC_SEQUENCE_COUNTER_MASK;
+ new_sequence = 1;
}
memset(s->horizontal_bs, 0, s->bs_width * s->bs_height);
@@ -2934,11 +2935,8 @@ static int hevc_frame_start(HEVCContext *s)
memset(s->is_pcm, 0, (sps->min_pu_width + 1) * (sps->min_pu_height + 1));
memset(s->tab_slice_address, -1, pic_size_in_ctb * sizeof(*s->tab_slice_address));
- if ((IS_IDR(s) || IS_BLA(s))) {
- s->seq_decode = (s->seq_decode + 1) & HEVC_SEQUENCE_COUNTER_MASK;
- if (IS_IDR(s))
- ff_hevc_clear_refs(s);
- }
+ if (IS_IDR(s))
+ ff_hevc_clear_refs(s);
s->slice_idx = 0;
s->first_nal_type = s->nal_unit_type;
@@ -2962,6 +2960,12 @@ static int hevc_frame_start(HEVCContext *s)
if (pps->tiles_enabled_flag)
s->local_ctx[0].end_of_tiles_x = pps->column_width[0] << sps->log2_ctb_size;
+ if (new_sequence) {
+ ret = ff_hevc_output_frames(s, 0, 0, s->sh.no_output_of_prior_pics_flag);
+ if (ret < 0)
+ return ret;
+ }
+
ret = export_stream_params_from_sei(s);
if (ret < 0)
return ret;
@@ -3014,10 +3018,8 @@ static int hevc_frame_start(HEVCContext *s)
s->cur_frame->f->pict_type = 3 - s->sh.slice_type;
- if (!IS_IRAP(s))
- ff_hevc_bump_frame(s);
-
- ret = ff_hevc_output_frame(s, 0);
+ ret = ff_hevc_output_frames(s, sps->temporal_layer[sps->max_sub_layers - 1].num_reorder_pics,
+ sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering, 0);
if (ret < 0)
goto fail;
@@ -3262,8 +3264,6 @@ static int decode_nal_unit(HEVCContext *s, const H2645NAL *nal)
break;
case HEVC_NAL_EOS_NUT:
case HEVC_NAL_EOB_NUT:
- s->seq_decode = (s->seq_decode + 1) & HEVC_SEQUENCE_COUNTER_MASK;
- break;
case HEVC_NAL_AUD:
case HEVC_NAL_FD_NUT:
case HEVC_NAL_UNSPEC62:
@@ -3420,7 +3420,7 @@ static int hevc_receive_frame(AVCodecContext *avctx, AVFrame *frame)
av_packet_unref(avpkt);
ret = ff_decode_get_packet(avctx, avpkt);
if (ret == AVERROR_EOF) {
- ret = ff_hevc_output_frame(s, 1);
+ ret = ff_hevc_output_frames(s, 0, 0, 0);
if (ret < 0)
return ret;
goto do_output;
@@ -3482,7 +3482,6 @@ static int hevc_ref_frame(HEVCFrame *dst, const HEVCFrame *src)
dst->poc = src->poc;
dst->ctb_count = src->ctb_count;
dst->flags = src->flags;
- dst->sequence = src->sequence;
ff_refstruct_replace(&dst->hwaccel_picture_private,
src->hwaccel_picture_private);
@@ -3606,8 +3605,6 @@ static int hevc_update_thread_context(AVCodecContext *dst,
if ((ret = set_sps(s, s0->ps.sps)) < 0)
return ret;
- s->seq_decode = s0->seq_decode;
- s->seq_output = s0->seq_output;
s->poc_tid0 = s0->poc_tid0;
s->eos = s0->eos;
s->no_rasl_output_flag = s0->no_rasl_output_flag;
@@ -3617,10 +3614,6 @@ static int hevc_update_thread_context(AVCodecContext *dst,
s->film_grain_warning_shown = s0->film_grain_warning_shown;
- if (s0->eos) {
- s->seq_decode = (s->seq_decode + 1) & HEVC_SEQUENCE_COUNTER_MASK;
- }
-
ret = ff_h2645_sei_ctx_replace(&s->sei.common, &s0->sei.common);
if (ret < 0)
return ret;
diff --git a/libavcodec/hevc/hevcdec.h b/libavcodec/hevc/hevcdec.h
index f2705b8de2..4293f1bd66 100644
--- a/libavcodec/hevc/hevcdec.h
+++ b/libavcodec/hevc/hevcdec.h
@@ -350,10 +350,7 @@ typedef struct DBParams {
#define HEVC_FRAME_FLAG_OUTPUT (1 << 0)
#define HEVC_FRAME_FLAG_SHORT_REF (1 << 1)
#define HEVC_FRAME_FLAG_LONG_REF (1 << 2)
-#define HEVC_FRAME_FLAG_BUMPING (1 << 3)
-
-#define HEVC_SEQUENCE_COUNTER_MASK 0xff
-#define HEVC_SEQUENCE_COUNTER_INVALID (HEVC_SEQUENCE_COUNTER_MASK + 1)
+#define HEVC_FRAME_FLAG_UNAVAILABLE (1 << 3)
typedef struct HEVCFrame {
union {
@@ -375,12 +372,6 @@ typedef struct HEVCFrame {
void *hwaccel_picture_private; ///< RefStruct reference
- /**
- * A sequence counter, so that old frames are output first
- * after a POC reset
- */
- uint16_t sequence;
-
/**
* A combination of HEVC_FRAME_FLAG_*
*/
@@ -519,13 +510,6 @@ typedef struct HEVCContext {
uint8_t *checksum_buf;
int checksum_buf_size;
- /**
- * Sequence counters for decoded and output frames, so that old
- * frames are output first after a POC reset
- */
- uint16_t seq_decode;
- uint16_t seq_output;
-
/** The target for the common_cabac_state of the local contexts. */
HEVCCABACState cabac;
@@ -634,12 +618,16 @@ static av_always_inline int ff_hevc_nal_is_nonref(enum HEVCNALUnitType type)
}
/**
- * Find next frame in output order and put a reference to it in frame.
- * @return 1 if a frame was output, 0 otherwise
+ * Find frames in the DPB that are ready for output and either write them to the
+ * output FIFO or drop their output flag, depending on the value of discard.
+ *
+ * @param max_output maximum number of output-pending frames that can be
+ * present in the DPB before output is triggered
+ * @param max_dpb maximum number of any frames that can be present in the DPB
+ * before output is triggered
*/
-int ff_hevc_output_frame(HEVCContext *s, int flush);
-
-void ff_hevc_bump_frame(HEVCContext *s);
+int ff_hevc_output_frames(HEVCContext *s, unsigned max_output,
+ unsigned max_dpb, int discard);
void ff_hevc_unref_frame(HEVCFrame *frame, int flags);
diff --git a/libavcodec/hevc/refs.c b/libavcodec/hevc/refs.c
index 44ee7c837a..50aa06b926 100644
--- a/libavcodec/hevc/refs.c
+++ b/libavcodec/hevc/refs.c
@@ -131,8 +131,7 @@ int ff_hevc_set_new_ref(HEVCContext *s, int poc)
for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
HEVCFrame *frame = &s->DPB[i];
- if (frame->f && frame->sequence == s->seq_decode &&
- frame->poc == poc) {
+ if (frame->f && frame->poc == poc) {
av_log(s->avctx, AV_LOG_ERROR, "Duplicate POC in a sequence: %d.\n",
poc);
return AVERROR_INVALIDDATA;
@@ -152,7 +151,6 @@ int ff_hevc_set_new_ref(HEVCContext *s, int poc)
ref->flags = HEVC_FRAME_FLAG_SHORT_REF;
ref->poc = poc;
- ref->sequence = s->seq_decode;
ref->f->crop_left = s->ps.sps->output_window.left_offset;
ref->f->crop_right = s->ps.sps->output_window.right_offset;
ref->f->crop_top = s->ps.sps->output_window.top_offset;
@@ -165,112 +163,49 @@ static void unref_missing_refs(HEVCContext *s)
{
for (int i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
HEVCFrame *frame = &s->DPB[i];
- if (frame->sequence == HEVC_SEQUENCE_COUNTER_INVALID) {
+ if (frame->flags & HEVC_FRAME_FLAG_UNAVAILABLE) {
ff_hevc_unref_frame(frame, ~0);
}
}
}
-int ff_hevc_output_frame(HEVCContext *s, int flush)
+int ff_hevc_output_frames(HEVCContext *s, unsigned max_output,
+ unsigned max_dpb, int discard)
{
- if (IS_IRAP(s) && s->no_rasl_output_flag == 1) {
- const static int mask = HEVC_FRAME_FLAG_BUMPING | HEVC_FRAME_FLAG_OUTPUT;
- for (int i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
- HEVCFrame *frame = &s->DPB[i];
- if ((frame->flags & mask) == HEVC_FRAME_FLAG_OUTPUT &&
- frame->sequence != s->seq_decode) {
- if (s->sh.no_output_of_prior_pics_flag == 1)
- ff_hevc_unref_frame(frame, HEVC_FRAME_FLAG_OUTPUT);
- else
- frame->flags |= HEVC_FRAME_FLAG_BUMPING;
- }
- }
- }
- do {
+ while (1) {
+ int nb_dpb = 0;
int nb_output = 0;
int min_poc = INT_MAX;
int i, min_idx, ret;
for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
HEVCFrame *frame = &s->DPB[i];
- if ((frame->flags & HEVC_FRAME_FLAG_OUTPUT) &&
- frame->sequence == s->seq_output) {
+ if (frame->flags & HEVC_FRAME_FLAG_OUTPUT) {
nb_output++;
if (frame->poc < min_poc || nb_output == 1) {
min_poc = frame->poc;
min_idx = i;
}
}
+ nb_dpb += !!frame->flags;
}
- /* wait for more frames before output */
- if (!flush && s->seq_output == s->seq_decode && s->ps.sps &&
- nb_output <= s->ps.sps->temporal_layer[s->ps.sps->max_sub_layers - 1].num_reorder_pics)
- return 0;
-
- if (nb_output) {
+ if (nb_output > max_output ||
+ (nb_output && nb_dpb > max_dpb)) {
HEVCFrame *frame = &s->DPB[min_idx];
- ret = ff_container_fifo_write(s->output_fifo,
+ ret = discard ? 0 :
+ ff_container_fifo_write(s->output_fifo,
frame->needs_fg ? frame->frame_grain : frame->f);
- if (frame->flags & HEVC_FRAME_FLAG_BUMPING)
- ff_hevc_unref_frame(frame, HEVC_FRAME_FLAG_OUTPUT | HEVC_FRAME_FLAG_BUMPING);
- else
- ff_hevc_unref_frame(frame, HEVC_FRAME_FLAG_OUTPUT);
+ ff_hevc_unref_frame(frame, HEVC_FRAME_FLAG_OUTPUT);
if (ret < 0)
return ret;
- av_log(s->avctx, AV_LOG_DEBUG,
- "Output frame with POC %d.\n", frame->poc);
- return 1;
+ av_log(s->avctx, AV_LOG_DEBUG, "%s frame with POC %d.\n",
+ discard ? "Discarded" : "Output", frame->poc);
+ continue;
}
-
- if (s->seq_output != s->seq_decode)
- s->seq_output = (s->seq_output + 1) & HEVC_SEQUENCE_COUNTER_MASK;
- else
- break;
- } while (1);
-
- return 0;
-}
-
-void ff_hevc_bump_frame(HEVCContext *s)
-{
- int dpb = 0;
- int min_poc = INT_MAX;
- int i;
-
- for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
- HEVCFrame *frame = &s->DPB[i];
- if ((frame->flags) &&
- frame->sequence == s->seq_output &&
- frame->poc != s->poc) {
- dpb++;
- }
- }
-
- if (s->ps.sps && dpb >= s->ps.sps->temporal_layer[s->ps.sps->max_sub_layers - 1].max_dec_pic_buffering) {
- for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
- HEVCFrame *frame = &s->DPB[i];
- if ((frame->flags) &&
- frame->sequence == s->seq_output &&
- frame->poc != s->poc) {
- if (frame->flags == HEVC_FRAME_FLAG_OUTPUT && frame->poc < min_poc) {
- min_poc = frame->poc;
- }
- }
- }
-
- for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
- HEVCFrame *frame = &s->DPB[i];
- if (frame->flags & HEVC_FRAME_FLAG_OUTPUT &&
- frame->sequence == s->seq_output &&
- frame->poc <= min_poc) {
- frame->flags |= HEVC_FRAME_FLAG_BUMPING;
- }
- }
-
- dpb--;
+ return 0;
}
}
@@ -384,7 +319,7 @@ static HEVCFrame *find_ref_idx(HEVCContext *s, int poc, uint8_t use_msb)
for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
HEVCFrame *ref = &s->DPB[i];
- if (ref->f && ref->sequence == s->seq_decode) {
+ if (ref->f) {
if ((ref->poc & mask) == poc && (use_msb || ref->poc != s->poc))
return ref;
}
@@ -427,8 +362,7 @@ static HEVCFrame *generate_missing_ref(HEVCContext *s, int poc)
}
frame->poc = poc;
- frame->sequence = HEVC_SEQUENCE_COUNTER_INVALID;
- frame->flags = 0;
+ frame->flags = HEVC_FRAME_FLAG_UNAVAILABLE;
if (s->avctx->active_thread_type == FF_THREAD_FRAME)
ff_progress_frame_report(&frame->tf, INT_MAX);
--
2.43.0
More information about the ffmpeg-devel
mailing list