avcodec/vaapi_encode: move the dpb logic from VAAPI to base layer

Move receive_packet function to base. This requires adding *alloc, *issue, *output, *free as hardware callbacks. HWBaseEncodePicture is introduced as the base layer structure. The related parameters in VAAPIEncodeContext are also extracted to HWBaseEncodeContext. Then DPB management logic can be fully extracted to base layer as-is.

Tong Wu committed 2 years ago

Commit aa823401 parent f303c2612 files · +985 −890Browse files

Showing the changes to libavcodec (12 of 12 files). Show all

Unified · Split

12 files

libavcodec/Makefile +1 −1
libavcodec/hw_base_encode.c +594 −0
libavcodec/hw_base_encode.h +124 −0
libavcodec/vaapi_encode.c +126 −667
libavcodec/vaapi_encode.h +1 −101
libavcodec/vaapi_encode_av1.c +19 −16
libavcodec/vaapi_encode_h264.c +44 −40
libavcodec/vaapi_encode_h265.c +28 −25
libavcodec/vaapi_encode_mjpeg.c +7 −6
libavcodec/vaapi_encode_mpeg2.c +18 −15
libavcodec/vaapi_encode_vp8.c +10 −8
libavcodec/vaapi_encode_vp9.c +13 −11

libavcodec/Makefile+1−1

@@ -165,7 +165,7 @@ OBJS-$(CONFIG_STARTCODE) += startcode.o · 164 unmodified lines
165	OBJS-$(CONFIG_TEXTUREDSP) += texturedsp.o	165	OBJS-$(CONFIG_TEXTUREDSP) += texturedsp.o
166	OBJS-$(CONFIG_TEXTUREDSPENC) += texturedspenc.o	166	OBJS-$(CONFIG_TEXTUREDSPENC) += texturedspenc.o
167	OBJS-$(CONFIG_TPELDSP) += tpeldsp.o	167	OBJS-$(CONFIG_TPELDSP) += tpeldsp.o
168	OBJS-$(CONFIG_VAAPI_ENCODE) += vaapi_encode.o	168	OBJS-$(CONFIG_VAAPI_ENCODE) += vaapi_encode.o hw_base_encode.o
169	OBJS-$(CONFIG_AV1_AMF_ENCODER) += amfenc_av1.o	169	OBJS-$(CONFIG_AV1_AMF_ENCODER) += amfenc_av1.o
170	OBJS-$(CONFIG_VC1DSP) += vc1dsp.o	170	OBJS-$(CONFIG_VC1DSP) += vc1dsp.o
171	OBJS-$(CONFIG_VIDEODSP) += videodsp.o	171	OBJS-$(CONFIG_VIDEODSP) += videodsp.o

libavcodec/hw_base_encode.c+594−0

@@ -0,0 +1,594 @@
	1	/*
	2	* This file is part of FFmpeg.
	3	*
	4	* FFmpeg is free software; you can redistribute it and/or
	5	* modify it under the terms of the GNU Lesser General Public
	6	* License as published by the Free Software Foundation; either
	7	* version 2.1 of the License, or (at your option) any later version.
	8	*
	9	* FFmpeg is distributed in the hope that it will be useful,
	10	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	11	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	12	* Lesser General Public License for more details.
	13	*
	14	* You should have received a copy of the GNU Lesser General Public
	15	* License along with FFmpeg; if not, write to the Free Software
	16	* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	17	*/
	18
	19	#include "libavutil/avassert.h"
	20	#include "libavutil/common.h"
	21	#include "libavutil/internal.h"
	22	#include "libavutil/log.h"
	23	#include "libavutil/mem.h"
	24	#include "libavutil/pixdesc.h"
	25
	26	#include "encode.h"
	27	#include "avcodec.h"
	28	#include "hw_base_encode.h"
	29
	30	static void hw_base_encode_add_ref(FFHWBaseEncodePicture *pic,
	31	FFHWBaseEncodePicture *target,
	32	int is_ref, int in_dpb, int prev)
	33	{
	34	int refs = 0;
	35
	36	if (is_ref) {
	37	av_assert0(pic != target);
	38	av_assert0(pic->nb_refs[0] < MAX_PICTURE_REFERENCES &&
	39	pic->nb_refs[1] < MAX_PICTURE_REFERENCES);
	40	if (target->display_order < pic->display_order)
	41	pic->refs[0][pic->nb_refs[0]++] = target;
	42	else
	43	pic->refs[1][pic->nb_refs[1]++] = target;
	44	++refs;
	45	}
	46
	47	if (in_dpb) {
	48	av_assert0(pic->nb_dpb_pics < MAX_DPB_SIZE);
	49	pic->dpb[pic->nb_dpb_pics++] = target;
	50	++refs;
	51	}
	52
	53	if (prev) {
	54	av_assert0(!pic->prev);
	55	pic->prev = target;
	56	++refs;
	57	}
	58
	59	target->ref_count[0] += refs;
	60	target->ref_count[1] += refs;
	61	}
	62
	63	static void hw_base_encode_remove_refs(FFHWBaseEncodePicture *pic, int level)
	64	{
	65	int i;
	66
	67	if (pic->ref_removed[level])
	68	return;
	69
	70	for (i = 0; i < pic->nb_refs[0]; i++) {
	71	av_assert0(pic->refs[0][i]);
	72	--pic->refs[0][i]->ref_count[level];
	73	av_assert0(pic->refs[0][i]->ref_count[level] >= 0);
	74	}
	75
	76	for (i = 0; i < pic->nb_refs[1]; i++) {
	77	av_assert0(pic->refs[1][i]);
	78	--pic->refs[1][i]->ref_count[level];
	79	av_assert0(pic->refs[1][i]->ref_count[level] >= 0);
	80	}
	81
	82	for (i = 0; i < pic->nb_dpb_pics; i++) {
	83	av_assert0(pic->dpb[i]);
	84	--pic->dpb[i]->ref_count[level];
	85	av_assert0(pic->dpb[i]->ref_count[level] >= 0);
	86	}
	87
	88	av_assert0(pic->prev \|\| pic->type == FF_HW_PICTURE_TYPE_IDR);
	89	if (pic->prev) {
	90	--pic->prev->ref_count[level];
	91	av_assert0(pic->prev->ref_count[level] >= 0);
	92	}
	93
	94	pic->ref_removed[level] = 1;
	95	}
	96
	97	static void hw_base_encode_set_b_pictures(AVCodecContext *avctx,
	98	FFHWBaseEncodePicture *start,
	99	FFHWBaseEncodePicture *end,
	100	FFHWBaseEncodePicture *prev,
	101	int current_depth,
	102	FFHWBaseEncodePicture **last)
	103	{
	104	FFHWBaseEncodeContext *ctx = avctx->priv_data;
	105	FFHWBaseEncodePicture pic, next, *ref;
	106	int i, len;
	107
	108	av_assert0(start && end && start != end && start->next != end);
	109
	110	// If we are at the maximum depth then encode all pictures as
	111	// non-referenced B-pictures. Also do this if there is exactly one
	112	// picture left, since there will be nothing to reference it.
	113	if (current_depth == ctx->max_b_depth \|\| start->next->next == end) {
	114	for (pic = start->next; pic; pic = pic->next) {
	115	if (pic == end)
	116	break;
	117	pic->type = FF_HW_PICTURE_TYPE_B;
	118	pic->b_depth = current_depth;
	119
	120	hw_base_encode_add_ref(pic, start, 1, 1, 0);
	121	hw_base_encode_add_ref(pic, end, 1, 1, 0);
	122	hw_base_encode_add_ref(pic, prev, 0, 0, 1);
	123
	124	for (ref = end->refs[1][0]; ref; ref = ref->refs[1][0])
	125	hw_base_encode_add_ref(pic, ref, 0, 1, 0);
	126	}
	127	*last = prev;
	128
	129	} else {
	130	// Split the current list at the midpoint with a referenced
	131	// B-picture, then descend into each side separately.
	132	len = 0;
	133	for (pic = start->next; pic != end; pic = pic->next)
	134	++len;
	135	for (pic = start->next, i = 1; 2 * i < len; pic = pic->next, i++);
	136
	137	pic->type = FF_HW_PICTURE_TYPE_B;
	138	pic->b_depth = current_depth;
	139
	140	pic->is_reference = 1;
	141
	142	hw_base_encode_add_ref(pic, pic, 0, 1, 0);
	143	hw_base_encode_add_ref(pic, start, 1, 1, 0);
	144	hw_base_encode_add_ref(pic, end, 1, 1, 0);
	145	hw_base_encode_add_ref(pic, prev, 0, 0, 1);
	146
	147	for (ref = end->refs[1][0]; ref; ref = ref->refs[1][0])
	148	hw_base_encode_add_ref(pic, ref, 0, 1, 0);
	149
	150	if (i > 1)
	151	hw_base_encode_set_b_pictures(avctx, start, pic, pic,
	152	current_depth + 1, &next);
	153	else
	154	next = pic;
	155
	156	hw_base_encode_set_b_pictures(avctx, pic, end, next,
	157	current_depth + 1, last);
	158	}
	159	}
	160
	161	static void hw_base_encode_add_next_prev(AVCodecContext *avctx,
	162	FFHWBaseEncodePicture *pic)
	163	{
	164	FFHWBaseEncodeContext *ctx = avctx->priv_data;
	165	int i;
	166
	167	if (!pic)
	168	return;
	169
	170	if (pic->type == FF_HW_PICTURE_TYPE_IDR) {
	171	for (i = 0; i < ctx->nb_next_prev; i++) {
	172	--ctx->next_prev[i]->ref_count[0];
	173	ctx->next_prev[i] = NULL;
	174	}
	175	ctx->next_prev[0] = pic;
	176	++pic->ref_count[0];
	177	ctx->nb_next_prev = 1;
	178
	179	return;
	180	}
	181
	182	if (ctx->nb_next_prev < MAX_PICTURE_REFERENCES) {
	183	ctx->next_prev[ctx->nb_next_prev++] = pic;
	184	++pic->ref_count[0];
	185	} else {
	186	--ctx->next_prev[0]->ref_count[0];
	187	for (i = 0; i < MAX_PICTURE_REFERENCES - 1; i++)
	188	ctx->next_prev[i] = ctx->next_prev[i + 1];
	189	ctx->next_prev[i] = pic;
	190	++pic->ref_count[0];
	191	}
	192	}
	193
	194	static int hw_base_encode_pick_next(AVCodecContext *avctx,
	195	FFHWBaseEncodePicture **pic_out)
	196	{
	197	FFHWBaseEncodeContext *ctx = avctx->priv_data;
	198	FFHWBaseEncodePicture pic = NULL, prev = NULL, next, start;
	199	int i, b_counter, closed_gop_end;
	200
	201	// If there are any B-frames already queued, the next one to encode
	202	// is the earliest not-yet-issued frame for which all references are
	203	// available.
	204	for (pic = ctx->pic_start; pic; pic = pic->next) {
	205	if (pic->encode_issued)
	206	continue;
	207	if (pic->type != FF_HW_PICTURE_TYPE_B)
	208	continue;
	209	for (i = 0; i < pic->nb_refs[0]; i++) {
	210	if (!pic->refs[0][i]->encode_issued)
	211	break;
	212	}
	213	if (i != pic->nb_refs[0])
	214	continue;
	215
	216	for (i = 0; i < pic->nb_refs[1]; i++) {
	217	if (!pic->refs[1][i]->encode_issued)
	218	break;
	219	}
	220	if (i == pic->nb_refs[1])
	221	break;
	222	}
	223
	224	if (pic) {
	225	av_log(avctx, AV_LOG_DEBUG, "Pick B-picture at depth %d to "
	226	"encode next.\n", pic->b_depth);
	227	*pic_out = pic;
	228	return 0;
	229	}
	230
	231	// Find the B-per-Pth available picture to become the next picture
	232	// on the top layer.
	233	start = NULL;
	234	b_counter = 0;
	235	closed_gop_end = ctx->closed_gop \|\|
	236	ctx->idr_counter == ctx->gop_per_idr;
	237	for (pic = ctx->pic_start; pic; pic = next) {
	238	next = pic->next;
	239	if (pic->encode_issued) {
	240	start = pic;
	241	continue;
	242	}
	243	// If the next available picture is force-IDR, encode it to start
	244	// a new GOP immediately.
	245	if (pic->force_idr)
	246	break;
	247	if (b_counter == ctx->b_per_p)
	248	break;
	249	// If this picture ends a closed GOP or starts a new GOP then it
	250	// needs to be in the top layer.
	251	if (ctx->gop_counter + b_counter + closed_gop_end >= ctx->gop_size)
	252	break;
	253	// If the picture after this one is force-IDR, we need to encode
	254	// this one in the top layer.
	255	if (next && next->force_idr)
	256	break;
	257	++b_counter;
	258	}
	259
	260	// At the end of the stream the last picture must be in the top layer.
	261	if (!pic && ctx->end_of_stream) {
	262	--b_counter;
	263	pic = ctx->pic_end;
	264	if (pic->encode_complete)
	265	return AVERROR_EOF;
	266	else if (pic->encode_issued)
	267	return AVERROR(EAGAIN);
	268	}
	269
	270	if (!pic) {
	271	av_log(avctx, AV_LOG_DEBUG, "Pick nothing to encode next - "
	272	"need more input for reference pictures.\n");
	273	return AVERROR(EAGAIN);
	274	}
	275	if (ctx->input_order <= ctx->decode_delay && !ctx->end_of_stream) {
	276	av_log(avctx, AV_LOG_DEBUG, "Pick nothing to encode next - "
	277	"need more input for timestamps.\n");
	278	return AVERROR(EAGAIN);
	279	}
	280
	281	if (pic->force_idr) {
	282	av_log(avctx, AV_LOG_DEBUG, "Pick forced IDR-picture to "
	283	"encode next.\n");
	284	pic->type = FF_HW_PICTURE_TYPE_IDR;
	285	ctx->idr_counter = 1;
	286	ctx->gop_counter = 1;
	287
	288	} else if (ctx->gop_counter + b_counter >= ctx->gop_size) {
	289	if (ctx->idr_counter == ctx->gop_per_idr) {
	290	av_log(avctx, AV_LOG_DEBUG, "Pick new-GOP IDR-picture to "
	291	"encode next.\n");
	292	pic->type = FF_HW_PICTURE_TYPE_IDR;
	293	ctx->idr_counter = 1;
	294	} else {
	295	av_log(avctx, AV_LOG_DEBUG, "Pick new-GOP I-picture to "
	296	"encode next.\n");
	297	pic->type = FF_HW_PICTURE_TYPE_I;
	298	++ctx->idr_counter;
	299	}
	300	ctx->gop_counter = 1;
	301
	302	} else {
	303	if (ctx->gop_counter + b_counter + closed_gop_end == ctx->gop_size) {
	304	av_log(avctx, AV_LOG_DEBUG, "Pick group-end P-picture to "
	305	"encode next.\n");
	306	} else {
	307	av_log(avctx, AV_LOG_DEBUG, "Pick normal P-picture to "
	308	"encode next.\n");
	309	}
	310	pic->type = FF_HW_PICTURE_TYPE_P;
	311	av_assert0(start);
	312	ctx->gop_counter += 1 + b_counter;
	313	}
	314	pic->is_reference = 1;
	315	*pic_out = pic;
	316
	317	hw_base_encode_add_ref(pic, pic, 0, 1, 0);
	318	if (pic->type != FF_HW_PICTURE_TYPE_IDR) {
	319	// TODO: apply both previous and forward multi reference for all vaapi encoders.
	320	// And L0/L1 reference frame number can be set dynamically through query
	321	// VAConfigAttribEncMaxRefFrames attribute.
	322	if (avctx->codec_id == AV_CODEC_ID_AV1) {
	323	for (i = 0; i < ctx->nb_next_prev; i++)
	324	hw_base_encode_add_ref(pic, ctx->next_prev[i],
	325	pic->type == FF_HW_PICTURE_TYPE_P,
	326	b_counter > 0, 0);
	327	} else
	328	hw_base_encode_add_ref(pic, start,
	329	pic->type == FF_HW_PICTURE_TYPE_P,
	330	b_counter > 0, 0);
	331
	332	hw_base_encode_add_ref(pic, ctx->next_prev[ctx->nb_next_prev - 1], 0, 0, 1);
	333	}
	334
	335	if (b_counter > 0) {
	336	hw_base_encode_set_b_pictures(avctx, start, pic, pic, 1,
	337	&prev);
	338	} else {
	339	prev = pic;
	340	}
	341	hw_base_encode_add_next_prev(avctx, prev);
	342
	343	return 0;
	344	}
	345
	346	static int hw_base_encode_clear_old(AVCodecContext *avctx)
	347	{
	348	FFHWBaseEncodeContext *ctx = avctx->priv_data;
	349	FFHWBaseEncodePicture pic, prev, *next;
	350
	351	av_assert0(ctx->pic_start);
	352
	353	// Remove direct references once each picture is complete.
	354	for (pic = ctx->pic_start; pic; pic = pic->next) {
	355	if (pic->encode_complete && pic->next)
	356	hw_base_encode_remove_refs(pic, 0);
	357	}
	358
	359	// Remove indirect references once a picture has no direct references.
	360	for (pic = ctx->pic_start; pic; pic = pic->next) {
	361	if (pic->encode_complete && pic->ref_count[0] == 0)
	362	hw_base_encode_remove_refs(pic, 1);
	363	}
	364
	365	// Clear out all complete pictures with no remaining references.
	366	prev = NULL;
	367	for (pic = ctx->pic_start; pic; pic = next) {
	368	next = pic->next;
	369	if (pic->encode_complete && pic->ref_count[1] == 0) {
	370	av_assert0(pic->ref_removed[0] && pic->ref_removed[1]);
	371	if (prev)
	372	prev->next = next;
	373	else
	374	ctx->pic_start = next;
	375	ctx->op->free(avctx, pic);
	376	} else {
	377	prev = pic;
	378	}
	379	}
	380
	381	return 0;
	382	}
	383
	384	static int hw_base_encode_check_frame(AVCodecContext *avctx,
	385	const AVFrame *frame)
	386	{
	387	FFHWBaseEncodeContext *ctx = avctx->priv_data;
	388
	389	if ((frame->crop_top \|\| frame->crop_bottom \|\|
	390	frame->crop_left \|\| frame->crop_right) && !ctx->crop_warned) {
	391	av_log(avctx, AV_LOG_WARNING, "Cropping information on input "
	392	"frames ignored due to lack of API support.\n");
	393	ctx->crop_warned = 1;
	394	}
	395
	396	if (!ctx->roi_allowed) {
	397	AVFrameSideData *sd =
	398	av_frame_get_side_data(frame, AV_FRAME_DATA_REGIONS_OF_INTEREST);
	399
	400	if (sd && !ctx->roi_warned) {
	401	av_log(avctx, AV_LOG_WARNING, "ROI side data on input "
	402	"frames ignored due to lack of driver support.\n");
	403	ctx->roi_warned = 1;
	404	}
	405	}
	406
	407	return 0;
	408	}
	409
	410	static int hw_base_encode_send_frame(AVCodecContext avctx, AVFrame frame)
	411	{
	412	FFHWBaseEncodeContext *ctx = avctx->priv_data;
	413	FFHWBaseEncodePicture *pic;
	414	int err;
	415
	416	if (frame) {
	417	av_log(avctx, AV_LOG_DEBUG, "Input frame: %ux%u (%"PRId64").\n",
	418	frame->width, frame->height, frame->pts);
	419
	420	err = hw_base_encode_check_frame(avctx, frame);
	421	if (err < 0)
	422	return err;
	423
	424	pic = ctx->op->alloc(avctx, frame);
	425	if (!pic)
	426	return AVERROR(ENOMEM);
	427
	428	pic->input_image = av_frame_alloc();
	429	if (!pic->input_image) {
	430	err = AVERROR(ENOMEM);
	431	goto fail;
	432	}
	433
	434	pic->recon_image = av_frame_alloc();
	435	if (!pic->recon_image) {
	436	err = AVERROR(ENOMEM);
	437	goto fail;
	438	}
	439
	440	if (ctx->input_order == 0 \|\| frame->pict_type == AV_PICTURE_TYPE_I)
	441	pic->force_idr = 1;
	442
	443	pic->pts = frame->pts;
	444	pic->duration = frame->duration;
	445
	446	if (avctx->flags & AV_CODEC_FLAG_COPY_OPAQUE) {
	447	err = av_buffer_replace(&pic->opaque_ref, frame->opaque_ref);
	448	if (err < 0)
	449	goto fail;
	450
	451	pic->opaque = frame->opaque;
	452	}
	453
	454	av_frame_move_ref(pic->input_image, frame);
	455
	456	if (ctx->input_order == 0)
	457	ctx->first_pts = pic->pts;
	458	if (ctx->input_order == ctx->decode_delay)
	459	ctx->dts_pts_diff = pic->pts - ctx->first_pts;
	460	if (ctx->output_delay > 0)
	461	ctx->ts_ring[ctx->input_order %
	462	(3 * ctx->output_delay + ctx->async_depth)] = pic->pts;
	463
	464	pic->display_order = ctx->input_order;
	465	++ctx->input_order;
	466
	467	if (ctx->pic_start) {
	468	ctx->pic_end->next = pic;
	469	ctx->pic_end = pic;
	470	} else {
	471	ctx->pic_start = pic;
	472	ctx->pic_end = pic;
	473	}
	474
	475	} else {
	476	ctx->end_of_stream = 1;
	477
	478	// Fix timestamps if we hit end-of-stream before the initial decode
	479	// delay has elapsed.
	480	if (ctx->input_order < ctx->decode_delay)
	481	ctx->dts_pts_diff = ctx->pic_end->pts - ctx->first_pts;
	482	}
	483
	484	return 0;
	485
	486	fail:
	487	ctx->op->free(avctx, pic);
	488	return err;
	489	}
	490
	491	int ff_hw_base_encode_receive_packet(AVCodecContext avctx, AVPacket pkt)
	492	{
	493	FFHWBaseEncodeContext *ctx = avctx->priv_data;
	494	FFHWBaseEncodePicture *pic = NULL;
	495	AVFrame *frame = ctx->frame;
	496	int err;
	497
	498	av_assert0(ctx->op && ctx->op->alloc && ctx->op->issue &&
	499	ctx->op->output && ctx->op->free);
	500
	501	start:
	502	/** if no B frame before repeat P frame, sent repeat P frame out. */
	503	if (ctx->tail_pkt->size) {
	504	for (FFHWBaseEncodePicture *tmp = ctx->pic_start; tmp; tmp = tmp->next) {
	505	if (tmp->type == FF_HW_PICTURE_TYPE_B && tmp->pts < ctx->tail_pkt->pts)
	506	break;
	507	else if (!tmp->next) {
	508	av_packet_move_ref(pkt, ctx->tail_pkt);
	509	goto end;
	510	}
	511	}
	512	}
	513
	514	err = ff_encode_get_frame(avctx, frame);
	515	if (err < 0 && err != AVERROR_EOF)
	516	return err;
	517
	518	if (err == AVERROR_EOF)
	519	frame = NULL;
	520
	521	err = hw_base_encode_send_frame(avctx, frame);
	522	if (err < 0)
	523	return err;
	524
	525	if (!ctx->pic_start) {
	526	if (ctx->end_of_stream)
	527	return AVERROR_EOF;
	528	else
	529	return AVERROR(EAGAIN);
	530	}
	531
	532	if (ctx->async_encode) {
	533	if (av_fifo_can_write(ctx->encode_fifo)) {
	534	err = hw_base_encode_pick_next(avctx, &pic);
	535	if (!err) {
	536	av_assert0(pic);
	537	pic->encode_order = ctx->encode_order +
	538	av_fifo_can_read(ctx->encode_fifo);
	539	err = ctx->op->issue(avctx, pic);
	540	if (err < 0) {
	541	av_log(avctx, AV_LOG_ERROR, "Encode failed: %d.\n", err);
	542	return err;
	543	}
	544	pic->encode_issued = 1;
	545	av_fifo_write(ctx->encode_fifo, &pic, 1);
	546	}
	547	}
	548
	549	if (!av_fifo_can_read(ctx->encode_fifo))
	550	return err;
	551
	552	// More frames can be buffered
	553	if (av_fifo_can_write(ctx->encode_fifo) && !ctx->end_of_stream)
	554	return AVERROR(EAGAIN);
	555
	556	av_fifo_read(ctx->encode_fifo, &pic, 1);
	557	ctx->encode_order = pic->encode_order + 1;
	558	} else {
	559	err = hw_base_encode_pick_next(avctx, &pic);
	560	if (err < 0)
	561	return err;
	562	av_assert0(pic);
	563
	564	pic->encode_order = ctx->encode_order++;
	565
	566	err = ctx->op->issue(avctx, pic);
	567	if (err < 0) {
	568	av_log(avctx, AV_LOG_ERROR, "Encode failed: %d.\n", err);
	569	return err;
	570	}
	571
	572	pic->encode_issued = 1;
	573	}
	574
	575	err = ctx->op->output(avctx, pic, pkt);
	576	if (err < 0) {
	577	av_log(avctx, AV_LOG_ERROR, "Output failed: %d.\n", err);
	578	return err;
	579	}
	580
	581	ctx->output_order = pic->encode_order;
	582	hw_base_encode_clear_old(avctx);
	583
	584	/** loop to get an available pkt in encoder flushing. */
	585	if (ctx->end_of_stream && !pkt->size)
	586	goto start;
	587
	588	end:
	589	if (pkt->size)
	590	av_log(avctx, AV_LOG_DEBUG, "Output packet: pts %"PRId64", dts %"PRId64", "
	591	"size %d bytes.\n", pkt->pts, pkt->dts, pkt->size);
	592
	593	return 0;
	594	}

libavcodec/hw_base_encode.h+124−0

@@ -19,6 +19,8 @@ · 18 unmodified lines
19	#ifndef AVCODEC_HW_BASE_ENCODE_H	19	#ifndef AVCODEC_HW_BASE_ENCODE_H
20	#define AVCODEC_HW_BASE_ENCODE_H	20	#define AVCODEC_HW_BASE_ENCODE_H
21		21
		22	#include "libavutil/fifo.h"
		23
22	#define MAX_DPB_SIZE 16	24	#define MAX_DPB_SIZE 16
23	#define MAX_PICTURE_REFERENCES 2	25	#define MAX_PICTURE_REFERENCES 2
24	#define MAX_REORDER_DELAY 16	26	#define MAX_REORDER_DELAY 16
@@ -54,13 +56,135 @@ enum { · 29 unmodified lines
54	FF_HW_FLAG_NON_IDR_KEY_PICTURES = 1 << 5,	56	FF_HW_FLAG_NON_IDR_KEY_PICTURES = 1 << 5,
55	};	57	};
56		58
		59	typedef struct FFHWBaseEncodePicture {
		60	struct FFHWBaseEncodePicture *next;
		61
		62	int64_t display_order;
		63	int64_t encode_order;
		64	int64_t pts;
		65	int64_t duration;
		66	int force_idr;
		67
		68	void *opaque;
		69	AVBufferRef *opaque_ref;
		70
		71	int type;
		72	int b_depth;
		73	int encode_issued;
		74	int encode_complete;
		75
		76	AVFrame *input_image;
		77	AVFrame *recon_image;
		78
		79	void *priv_data;
		80
		81	// Whether this picture is a reference picture.
		82	int is_reference;
		83
		84	// The contents of the DPB after this picture has been decoded.
		85	// This will contain the picture itself if it is a reference picture,
		86	// but not if it isn't.
		87	int nb_dpb_pics;
		88	struct FFHWBaseEncodePicture *dpb[MAX_DPB_SIZE];
		89	// The reference pictures used in decoding this picture. If they are
		90	// used by later pictures they will also appear in the DPB. ref[0][] for
		91	// previous reference frames. ref[1][] for future reference frames.
		92	int nb_refs[MAX_REFERENCE_LIST_NUM];
		93	struct FFHWBaseEncodePicture *refs[MAX_REFERENCE_LIST_NUM][MAX_PICTURE_REFERENCES];
		94	// The previous reference picture in encode order. Must be in at least
		95	// one of the reference list and DPB list.
		96	struct FFHWBaseEncodePicture *prev;
		97	// Reference count for other pictures referring to this one through
		98	// the above pointers, directly from incomplete pictures and indirectly
		99	// through completed pictures.
		100	int ref_count[2];
		101	int ref_removed[2];
		102	} FFHWBaseEncodePicture;
		103
		104	typedef struct FFHWEncodePictureOperation {
		105	// Alloc memory for the picture structure and initialize the API-specific internals
		106	// based of the given frame.
		107	FFHWBaseEncodePicture * (alloc)(AVCodecContext avctx, const AVFrame *frame);
		108	// Issue the picture structure, which will send the frame surface to HW Encode API.
		109	int (issue)(AVCodecContext avctx, const FFHWBaseEncodePicture *base_pic);
		110	// Get the output AVPacket.
		111	int (output)(AVCodecContext avctx, const FFHWBaseEncodePicture base_pic, AVPacket pkt);
		112	// Free the picture structure.
		113	int (free)(AVCodecContext avctx, FFHWBaseEncodePicture *base_pic);
		114	} FFHWEncodePictureOperation;
		115
57	typedef struct FFHWBaseEncodeContext {	116	typedef struct FFHWBaseEncodeContext {
58	const AVClass *class;	117	const AVClass *class;
59		118
		119	// Hardware-specific hooks.
		120	const struct FFHWEncodePictureOperation *op;
		121
		122	// Current encoding window, in display (input) order.
		123	FFHWBaseEncodePicture pic_start, pic_end;
		124	// The next picture to use as the previous reference picture in
		125	// encoding order. Order from small to large in encoding order.
		126	FFHWBaseEncodePicture *next_prev[MAX_PICTURE_REFERENCES];
		127	int nb_next_prev;
		128
		129	// Next input order index (display order).
		130	int64_t input_order;
		131	// Number of frames that output is behind input.
		132	int64_t output_delay;
		133	// Next encode order index.
		134	int64_t encode_order;
		135	// Number of frames decode output will need to be delayed.
		136	int64_t decode_delay;
		137	// Next output order index (in encode order).
		138	int64_t output_order;
		139
		140	// Timestamp handling.
		141	int64_t first_pts;
		142	int64_t dts_pts_diff;
		143	int64_t ts_ring[MAX_REORDER_DELAY * 3 +
		144	MAX_ASYNC_DEPTH];
		145
		146	// Frame type decision.
		147	int gop_size;
		148	int closed_gop;
		149	int gop_per_idr;
		150	int p_per_i;
		151	int max_b_depth;
		152	int b_per_p;
		153	int force_idr;
		154	int idr_counter;
		155	int gop_counter;
		156	int end_of_stream;
		157	int p_to_gpb;
		158
		159	// Whether the driver supports ROI at all.
		160	int roi_allowed;
		161
		162	// The encoder does not support cropping information, so warn about
		163	// it the first time we encounter any nonzero crop fields.
		164	int crop_warned;
		165	// If the driver does not support ROI then warn the first time we
		166	// encounter a frame with ROI side data.
		167	int roi_warned;
		168
		169	// The frame to be filled with data.
		170	AVFrame *frame;
		171
		172	// Whether the HW supports sync buffer function.
		173	// If supported, encode_fifo/async_depth will be used together.
		174	// Used for output buffer synchronization.
		175	int async_encode;
		176
		177	// Store buffered pic.
		178	AVFifo *encode_fifo;
60	// Max number of frame buffered in encoder.	179	// Max number of frame buffered in encoder.
61	int async_depth;	180	int async_depth;
		181
		182	/** Tail data of a pic, now only used for av1 repeat frame header. */
		183	AVPacket *tail_pkt;
62	} FFHWBaseEncodeContext;	184	} FFHWBaseEncodeContext;
63		185
		186	int ff_hw_base_encode_receive_packet(AVCodecContext avctx, AVPacket pkt);
		187
64	#define HW_BASE_ENCODE_COMMON_OPTIONS \	188	#define HW_BASE_ENCODE_COMMON_OPTIONS \
65	{ "async_depth", "Maximum processing parallelism. " \	189	{ "async_depth", "Maximum processing parallelism. " \
66	"Increase this to improve single channel performance.", \	190	"Increase this to improve single channel performance.", \

libavcodec/vaapi_encode.c+126−667

@@ -138,22 +138,26 @@ static int vaapi_encode_make_misc_param_buffer(AVCodecContext *avctx, · 137 unmodified lines
138	static int vaapi_encode_wait(AVCodecContext *avctx,	138	static int vaapi_encode_wait(AVCodecContext *avctx,
139	VAAPIEncodePicture *pic)	139	VAAPIEncodePicture *pic)
140	{	140	{
		141	#if VA_CHECK_VERSION(1, 9, 0)
		142	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;
		143	#endif
141	VAAPIEncodeContext *ctx = avctx->priv_data;	144	VAAPIEncodeContext *ctx = avctx->priv_data;
		145	FFHWBaseEncodePicture *base_pic = &pic->base;
142	VAStatus vas;	146	VAStatus vas;
143		147
144	av_assert0(pic->encode_issued);	148	av_assert0(base_pic->encode_issued);
145		149
146	if (pic->encode_complete) {	150	if (base_pic->encode_complete) {
147	// Already waited for this picture.	151	// Already waited for this picture.
148	return 0;	152	return 0;
149	}	153	}
150		154
151	av_log(avctx, AV_LOG_DEBUG, "Sync to pic %"PRId64"/%"PRId64" "	155	av_log(avctx, AV_LOG_DEBUG, "Sync to pic %"PRId64"/%"PRId64" "
152	"(input surface %#x).\n", pic->display_order,	156	"(input surface %#x).\n", base_pic->display_order,
153	pic->encode_order, pic->input_surface);	157	base_pic->encode_order, pic->input_surface);
154		158
155	#if VA_CHECK_VERSION(1, 9, 0)	159	#if VA_CHECK_VERSION(1, 9, 0)
156	if (ctx->has_sync_buffer_func) {	160	if (base_ctx->async_encode) {
157	vas = vaSyncBuffer(ctx->hwctx->display,	161	vas = vaSyncBuffer(ctx->hwctx->display,
158	pic->output_buffer,	162	pic->output_buffer,
159	VA_TIMEOUT_INFINITE);	163	VA_TIMEOUT_INFINITE);
@@ -174,9 +178,9 @@ static int vaapi_encode_wait(AVCodecContext *avctx, · 14 unmodified lines
174	}	178	}
175		179
176	// Input is definitely finished with now.	180	// Input is definitely finished with now.
177	av_frame_free(&pic->input_image);	181	av_frame_free(&base_pic->input_image);
178		182
179	pic->encode_complete = 1;	183	base_pic->encode_complete = 1;
180	return 0;	184	return 0;
181	}	185	}
182		186
@@ -263,9 +267,11 @@ static int vaapi_encode_make_tile_slice(AVCodecContext *avctx, · 80 unmodified lines
263	}	267	}
264		268
265	static int vaapi_encode_issue(AVCodecContext *avctx,	269	static int vaapi_encode_issue(AVCodecContext *avctx,
266	VAAPIEncodePicture *pic)	270	const FFHWBaseEncodePicture *base_pic)
267	{	271	{
268	VAAPIEncodeContext *ctx = avctx->priv_data;	272	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;
		273	VAAPIEncodeContext *ctx = avctx->priv_data;
		274	VAAPIEncodePicture pic = (VAAPIEncodePicture)base_pic;
269	VAAPIEncodeSlice *slice;	275	VAAPIEncodeSlice *slice;
270	VAStatus vas;	276	VAStatus vas;
271	int err, i;	277	int err, i;
@@ -274,52 +280,46 @@ static int vaapi_encode_issue(AVCodecContext *avctx, · 2 unmodified lines
274	av_unused AVFrameSideData *sd;	280	av_unused AVFrameSideData *sd;
275		281
276	av_log(avctx, AV_LOG_DEBUG, "Issuing encode for pic %"PRId64"/%"PRId64" "	282	av_log(avctx, AV_LOG_DEBUG, "Issuing encode for pic %"PRId64"/%"PRId64" "
277	"as type %s.\n", pic->display_order, pic->encode_order,	283	"as type %s.\n", base_pic->display_order, base_pic->encode_order,
278	ff_hw_base_encode_get_pictype_name(pic->type));	284	ff_hw_base_encode_get_pictype_name(base_pic->type));
279	if (pic->nb_refs[0] == 0 && pic->nb_refs[1] == 0) {	285	if (base_pic->nb_refs[0] == 0 && base_pic->nb_refs[1] == 0) {
280	av_log(avctx, AV_LOG_DEBUG, "No reference pictures.\n");	286	av_log(avctx, AV_LOG_DEBUG, "No reference pictures.\n");
281	} else {	287	} else {
282	av_log(avctx, AV_LOG_DEBUG, "L0 refers to");	288	av_log(avctx, AV_LOG_DEBUG, "L0 refers to");
283	for (i = 0; i < pic->nb_refs[0]; i++) {	289	for (i = 0; i < base_pic->nb_refs[0]; i++) {
284	av_log(avctx, AV_LOG_DEBUG, " %"PRId64"/%"PRId64,	290	av_log(avctx, AV_LOG_DEBUG, " %"PRId64"/%"PRId64,
285	pic->refs[0][i]->display_order, pic->refs[0][i]->encode_order);	291	base_pic->refs[0][i]->display_order, base_pic->refs[0][i]->encode_order);
286	}	292	}
287	av_log(avctx, AV_LOG_DEBUG, ".\n");	293	av_log(avctx, AV_LOG_DEBUG, ".\n");
288		294
289	if (pic->nb_refs[1]) {	295	if (base_pic->nb_refs[1]) {
290	av_log(avctx, AV_LOG_DEBUG, "L1 refers to");	296	av_log(avctx, AV_LOG_DEBUG, "L1 refers to");
291	for (i = 0; i < pic->nb_refs[1]; i++) {	297	for (i = 0; i < base_pic->nb_refs[1]; i++) {
292	av_log(avctx, AV_LOG_DEBUG, " %"PRId64"/%"PRId64,	298	av_log(avctx, AV_LOG_DEBUG, " %"PRId64"/%"PRId64,
293	pic->refs[1][i]->display_order, pic->refs[1][i]->encode_order);	299	base_pic->refs[1][i]->display_order, base_pic->refs[1][i]->encode_order);
294	}	300	}
295	av_log(avctx, AV_LOG_DEBUG, ".\n");	301	av_log(avctx, AV_LOG_DEBUG, ".\n");
296	}	302	}
297	}	303	}
298		304
299	av_assert0(!pic->encode_issued);	305	av_assert0(!base_pic->encode_issued);
300	for (i = 0; i < pic->nb_refs[0]; i++) {	306	for (i = 0; i < base_pic->nb_refs[0]; i++) {
301	av_assert0(pic->refs[0][i]);	307	av_assert0(base_pic->refs[0][i]);
302	av_assert0(pic->refs[0][i]->encode_issued);	308	av_assert0(base_pic->refs[0][i]->encode_issued);
303	}	309	}
304	for (i = 0; i < pic->nb_refs[1]; i++) {	310	for (i = 0; i < base_pic->nb_refs[1]; i++) {
305	av_assert0(pic->refs[1][i]);	311	av_assert0(base_pic->refs[1][i]);
306	av_assert0(pic->refs[1][i]->encode_issued);	312	av_assert0(base_pic->refs[1][i]->encode_issued);
307	}	313	}
308		314
309	av_log(avctx, AV_LOG_DEBUG, "Input surface is %#x.\n", pic->input_surface);	315	av_log(avctx, AV_LOG_DEBUG, "Input surface is %#x.\n", pic->input_surface);
310		316
311	pic->recon_image = av_frame_alloc();	317	err = av_hwframe_get_buffer(ctx->recon_frames_ref, base_pic->recon_image, 0);
312	if (!pic->recon_image) {
313	err = AVERROR(ENOMEM);
314	goto fail;
315	}
316
317	err = av_hwframe_get_buffer(ctx->recon_frames_ref, pic->recon_image, 0);
318	if (err < 0) {	318	if (err < 0) {
319	err = AVERROR(ENOMEM);	319	err = AVERROR(ENOMEM);
320	goto fail;	320	goto fail;
321	}	321	}
322	pic->recon_surface = (VASurfaceID)(uintptr_t)pic->recon_image->data[3];	322	pic->recon_surface = (VASurfaceID)(uintptr_t)base_pic->recon_image->data[3];
323	av_log(avctx, AV_LOG_DEBUG, "Recon surface is %#x.\n", pic->recon_surface);	323	av_log(avctx, AV_LOG_DEBUG, "Recon surface is %#x.\n", pic->recon_surface);
324		324
325	pic->output_buffer_ref = ff_refstruct_pool_get(ctx->output_buffer_pool);	325	pic->output_buffer_ref = ff_refstruct_pool_get(ctx->output_buffer_pool);
@@ -343,7 +343,7 @@ static int vaapi_encode_issue(AVCodecContext *avctx, · 17 unmodified lines
343		343
344	pic->nb_param_buffers = 0;	344	pic->nb_param_buffers = 0;
345		345
346	if (pic->type == FF_HW_PICTURE_TYPE_IDR && ctx->codec->init_sequence_params) {	346	if (base_pic->type == FF_HW_PICTURE_TYPE_IDR && ctx->codec->init_sequence_params) {
347	err = vaapi_encode_make_param_buffer(avctx, pic,	347	err = vaapi_encode_make_param_buffer(avctx, pic,
348	VAEncSequenceParameterBufferType,	348	VAEncSequenceParameterBufferType,
349	ctx->codec_sequence_params,	349	ctx->codec_sequence_params,
@@ -352,7 +352,7 @@ static int vaapi_encode_issue(AVCodecContext *avctx, · 2 unmodified lines
352	goto fail;	352	goto fail;
353	}	353	}
354		354
355	if (pic->type == FF_HW_PICTURE_TYPE_IDR) {	355	if (base_pic->type == FF_HW_PICTURE_TYPE_IDR) {
356	for (i = 0; i < ctx->nb_global_params; i++) {	356	for (i = 0; i < ctx->nb_global_params; i++) {
357	err = vaapi_encode_make_misc_param_buffer(avctx, pic,	357	err = vaapi_encode_make_misc_param_buffer(avctx, pic,
358	ctx->global_params_type[i],	358	ctx->global_params_type[i],
@@ -389,7 +389,7 @@ static int vaapi_encode_issue(AVCodecContext *avctx, · 30 unmodified lines
389	}	389	}
390	#endif	390	#endif
391		391
392	if (pic->type == FF_HW_PICTURE_TYPE_IDR) {	392	if (base_pic->type == FF_HW_PICTURE_TYPE_IDR) {
393	if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE &&	393	if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE &&
394	ctx->codec->write_sequence_header) {	394	ctx->codec->write_sequence_header) {
395	bit_len = 8 * sizeof(data);	395	bit_len = 8 * sizeof(data);
@@ -529,9 +529,9 @@ static int vaapi_encode_issue(AVCodecContext *avctx, · 133 unmodified lines
529	}	529	}
530		530
531	#if VA_CHECK_VERSION(1, 0, 0)	531	#if VA_CHECK_VERSION(1, 0, 0)
532	sd = av_frame_get_side_data(pic->input_image,	532	sd = av_frame_get_side_data(base_pic->input_image,
533	AV_FRAME_DATA_REGIONS_OF_INTEREST);	533	AV_FRAME_DATA_REGIONS_OF_INTEREST);
534	if (sd && ctx->roi_allowed) {	534	if (sd && base_ctx->roi_allowed) {
535	const AVRegionOfInterest *roi;	535	const AVRegionOfInterest *roi;
536	uint32_t roi_size;	536	uint32_t roi_size;
537	VAEncMiscParameterBufferROI param_roi;	537	VAEncMiscParameterBufferROI param_roi;
@@ -542,11 +542,11 @@ static int vaapi_encode_issue(AVCodecContext *avctx, · 4 unmodified lines
542	av_assert0(roi_size && sd->size % roi_size == 0);	542	av_assert0(roi_size && sd->size % roi_size == 0);
543	nb_roi = sd->size / roi_size;	543	nb_roi = sd->size / roi_size;
544	if (nb_roi > ctx->roi_max_regions) {	544	if (nb_roi > ctx->roi_max_regions) {
545	if (!ctx->roi_warned) {	545	if (!base_ctx->roi_warned) {
546	av_log(avctx, AV_LOG_WARNING, "More ROIs set than "	546	av_log(avctx, AV_LOG_WARNING, "More ROIs set than "
547	"supported by driver (%d > %d).\n",	547	"supported by driver (%d > %d).\n",
548	nb_roi, ctx->roi_max_regions);	548	nb_roi, ctx->roi_max_regions);
549	ctx->roi_warned = 1;	549	base_ctx->roi_warned = 1;
550	}	550	}
551	nb_roi = ctx->roi_max_regions;	551	nb_roi = ctx->roi_max_regions;
552	}	552	}
@@ -639,8 +639,6 @@ static int vaapi_encode_issue(AVCodecContext *avctx, · 86 unmodified lines
639	}	639	}
640	}	640	}
641		641
642	pic->encode_issued = 1;
643
644	return 0;	642	return 0;
645		643
646	fail_with_picture:	644	fail_with_picture:
@@ -657,14 +655,13 @@ fail_at_end: · 10 unmodified lines
657	av_freep(&pic->param_buffers);	655	av_freep(&pic->param_buffers);
658	av_freep(&pic->slices);	656	av_freep(&pic->slices);
659	av_freep(&pic->roi);	657	av_freep(&pic->roi);
660	av_frame_free(&pic->recon_image);
661	ff_refstruct_unref(&pic->output_buffer_ref);	658	ff_refstruct_unref(&pic->output_buffer_ref);
662	pic->output_buffer = VA_INVALID_ID;	659	pic->output_buffer = VA_INVALID_ID;
663	return err;	660	return err;
664	}	661	}
665		662
666	static int vaapi_encode_set_output_property(AVCodecContext *avctx,	663	static int vaapi_encode_set_output_property(AVCodecContext *avctx,
667	VAAPIEncodePicture *pic,	664	FFHWBaseEncodePicture *pic,
668	AVPacket *pkt)	665	AVPacket *pkt)
669	{	666	{
670	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;	667	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;
@@ -689,16 +686,16 @@ static int vaapi_encode_set_output_property(AVCodecContext *avctx, · 18 unmodified lines
689	return 0;	686	return 0;
690	}	687	}
691		688
692	if (ctx->output_delay == 0) {	689	if (base_ctx->output_delay == 0) {
693	pkt->dts = pkt->pts;	690	pkt->dts = pkt->pts;
694	} else if (pic->encode_order < ctx->decode_delay) {	691	} else if (pic->encode_order < base_ctx->decode_delay) {
695	if (ctx->ts_ring[pic->encode_order] < INT64_MIN + ctx->dts_pts_diff)	692	if (base_ctx->ts_ring[pic->encode_order] < INT64_MIN + base_ctx->dts_pts_diff)
696	pkt->dts = INT64_MIN;	693	pkt->dts = INT64_MIN;
697	else	694	else
698	pkt->dts = ctx->ts_ring[pic->encode_order] - ctx->dts_pts_diff;	695	pkt->dts = base_ctx->ts_ring[pic->encode_order] - base_ctx->dts_pts_diff;
699	} else {	696	} else {
700	pkt->dts = ctx->ts_ring[(pic->encode_order - ctx->decode_delay) %	697	pkt->dts = base_ctx->ts_ring[(pic->encode_order - base_ctx->decode_delay) %
701	(3 * ctx->output_delay + base_ctx->async_depth)];	698	(3 * base_ctx->output_delay + base_ctx->async_depth)];
702	}	699	}
703		700
704	return 0;	701	return 0;
@@ -817,9 +814,11 @@ end: · 112 unmodified lines
817	}	814	}
818		815
819	static int vaapi_encode_output(AVCodecContext *avctx,	816	static int vaapi_encode_output(AVCodecContext *avctx,
820	VAAPIEncodePicture pic, AVPacket pkt)	817	const FFHWBaseEncodePicture base_pic, AVPacket pkt)
821	{	818	{
822	VAAPIEncodeContext *ctx = avctx->priv_data;	819	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;
		820	VAAPIEncodeContext *ctx = avctx->priv_data;
		821	VAAPIEncodePicture pic = (VAAPIEncodePicture)base_pic;
823	AVPacket *pkt_ptr = pkt;	822	AVPacket *pkt_ptr = pkt;
824	int err;	823	int err;
825		824
@@ -832,17 +831,17 @@ static int vaapi_encode_output(AVCodecContext *avctx, · 6 unmodified lines
832	ctx->coded_buffer_ref = ff_refstruct_ref(pic->output_buffer_ref);	831	ctx->coded_buffer_ref = ff_refstruct_ref(pic->output_buffer_ref);
833		832
834	if (pic->tail_size) {	833	if (pic->tail_size) {
835	if (ctx->tail_pkt->size) {	834	if (base_ctx->tail_pkt->size) {
836	err = AVERROR_BUG;	835	err = AVERROR_BUG;
837	goto end;	836	goto end;
838	}	837	}
839		838
840	err = ff_get_encode_buffer(avctx, ctx->tail_pkt, pic->tail_size, 0);	839	err = ff_get_encode_buffer(avctx, base_ctx->tail_pkt, pic->tail_size, 0);
841	if (err < 0)	840	if (err < 0)
842	goto end;	841	goto end;
843		842
844	memcpy(ctx->tail_pkt->data, pic->tail_data, pic->tail_size);	843	memcpy(base_ctx->tail_pkt->data, pic->tail_data, pic->tail_size);
845	pkt_ptr = ctx->tail_pkt;	844	pkt_ptr = base_ctx->tail_pkt;
846	}	845	}
847	} else {	846	} else {
848	err = vaapi_encode_get_coded_data(avctx, pic, pkt);	847	err = vaapi_encode_get_coded_data(avctx, pic, pkt);
@@ -851,9 +850,9 @@ static int vaapi_encode_output(AVCodecContext *avctx, · 2 unmodified lines
851	}	850	}
852		851
853	av_log(avctx, AV_LOG_DEBUG, "Output read for pic %"PRId64"/%"PRId64".\n",	852	av_log(avctx, AV_LOG_DEBUG, "Output read for pic %"PRId64"/%"PRId64".\n",
854	pic->display_order, pic->encode_order);	853	base_pic->display_order, base_pic->encode_order);
855		854
856	vaapi_encode_set_output_property(avctx, pic, pkt_ptr);	855	vaapi_encode_set_output_property(avctx, (FFHWBaseEncodePicture*)pic, pkt_ptr);
857		856
858	end:	857	end:
859	ff_refstruct_unref(&pic->output_buffer_ref);	858	ff_refstruct_unref(&pic->output_buffer_ref);
@@ -864,12 +863,14 @@ end: · 4 unmodified lines
864	static int vaapi_encode_discard(AVCodecContext *avctx,	863	static int vaapi_encode_discard(AVCodecContext *avctx,
865	VAAPIEncodePicture *pic)	864	VAAPIEncodePicture *pic)
866	{	865	{
		866	FFHWBaseEncodePicture *base_pic = &pic->base;
		867
867	vaapi_encode_wait(avctx, pic);	868	vaapi_encode_wait(avctx, pic);
868		869
869	if (pic->output_buffer_ref) {	870	if (pic->output_buffer_ref) {
870	av_log(avctx, AV_LOG_DEBUG, "Discard output for pic "	871	av_log(avctx, AV_LOG_DEBUG, "Discard output for pic "
871	"%"PRId64"/%"PRId64".\n",	872	"%"PRId64"/%"PRId64".\n",
872	pic->display_order, pic->encode_order);	873	base_pic->display_order, base_pic->encode_order);
873		874
874	ff_refstruct_unref(&pic->output_buffer_ref);	875	ff_refstruct_unref(&pic->output_buffer_ref);
875	pic->output_buffer = VA_INVALID_ID;	876	pic->output_buffer = VA_INVALID_ID;
@@ -878,8 +879,8 @@ static int vaapi_encode_discard(AVCodecContext *avctx, · 2 unmodified lines
878	return 0;	879	return 0;
879	}	880	}
880		881
881	static VAAPIEncodePicture vaapi_encode_alloc(AVCodecContext avctx,	882	static FFHWBaseEncodePicture vaapi_encode_alloc(AVCodecContext avctx,
882	const AVFrame *frame)	883	const AVFrame *frame)
883	{	884	{
884	VAAPIEncodeContext *ctx = avctx->priv_data;	885	VAAPIEncodeContext *ctx = avctx->priv_data;
885	VAAPIEncodePicture *pic;	886	VAAPIEncodePicture *pic;
@@ -889,8 +890,8 @@ static VAAPIEncodePicture vaapi_encode_alloc(AVCodecContext avctx, · 3 unmodified lines
889	return NULL;	890	return NULL;
890		891
891	if (ctx->codec->picture_priv_data_size > 0) {	892	if (ctx->codec->picture_priv_data_size > 0) {
892	pic->priv_data = av_mallocz(ctx->codec->picture_priv_data_size);	893	pic->base.priv_data = av_mallocz(ctx->codec->picture_priv_data_size);
893	if (!pic->priv_data) {	894	if (!pic->base.priv_data) {
894	av_freep(&pic);	895	av_freep(&pic);
895	return NULL;	896	return NULL;
896	}	897	}
@@ -900,15 +901,16 @@ static VAAPIEncodePicture vaapi_encode_alloc(AVCodecContext avctx, · 3 unmodified lines
900	pic->recon_surface = VA_INVALID_ID;	901	pic->recon_surface = VA_INVALID_ID;
901	pic->output_buffer = VA_INVALID_ID;	902	pic->output_buffer = VA_INVALID_ID;
902		903
903	return pic;	904	return &pic->base;
904	}	905	}
905		906
906	static int vaapi_encode_free(AVCodecContext *avctx,	907	static int vaapi_encode_free(AVCodecContext *avctx,
907	VAAPIEncodePicture *pic)	908	FFHWBaseEncodePicture *base_pic)
908	{	909	{
		910	VAAPIEncodePicture pic = (VAAPIEncodePicture)base_pic;
909	int i;	911	int i;
910		912
911	if (pic->encode_issued)	913	if (base_pic->encode_issued)
912	vaapi_encode_discard(avctx, pic);	914	vaapi_encode_discard(avctx, pic);
913		915
914	if (pic->slices) {	916	if (pic->slices) {
@@ -916,17 +918,17 @@ static int vaapi_encode_free(AVCodecContext *avctx, · 1 unmodified line
916	av_freep(&pic->slices[i].codec_slice_params);	918	av_freep(&pic->slices[i].codec_slice_params);
917	}	919	}
918		920
919	av_frame_free(&pic->input_image);	921	av_frame_free(&base_pic->input_image);
920	av_frame_free(&pic->recon_image);	922	av_frame_free(&base_pic->recon_image);
921		923
922	av_buffer_unref(&pic->opaque_ref);	924	av_buffer_unref(&base_pic->opaque_ref);
923		925
924	av_freep(&pic->param_buffers);	926	av_freep(&pic->param_buffers);
925	av_freep(&pic->slices);	927	av_freep(&pic->slices);
926	// Output buffer should already be destroyed.	928	// Output buffer should already be destroyed.
927	av_assert0(pic->output_buffer == VA_INVALID_ID);	929	av_assert0(pic->output_buffer == VA_INVALID_ID);
928		930
929	av_freep(&pic->priv_data);	931	av_freep(&base_pic->priv_data);
930	av_freep(&pic->codec_picture_params);	932	av_freep(&pic->codec_picture_params);
931	av_freep(&pic->roi);	933	av_freep(&pic->roi);
932		934
@@ -935,564 +937,6 @@ static int vaapi_encode_free(AVCodecContext *avctx, · 2 unmodified lines
935	return 0;	937	return 0;
936	}	938	}
937		939
938	static void vaapi_encode_add_ref(AVCodecContext *avctx,
939	VAAPIEncodePicture *pic,
940	VAAPIEncodePicture *target,
941	int is_ref, int in_dpb, int prev)
942	{
943	int refs = 0;
944
945	if (is_ref) {
946	av_assert0(pic != target);
947	av_assert0(pic->nb_refs[0] < MAX_PICTURE_REFERENCES &&
948	pic->nb_refs[1] < MAX_PICTURE_REFERENCES);
949	if (target->display_order < pic->display_order)
950	pic->refs[0][pic->nb_refs[0]++] = target;
951	else
952	pic->refs[1][pic->nb_refs[1]++] = target;
953	++refs;
954	}
955
956	if (in_dpb) {
957	av_assert0(pic->nb_dpb_pics < MAX_DPB_SIZE);
958	pic->dpb[pic->nb_dpb_pics++] = target;
959	++refs;
960	}
961
962	if (prev) {
963	av_assert0(!pic->prev);
964	pic->prev = target;
965	++refs;
966	}
967
968	target->ref_count[0] += refs;
969	target->ref_count[1] += refs;
970	}
971
972	static void vaapi_encode_remove_refs(AVCodecContext *avctx,
973	VAAPIEncodePicture *pic,
974	int level)
975	{
976	int i;
977
978	if (pic->ref_removed[level])
979	return;
980
981	for (i = 0; i < pic->nb_refs[0]; i++) {
982	av_assert0(pic->refs[0][i]);
983	--pic->refs[0][i]->ref_count[level];
984	av_assert0(pic->refs[0][i]->ref_count[level] >= 0);
985	}
986
987	for (i = 0; i < pic->nb_refs[1]; i++) {
988	av_assert0(pic->refs[1][i]);
989	--pic->refs[1][i]->ref_count[level];
990	av_assert0(pic->refs[1][i]->ref_count[level] >= 0);
991	}
992
993	for (i = 0; i < pic->nb_dpb_pics; i++) {
994	av_assert0(pic->dpb[i]);
995	--pic->dpb[i]->ref_count[level];
996	av_assert0(pic->dpb[i]->ref_count[level] >= 0);
997	}
998
999	av_assert0(pic->prev \|\| pic->type == FF_HW_PICTURE_TYPE_IDR);
1000	if (pic->prev) {
1001	--pic->prev->ref_count[level];
1002	av_assert0(pic->prev->ref_count[level] >= 0);
1003	}
1004
1005	pic->ref_removed[level] = 1;
1006	}
1007
1008	static void vaapi_encode_set_b_pictures(AVCodecContext *avctx,
1009	VAAPIEncodePicture *start,
1010	VAAPIEncodePicture *end,
1011	VAAPIEncodePicture *prev,
1012	int current_depth,
1013	VAAPIEncodePicture **last)
1014	{
1015	VAAPIEncodeContext *ctx = avctx->priv_data;
1016	VAAPIEncodePicture pic, next, *ref;
1017	int i, len;
1018
1019	av_assert0(start && end && start != end && start->next != end);
1020
1021	// If we are at the maximum depth then encode all pictures as
1022	// non-referenced B-pictures. Also do this if there is exactly one
1023	// picture left, since there will be nothing to reference it.
1024	if (current_depth == ctx->max_b_depth \|\| start->next->next == end) {
1025	for (pic = start->next; pic; pic = pic->next) {
1026	if (pic == end)
1027	break;
1028	pic->type = FF_HW_PICTURE_TYPE_B;
1029	pic->b_depth = current_depth;
1030
1031	vaapi_encode_add_ref(avctx, pic, start, 1, 1, 0);
1032	vaapi_encode_add_ref(avctx, pic, end, 1, 1, 0);
1033	vaapi_encode_add_ref(avctx, pic, prev, 0, 0, 1);
1034
1035	for (ref = end->refs[1][0]; ref; ref = ref->refs[1][0])
1036	vaapi_encode_add_ref(avctx, pic, ref, 0, 1, 0);
1037	}
1038	*last = prev;
1039
1040	} else {
1041	// Split the current list at the midpoint with a referenced
1042	// B-picture, then descend into each side separately.
1043	len = 0;
1044	for (pic = start->next; pic != end; pic = pic->next)
1045	++len;
1046	for (pic = start->next, i = 1; 2 * i < len; pic = pic->next, i++);
1047
1048	pic->type = FF_HW_PICTURE_TYPE_B;
1049	pic->b_depth = current_depth;
1050
1051	pic->is_reference = 1;
1052
1053	vaapi_encode_add_ref(avctx, pic, pic, 0, 1, 0);
1054	vaapi_encode_add_ref(avctx, pic, start, 1, 1, 0);
1055	vaapi_encode_add_ref(avctx, pic, end, 1, 1, 0);
1056	vaapi_encode_add_ref(avctx, pic, prev, 0, 0, 1);
1057
1058	for (ref = end->refs[1][0]; ref; ref = ref->refs[1][0])
1059	vaapi_encode_add_ref(avctx, pic, ref, 0, 1, 0);
1060
1061	if (i > 1)
1062	vaapi_encode_set_b_pictures(avctx, start, pic, pic,
1063	current_depth + 1, &next);
1064	else
1065	next = pic;
1066
1067	vaapi_encode_set_b_pictures(avctx, pic, end, next,
1068	current_depth + 1, last);
1069	}
1070	}
1071
1072	static void vaapi_encode_add_next_prev(AVCodecContext *avctx,
1073	VAAPIEncodePicture *pic)
1074	{
1075	VAAPIEncodeContext *ctx = avctx->priv_data;
1076	int i;
1077
1078	if (!pic)
1079	return;
1080
1081	if (pic->type == FF_HW_PICTURE_TYPE_IDR) {
1082	for (i = 0; i < ctx->nb_next_prev; i++) {
1083	--ctx->next_prev[i]->ref_count[0];
1084	ctx->next_prev[i] = NULL;
1085	}
1086	ctx->next_prev[0] = pic;
1087	++pic->ref_count[0];
1088	ctx->nb_next_prev = 1;
1089
1090	return;
1091	}
1092
1093	if (ctx->nb_next_prev < MAX_PICTURE_REFERENCES) {
1094	ctx->next_prev[ctx->nb_next_prev++] = pic;
1095	++pic->ref_count[0];
1096	} else {
1097	--ctx->next_prev[0]->ref_count[0];
1098	for (i = 0; i < MAX_PICTURE_REFERENCES - 1; i++)
1099	ctx->next_prev[i] = ctx->next_prev[i + 1];
1100	ctx->next_prev[i] = pic;
1101	++pic->ref_count[0];
1102	}
1103	}
1104
1105	static int vaapi_encode_pick_next(AVCodecContext *avctx,
1106	VAAPIEncodePicture **pic_out)
1107	{
1108	VAAPIEncodeContext *ctx = avctx->priv_data;
1109	VAAPIEncodePicture pic = NULL, prev = NULL, next, start;
1110	int i, b_counter, closed_gop_end;
1111
1112	// If there are any B-frames already queued, the next one to encode
1113	// is the earliest not-yet-issued frame for which all references are
1114	// available.
1115	for (pic = ctx->pic_start; pic; pic = pic->next) {
1116	if (pic->encode_issued)
1117	continue;
1118	if (pic->type != FF_HW_PICTURE_TYPE_B)
1119	continue;
1120	for (i = 0; i < pic->nb_refs[0]; i++) {
1121	if (!pic->refs[0][i]->encode_issued)
1122	break;
1123	}
1124	if (i != pic->nb_refs[0])
1125	continue;
1126
1127	for (i = 0; i < pic->nb_refs[1]; i++) {
1128	if (!pic->refs[1][i]->encode_issued)
1129	break;
1130	}
1131	if (i == pic->nb_refs[1])
1132	break;
1133	}
1134
1135	if (pic) {
1136	av_log(avctx, AV_LOG_DEBUG, "Pick B-picture at depth %d to "
1137	"encode next.\n", pic->b_depth);
1138	*pic_out = pic;
1139	return 0;
1140	}
1141
1142	// Find the B-per-Pth available picture to become the next picture
1143	// on the top layer.
1144	start = NULL;
1145	b_counter = 0;
1146	closed_gop_end = ctx->closed_gop \|\|
1147	ctx->idr_counter == ctx->gop_per_idr;
1148	for (pic = ctx->pic_start; pic; pic = next) {
1149	next = pic->next;
1150	if (pic->encode_issued) {
1151	start = pic;
1152	continue;
1153	}
1154	// If the next available picture is force-IDR, encode it to start
1155	// a new GOP immediately.
1156	if (pic->force_idr)
1157	break;
1158	if (b_counter == ctx->b_per_p)
1159	break;
1160	// If this picture ends a closed GOP or starts a new GOP then it
1161	// needs to be in the top layer.
1162	if (ctx->gop_counter + b_counter + closed_gop_end >= ctx->gop_size)
1163	break;
1164	// If the picture after this one is force-IDR, we need to encode
1165	// this one in the top layer.
1166	if (next && next->force_idr)
1167	break;
1168	++b_counter;
1169	}
1170
1171	// At the end of the stream the last picture must be in the top layer.
1172	if (!pic && ctx->end_of_stream) {
1173	--b_counter;
1174	pic = ctx->pic_end;
1175	if (pic->encode_complete)
1176	return AVERROR_EOF;
1177	else if (pic->encode_issued)
1178	return AVERROR(EAGAIN);
1179	}
1180
1181	if (!pic) {
1182	av_log(avctx, AV_LOG_DEBUG, "Pick nothing to encode next - "
1183	"need more input for reference pictures.\n");
1184	return AVERROR(EAGAIN);
1185	}
1186	if (ctx->input_order <= ctx->decode_delay && !ctx->end_of_stream) {
1187	av_log(avctx, AV_LOG_DEBUG, "Pick nothing to encode next - "
1188	"need more input for timestamps.\n");
1189	return AVERROR(EAGAIN);
1190	}
1191
1192	if (pic->force_idr) {
1193	av_log(avctx, AV_LOG_DEBUG, "Pick forced IDR-picture to "
1194	"encode next.\n");
1195	pic->type = FF_HW_PICTURE_TYPE_IDR;
1196	ctx->idr_counter = 1;
1197	ctx->gop_counter = 1;
1198
1199	} else if (ctx->gop_counter + b_counter >= ctx->gop_size) {
1200	if (ctx->idr_counter == ctx->gop_per_idr) {
1201	av_log(avctx, AV_LOG_DEBUG, "Pick new-GOP IDR-picture to "
1202	"encode next.\n");
1203	pic->type = FF_HW_PICTURE_TYPE_IDR;
1204	ctx->idr_counter = 1;
1205	} else {
1206	av_log(avctx, AV_LOG_DEBUG, "Pick new-GOP I-picture to "
1207	"encode next.\n");
1208	pic->type = FF_HW_PICTURE_TYPE_I;
1209	++ctx->idr_counter;
1210	}
1211	ctx->gop_counter = 1;
1212
1213	} else {
1214	if (ctx->gop_counter + b_counter + closed_gop_end == ctx->gop_size) {
1215	av_log(avctx, AV_LOG_DEBUG, "Pick group-end P-picture to "
1216	"encode next.\n");
1217	} else {
1218	av_log(avctx, AV_LOG_DEBUG, "Pick normal P-picture to "
1219	"encode next.\n");
1220	}
1221	pic->type = FF_HW_PICTURE_TYPE_P;
1222	av_assert0(start);
1223	ctx->gop_counter += 1 + b_counter;
1224	}
1225	pic->is_reference = 1;
1226	*pic_out = pic;
1227
1228	vaapi_encode_add_ref(avctx, pic, pic, 0, 1, 0);
1229	if (pic->type != FF_HW_PICTURE_TYPE_IDR) {
1230	// TODO: apply both previous and forward multi reference for all vaapi encoders.
1231	// And L0/L1 reference frame number can be set dynamically through query
1232	// VAConfigAttribEncMaxRefFrames attribute.
1233	if (avctx->codec_id == AV_CODEC_ID_AV1) {
1234	for (i = 0; i < ctx->nb_next_prev; i++)
1235	vaapi_encode_add_ref(avctx, pic, ctx->next_prev[i],
1236	pic->type == FF_HW_PICTURE_TYPE_P,
1237	b_counter > 0, 0);
1238	} else
1239	vaapi_encode_add_ref(avctx, pic, start,
1240	pic->type == FF_HW_PICTURE_TYPE_P,
1241	b_counter > 0, 0);
1242
1243	vaapi_encode_add_ref(avctx, pic, ctx->next_prev[ctx->nb_next_prev - 1], 0, 0, 1);
1244	}
1245
1246	if (b_counter > 0) {
1247	vaapi_encode_set_b_pictures(avctx, start, pic, pic, 1,
1248	&prev);
1249	} else {
1250	prev = pic;
1251	}
1252	vaapi_encode_add_next_prev(avctx, prev);
1253
1254	return 0;
1255	}
1256
1257	static int vaapi_encode_clear_old(AVCodecContext *avctx)
1258	{
1259	VAAPIEncodeContext *ctx = avctx->priv_data;
1260	VAAPIEncodePicture pic, prev, *next;
1261
1262	av_assert0(ctx->pic_start);
1263
1264	// Remove direct references once each picture is complete.
1265	for (pic = ctx->pic_start; pic; pic = pic->next) {
1266	if (pic->encode_complete && pic->next)
1267	vaapi_encode_remove_refs(avctx, pic, 0);
1268	}
1269
1270	// Remove indirect references once a picture has no direct references.
1271	for (pic = ctx->pic_start; pic; pic = pic->next) {
1272	if (pic->encode_complete && pic->ref_count[0] == 0)
1273	vaapi_encode_remove_refs(avctx, pic, 1);
1274	}
1275
1276	// Clear out all complete pictures with no remaining references.
1277	prev = NULL;
1278	for (pic = ctx->pic_start; pic; pic = next) {
1279	next = pic->next;
1280	if (pic->encode_complete && pic->ref_count[1] == 0) {
1281	av_assert0(pic->ref_removed[0] && pic->ref_removed[1]);
1282	if (prev)
1283	prev->next = next;
1284	else
1285	ctx->pic_start = next;
1286	vaapi_encode_free(avctx, pic);
1287	} else {
1288	prev = pic;
1289	}
1290	}
1291
1292	return 0;
1293	}
1294
1295	static int vaapi_encode_check_frame(AVCodecContext *avctx,
1296	const AVFrame *frame)
1297	{
1298	VAAPIEncodeContext *ctx = avctx->priv_data;
1299
1300	if ((frame->crop_top \|\| frame->crop_bottom \|\|
1301	frame->crop_left \|\| frame->crop_right) && !ctx->crop_warned) {
1302	av_log(avctx, AV_LOG_WARNING, "Cropping information on input "
1303	"frames ignored due to lack of API support.\n");
1304	ctx->crop_warned = 1;
1305	}
1306
1307	if (!ctx->roi_allowed) {
1308	AVFrameSideData *sd =
1309	av_frame_get_side_data(frame, AV_FRAME_DATA_REGIONS_OF_INTEREST);
1310
1311	if (sd && !ctx->roi_warned) {
1312	av_log(avctx, AV_LOG_WARNING, "ROI side data on input "
1313	"frames ignored due to lack of driver support.\n");
1314	ctx->roi_warned = 1;
1315	}
1316	}
1317
1318	return 0;
1319	}
1320
1321	static int vaapi_encode_send_frame(AVCodecContext avctx, AVFrame frame)
1322	{
1323	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;
1324	VAAPIEncodeContext *ctx = avctx->priv_data;
1325	VAAPIEncodePicture *pic;
1326	int err;
1327
1328	if (frame) {
1329	av_log(avctx, AV_LOG_DEBUG, "Input frame: %ux%u (%"PRId64").\n",
1330	frame->width, frame->height, frame->pts);
1331
1332	err = vaapi_encode_check_frame(avctx, frame);
1333	if (err < 0)
1334	return err;
1335
1336	pic = vaapi_encode_alloc(avctx, frame);
1337	if (!pic)
1338	return AVERROR(ENOMEM);
1339
1340	pic->input_image = av_frame_alloc();
1341	if (!pic->input_image) {
1342	err = AVERROR(ENOMEM);
1343	goto fail;
1344	}
1345
1346	if (ctx->input_order == 0 \|\| frame->pict_type == AV_PICTURE_TYPE_I)
1347	pic->force_idr = 1;
1348
1349	pic->pts = frame->pts;
1350	pic->duration = frame->duration;
1351
1352	if (avctx->flags & AV_CODEC_FLAG_COPY_OPAQUE) {
1353	err = av_buffer_replace(&pic->opaque_ref, frame->opaque_ref);
1354	if (err < 0)
1355	goto fail;
1356
1357	pic->opaque = frame->opaque;
1358	}
1359
1360	av_frame_move_ref(pic->input_image, frame);
1361
1362	if (ctx->input_order == 0)
1363	ctx->first_pts = pic->pts;
1364	if (ctx->input_order == ctx->decode_delay)
1365	ctx->dts_pts_diff = pic->pts - ctx->first_pts;
1366	if (ctx->output_delay > 0)
1367	ctx->ts_ring[ctx->input_order %
1368	(3 * ctx->output_delay + base_ctx->async_depth)] = pic->pts;
1369
1370	pic->display_order = ctx->input_order;
1371	++ctx->input_order;
1372
1373	if (ctx->pic_start) {
1374	ctx->pic_end->next = pic;
1375	ctx->pic_end = pic;
1376	} else {
1377	ctx->pic_start = pic;
1378	ctx->pic_end = pic;
1379	}
1380
1381	} else {
1382	ctx->end_of_stream = 1;
1383
1384	// Fix timestamps if we hit end-of-stream before the initial decode
1385	// delay has elapsed.
1386	if (ctx->input_order < ctx->decode_delay)
1387	ctx->dts_pts_diff = ctx->pic_end->pts - ctx->first_pts;
1388	}
1389
1390	return 0;
1391
1392	fail:
1393	vaapi_encode_free(avctx, pic);
1394	return err;
1395	}
1396
1397	int ff_vaapi_encode_receive_packet(AVCodecContext avctx, AVPacket pkt)
1398	{
1399	VAAPIEncodeContext *ctx = avctx->priv_data;
1400	VAAPIEncodePicture *pic = NULL;
1401	AVFrame *frame = ctx->frame;
1402	int err;
1403
1404	start:
1405	/** if no B frame before repeat P frame, sent repeat P frame out. */
1406	if (ctx->tail_pkt->size) {
1407	for (VAAPIEncodePicture *tmp = ctx->pic_start; tmp; tmp = tmp->next) {
1408	if (tmp->type == FF_HW_PICTURE_TYPE_B && tmp->pts < ctx->tail_pkt->pts)
1409	break;
1410	else if (!tmp->next) {
1411	av_packet_move_ref(pkt, ctx->tail_pkt);
1412	goto end;
1413	}
1414	}
1415	}
1416
1417	err = ff_encode_get_frame(avctx, frame);
1418	if (err < 0 && err != AVERROR_EOF)
1419	return err;
1420
1421	if (err == AVERROR_EOF)
1422	frame = NULL;
1423
1424	err = vaapi_encode_send_frame(avctx, frame);
1425	if (err < 0)
1426	return err;
1427
1428	if (!ctx->pic_start) {
1429	if (ctx->end_of_stream)
1430	return AVERROR_EOF;
1431	else
1432	return AVERROR(EAGAIN);
1433	}
1434
1435	if (ctx->has_sync_buffer_func) {
1436	if (av_fifo_can_write(ctx->encode_fifo)) {
1437	err = vaapi_encode_pick_next(avctx, &pic);
1438	if (!err) {
1439	av_assert0(pic);
1440	pic->encode_order = ctx->encode_order +
1441	av_fifo_can_read(ctx->encode_fifo);
1442	err = vaapi_encode_issue(avctx, pic);
1443	if (err < 0) {
1444	av_log(avctx, AV_LOG_ERROR, "Encode failed: %d.\n", err);
1445	return err;
1446	}
1447	av_fifo_write(ctx->encode_fifo, &pic, 1);
1448	}
1449	}
1450
1451	if (!av_fifo_can_read(ctx->encode_fifo))
1452	return err;
1453
1454	// More frames can be buffered
1455	if (av_fifo_can_write(ctx->encode_fifo) && !ctx->end_of_stream)
1456	return AVERROR(EAGAIN);
1457
1458	av_fifo_read(ctx->encode_fifo, &pic, 1);
1459	ctx->encode_order = pic->encode_order + 1;
1460	} else {
1461	err = vaapi_encode_pick_next(avctx, &pic);
1462	if (err < 0)
1463	return err;
1464	av_assert0(pic);
1465
1466	pic->encode_order = ctx->encode_order++;
1467
1468	err = vaapi_encode_issue(avctx, pic);
1469	if (err < 0) {
1470	av_log(avctx, AV_LOG_ERROR, "Encode failed: %d.\n", err);
1471	return err;
1472	}
1473	}
1474
1475	err = vaapi_encode_output(avctx, pic, pkt);
1476	if (err < 0) {
1477	av_log(avctx, AV_LOG_ERROR, "Output failed: %d.\n", err);
1478	return err;
1479	}
1480
1481	ctx->output_order = pic->encode_order;
1482	vaapi_encode_clear_old(avctx);
1483
1484	/** loop to get an available pkt in encoder flushing. */
1485	if (ctx->end_of_stream && !pkt->size)
1486	goto start;
1487
1488	end:
1489	if (pkt->size)
1490	av_log(avctx, AV_LOG_DEBUG, "Output packet: pts %"PRId64", dts %"PRId64", "
1491	"size %d bytes.\n", pkt->pts, pkt->dts, pkt->size);
1492
1493	return 0;
1494	}
1495
1496	static av_cold void vaapi_encode_add_global_param(AVCodecContext *avctx, int type,	940	static av_cold void vaapi_encode_add_global_param(AVCodecContext *avctx, int type,
1497	void *buffer, size_t size)	941	void *buffer, size_t size)
1498	{	942	{
@@ -2188,7 +1632,8 @@ static av_cold int vaapi_encode_init_max_frame_size(AVCodecContext *avctx) · 689 unmodified lines
2188		1632
2189	static av_cold int vaapi_encode_init_gop_structure(AVCodecContext *avctx)	1633	static av_cold int vaapi_encode_init_gop_structure(AVCodecContext *avctx)
2190	{	1634	{
2191	VAAPIEncodeContext *ctx = avctx->priv_data;	1635	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;
		1636	VAAPIEncodeContext *ctx = avctx->priv_data;
2192	VAStatus vas;	1637	VAStatus vas;
2193	VAConfigAttrib attr = { VAConfigAttribEncMaxRefFrames };	1638	VAConfigAttrib attr = { VAConfigAttribEncMaxRefFrames };
2194	uint32_t ref_l0, ref_l1;	1639	uint32_t ref_l0, ref_l1;
@@ -2211,7 +1656,7 @@ static av_cold int vaapi_encode_init_gop_structure(AVCodecContext *avctx) · 16 unmodified lines
2211	ref_l1 = attr.value >> 16 & 0xffff;	1656	ref_l1 = attr.value >> 16 & 0xffff;
2212	}	1657	}
2213		1658
2214	ctx->p_to_gpb = 0;	1659	base_ctx->p_to_gpb = 0;
2215	prediction_pre_only = 0;	1660	prediction_pre_only = 0;
2216		1661
2217	#if VA_CHECK_VERSION(1, 9, 0)	1662	#if VA_CHECK_VERSION(1, 9, 0)
@@ -2247,7 +1692,7 @@ static av_cold int vaapi_encode_init_gop_structure(AVCodecContext *avctx) · 29 unmodified lines
2247		1692
2248	if (attr.value & VA_PREDICTION_DIRECTION_BI_NOT_EMPTY) {	1693	if (attr.value & VA_PREDICTION_DIRECTION_BI_NOT_EMPTY) {
2249	if (ref_l0 > 0 && ref_l1 > 0) {	1694	if (ref_l0 > 0 && ref_l1 > 0) {
2250	ctx->p_to_gpb = 1;	1695	base_ctx->p_to_gpb = 1;
2251	av_log(avctx, AV_LOG_VERBOSE, "Driver does not support P-frames, "	1696	av_log(avctx, AV_LOG_VERBOSE, "Driver does not support P-frames, "
2252	"replacing them with B-frames.\n");	1697	"replacing them with B-frames.\n");
2253	}	1698	}
@@ -2259,7 +1704,7 @@ static av_cold int vaapi_encode_init_gop_structure(AVCodecContext *avctx) · 5 unmodified lines
2259	if (ctx->codec->flags & FF_HW_FLAG_INTRA_ONLY \|\|	1704	if (ctx->codec->flags & FF_HW_FLAG_INTRA_ONLY \|\|
2260	avctx->gop_size <= 1) {	1705	avctx->gop_size <= 1) {
2261	av_log(avctx, AV_LOG_VERBOSE, "Using intra frames only.\n");	1706	av_log(avctx, AV_LOG_VERBOSE, "Using intra frames only.\n");
2262	ctx->gop_size = 1;	1707	base_ctx->gop_size = 1;
2263	} else if (ref_l0 < 1) {	1708	} else if (ref_l0 < 1) {
2264	av_log(avctx, AV_LOG_ERROR, "Driver does not support any "	1709	av_log(avctx, AV_LOG_ERROR, "Driver does not support any "
2265	"reference frames.\n");	1710	"reference frames.\n");
@@ -2267,41 +1712,41 @@ static av_cold int vaapi_encode_init_gop_structure(AVCodecContext *avctx) · 1 unmodified line
2267	} else if (!(ctx->codec->flags & FF_HW_FLAG_B_PICTURES) \|\|	1712	} else if (!(ctx->codec->flags & FF_HW_FLAG_B_PICTURES) \|\|
2268	ref_l1 < 1 \|\| avctx->max_b_frames < 1 \|\|	1713	ref_l1 < 1 \|\| avctx->max_b_frames < 1 \|\|
2269	prediction_pre_only) {	1714	prediction_pre_only) {
2270	if (ctx->p_to_gpb)	1715	if (base_ctx->p_to_gpb)
2271	av_log(avctx, AV_LOG_VERBOSE, "Using intra and B-frames "	1716	av_log(avctx, AV_LOG_VERBOSE, "Using intra and B-frames "
2272	"(supported references: %d / %d).\n",	1717	"(supported references: %d / %d).\n",
2273	ref_l0, ref_l1);	1718	ref_l0, ref_l1);
2274	else	1719	else
2275	av_log(avctx, AV_LOG_VERBOSE, "Using intra and P-frames "	1720	av_log(avctx, AV_LOG_VERBOSE, "Using intra and P-frames "
2276	"(supported references: %d / %d).\n", ref_l0, ref_l1);	1721	"(supported references: %d / %d).\n", ref_l0, ref_l1);
2277	ctx->gop_size = avctx->gop_size;	1722	base_ctx->gop_size = avctx->gop_size;
2278	ctx->p_per_i = INT_MAX;	1723	base_ctx->p_per_i = INT_MAX;
2279	ctx->b_per_p = 0;	1724	base_ctx->b_per_p = 0;
2280	} else {	1725	} else {
2281	if (ctx->p_to_gpb)	1726	if (base_ctx->p_to_gpb)
2282	av_log(avctx, AV_LOG_VERBOSE, "Using intra and B-frames "	1727	av_log(avctx, AV_LOG_VERBOSE, "Using intra and B-frames "
2283	"(supported references: %d / %d).\n",	1728	"(supported references: %d / %d).\n",
2284	ref_l0, ref_l1);	1729	ref_l0, ref_l1);
2285	else	1730	else
2286	av_log(avctx, AV_LOG_VERBOSE, "Using intra, P- and B-frames "	1731	av_log(avctx, AV_LOG_VERBOSE, "Using intra, P- and B-frames "
2287	"(supported references: %d / %d).\n", ref_l0, ref_l1);	1732	"(supported references: %d / %d).\n", ref_l0, ref_l1);
2288	ctx->gop_size = avctx->gop_size;	1733	base_ctx->gop_size = avctx->gop_size;
2289	ctx->p_per_i = INT_MAX;	1734	base_ctx->p_per_i = INT_MAX;
2290	ctx->b_per_p = avctx->max_b_frames;	1735	base_ctx->b_per_p = avctx->max_b_frames;
2291	if (ctx->codec->flags & FF_HW_FLAG_B_PICTURE_REFERENCES) {	1736	if (ctx->codec->flags & FF_HW_FLAG_B_PICTURE_REFERENCES) {
2292	ctx->max_b_depth = FFMIN(ctx->desired_b_depth,	1737	base_ctx->max_b_depth = FFMIN(ctx->desired_b_depth,
2293	av_log2(ctx->b_per_p) + 1);	1738	av_log2(base_ctx->b_per_p) + 1);
2294	} else {	1739	} else {
2295	ctx->max_b_depth = 1;	1740	base_ctx->max_b_depth = 1;
2296	}	1741	}
2297	}	1742	}
2298		1743
2299	if (ctx->codec->flags & FF_HW_FLAG_NON_IDR_KEY_PICTURES) {	1744	if (ctx->codec->flags & FF_HW_FLAG_NON_IDR_KEY_PICTURES) {
2300	ctx->closed_gop = !!(avctx->flags & AV_CODEC_FLAG_CLOSED_GOP);	1745	base_ctx->closed_gop = !!(avctx->flags & AV_CODEC_FLAG_CLOSED_GOP);
2301	ctx->gop_per_idr = ctx->idr_interval + 1;	1746	base_ctx->gop_per_idr = ctx->idr_interval + 1;
2302	} else {	1747	} else {
2303	ctx->closed_gop = 1;	1748	base_ctx->closed_gop = 1;
2304	ctx->gop_per_idr = 1;	1749	base_ctx->gop_per_idr = 1;
2305	}	1750	}
2306		1751
2307	return 0;	1752	return 0;
@@ -2614,7 +2059,8 @@ static av_cold int vaapi_encode_init_quality(AVCodecContext *avctx) · 306 unmodified lines
2614	static av_cold int vaapi_encode_init_roi(AVCodecContext *avctx)	2059	static av_cold int vaapi_encode_init_roi(AVCodecContext *avctx)
2615	{	2060	{
2616	#if VA_CHECK_VERSION(1, 0, 0)	2061	#if VA_CHECK_VERSION(1, 0, 0)
2617	VAAPIEncodeContext *ctx = avctx->priv_data;	2062	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;
		2063	VAAPIEncodeContext *ctx = avctx->priv_data;
2618	VAStatus vas;	2064	VAStatus vas;
2619	VAConfigAttrib attr = { VAConfigAttribEncROI };	2065	VAConfigAttrib attr = { VAConfigAttribEncROI };
2620		2066
@@ -2629,14 +2075,14 @@ static av_cold int vaapi_encode_init_roi(AVCodecContext *avctx) · 8 unmodified lines
2629	}	2075	}
2630		2076
2631	if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {	2077	if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
2632	ctx->roi_allowed = 0;	2078	base_ctx->roi_allowed = 0;
2633	} else {	2079	} else {
2634	VAConfigAttribValEncROI roi = {	2080	VAConfigAttribValEncROI roi = {
2635	.value = attr.value,	2081	.value = attr.value,
2636	};	2082	};
2637		2083
2638	ctx->roi_max_regions = roi.bits.num_roi_regions;	2084	ctx->roi_max_regions = roi.bits.num_roi_regions;
2639	ctx->roi_allowed = ctx->roi_max_regions > 0 &&	2085	base_ctx->roi_allowed = ctx->roi_max_regions > 0 &&
2640	(ctx->va_rc_mode == VA_RC_CQP \|\|	2086	(ctx->va_rc_mode == VA_RC_CQP \|\|
2641	roi.bits.roi_rc_qp_delta_support);	2087	roi.bits.roi_rc_qp_delta_support);
2642	}	2088	}
@@ -2771,6 +2217,16 @@ static av_cold int vaapi_encode_create_recon_frames(AVCodecContext *avctx) · 128 unmodified lines
2771	return err;	2217	return err;
2772	}	2218	}
2773		2219
		2220	static const FFHWEncodePictureOperation vaapi_op = {
		2221	.alloc = &vaapi_encode_alloc,
		2222
		2223	.issue = &vaapi_encode_issue,
		2224
		2225	.output = &vaapi_encode_output,
		2226
		2227	.free = &vaapi_encode_free,
		2228	};
		2229
2774	av_cold int ff_vaapi_encode_init(AVCodecContext *avctx)	2230	av_cold int ff_vaapi_encode_init(AVCodecContext *avctx)
2775	{	2231	{
2776	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;	2232	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;
@@ -2782,10 +2238,12 @@ av_cold int ff_vaapi_encode_init(AVCodecContext *avctx) · 5 unmodified lines
2782	ctx->va_config = VA_INVALID_ID;	2238	ctx->va_config = VA_INVALID_ID;
2783	ctx->va_context = VA_INVALID_ID;	2239	ctx->va_context = VA_INVALID_ID;
2784		2240
		2241	base_ctx->op = &vaapi_op;
		2242
2785	/* If you add something that can fail above this av_frame_alloc(),	2243	/* If you add something that can fail above this av_frame_alloc(),
2786	* modify ff_vaapi_encode_close() accordingly. */	2244	* modify ff_vaapi_encode_close() accordingly. */
2787	ctx->frame = av_frame_alloc();	2245	base_ctx->frame = av_frame_alloc();
2788	if (!ctx->frame) {	2246	if (!base_ctx->frame) {
2789	return AVERROR(ENOMEM);	2247	return AVERROR(ENOMEM);
2790	}	2248	}
2791		2249
@@ -2810,8 +2268,8 @@ av_cold int ff_vaapi_encode_init(AVCodecContext *avctx) · 18 unmodified lines
2810	ctx->device = (AVHWDeviceContext*)ctx->device_ref->data;	2268	ctx->device = (AVHWDeviceContext*)ctx->device_ref->data;
2811	ctx->hwctx = ctx->device->hwctx;	2269	ctx->hwctx = ctx->device->hwctx;
2812		2270
2813	ctx->tail_pkt = av_packet_alloc();	2271	base_ctx->tail_pkt = av_packet_alloc();
2814	if (!ctx->tail_pkt) {	2272	if (!base_ctx->tail_pkt) {
2815	err = AVERROR(ENOMEM);	2273	err = AVERROR(ENOMEM);
2816	goto fail;	2274	goto fail;
2817	}	2275	}
@@ -2910,8 +2368,8 @@ av_cold int ff_vaapi_encode_init(AVCodecContext *avctx) · 92 unmodified lines
2910	goto fail;	2368	goto fail;
2911	}	2369	}
2912		2370
2913	ctx->output_delay = ctx->b_per_p;	2371	base_ctx->output_delay = base_ctx->b_per_p;
2914	ctx->decode_delay = ctx->max_b_depth;	2372	base_ctx->decode_delay = base_ctx->max_b_depth;
2915		2373
2916	if (ctx->codec->sequence_params_size > 0) {	2374	if (ctx->codec->sequence_params_size > 0) {
2917	ctx->codec_sequence_params =	2375	ctx->codec_sequence_params =
@@ -2966,11 +2424,11 @@ av_cold int ff_vaapi_encode_init(AVCodecContext *avctx) · 48 unmodified lines
2966	// check vaSyncBuffer function	2424	// check vaSyncBuffer function
2967	vas = vaSyncBuffer(ctx->hwctx->display, VA_INVALID_ID, 0);	2425	vas = vaSyncBuffer(ctx->hwctx->display, VA_INVALID_ID, 0);
2968	if (vas != VA_STATUS_ERROR_UNIMPLEMENTED) {	2426	if (vas != VA_STATUS_ERROR_UNIMPLEMENTED) {
2969	ctx->has_sync_buffer_func = 1;	2427	base_ctx->async_encode = 1;
2970	ctx->encode_fifo = av_fifo_alloc2(base_ctx->async_depth,	2428	base_ctx->encode_fifo = av_fifo_alloc2(base_ctx->async_depth,
2971	sizeof(VAAPIEncodePicture *),	2429	sizeof(VAAPIEncodePicture*),
2972	0);	2430	0);
2973	if (!ctx->encode_fifo)	2431	if (!base_ctx->encode_fifo)
2974	return AVERROR(ENOMEM);	2432	return AVERROR(ENOMEM);
2975	}	2433	}
2976	#endif	2434	#endif
@@ -2983,15 +2441,16 @@ fail: · 6 unmodified lines
2983		2441
2984	av_cold int ff_vaapi_encode_close(AVCodecContext *avctx)	2442	av_cold int ff_vaapi_encode_close(AVCodecContext *avctx)
2985	{	2443	{
2986	VAAPIEncodeContext *ctx = avctx->priv_data;	2444	FFHWBaseEncodeContext *base_ctx = avctx->priv_data;
2987	VAAPIEncodePicture pic, next;	2445	VAAPIEncodeContext *ctx = avctx->priv_data;
		2446	FFHWBaseEncodePicture pic, next;
2988		2447
2989	/* We check ctx->frame to know whether ff_vaapi_encode_init()	2448	/* We check ctx->frame to know whether ff_vaapi_encode_init()
2990	* has been called and va_config/va_context initialized. */	2449	* has been called and va_config/va_context initialized. */
2991	if (!ctx->frame)	2450	if (!base_ctx->frame)
2992	return 0;	2451	return 0;
2993		2452
2994	for (pic = ctx->pic_start; pic; pic = next) {	2453	for (pic = base_ctx->pic_start; pic; pic = next) {
2995	next = pic->next;	2454	next = pic->next;
2996	vaapi_encode_free(avctx, pic);	2455	vaapi_encode_free(avctx, pic);
2997	}	2456	}
@@ -3008,12 +2467,12 @@ av_cold int ff_vaapi_encode_close(AVCodecContext *avctx) · 10 unmodified lines
3008	ctx->va_config = VA_INVALID_ID;	2467	ctx->va_config = VA_INVALID_ID;
3009	}	2468	}
3010		2469
3011	av_frame_free(&ctx->frame);	2470	av_frame_free(&base_ctx->frame);
3012	av_packet_free(&ctx->tail_pkt);	2471	av_packet_free(&base_ctx->tail_pkt);
3013		2472
3014	av_freep(&ctx->codec_sequence_params);	2473	av_freep(&ctx->codec_sequence_params);
3015	av_freep(&ctx->codec_picture_params);	2474	av_freep(&ctx->codec_picture_params);
3016	av_fifo_freep2(&ctx->encode_fifo);	2475	av_fifo_freep2(&base_ctx->encode_fifo);
3017		2476
3018	av_buffer_unref(&ctx->recon_frames_ref);	2477	av_buffer_unref(&ctx->recon_frames_ref);
3019	av_buffer_unref(&ctx->input_frames_ref);	2478	av_buffer_unref(&ctx->input_frames_ref);

libavcodec/vaapi_encode.h+1−101

@@ -29,7 +29,6 @@ · 28 unmodified lines
29		29
30	#include "libavutil/hwcontext.h"	30	#include "libavutil/hwcontext.h"
31	#include "libavutil/hwcontext_vaapi.h"	31	#include "libavutil/hwcontext_vaapi.h"
32	#include "libavutil/fifo.h"
33		32
34	#include "avcodec.h"	33	#include "avcodec.h"
35	#include "hwconfig.h"	34	#include "hwconfig.h"
@@ -64,16 +63,7 @@ typedef struct VAAPIEncodeSlice { · 28 unmodified lines
64	} VAAPIEncodeSlice;	63	} VAAPIEncodeSlice;
65		64
66	typedef struct VAAPIEncodePicture {	65	typedef struct VAAPIEncodePicture {
67	struct VAAPIEncodePicture *next;	66	FFHWBaseEncodePicture base;
68
69	int64_t display_order;
70	int64_t encode_order;
71	int64_t pts;
72	int64_t duration;
73	int force_idr;
74
75	void *opaque;
76	AVBufferRef *opaque_ref;
77		67
78	#if VA_CHECK_VERSION(1, 0, 0)	68	#if VA_CHECK_VERSION(1, 0, 0)
79	// ROI regions.	69	// ROI regions.
@@ -82,15 +72,7 @@ typedef struct VAAPIEncodePicture { · 2 unmodified lines
82	void *roi;	72	void *roi;
83	#endif	73	#endif
84		74
85	int type;
86	int b_depth;
87	int encode_issued;
88	int encode_complete;
89
90	AVFrame *input_image;
91	VASurfaceID input_surface;	75	VASurfaceID input_surface;
92
93	AVFrame *recon_image;
94	VASurfaceID recon_surface;	76	VASurfaceID recon_surface;
95		77
96	int nb_param_buffers;	78	int nb_param_buffers;
@@ -100,31 +82,8 @@ typedef struct VAAPIEncodePicture { · 3 unmodified lines
100	VABufferID *output_buffer_ref;	82	VABufferID *output_buffer_ref;
101	VABufferID output_buffer;	83	VABufferID output_buffer;
102		84
103	void *priv_data;
104	void *codec_picture_params;	85	void *codec_picture_params;
105		86
106	// Whether this picture is a reference picture.
107	int is_reference;
108
109	// The contents of the DPB after this picture has been decoded.
110	// This will contain the picture itself if it is a reference picture,
111	// but not if it isn't.
112	int nb_dpb_pics;
113	struct VAAPIEncodePicture *dpb[MAX_DPB_SIZE];
114	// The reference pictures used in decoding this picture. If they are
115	// used by later pictures they will also appear in the DPB. ref[0][] for
116	// previous reference frames. ref[1][] for future reference frames.
117	int nb_refs[MAX_REFERENCE_LIST_NUM];
118	struct VAAPIEncodePicture *refs[MAX_REFERENCE_LIST_NUM][MAX_PICTURE_REFERENCES];
119	// The previous reference picture in encode order. Must be in at least
120	// one of the reference list and DPB list.
121	struct VAAPIEncodePicture *prev;
122	// Reference count for other pictures referring to this one through
123	// the above pointers, directly from incomplete pictures and indirectly
124	// through completed pictures.
125	int ref_count[2];
126	int ref_removed[2];
127
128	int nb_slices;	87	int nb_slices;
129	VAAPIEncodeSlice *slices;	88	VAAPIEncodeSlice *slices;
130		89
@@ -298,30 +257,6 @@ typedef struct VAAPIEncodeContext { · 167 unmodified lines
298	// structure (VAEncPictureParameterBuffer*).	257	// structure (VAEncPictureParameterBuffer*).
299	void *codec_picture_params;	258	void *codec_picture_params;
300		259
301	// Current encoding window, in display (input) order.
302	VAAPIEncodePicture pic_start, pic_end;
303	// The next picture to use as the previous reference picture in
304	// encoding order. Order from small to large in encoding order.
305	VAAPIEncodePicture *next_prev[MAX_PICTURE_REFERENCES];
306	int nb_next_prev;
307
308	// Next input order index (display order).
309	int64_t input_order;
310	// Number of frames that output is behind input.
311	int64_t output_delay;
312	// Next encode order index.
313	int64_t encode_order;
314	// Number of frames decode output will need to be delayed.
315	int64_t decode_delay;
316	// Next output order index (in encode order).
317	int64_t output_order;
318
319	// Timestamp handling.
320	int64_t first_pts;
321	int64_t dts_pts_diff;
322	int64_t ts_ring[MAX_REORDER_DELAY * 3 +
323	MAX_ASYNC_DEPTH];
324
325	// Slice structure.	260	// Slice structure.
326	int slice_block_rows;	261	int slice_block_rows;
327	int slice_block_cols;	262	int slice_block_cols;
@@ -340,41 +275,12 @@ typedef struct VAAPIEncodeContext { · 12 unmodified lines
340	// Location of the i-th tile row boundary.	275	// Location of the i-th tile row boundary.
341	int row_bd[MAX_TILE_ROWS + 1];	276	int row_bd[MAX_TILE_ROWS + 1];
342		277
343	// Frame type decision.
344	int gop_size;
345	int closed_gop;
346	int gop_per_idr;
347	int p_per_i;
348	int max_b_depth;
349	int b_per_p;
350	int force_idr;
351	int idr_counter;
352	int gop_counter;
353	int end_of_stream;
354	int p_to_gpb;
355
356	// Whether the driver supports ROI at all.
357	int roi_allowed;
358	// Maximum number of regions supported by the driver.	278	// Maximum number of regions supported by the driver.
359	int roi_max_regions;	279	int roi_max_regions;
360	// Quantisation range for offset calculations. Set by codec-specific	280	// Quantisation range for offset calculations. Set by codec-specific
361	// code, as it may change based on parameters.	281	// code, as it may change based on parameters.
362	int roi_quant_range;	282	int roi_quant_range;
363		283
364	// The encoder does not support cropping information, so warn about
365	// it the first time we encounter any nonzero crop fields.
366	int crop_warned;
367	// If the driver does not support ROI then warn the first time we
368	// encounter a frame with ROI side data.
369	int roi_warned;
370
371	AVFrame *frame;
372
373	// Whether the driver support vaSyncBuffer
374	int has_sync_buffer_func;
375	// Store buffered pic
376	AVFifo *encode_fifo;
377
378	/** Head data for current output pkt, used only for AV1. */	284	/** Head data for current output pkt, used only for AV1. */
379	//void *header_data;	285	//void *header_data;
380	//size_t header_data_size;	286	//size_t header_data_size;
@@ -384,9 +290,6 @@ typedef struct VAAPIEncodeContext { · 3 unmodified lines
384	* This is a RefStruct reference.	290	* This is a RefStruct reference.
385	*/	291	*/
386	VABufferID *coded_buffer_ref;	292	VABufferID *coded_buffer_ref;
387
388	/** Tail data of a pic, now only used for av1 repeat frame header. */
389	AVPacket *tail_pkt;
390	} VAAPIEncodeContext;	293	} VAAPIEncodeContext;
391		294
392	typedef struct VAAPIEncodeType {	295	typedef struct VAAPIEncodeType {
@@ -468,9 +371,6 @@ typedef struct VAAPIEncodeType { · 75 unmodified lines
468	char data, size_t data_len);	371	char data, size_t data_len);
469	} VAAPIEncodeType;	372	} VAAPIEncodeType;
470		373
471
472	int ff_vaapi_encode_receive_packet(AVCodecContext avctx, AVPacket pkt);
473
474	int ff_vaapi_encode_init(AVCodecContext *avctx);	374	int ff_vaapi_encode_init(AVCodecContext *avctx);
475	int ff_vaapi_encode_close(AVCodecContext *avctx);	375	int ff_vaapi_encode_close(AVCodecContext *avctx);
476		376

libavcodec/vaapi_encode_av1.c+19−16

 static int vaapi_encode_av1_init_sequence_params(AVCodecContext *avctx)
 {
+    FFHWBaseEncodeContext       *base_ctx = avctx->priv_data;
     VAAPIEncodeContext               *ctx = avctx->priv_data;
     VAAPIEncodeAV1Context           *priv = avctx->priv_data;
     AV1RawOBU                     *sh_obu = &priv->sh;
     vseq->seq_level_idx           = sh->seq_level_idx[0];
     vseq->seq_tier                = sh->seq_tier[0];
     vseq->order_hint_bits_minus_1 = sh->order_hint_bits_minus_1;
-    vseq->intra_period            = ctx->gop_size;
+    vseq->intra_period            = base_ctx->gop_size;
-    vseq->ip_period               = ctx->b_per_p + 1;
+    vseq->ip_period               = base_ctx->b_per_p + 1;
     vseq->seq_fields.bits.enable_order_hint = sh->enable_order_hint;
 }
 static int vaapi_encode_av1_init_picture_params(AVCodecContext *avctx,
-                                                VAAPIEncodePicture *pic)
+                                                VAAPIEncodePicture *vaapi_pic)
 {
     VAAPIEncodeContext              *ctx = avctx->priv_data;
     VAAPIEncodeAV1Context          *priv = avctx->priv_data;
+    const FFHWBaseEncodePicture     *pic = &vaapi_pic->base;
     VAAPIEncodeAV1Picture          *hpic = pic->priv_data;
     AV1RawOBU                    *fh_obu = &priv->fh;
     AV1RawFrameHeader                *fh = &fh_obu->obu.frame.header;
-    VAEncPictureParameterBufferAV1 *vpic = pic->codec_picture_params;
+    VAEncPictureParameterBufferAV1 *vpic = vaapi_pic->codec_picture_params;
     CodedBitstreamFragment          *obu = &priv->current_obu;
-    VAAPIEncodePicture    *ref;
+    FFHWBaseEncodePicture *ref;
     VAAPIEncodeAV1Picture *href;
     int slot, i;
     int ret;
         { 1, 0, 0, 0, -1, 0, -1, -1 };
     memset(fh_obu, 0, sizeof(*fh_obu));
-    pic->nb_slices = priv->tile_groups;
+    vaapi_pic->nb_slices = priv->tile_groups;
-    pic->non_independent_frame = pic->encode_order < pic->display_order;
+    vaapi_pic->non_independent_frame = pic->encode_order < pic->display_order;
     fh_obu->header.obu_type = AV1_OBU_FRAME_HEADER;
     fh_obu->header.obu_has_size_field = 1;
     vpic->frame_width_minus_1  = fh->frame_width_minus_1;
     vpic->frame_height_minus_1 = fh->frame_height_minus_1;
     vpic->primary_ref_frame    = fh->primary_ref_frame;
-    vpic->reconstructed_frame  = pic->recon_surface;
+    vpic->reconstructed_frame  = vaapi_pic->recon_surface;
-    vpic->coded_buf            = pic->output_buffer;
+    vpic->coded_buf            = vaapi_pic->output_buffer;
     vpic->tile_cols            = fh->tile_cols;
     vpic->tile_rows            = fh->tile_rows;
     vpic->order_hint           = fh->order_hint;
     for (i = 0; i < MAX_REFERENCE_LIST_NUM; i++) {
         for (int j = 0; j < pic->nb_refs[i]; j++) {
-            VAAPIEncodePicture *ref_pic = pic->refs[i][j];
+            FFHWBaseEncodePicture *ref_pic = pic->refs[i][j];
             slot = ((VAAPIEncodeAV1Picture*)ref_pic->priv_data)->slot;
             av_assert0(vpic->reference_frames[slot] == VA_INVALID_SURFACE);
-            vpic->reference_frames[slot] = ref_pic->recon_surface;
+            vpic->reference_frames[slot] = ((VAAPIEncodePicture *)ref_pic)->recon_surface;
         }
     }
 }
 static int vaapi_encode_av1_write_picture_header(AVCodecContext *avctx,
-                                                 VAAPIEncodePicture *pic,
+                                                 VAAPIEncodePicture *vaapi_pic,
                                                  char *data, size_t *data_len)
 {
     VAAPIEncodeAV1Context     *priv = avctx->priv_data;
     CodedBitstreamAV1Context *cbctx = priv->cbc->priv_data;
     AV1RawOBU               *fh_obu = &priv->fh;
     AV1RawFrameHeader       *rep_fh = &fh_obu->obu.frame_header;
+    const FFHWBaseEncodePicture *pic = &vaapi_pic->base;
     VAAPIEncodeAV1Picture *href;
     int ret = 0;
-    pic->tail_size = 0;
+    vaapi_pic->tail_size = 0;
     /** Pack repeat frame header. */
     if (pic->display_order > pic->encode_order) {
         memset(fh_obu, 0, sizeof(*fh_obu));
         if (ret < 0)
             goto end;
-        ret = vaapi_encode_av1_write_obu(avctx, pic->tail_data, &pic->tail_size, obu);
+        ret = vaapi_encode_av1_write_obu(avctx, vaapi_pic->tail_data, &vaapi_pic->tail_size, obu);
         if (ret < 0)
             goto end;
-        pic->tail_size /= 8;
+        vaapi_pic->tail_size /= 8;
     }
     memcpy(data, &priv->fh_data, MAX_PARAM_BUFFER_SIZE * sizeof(char));
     .p.id           = AV_CODEC_ID_AV1,
     .priv_data_size = sizeof(VAAPIEncodeAV1Context),
     .init           = &vaapi_encode_av1_init,
-    FF_CODEC_RECEIVE_PACKET_CB(&ff_vaapi_encode_receive_packet),
+    FF_CODEC_RECEIVE_PACKET_CB(&ff_hw_base_encode_receive_packet),
     .close          = &vaapi_encode_av1_close,
     .p.priv_class   = &vaapi_encode_av1_class,
     .p.capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_HARDWARE |

libavcodec/vaapi_encode_h264.c+44−40

                 goto fail;
         }
         if (priv->sei_needed & SEI_TIMING) {
-            if (pic->type == FF_HW_PICTURE_TYPE_IDR) {
+            if (pic->base.type == FF_HW_PICTURE_TYPE_IDR) {
                 err = ff_cbs_sei_add_message(priv->cbc, au, 1,
                                              SEI_TYPE_BUFFERING_PERIOD,
                                              &priv->sei_buffering_period, NULL);
 static int vaapi_encode_h264_init_sequence_params(AVCodecContext *avctx)
 {
+    FFHWBaseEncodeContext        *base_ctx = avctx->priv_data;
     VAAPIEncodeContext                *ctx = avctx->priv_data;
     VAAPIEncodeH264Context           *priv = avctx->priv_data;
     H264RawSPS                        *sps = &priv->raw_sps;
         sps->constraint_set1_flag = 1;
     if (avctx->profile == AV_PROFILE_H264_HIGH || avctx->profile == AV_PROFILE_H264_HIGH_10)
-        sps->constraint_set3_flag = ctx->gop_size == 1;
+        sps->constraint_set3_flag = base_ctx->gop_size == 1;
     if (avctx->profile == AV_PROFILE_H264_MAIN ||
         avctx->profile == AV_PROFILE_H264_HIGH || avctx->profile == AV_PROFILE_H264_HIGH_10) {
         sps->constraint_set4_flag = 1;
-        sps->constraint_set5_flag = ctx->b_per_p == 0;
+        sps->constraint_set5_flag = base_ctx->b_per_p == 0;
     }
-    if (ctx->gop_size == 1)
+    if (base_ctx->gop_size == 1)
         priv->dpb_frames = 0;
     else
-        priv->dpb_frames = 1 + ctx->max_b_depth;
+        priv->dpb_frames = 1 + base_ctx->max_b_depth;
     if (avctx->level != AV_LEVEL_UNKNOWN) {
         sps->level_idc = avctx->level;
     sps->bit_depth_chroma_minus8 = bit_depth - 8;
     sps->log2_max_frame_num_minus4 = 4;
-    sps->pic_order_cnt_type        = ctx->max_b_depth ? 0 : 2;
+    sps->pic_order_cnt_type        = base_ctx->max_b_depth ? 0 : 2;
     if (sps->pic_order_cnt_type == 0) {
         sps->log2_max_pic_order_cnt_lsb_minus4 = 4;
     }
     sps->vui.motion_vectors_over_pic_boundaries_flag = 1;
     sps->vui.log2_max_mv_length_horizontal = 15;
     sps->vui.log2_max_mv_length_vertical   = 15;
-    sps->vui.max_num_reorder_frames        = ctx->max_b_depth;
+    sps->vui.max_num_reorder_frames        = base_ctx->max_b_depth;
-    sps->vui.max_dec_frame_buffering       = ctx->max_b_depth + 1;
+    sps->vui.max_dec_frame_buffering       = base_ctx->max_b_depth + 1;
     pps->nal_unit_header.nal_ref_idc = 3;
     pps->nal_unit_header.nal_unit_type = H264_NAL_PPS;
     *vseq = (VAEncSequenceParameterBufferH264) {
         .seq_parameter_set_id = sps->seq_parameter_set_id,
         .level_idc        = sps->level_idc,
-        .intra_period     = ctx->gop_size,
+        .intra_period     = base_ctx->gop_size,
-        .intra_idr_period = ctx->gop_size,
+        .intra_idr_period = base_ctx->gop_size,
-        .ip_period        = ctx->b_per_p + 1,
+        .ip_period        = base_ctx->b_per_p + 1,
         .bits_per_second       = ctx->va_bit_rate,
         .max_num_ref_frames    = sps->max_num_ref_frames,
 }
 static int vaapi_encode_h264_init_picture_params(AVCodecContext *avctx,
-                                                 VAAPIEncodePicture *pic)
+                                                 VAAPIEncodePicture *vaapi_pic)
 {
-    VAAPIEncodeContext               *ctx = avctx->priv_data;
+    FFHWBaseEncodeContext       *base_ctx = avctx->priv_data;
     VAAPIEncodeH264Context          *priv = avctx->priv_data;
+    const FFHWBaseEncodePicture      *pic = &vaapi_pic->base;
     VAAPIEncodeH264Picture          *hpic = pic->priv_data;
-    VAAPIEncodePicture              *prev = pic->prev;
+    FFHWBaseEncodePicture           *prev = pic->prev;
     VAAPIEncodeH264Picture         *hprev = prev ? prev->priv_data : NULL;
-    VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params;
+    VAEncPictureParameterBufferH264 *vpic = vaapi_pic->codec_picture_params;
     int i, j = 0;
     if (pic->type == FF_HW_PICTURE_TYPE_IDR) {
         hpic->pic_order_cnt *= 2;
     }
-    hpic->dpb_delay     = pic->display_order - pic->encode_order + ctx->max_b_depth;
+    hpic->dpb_delay     = pic->display_order - pic->encode_order + base_ctx->max_b_depth;
     hpic->cpb_delay     = pic->encode_order - hpic->last_idr_frame;
     if (priv->aud) {
         priv->sei_recovery_point = (H264RawSEIRecoveryPoint) {
             .recovery_frame_cnt = 0,
             .exact_match_flag   = 1,
-            .broken_link_flag   = ctx->b_per_p > 0,
+            .broken_link_flag   = base_ctx->b_per_p > 0,
         };
         priv->sei_needed |= SEI_RECOVERY_POINT;
     }
     vpic->CurrPic = (VAPictureH264) {
-        .picture_id          = pic->recon_surface,
+        .picture_id          = vaapi_pic->recon_surface,
         .frame_idx           = hpic->frame_num,
         .flags               = 0,
         .TopFieldOrderCnt    = hpic->pic_order_cnt,
     };
     for (int k = 0; k < MAX_REFERENCE_LIST_NUM; k++) {
         for (i = 0; i < pic->nb_refs[k]; i++) {
-            VAAPIEncodePicture      *ref = pic->refs[k][i];
+            FFHWBaseEncodePicture *ref = pic->refs[k][i];
             VAAPIEncodeH264Picture *href;
             av_assert0(ref && ref->encode_order < pic->encode_order);
             href = ref->priv_data;
             vpic->ReferenceFrames[j++] = (VAPictureH264) {
-                .picture_id          = ref->recon_surface,
+                .picture_id          = ((VAAPIEncodePicture *)ref)->recon_surface,
                 .frame_idx           = href->frame_num,
                 .flags               = VA_PICTURE_H264_SHORT_TERM_REFERENCE,
                 .TopFieldOrderCnt    = href->pic_order_cnt,
         };
     }
-    vpic->coded_buf = pic->output_buffer;
+    vpic->coded_buf = vaapi_pic->output_buffer;
     vpic->frame_num = hpic->frame_num;
 }
 static void vaapi_encode_h264_default_ref_pic_list(AVCodecContext *avctx,
-                                                   VAAPIEncodePicture *pic,
+                                                   VAAPIEncodePicture *vaapi_pic,
                                                    VAAPIEncodePicture **rpl0,
                                                    VAAPIEncodePicture **rpl1,
                                                    int *rpl_size)
 {
-    VAAPIEncodePicture *prev;
+    FFHWBaseEncodePicture *pic = &vaapi_pic->base;
+    FFHWBaseEncodePicture *prev;
     VAAPIEncodeH264Picture *hp, *hn, *hc;
     int i, j, n = 0;
         if (pic->type == FF_HW_PICTURE_TYPE_P) {
             for (j = n; j > 0; j--) {
-                hc = rpl0[j - 1]->priv_data;
+                hc = rpl0[j - 1]->base.priv_data;
                 av_assert0(hc->frame_num != hn->frame_num);
                 if (hc->frame_num > hn->frame_num)
                     break;
                 rpl0[j] = rpl0[j - 1];
             }
-            rpl0[j] = prev->dpb[i];
+            rpl0[j] = (VAAPIEncodePicture *)prev->dpb[i];
         } else if (pic->type == FF_HW_PICTURE_TYPE_B) {
             for (j = n; j > 0; j--) {
-                hc = rpl0[j - 1]->priv_data;
+                hc = rpl0[j - 1]->base.priv_data;
                 av_assert0(hc->pic_order_cnt != hp->pic_order_cnt);
                 if (hc->pic_order_cnt < hp->pic_order_cnt) {
                     if (hn->pic_order_cnt > hp->pic_order_cnt ||
                 }
                 rpl0[j] = rpl0[j - 1];
             }
-            rpl0[j] = prev->dpb[i];
+            rpl0[j] = (VAAPIEncodePicture *)prev->dpb[i];
             for (j = n; j > 0; j--) {
-                hc = rpl1[j - 1]->priv_data;
+                hc = rpl1[j - 1]->base.priv_data;
                 av_assert0(hc->pic_order_cnt != hp->pic_order_cnt);
                 if (hc->pic_order_cnt > hp->pic_order_cnt) {
                     if (hn->pic_order_cnt < hp->pic_order_cnt ||
                 }
                 rpl1[j] = rpl1[j - 1];
             }
-            rpl1[j] = prev->dpb[i];
+            rpl1[j] = (VAAPIEncodePicture *)prev->dpb[i];
         }
         ++n;
         av_log(avctx, AV_LOG_DEBUG, "Default RefPicList0 for fn=%d/poc=%d:",
                hp->frame_num, hp->pic_order_cnt);
         for (i = 0; i < n; i++) {
-            hn = rpl0[i]->priv_data;
+            hn = rpl0[i]->base.priv_data;
             av_log(avctx, AV_LOG_DEBUG, "  fn=%d/poc=%d",
                    hn->frame_num, hn->pic_order_cnt);
         }
         av_log(avctx, AV_LOG_DEBUG, "Default RefPicList1 for fn=%d/poc=%d:",
                hp->frame_num, hp->pic_order_cnt);
         for (i = 0; i < n; i++) {
-            hn = rpl1[i]->priv_data;
+            hn = rpl1[i]->base.priv_data;
             av_log(avctx, AV_LOG_DEBUG, "  fn=%d/poc=%d",
                    hn->frame_num, hn->pic_order_cnt);
         }
 }
 static int vaapi_encode_h264_init_slice_params(AVCodecContext *avctx,
-                                               VAAPIEncodePicture *pic,
+                                               VAAPIEncodePicture *vaapi_pic,
                                                VAAPIEncodeSlice *slice)
 {
     VAAPIEncodeH264Context          *priv = avctx->priv_data;
+    const FFHWBaseEncodePicture      *pic = &vaapi_pic->base;
     VAAPIEncodeH264Picture          *hpic = pic->priv_data;
-    VAAPIEncodePicture              *prev = pic->prev;
+    FFHWBaseEncodePicture           *prev = pic->prev;
     H264RawSPS                       *sps = &priv->raw_sps;
     H264RawPPS                       *pps = &priv->raw_pps;
     H264RawSliceHeader                *sh = &priv->raw_slice.header;
-    VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params;
+    VAEncPictureParameterBufferH264 *vpic = vaapi_pic->codec_picture_params;
     VAEncSliceParameterBufferH264 *vslice = slice->codec_slice_params;
     int i, j;
         sh->slice_qp_delta = priv->fixed_qp_idr - (pps->pic_init_qp_minus26 + 26);
     if (pic->is_reference && pic->type != FF_HW_PICTURE_TYPE_IDR) {
-        VAAPIEncodePicture *discard_list[MAX_DPB_SIZE];
+        FFHWBaseEncodePicture *discard_list[MAX_DPB_SIZE];
         int discard = 0, keep = 0;
         // Discard everything which is in the DPB of the previous frame but
         VAAPIEncodeH264Picture *href;
         int n;
-        vaapi_encode_h264_default_ref_pic_list(avctx, pic,
+        vaapi_encode_h264_default_ref_pic_list(avctx, vaapi_pic,
                                                def_l0, def_l1, &n);
         if (pic->type == FF_HW_PICTURE_TYPE_P) {
             int need_rplm = 0;
             for (i = 0; i < pic->nb_refs[0]; i++) {
                 av_assert0(pic->refs[0][i]);
-                if (pic->refs[0][i] != def_l0[i])
+                if (pic->refs[0][i] != (FFHWBaseEncodePicture *)def_l0[i])
                     need_rplm = 1;
             }
                 av_assert0(pic->refs[0][i]);
                 href = pic->refs[0][i]->priv_data;
                 av_assert0(href->pic_order_cnt < hpic->pic_order_cnt);
-                if (pic->refs[0][i] != def_l0[n0])
+                if (pic->refs[0][i] != (FFHWBaseEncodePicture *)def_l0[n0])
                     need_rplm_l0 = 1;
                 ++n0;
             }
                 av_assert0(pic->refs[1][i]);
                 href = pic->refs[1][i]->priv_data;
                 av_assert0(href->pic_order_cnt > hpic->pic_order_cnt);
-                if (pic->refs[1][i] != def_l1[n1])
+                if (pic->refs[1][i] != (FFHWBaseEncodePicture *)def_l1[n1])
                     need_rplm_l1 = 1;
                 ++n1;
             }
     .p.id           = AV_CODEC_ID_H264,
     .priv_data_size = sizeof(VAAPIEncodeH264Context),
     .init           = &vaapi_encode_h264_init,
-    FF_CODEC_RECEIVE_PACKET_CB(&ff_vaapi_encode_receive_packet),
+    FF_CODEC_RECEIVE_PACKET_CB(&ff_hw_base_encode_receive_packet),
     .close          = &vaapi_encode_h264_close,
     .p.priv_class   = &vaapi_encode_h264_class,
     .p.capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_HARDWARE |

libavcodec/vaapi_encode_h265.c+28−25

 static int vaapi_encode_h265_init_sequence_params(AVCodecContext *avctx)
 {
+    FFHWBaseEncodeContext        *base_ctx = avctx->priv_data;
     VAAPIEncodeContext                *ctx = avctx->priv_data;
     VAAPIEncodeH265Context           *priv = avctx->priv_data;
     H265RawVPS                        *vps = &priv->raw_vps;
     ptl->general_max_420chroma_constraint_flag  = chroma_format <= 1;
     ptl->general_max_monochrome_constraint_flag = chroma_format == 0;
-    ptl->general_intra_constraint_flag = ctx->gop_size == 1;
+    ptl->general_intra_constraint_flag = base_ctx->gop_size == 1;
     ptl->general_one_picture_only_constraint_flag = 0;
     ptl->general_lower_bit_rate_constraint_flag = 1;
         level = ff_h265_guess_level(ptl, avctx->bit_rate,
                                     ctx->surface_width, ctx->surface_height,
                                     ctx->nb_slices, ctx->tile_rows, ctx->tile_cols,
-                                    (ctx->b_per_p > 0) + 1);
+                                    (base_ctx->b_per_p > 0) + 1);
         if (level) {
             av_log(avctx, AV_LOG_VERBOSE, "Using level %s.\n", level->name);
             ptl->general_level_idc = level->level_idc;
     }
     vps->vps_sub_layer_ordering_info_present_flag = 0;
-    vps->vps_max_dec_pic_buffering_minus1[0]      = ctx->max_b_depth + 1;
+    vps->vps_max_dec_pic_buffering_minus1[0]      = base_ctx->max_b_depth + 1;
-    vps->vps_max_num_reorder_pics[0]              = ctx->max_b_depth;
+    vps->vps_max_num_reorder_pics[0]              = base_ctx->max_b_depth;
     vps->vps_max_latency_increase_plus1[0]        = 0;
     vps->vps_max_layer_id             = 0;
         .general_level_idc   = vps->profile_tier_level.general_level_idc,
         .general_tier_flag   = vps->profile_tier_level.general_tier_flag,
-        .intra_period     = ctx->gop_size,
+        .intra_period     = base_ctx->gop_size,
-        .intra_idr_period = ctx->gop_size,
+        .intra_idr_period = base_ctx->gop_size,
-        .ip_period        = ctx->b_per_p + 1,
+        .ip_period        = base_ctx->b_per_p + 1,
         .bits_per_second  = ctx->va_bit_rate,
         .pic_width_in_luma_samples  = sps->pic_width_in_luma_samples,
 }
 static int vaapi_encode_h265_init_picture_params(AVCodecContext *avctx,
-                                                 VAAPIEncodePicture *pic)
+                                                 VAAPIEncodePicture *vaapi_pic)
 {
-    VAAPIEncodeContext               *ctx = avctx->priv_data;
+    FFHWBaseEncodeContext       *base_ctx = avctx->priv_data;
     VAAPIEncodeH265Context          *priv = avctx->priv_data;
+    FFHWBaseEncodePicture            *pic = &vaapi_pic->base;
     VAAPIEncodeH265Picture          *hpic = pic->priv_data;
-    VAAPIEncodePicture              *prev = pic->prev;
+    FFHWBaseEncodePicture           *prev = pic->prev;
     VAAPIEncodeH265Picture         *hprev = prev ? prev->priv_data : NULL;
-    VAEncPictureParameterBufferHEVC *vpic = pic->codec_picture_params;
+    VAEncPictureParameterBufferHEVC *vpic = vaapi_pic->codec_picture_params;
     int i, j = 0;
     if (pic->type == FF_HW_PICTURE_TYPE_IDR) {
             hpic->slice_type     = HEVC_SLICE_P;
             hpic->pic_type       = 1;
         } else {
-            VAAPIEncodePicture *irap_ref;
+            FFHWBaseEncodePicture *irap_ref;
             av_assert0(pic->refs[0][0] && pic->refs[1][0]);
             for (irap_ref = pic; irap_ref; irap_ref = irap_ref->refs[1][0]) {
                 if (irap_ref->type == FF_HW_PICTURE_TYPE_I)
                     break;
             }
-            if (pic->b_depth == ctx->max_b_depth) {
+            if (pic->b_depth == base_ctx->max_b_depth) {
                 hpic->slice_nal_unit = irap_ref ? HEVC_NAL_RASL_N
                                                 : HEVC_NAL_TRAIL_N;
             } else {
     }
     vpic->decoded_curr_pic = (VAPictureHEVC) {
-        .picture_id    = pic->recon_surface,
+        .picture_id    = vaapi_pic->recon_surface,
         .pic_order_cnt = hpic->pic_order_cnt,
         .flags         = 0,
     };
     for (int k = 0; k < MAX_REFERENCE_LIST_NUM; k++) {
         for (i = 0; i < pic->nb_refs[k]; i++) {
-            VAAPIEncodePicture      *ref = pic->refs[k][i];
+            FFHWBaseEncodePicture *ref = pic->refs[k][i];
             VAAPIEncodeH265Picture *href;
             av_assert0(ref && ref->encode_order < pic->encode_order);
             href = ref->priv_data;
             vpic->reference_frames[j++] = (VAPictureHEVC) {
-                .picture_id    = ref->recon_surface,
+                .picture_id    = ((VAAPIEncodePicture *)ref)->recon_surface,
                 .pic_order_cnt = href->pic_order_cnt,
                 .flags = (ref->display_order < pic->display_order ?
                           VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE : 0) |
         };
     }
-    vpic->coded_buf = pic->output_buffer;
+    vpic->coded_buf = vaapi_pic->output_buffer;
     vpic->nal_unit_type = hpic->slice_nal_unit;
 }
 static int vaapi_encode_h265_init_slice_params(AVCodecContext *avctx,
-                                               VAAPIEncodePicture *pic,
+                                               VAAPIEncodePicture *vaapi_pic,
                                                VAAPIEncodeSlice *slice)
 {
-    VAAPIEncodeContext                *ctx = avctx->priv_data;
+    FFHWBaseEncodeContext        *base_ctx = avctx->priv_data;
     VAAPIEncodeH265Context           *priv = avctx->priv_data;
+    const FFHWBaseEncodePicture       *pic = &vaapi_pic->base;
     VAAPIEncodeH265Picture           *hpic = pic->priv_data;
     const H265RawSPS                  *sps = &priv->raw_sps;
     const H265RawPPS                  *pps = &priv->raw_pps;
     H265RawSliceHeader                 *sh = &priv->raw_slice.header;
-    VAEncPictureParameterBufferHEVC  *vpic = pic->codec_picture_params;
+    VAEncPictureParameterBufferHEVC  *vpic = vaapi_pic->codec_picture_params;
     VAEncSliceParameterBufferHEVC  *vslice = slice->codec_slice_params;
     int i;
     sh->slice_type = hpic->slice_type;
-    if (sh->slice_type == HEVC_SLICE_P && ctx->p_to_gpb)
+    if (sh->slice_type == HEVC_SLICE_P && base_ctx->p_to_gpb)
         sh->slice_type = HEVC_SLICE_B;
     sh->slice_pic_order_cnt_lsb = hpic->pic_order_cnt &
         .slice_tc_offset_div2   = sh->slice_tc_offset_div2,
         .slice_fields.bits = {
-            .last_slice_of_pic_flag       = slice->index == pic->nb_slices - 1,
+            .last_slice_of_pic_flag       = slice->index == vaapi_pic->nb_slices - 1,
             .dependent_slice_segment_flag = sh->dependent_slice_segment_flag,
             .colour_plane_id              = sh->colour_plane_id,
             .slice_temporal_mvp_enabled_flag =
         av_assert0(pic->type == FF_HW_PICTURE_TYPE_P ||
                    pic->type == FF_HW_PICTURE_TYPE_B);
         vslice->ref_pic_list0[0] = vpic->reference_frames[0];
-        if (ctx->p_to_gpb && pic->type == FF_HW_PICTURE_TYPE_P)
+        if (base_ctx->p_to_gpb && pic->type == FF_HW_PICTURE_TYPE_P)
             // Reference for GPB B-frame, L0 == L1
             vslice->ref_pic_list1[0] = vpic->reference_frames[0];
     }
         vslice->ref_pic_list1[0] = vpic->reference_frames[1];
     }
-    if (pic->type == FF_HW_PICTURE_TYPE_P && ctx->p_to_gpb) {
+    if (pic->type == FF_HW_PICTURE_TYPE_P && base_ctx->p_to_gpb) {
         vslice->slice_type = HEVC_SLICE_B;
         for (i = 0; i < FF_ARRAY_ELEMS(vslice->ref_pic_list0); i++) {
             vslice->ref_pic_list1[i].picture_id = vslice->ref_pic_list0[i].picture_id;
     .p.id           = AV_CODEC_ID_HEVC,
     .priv_data_size = sizeof(VAAPIEncodeH265Context),
     .init           = &vaapi_encode_h265_init,
-    FF_CODEC_RECEIVE_PACKET_CB(&ff_vaapi_encode_receive_packet),
+    FF_CODEC_RECEIVE_PACKET_CB(&ff_hw_base_encode_receive_packet),
     .close          = &vaapi_encode_h265_close,
     .p.priv_class   = &vaapi_encode_h265_class,
     .p.capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_HARDWARE |

libavcodec/vaapi_encode_mjpeg.c+7−6

 }
 static int vaapi_encode_mjpeg_init_picture_params(AVCodecContext *avctx,
-                                                  VAAPIEncodePicture *pic)
+                                                  VAAPIEncodePicture *vaapi_pic)
 {
     VAAPIEncodeMJPEGContext         *priv = avctx->priv_data;
+    const FFHWBaseEncodePicture      *pic = &vaapi_pic->base;
     JPEGRawFrameHeader                *fh = &priv->frame_header;
     JPEGRawScanHeader                 *sh = &priv->scan.header;
-    VAEncPictureParameterBufferJPEG *vpic = pic->codec_picture_params;
+    VAEncPictureParameterBufferJPEG *vpic = vaapi_pic->codec_picture_params;
     const AVPixFmtDescriptor *desc;
     const uint8_t components_rgb[3] = { 'R', 'G', 'B' };
     const uint8_t components_yuv[3] = {  1,   2,   3  };
     *vpic = (VAEncPictureParameterBufferJPEG) {
-        .reconstructed_picture = pic->recon_surface,
+        .reconstructed_picture = vaapi_pic->recon_surface,
-        .coded_buf             = pic->output_buffer,
+        .coded_buf             = vaapi_pic->output_buffer,
         .picture_width  = fh->X,
         .picture_height = fh->Y,
         vpic->quantiser_table_selector[i] = fh->Tq[i];
     }
-    pic->nb_slices = 1;
+    vaapi_pic->nb_slices = 1;
     return 0;
 }
     .p.id           = AV_CODEC_ID_MJPEG,
     .priv_data_size = sizeof(VAAPIEncodeMJPEGContext),
     .init           = &vaapi_encode_mjpeg_init,
-    FF_CODEC_RECEIVE_PACKET_CB(&ff_vaapi_encode_receive_packet),
+    FF_CODEC_RECEIVE_PACKET_CB(&ff_hw_base_encode_receive_packet),
     .close          = &vaapi_encode_mjpeg_close,
     .p.priv_class   = &vaapi_encode_mjpeg_class,
     .p.capabilities = AV_CODEC_CAP_HARDWARE | AV_CODEC_CAP_DR1 |

libavcodec/vaapi_encode_mpeg2.c+18−15

 static int vaapi_encode_mpeg2_init_sequence_params(AVCodecContext *avctx)
 {
+    FFHWBaseEncodeContext         *base_ctx = avctx->priv_data;
     VAAPIEncodeContext                 *ctx = avctx->priv_data;
     VAAPIEncodeMPEG2Context           *priv = avctx->priv_data;
     MPEG2RawSequenceHeader              *sh = &priv->sequence_header;
     se->bit_rate_extension        = priv->bit_rate >> 18;
     se->vbv_buffer_size_extension = priv->vbv_buffer_size >> 10;
-    se->low_delay                 = ctx->b_per_p == 0;
+    se->low_delay                 = base_ctx->b_per_p == 0;
     se->frame_rate_extension_n = ext_n;
     se->frame_rate_extension_d = ext_d;
     *vseq = (VAEncSequenceParameterBufferMPEG2) {
-        .intra_period = ctx->gop_size,
+        .intra_period = base_ctx->gop_size,
-        .ip_period    = ctx->b_per_p + 1,
+        .ip_period    = base_ctx->b_per_p + 1,
         .picture_width  = avctx->width,
         .picture_height = avctx->height,
 }
 static int vaapi_encode_mpeg2_init_picture_params(AVCodecContext *avctx,
-                                                 VAAPIEncodePicture *pic)
+                                                  VAAPIEncodePicture *vaapi_pic)
 {
     VAAPIEncodeMPEG2Context          *priv = avctx->priv_data;
+    const FFHWBaseEncodePicture       *pic = &vaapi_pic->base;
     MPEG2RawPictureHeader              *ph = &priv->picture_header;
     MPEG2RawPictureCodingExtension    *pce = &priv->picture_coding_extension.data.picture_coding;
-    VAEncPictureParameterBufferMPEG2 *vpic = pic->codec_picture_params;
+    VAEncPictureParameterBufferMPEG2 *vpic = vaapi_pic->codec_picture_params;
     if (pic->type == FF_HW_PICTURE_TYPE_IDR || pic->type == FF_HW_PICTURE_TYPE_I) {
         ph->temporal_reference  = 0;
         pce->f_code[1][1] = 15;
     }
-    vpic->reconstructed_picture = pic->recon_surface;
+    vpic->reconstructed_picture = vaapi_pic->recon_surface;
-    vpic->coded_buf             = pic->output_buffer;
+    vpic->coded_buf             = vaapi_pic->output_buffer;
     switch (pic->type) {
     case FF_HW_PICTURE_TYPE_IDR:
         break;
     case FF_HW_PICTURE_TYPE_P:
         vpic->picture_type = VAEncPictureTypePredictive;
-        vpic->forward_reference_picture = pic->refs[0][0]->recon_surface;
+        vpic->forward_reference_picture = ((VAAPIEncodePicture *)pic->refs[0][0])->recon_surface;
         break;
     case FF_HW_PICTURE_TYPE_B:
         vpic->picture_type = VAEncPictureTypeBidirectional;
-        vpic->forward_reference_picture  = pic->refs[0][0]->recon_surface;
+        vpic->forward_reference_picture  = ((VAAPIEncodePicture *)pic->refs[0][0])->recon_surface;
-        vpic->backward_reference_picture = pic->refs[1][0]->recon_surface;
+        vpic->backward_reference_picture = ((VAAPIEncodePicture *)pic->refs[1][0])->recon_surface;
         break;
     default:
         av_assert0(0 && "invalid picture type");
 }
 static int vaapi_encode_mpeg2_init_slice_params(AVCodecContext *avctx,
-                                               VAAPIEncodePicture *pic,
+                                                VAAPIEncodePicture *vaapi_pic,
-                                               VAAPIEncodeSlice *slice)
+                                                VAAPIEncodeSlice *slice)
 {
-    VAAPIEncodeMPEG2Context            *priv = avctx->priv_data;
+    const FFHWBaseEncodePicture       *pic = &vaapi_pic->base;
-    VAEncSliceParameterBufferMPEG2   *vslice = slice->codec_slice_params;
+    VAAPIEncodeMPEG2Context          *priv = avctx->priv_data;
+    VAEncSliceParameterBufferMPEG2 *vslice = slice->codec_slice_params;
     int qp;
     vslice->macroblock_address = slice->block_start;
     .p.id           = AV_CODEC_ID_MPEG2VIDEO,
     .priv_data_size = sizeof(VAAPIEncodeMPEG2Context),
     .init           = &vaapi_encode_mpeg2_init,
-    FF_CODEC_RECEIVE_PACKET_CB(&ff_vaapi_encode_receive_packet),
+    FF_CODEC_RECEIVE_PACKET_CB(&ff_hw_base_encode_receive_packet),
     .close          = &vaapi_encode_mpeg2_close,
     .p.priv_class   = &vaapi_encode_mpeg2_class,
     .p.capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_HARDWARE |

libavcodec/vaapi_encode_vp8.c+10−8

 static int vaapi_encode_vp8_init_sequence_params(AVCodecContext *avctx)
 {
+    FFHWBaseEncodeContext       *base_ctx = avctx->priv_data;
     VAAPIEncodeContext               *ctx = avctx->priv_data;
     VAEncSequenceParameterBufferVP8 *vseq = ctx->codec_sequence_params;
     if (!(ctx->va_rc_mode & VA_RC_CQP)) {
         vseq->bits_per_second = ctx->va_bit_rate;
-        vseq->intra_period    = ctx->gop_size;
+        vseq->intra_period    = base_ctx->gop_size;
     }
     return 0;
 }
 static int vaapi_encode_vp8_init_picture_params(AVCodecContext *avctx,
-                                                VAAPIEncodePicture *pic)
+                                                VAAPIEncodePicture *vaapi_pic)
 {
+    const FFHWBaseEncodePicture     *pic = &vaapi_pic->base;
     VAAPIEncodeVP8Context          *priv = avctx->priv_data;
-    VAEncPictureParameterBufferVP8 *vpic = pic->codec_picture_params;
+    VAEncPictureParameterBufferVP8 *vpic = vaapi_pic->codec_picture_params;
     int i;
-    vpic->reconstructed_frame = pic->recon_surface;
+    vpic->reconstructed_frame = vaapi_pic->recon_surface;
-    vpic->coded_buf = pic->output_buffer;
+    vpic->coded_buf = vaapi_pic->output_buffer;
     switch (pic->type) {
     case FF_HW_PICTURE_TYPE_IDR:
         vpic->ref_last_frame =
         vpic->ref_gf_frame   =
         vpic->ref_arf_frame  =
-            pic->refs[0][0]->recon_surface;
+            ((VAAPIEncodePicture *)pic->refs[0][0])->recon_surface;
         break;
     default:
         av_assert0(0 && "invalid picture type");
     memset(&quant, 0, sizeof(quant));
-    if (pic->type == FF_HW_PICTURE_TYPE_P)
+    if (pic->base.type == FF_HW_PICTURE_TYPE_P)
         q = priv->q_index_p;
     else
         q = priv->q_index_i;
     .p.id           = AV_CODEC_ID_VP8,
     .priv_data_size = sizeof(VAAPIEncodeVP8Context),
     .init           = &vaapi_encode_vp8_init,
-    FF_CODEC_RECEIVE_PACKET_CB(&ff_vaapi_encode_receive_packet),
+    FF_CODEC_RECEIVE_PACKET_CB(&ff_hw_base_encode_receive_packet),
     .close          = &ff_vaapi_encode_close,
     .p.priv_class   = &vaapi_encode_vp8_class,
     .p.capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_HARDWARE |

libavcodec/vaapi_encode_vp9.c+13−11

 static int vaapi_encode_vp9_init_sequence_params(AVCodecContext *avctx)
 {
+    FFHWBaseEncodeContext       *base_ctx = avctx->priv_data;
     VAAPIEncodeContext               *ctx = avctx->priv_data;
     VAEncSequenceParameterBufferVP9 *vseq = ctx->codec_sequence_params;
     VAEncPictureParameterBufferVP9  *vpic = ctx->codec_picture_params;
     if (!(ctx->va_rc_mode & VA_RC_CQP)) {
         vseq->bits_per_second = ctx->va_bit_rate;
-        vseq->intra_period    = ctx->gop_size;
+        vseq->intra_period    = base_ctx->gop_size;
     }
     vpic->frame_width_src  = avctx->width;
 }
 static int vaapi_encode_vp9_init_picture_params(AVCodecContext *avctx,
-                                                VAAPIEncodePicture *pic)
+                                                VAAPIEncodePicture *vaapi_pic)
 {
-    VAAPIEncodeContext              *ctx = avctx->priv_data;
+    FFHWBaseEncodeContext      *base_ctx = avctx->priv_data;
     VAAPIEncodeVP9Context          *priv = avctx->priv_data;
+    const FFHWBaseEncodePicture     *pic = &vaapi_pic->base;
     VAAPIEncodeVP9Picture          *hpic = pic->priv_data;
-    VAEncPictureParameterBufferVP9 *vpic = pic->codec_picture_params;
+    VAEncPictureParameterBufferVP9 *vpic = vaapi_pic->codec_picture_params;
     int i;
     int num_tile_columns;
-    vpic->reconstructed_frame = pic->recon_surface;
+    vpic->reconstructed_frame = vaapi_pic->recon_surface;
-    vpic->coded_buf = pic->output_buffer;
+    vpic->coded_buf = vaapi_pic->output_buffer;
     // Maximum width of a tile in units of superblocks is MAX_TILE_WIDTH_B64(64)
     // So the number of tile columns is related to the width of the picture.
             VAAPIEncodeVP9Picture *href = pic->refs[0][0]->priv_data;
             av_assert0(href->slot == 0 || href->slot == 1);
-            if (ctx->max_b_depth > 0) {
+            if (base_ctx->max_b_depth > 0) {
                 hpic->slot = !href->slot;
                 vpic->refresh_frame_flags = 1 << hpic->slot | 0xfc;
             } else {
             av_assert0(href0->slot < pic->b_depth + 1 &&
                        href1->slot < pic->b_depth + 1);
-            if (pic->b_depth == ctx->max_b_depth) {
+            if (pic->b_depth == base_ctx->max_b_depth) {
                 // Unreferenced frame.
                 vpic->refresh_frame_flags = 0x00;
                 hpic->slot = 8;
     for (i = 0; i < MAX_REFERENCE_LIST_NUM; i++) {
         for (int j = 0; j < pic->nb_refs[i]; j++) {
-            VAAPIEncodePicture *ref_pic = pic->refs[i][j];
+            FFHWBaseEncodePicture *ref_pic = pic->refs[i][j];
             int slot;
             slot = ((VAAPIEncodeVP9Picture*)ref_pic->priv_data)->slot;
             av_assert0(vpic->reference_frames[slot] == VA_INVALID_SURFACE);
-            vpic->reference_frames[slot] = ref_pic->recon_surface;
+            vpic->reference_frames[slot] = ((VAAPIEncodePicture *)ref_pic)->recon_surface;
         }
     }
     .p.id           = AV_CODEC_ID_VP9,
     .priv_data_size = sizeof(VAAPIEncodeVP9Context),
     .init           = &vaapi_encode_vp9_init,
-    FF_CODEC_RECEIVE_PACKET_CB(&ff_vaapi_encode_receive_packet),
+    FF_CODEC_RECEIVE_PACKET_CB(&ff_hw_base_encode_receive_packet),
     .close          = &ff_vaapi_encode_close,
     .p.priv_class   = &vaapi_encode_vp9_class,
     .p.capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_HARDWARE |

avcodec/vaapi_encode: move the dpb logic from VAAPI to base layer

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code