2 * This file is part of FFmpeg.
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.
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.
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
22 #include "libavutil/avassert.h"
23 #include "libavutil/common.h"
24 #include "libavutil/log.h"
25 #include "libavutil/pixdesc.h"
27 #include "vaapi_encode.h"
30 static const char * const picture_type_name[] = { "IDR", "I", "P", "B" };
32 static int vaapi_encode_make_packed_header(AVCodecContext *avctx,
33 VAAPIEncodePicture *pic,
34 int type, char *data, size_t bit_len)
36 VAAPIEncodeContext *ctx = avctx->priv_data;
38 VABufferID param_buffer, data_buffer;
40 VAEncPackedHeaderParameterBuffer params = {
42 .bit_length = bit_len,
43 .has_emulation_bytes = 1,
46 tmp = av_realloc_array(pic->param_buffers, sizeof(*tmp), pic->nb_param_buffers + 2);
48 return AVERROR(ENOMEM);
49 pic->param_buffers = tmp;
51 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
52 VAEncPackedHeaderParameterBufferType,
53 sizeof(params), 1, ¶ms, ¶m_buffer);
54 if (vas != VA_STATUS_SUCCESS) {
55 av_log(avctx, AV_LOG_ERROR, "Failed to create parameter buffer "
56 "for packed header (type %d): %d (%s).\n",
57 type, vas, vaErrorStr(vas));
60 pic->param_buffers[pic->nb_param_buffers++] = param_buffer;
62 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
63 VAEncPackedHeaderDataBufferType,
64 (bit_len + 7) / 8, 1, data, &data_buffer);
65 if (vas != VA_STATUS_SUCCESS) {
66 av_log(avctx, AV_LOG_ERROR, "Failed to create data buffer "
67 "for packed header (type %d): %d (%s).\n",
68 type, vas, vaErrorStr(vas));
71 pic->param_buffers[pic->nb_param_buffers++] = data_buffer;
73 av_log(avctx, AV_LOG_DEBUG, "Packed header buffer (%d) is %#x/%#x "
74 "(%zu bits).\n", type, param_buffer, data_buffer, bit_len);
78 static int vaapi_encode_make_param_buffer(AVCodecContext *avctx,
79 VAAPIEncodePicture *pic,
80 int type, char *data, size_t len)
82 VAAPIEncodeContext *ctx = avctx->priv_data;
87 tmp = av_realloc_array(pic->param_buffers, sizeof(*tmp), pic->nb_param_buffers + 1);
89 return AVERROR(ENOMEM);
90 pic->param_buffers = tmp;
92 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
93 type, len, 1, data, &buffer);
94 if (vas != VA_STATUS_SUCCESS) {
95 av_log(avctx, AV_LOG_ERROR, "Failed to create parameter buffer "
96 "(type %d): %d (%s).\n", type, vas, vaErrorStr(vas));
99 pic->param_buffers[pic->nb_param_buffers++] = buffer;
101 av_log(avctx, AV_LOG_DEBUG, "Param buffer (%d) is %#x.\n",
106 static int vaapi_encode_wait(AVCodecContext *avctx,
107 VAAPIEncodePicture *pic)
109 VAAPIEncodeContext *ctx = avctx->priv_data;
112 av_assert0(pic->encode_issued);
114 if (pic->encode_complete) {
115 // Already waited for this picture.
119 av_log(avctx, AV_LOG_DEBUG, "Sync to pic %"PRId64"/%"PRId64" "
120 "(input surface %#x).\n", pic->display_order,
121 pic->encode_order, pic->input_surface);
123 vas = vaSyncSurface(ctx->hwctx->display, pic->input_surface);
124 if (vas != VA_STATUS_SUCCESS) {
125 av_log(avctx, AV_LOG_ERROR, "Failed to sync to picture completion: "
126 "%d (%s).\n", vas, vaErrorStr(vas));
130 // Input is definitely finished with now.
131 av_frame_free(&pic->input_image);
133 pic->encode_complete = 1;
137 static int vaapi_encode_issue(AVCodecContext *avctx,
138 VAAPIEncodePicture *pic)
140 VAAPIEncodeContext *ctx = avctx->priv_data;
141 VAAPIEncodeSlice *slice;
144 char data[MAX_PARAM_BUFFER_SIZE];
147 av_log(avctx, AV_LOG_DEBUG, "Issuing encode for pic %"PRId64"/%"PRId64" "
148 "as type %s.\n", pic->display_order, pic->encode_order,
149 picture_type_name[pic->type]);
150 if (pic->nb_refs == 0) {
151 av_log(avctx, AV_LOG_DEBUG, "No reference pictures.\n");
153 av_log(avctx, AV_LOG_DEBUG, "Refers to:");
154 for (i = 0; i < pic->nb_refs; i++) {
155 av_log(avctx, AV_LOG_DEBUG, " %"PRId64"/%"PRId64,
156 pic->refs[i]->display_order, pic->refs[i]->encode_order);
158 av_log(avctx, AV_LOG_DEBUG, ".\n");
161 av_assert0(pic->input_available && !pic->encode_issued);
162 for (i = 0; i < pic->nb_refs; i++) {
163 av_assert0(pic->refs[i]);
164 // If we are serialised then the references must have already
165 // completed. If not, they must have been issued but need not
166 // have completed yet.
167 if (ctx->issue_mode == ISSUE_MODE_SERIALISE_EVERYTHING)
168 av_assert0(pic->refs[i]->encode_complete);
170 av_assert0(pic->refs[i]->encode_issued);
173 av_log(avctx, AV_LOG_DEBUG, "Input surface is %#x.\n", pic->input_surface);
175 pic->recon_image = av_frame_alloc();
176 if (!pic->recon_image) {
177 err = AVERROR(ENOMEM);
181 err = av_hwframe_get_buffer(ctx->recon_frames_ref, pic->recon_image, 0);
183 err = AVERROR(ENOMEM);
186 pic->recon_surface = (VASurfaceID)(uintptr_t)pic->recon_image->data[3];
187 av_log(avctx, AV_LOG_DEBUG, "Recon surface is %#x.\n", pic->recon_surface);
189 pic->output_buffer_ref = av_buffer_pool_get(ctx->output_buffer_pool);
190 if (!pic->output_buffer_ref) {
191 err = AVERROR(ENOMEM);
194 pic->output_buffer = (VABufferID)(uintptr_t)pic->output_buffer_ref->data;
195 av_log(avctx, AV_LOG_DEBUG, "Output buffer is %#x.\n",
198 if (ctx->codec->picture_params_size > 0) {
199 pic->codec_picture_params = av_malloc(ctx->codec->picture_params_size);
200 if (!pic->codec_picture_params)
202 memcpy(pic->codec_picture_params, ctx->codec_picture_params,
203 ctx->codec->picture_params_size);
205 av_assert0(!ctx->codec_picture_params);
208 pic->nb_param_buffers = 0;
210 if (pic->type == PICTURE_TYPE_IDR && ctx->codec->init_sequence_params) {
211 err = vaapi_encode_make_param_buffer(avctx, pic,
212 VAEncSequenceParameterBufferType,
213 ctx->codec_sequence_params,
214 ctx->codec->sequence_params_size);
219 if (pic->type == PICTURE_TYPE_IDR) {
220 for (i = 0; i < ctx->nb_global_params; i++) {
221 err = vaapi_encode_make_param_buffer(avctx, pic,
222 VAEncMiscParameterBufferType,
223 (char*)ctx->global_params[i],
224 ctx->global_params_size[i]);
230 if (ctx->codec->init_picture_params) {
231 err = ctx->codec->init_picture_params(avctx, pic);
233 av_log(avctx, AV_LOG_ERROR, "Failed to initialise picture "
234 "parameters: %d.\n", err);
237 err = vaapi_encode_make_param_buffer(avctx, pic,
238 VAEncPictureParameterBufferType,
239 pic->codec_picture_params,
240 ctx->codec->picture_params_size);
245 if (pic->type == PICTURE_TYPE_IDR) {
246 if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE &&
247 ctx->codec->write_sequence_header) {
248 bit_len = 8 * sizeof(data);
249 err = ctx->codec->write_sequence_header(avctx, data, &bit_len);
251 av_log(avctx, AV_LOG_ERROR, "Failed to write per-sequence "
252 "header: %d.\n", err);
255 err = vaapi_encode_make_packed_header(avctx, pic,
256 ctx->codec->sequence_header_type,
263 if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_PICTURE &&
264 ctx->codec->write_picture_header) {
265 bit_len = 8 * sizeof(data);
266 err = ctx->codec->write_picture_header(avctx, pic, data, &bit_len);
268 av_log(avctx, AV_LOG_ERROR, "Failed to write per-picture "
269 "header: %d.\n", err);
272 err = vaapi_encode_make_packed_header(avctx, pic,
273 ctx->codec->picture_header_type,
279 if (ctx->codec->write_extra_buffer) {
281 size_t len = sizeof(data);
283 err = ctx->codec->write_extra_buffer(avctx, pic, i, &type,
285 if (err == AVERROR_EOF)
288 av_log(avctx, AV_LOG_ERROR, "Failed to write extra "
289 "buffer %d: %d.\n", i, err);
293 err = vaapi_encode_make_param_buffer(avctx, pic, type,
300 if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_MISC &&
301 ctx->codec->write_extra_header) {
304 bit_len = 8 * sizeof(data);
305 err = ctx->codec->write_extra_header(avctx, pic, i, &type,
307 if (err == AVERROR_EOF)
310 av_log(avctx, AV_LOG_ERROR, "Failed to write extra "
311 "header %d: %d.\n", i, err);
315 err = vaapi_encode_make_packed_header(avctx, pic, type,
322 if (pic->nb_slices == 0)
323 pic->nb_slices = ctx->nb_slices;
324 if (pic->nb_slices > 0) {
327 pic->slices = av_mallocz_array(pic->nb_slices, sizeof(*pic->slices));
329 err = AVERROR(ENOMEM);
333 for (i = 0; i < pic->nb_slices; i++)
334 pic->slices[i].row_size = ctx->slice_size;
336 rounding = ctx->slice_block_rows - ctx->nb_slices * ctx->slice_size;
338 // Place rounding error at top and bottom of frame.
339 av_assert0(rounding < pic->nb_slices);
340 // Some Intel drivers contain a bug where the encoder will fail
341 // if the last slice is smaller than the one before it. Since
342 // that's straightforward to avoid here, just do so.
344 for (i = 0; i < rounding; i++)
345 ++pic->slices[i].row_size;
347 for (i = 0; i < (rounding + 1) / 2; i++)
348 ++pic->slices[pic->nb_slices - i - 1].row_size;
349 for (i = 0; i < rounding / 2; i++)
350 ++pic->slices[i].row_size;
352 } else if (rounding < 0) {
353 // Remove rounding error from last slice only.
354 av_assert0(rounding < ctx->slice_size);
355 pic->slices[pic->nb_slices - 1].row_size += rounding;
358 for (i = 0; i < pic->nb_slices; i++) {
359 slice = &pic->slices[i];
362 slice->row_start = 0;
363 slice->block_start = 0;
365 const VAAPIEncodeSlice *prev = &pic->slices[i - 1];
366 slice->row_start = prev->row_start + prev->row_size;
367 slice->block_start = prev->block_start + prev->block_size;
369 slice->block_size = slice->row_size * ctx->slice_block_cols;
371 av_log(avctx, AV_LOG_DEBUG, "Slice %d: %d-%d (%d rows), "
372 "%d-%d (%d blocks).\n", i, slice->row_start,
373 slice->row_start + slice->row_size - 1, slice->row_size,
374 slice->block_start, slice->block_start + slice->block_size - 1,
377 if (ctx->codec->slice_params_size > 0) {
378 slice->codec_slice_params = av_mallocz(ctx->codec->slice_params_size);
379 if (!slice->codec_slice_params) {
380 err = AVERROR(ENOMEM);
385 if (ctx->codec->init_slice_params) {
386 err = ctx->codec->init_slice_params(avctx, pic, slice);
388 av_log(avctx, AV_LOG_ERROR, "Failed to initialise slice "
389 "parameters: %d.\n", err);
394 if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SLICE &&
395 ctx->codec->write_slice_header) {
396 bit_len = 8 * sizeof(data);
397 err = ctx->codec->write_slice_header(avctx, pic, slice,
400 av_log(avctx, AV_LOG_ERROR, "Failed to write per-slice "
401 "header: %d.\n", err);
404 err = vaapi_encode_make_packed_header(avctx, pic,
405 ctx->codec->slice_header_type,
411 if (ctx->codec->init_slice_params) {
412 err = vaapi_encode_make_param_buffer(avctx, pic,
413 VAEncSliceParameterBufferType,
414 slice->codec_slice_params,
415 ctx->codec->slice_params_size);
421 vas = vaBeginPicture(ctx->hwctx->display, ctx->va_context,
423 if (vas != VA_STATUS_SUCCESS) {
424 av_log(avctx, AV_LOG_ERROR, "Failed to begin picture encode issue: "
425 "%d (%s).\n", vas, vaErrorStr(vas));
427 goto fail_with_picture;
430 vas = vaRenderPicture(ctx->hwctx->display, ctx->va_context,
431 pic->param_buffers, pic->nb_param_buffers);
432 if (vas != VA_STATUS_SUCCESS) {
433 av_log(avctx, AV_LOG_ERROR, "Failed to upload encode parameters: "
434 "%d (%s).\n", vas, vaErrorStr(vas));
436 goto fail_with_picture;
439 vas = vaEndPicture(ctx->hwctx->display, ctx->va_context);
440 if (vas != VA_STATUS_SUCCESS) {
441 av_log(avctx, AV_LOG_ERROR, "Failed to end picture encode issue: "
442 "%d (%s).\n", vas, vaErrorStr(vas));
444 // vaRenderPicture() has been called here, so we should not destroy
445 // the parameter buffers unless separate destruction is required.
446 if (CONFIG_VAAPI_1 || ctx->hwctx->driver_quirks &
447 AV_VAAPI_DRIVER_QUIRK_RENDER_PARAM_BUFFERS)
453 if (CONFIG_VAAPI_1 || ctx->hwctx->driver_quirks &
454 AV_VAAPI_DRIVER_QUIRK_RENDER_PARAM_BUFFERS) {
455 for (i = 0; i < pic->nb_param_buffers; i++) {
456 vas = vaDestroyBuffer(ctx->hwctx->display,
457 pic->param_buffers[i]);
458 if (vas != VA_STATUS_SUCCESS) {
459 av_log(avctx, AV_LOG_ERROR, "Failed to destroy "
460 "param buffer %#x: %d (%s).\n",
461 pic->param_buffers[i], vas, vaErrorStr(vas));
467 pic->encode_issued = 1;
469 if (ctx->issue_mode == ISSUE_MODE_SERIALISE_EVERYTHING)
470 return vaapi_encode_wait(avctx, pic);
475 vaEndPicture(ctx->hwctx->display, ctx->va_context);
477 for(i = 0; i < pic->nb_param_buffers; i++)
478 vaDestroyBuffer(ctx->hwctx->display, pic->param_buffers[i]);
479 for (i = 0; i < pic->nb_slices; i++) {
481 av_freep(&pic->slices[i].priv_data);
482 av_freep(&pic->slices[i].codec_slice_params);
486 av_freep(&pic->codec_picture_params);
487 av_freep(&pic->param_buffers);
488 av_freep(&pic->slices);
489 av_frame_free(&pic->recon_image);
490 av_buffer_unref(&pic->output_buffer_ref);
491 pic->output_buffer = VA_INVALID_ID;
495 static int vaapi_encode_output(AVCodecContext *avctx,
496 VAAPIEncodePicture *pic, AVPacket *pkt)
498 VAAPIEncodeContext *ctx = avctx->priv_data;
499 VACodedBufferSegment *buf_list, *buf;
503 err = vaapi_encode_wait(avctx, pic);
508 vas = vaMapBuffer(ctx->hwctx->display, pic->output_buffer,
510 if (vas != VA_STATUS_SUCCESS) {
511 av_log(avctx, AV_LOG_ERROR, "Failed to map output buffers: "
512 "%d (%s).\n", vas, vaErrorStr(vas));
517 for (buf = buf_list; buf; buf = buf->next) {
518 av_log(avctx, AV_LOG_DEBUG, "Output buffer: %u bytes "
519 "(status %08x).\n", buf->size, buf->status);
521 err = av_new_packet(pkt, buf->size);
525 memcpy(pkt->data, buf->buf, buf->size);
528 if (pic->type == PICTURE_TYPE_IDR)
529 pkt->flags |= AV_PKT_FLAG_KEY;
533 vas = vaUnmapBuffer(ctx->hwctx->display, pic->output_buffer);
534 if (vas != VA_STATUS_SUCCESS) {
535 av_log(avctx, AV_LOG_ERROR, "Failed to unmap output buffers: "
536 "%d (%s).\n", vas, vaErrorStr(vas));
541 av_buffer_unref(&pic->output_buffer_ref);
542 pic->output_buffer = VA_INVALID_ID;
544 av_log(avctx, AV_LOG_DEBUG, "Output read for pic %"PRId64"/%"PRId64".\n",
545 pic->display_order, pic->encode_order);
549 vaUnmapBuffer(ctx->hwctx->display, pic->output_buffer);
551 av_buffer_unref(&pic->output_buffer_ref);
552 pic->output_buffer = VA_INVALID_ID;
556 static int vaapi_encode_discard(AVCodecContext *avctx,
557 VAAPIEncodePicture *pic)
559 vaapi_encode_wait(avctx, pic);
561 if (pic->output_buffer_ref) {
562 av_log(avctx, AV_LOG_DEBUG, "Discard output for pic "
563 "%"PRId64"/%"PRId64".\n",
564 pic->display_order, pic->encode_order);
566 av_buffer_unref(&pic->output_buffer_ref);
567 pic->output_buffer = VA_INVALID_ID;
573 static VAAPIEncodePicture *vaapi_encode_alloc(AVCodecContext *avctx)
575 VAAPIEncodeContext *ctx = avctx->priv_data;
576 VAAPIEncodePicture *pic;
578 pic = av_mallocz(sizeof(*pic));
582 if (ctx->codec->picture_priv_data_size > 0) {
583 pic->priv_data = av_mallocz(ctx->codec->picture_priv_data_size);
584 if (!pic->priv_data) {
590 pic->input_surface = VA_INVALID_ID;
591 pic->recon_surface = VA_INVALID_ID;
592 pic->output_buffer = VA_INVALID_ID;
597 static int vaapi_encode_free(AVCodecContext *avctx,
598 VAAPIEncodePicture *pic)
602 if (pic->encode_issued)
603 vaapi_encode_discard(avctx, pic);
605 for (i = 0; i < pic->nb_slices; i++) {
607 av_freep(&pic->slices[i].priv_data);
608 av_freep(&pic->slices[i].codec_slice_params);
611 av_freep(&pic->codec_picture_params);
613 av_frame_free(&pic->input_image);
614 av_frame_free(&pic->recon_image);
616 av_freep(&pic->param_buffers);
617 av_freep(&pic->slices);
618 // Output buffer should already be destroyed.
619 av_assert0(pic->output_buffer == VA_INVALID_ID);
621 av_freep(&pic->priv_data);
622 av_freep(&pic->codec_picture_params);
629 static int vaapi_encode_step(AVCodecContext *avctx,
630 VAAPIEncodePicture *target)
632 VAAPIEncodeContext *ctx = avctx->priv_data;
633 VAAPIEncodePicture *pic;
636 if (ctx->issue_mode == ISSUE_MODE_SERIALISE_EVERYTHING ||
637 ctx->issue_mode == ISSUE_MODE_MINIMISE_LATENCY) {
638 // These two modes are equivalent, except that we wait for
639 // immediate completion on each operation if serialised.
642 // No target, nothing to do yet.
646 if (target->encode_complete) {
652 for (i = 0; i < pic->nb_refs; i++) {
653 if (!pic->refs[i]->encode_complete) {
654 err = vaapi_encode_step(avctx, pic->refs[i]);
660 err = vaapi_encode_issue(avctx, pic);
664 } else if (ctx->issue_mode == ISSUE_MODE_MAXIMISE_THROUGHPUT) {
667 // Run through the list of all available pictures repeatedly
668 // and issue the first one found which has all dependencies
669 // available (including previously-issued but not necessarily
670 // completed pictures).
673 for (pic = ctx->pic_start; pic; pic = pic->next) {
674 if (!pic->input_available || pic->encode_issued)
676 for (i = 0; i < pic->nb_refs; i++) {
677 if (!pic->refs[i]->encode_issued)
680 if (i < pic->nb_refs)
682 err = vaapi_encode_issue(avctx, pic);
686 // Start again from the beginning of the list,
687 // because issuing this picture may have satisfied
688 // forward dependencies of earlier ones.
693 // If we had a defined target for this step then it will
694 // always have been issued by now.
696 av_assert0(target->encode_issued && "broken dependencies?");
706 static int vaapi_encode_get_next(AVCodecContext *avctx,
707 VAAPIEncodePicture **pic_out)
709 VAAPIEncodeContext *ctx = avctx->priv_data;
710 VAAPIEncodePicture *start, *end, *pic;
713 for (pic = ctx->pic_start; pic; pic = pic->next) {
715 av_assert0(pic->display_order + 1 == pic->next->display_order);
716 if (pic->display_order == ctx->input_order) {
722 pic = vaapi_encode_alloc(avctx);
724 return AVERROR(ENOMEM);
726 if (ctx->input_order == 0 || ctx->force_idr ||
727 ctx->gop_counter >= ctx->gop_size) {
728 pic->type = PICTURE_TYPE_IDR;
730 ctx->gop_counter = 1;
732 } else if (ctx->p_counter >= ctx->p_per_i) {
733 pic->type = PICTURE_TYPE_I;
737 pic->type = PICTURE_TYPE_P;
738 pic->refs[0] = ctx->pic_end;
745 if (pic->type != PICTURE_TYPE_IDR) {
746 // If that was not an IDR frame, add B-frames display-before and
747 // encode-after it, but not exceeding the GOP size.
749 for (i = 0; i < ctx->b_per_p &&
750 ctx->gop_counter < ctx->gop_size; i++) {
751 pic = vaapi_encode_alloc(avctx);
755 pic->type = PICTURE_TYPE_B;
756 pic->refs[0] = ctx->pic_end;
761 pic->display_order = ctx->input_order + ctx->b_per_p - i - 1;
762 pic->encode_order = pic->display_order + 1;
769 if (ctx->input_order == 0) {
770 pic->display_order = 0;
771 pic->encode_order = 0;
773 ctx->pic_start = ctx->pic_end = pic;
776 for (i = 0, pic = start; pic; i++, pic = pic->next) {
777 pic->display_order = ctx->input_order + i;
778 if (end->type == PICTURE_TYPE_IDR)
779 pic->encode_order = ctx->input_order + i;
781 pic->encode_order = ctx->input_order;
783 pic->encode_order = ctx->input_order + i + 1;
786 av_assert0(ctx->pic_end);
787 ctx->pic_end->next = start;
792 av_log(avctx, AV_LOG_DEBUG, "Pictures:");
793 for (pic = ctx->pic_start; pic; pic = pic->next) {
794 av_log(avctx, AV_LOG_DEBUG, " %s (%"PRId64"/%"PRId64")",
795 picture_type_name[pic->type],
796 pic->display_order, pic->encode_order);
798 av_log(avctx, AV_LOG_DEBUG, "\n");
805 vaapi_encode_free(avctx, start);
808 return AVERROR(ENOMEM);
811 static int vaapi_encode_truncate_gop(AVCodecContext *avctx)
813 VAAPIEncodeContext *ctx = avctx->priv_data;
814 VAAPIEncodePicture *pic, *last_pic, *next;
816 av_assert0(!ctx->pic_start || ctx->pic_start->input_available);
818 // Find the last picture we actually have input for.
819 for (pic = ctx->pic_start; pic; pic = pic->next) {
820 if (!pic->input_available)
826 if (last_pic->type == PICTURE_TYPE_B) {
827 // Some fixing up is required. Change the type of this
828 // picture to P, then modify preceding B references which
829 // point beyond it to point at it instead.
831 last_pic->type = PICTURE_TYPE_P;
832 last_pic->encode_order = last_pic->refs[1]->encode_order;
834 for (pic = ctx->pic_start; pic != last_pic; pic = pic->next) {
835 if (pic->type == PICTURE_TYPE_B &&
836 pic->refs[1] == last_pic->refs[1])
837 pic->refs[1] = last_pic;
840 last_pic->nb_refs = 1;
841 last_pic->refs[1] = NULL;
843 // We can use the current structure (no references point
844 // beyond the end), but there are unused pics to discard.
847 // Discard all following pics, they will never be used.
848 for (pic = last_pic->next; pic; pic = next) {
850 vaapi_encode_free(avctx, pic);
853 last_pic->next = NULL;
854 ctx->pic_end = last_pic;
857 // Input is available for all pictures, so we don't need to
861 av_log(avctx, AV_LOG_DEBUG, "Pictures ending truncated GOP:");
862 for (pic = ctx->pic_start; pic; pic = pic->next) {
863 av_log(avctx, AV_LOG_DEBUG, " %s (%"PRId64"/%"PRId64")",
864 picture_type_name[pic->type],
865 pic->display_order, pic->encode_order);
867 av_log(avctx, AV_LOG_DEBUG, "\n");
872 static int vaapi_encode_clear_old(AVCodecContext *avctx)
874 VAAPIEncodeContext *ctx = avctx->priv_data;
875 VAAPIEncodePicture *pic, *old;
878 while (ctx->pic_start != ctx->pic_end) {
879 old = ctx->pic_start;
880 if (old->encode_order > ctx->output_order)
883 for (pic = old->next; pic; pic = pic->next) {
884 if (pic->encode_complete)
886 for (i = 0; i < pic->nb_refs; i++) {
887 if (pic->refs[i] == old) {
888 // We still need this picture because it's referred to
889 // directly by a later one, so it and all following
890 // pictures have to stay.
896 pic = ctx->pic_start;
897 ctx->pic_start = pic->next;
898 vaapi_encode_free(avctx, pic);
904 int ff_vaapi_encode2(AVCodecContext *avctx, AVPacket *pkt,
905 const AVFrame *input_image, int *got_packet)
907 VAAPIEncodeContext *ctx = avctx->priv_data;
908 VAAPIEncodePicture *pic;
912 av_log(avctx, AV_LOG_DEBUG, "Encode frame: %ux%u (%"PRId64").\n",
913 input_image->width, input_image->height, input_image->pts);
915 if (input_image->pict_type == AV_PICTURE_TYPE_I) {
916 err = vaapi_encode_truncate_gop(avctx);
922 err = vaapi_encode_get_next(avctx, &pic);
924 av_log(avctx, AV_LOG_ERROR, "Input setup failed: %d.\n", err);
928 pic->input_image = av_frame_alloc();
929 if (!pic->input_image) {
930 err = AVERROR(ENOMEM);
933 err = av_frame_ref(pic->input_image, input_image);
936 pic->input_surface = (VASurfaceID)(uintptr_t)input_image->data[3];
937 pic->pts = input_image->pts;
939 if (ctx->input_order == 0)
940 ctx->first_pts = pic->pts;
941 if (ctx->input_order == ctx->decode_delay)
942 ctx->dts_pts_diff = pic->pts - ctx->first_pts;
943 if (ctx->output_delay > 0)
944 ctx->ts_ring[ctx->input_order % (3 * ctx->output_delay)] = pic->pts;
946 pic->input_available = 1;
949 if (!ctx->end_of_stream) {
950 err = vaapi_encode_truncate_gop(avctx);
953 ctx->end_of_stream = 1;
959 av_assert0(ctx->output_order + ctx->output_delay + 1 == ctx->input_order);
961 for (pic = ctx->pic_start; pic; pic = pic->next)
962 if (pic->encode_order == ctx->output_order)
965 // pic can be null here if we don't have a specific target in this
966 // iteration. We might still issue encodes if things can be overlapped,
967 // even though we don't intend to output anything.
969 err = vaapi_encode_step(avctx, pic);
971 av_log(avctx, AV_LOG_ERROR, "Encode failed: %d.\n", err);
978 err = vaapi_encode_output(avctx, pic, pkt);
980 av_log(avctx, AV_LOG_ERROR, "Output failed: %d.\n", err);
984 if (ctx->output_delay == 0) {
986 } else if (ctx->output_order < ctx->decode_delay) {
987 if (ctx->ts_ring[ctx->output_order] < INT64_MIN + ctx->dts_pts_diff)
988 pkt->dts = INT64_MIN;
990 pkt->dts = ctx->ts_ring[ctx->output_order] - ctx->dts_pts_diff;
992 pkt->dts = ctx->ts_ring[(ctx->output_order - ctx->decode_delay) %
993 (3 * ctx->output_delay)];
999 err = vaapi_encode_clear_old(avctx);
1001 av_log(avctx, AV_LOG_ERROR, "List clearing failed: %d.\n", err);
1008 // Unclear what to clean up on failure. There are probably some things we
1009 // could do usefully clean up here, but for now just leave them for uninit()
1014 static av_cold void vaapi_encode_add_global_param(AVCodecContext *avctx,
1015 VAEncMiscParameterBuffer *buffer,
1018 VAAPIEncodeContext *ctx = avctx->priv_data;
1020 av_assert0(ctx->nb_global_params < MAX_GLOBAL_PARAMS);
1022 ctx->global_params [ctx->nb_global_params] = buffer;
1023 ctx->global_params_size[ctx->nb_global_params] = size;
1025 ++ctx->nb_global_params;
1028 typedef struct VAAPIEncodeRTFormat {
1035 } VAAPIEncodeRTFormat;
1037 static const VAAPIEncodeRTFormat vaapi_encode_rt_formats[] = {
1038 { "YUV400", VA_RT_FORMAT_YUV400, 8, 1, },
1039 { "YUV420", VA_RT_FORMAT_YUV420, 8, 3, 1, 1 },
1040 { "YUV422", VA_RT_FORMAT_YUV422, 8, 3, 1, 0 },
1041 { "YUV444", VA_RT_FORMAT_YUV444, 8, 3, 0, 0 },
1042 { "YUV411", VA_RT_FORMAT_YUV411, 8, 3, 2, 0 },
1043 #if VA_CHECK_VERSION(0, 38, 1)
1044 { "YUV420_10", VA_RT_FORMAT_YUV420_10BPP, 10, 3, 1, 1 },
1048 static const VAEntrypoint vaapi_encode_entrypoints_normal[] = {
1049 VAEntrypointEncSlice,
1050 VAEntrypointEncPicture,
1051 #if VA_CHECK_VERSION(0, 39, 2)
1052 VAEntrypointEncSliceLP,
1056 #if VA_CHECK_VERSION(0, 39, 2)
1057 static const VAEntrypoint vaapi_encode_entrypoints_low_power[] = {
1058 VAEntrypointEncSliceLP,
1063 static av_cold int vaapi_encode_profile_entrypoint(AVCodecContext *avctx)
1065 VAAPIEncodeContext *ctx = avctx->priv_data;
1066 VAProfile *va_profiles = NULL;
1067 VAEntrypoint *va_entrypoints = NULL;
1069 const VAEntrypoint *usable_entrypoints;
1070 const VAAPIEncodeProfile *profile;
1071 const AVPixFmtDescriptor *desc;
1072 VAConfigAttrib rt_format_attr;
1073 const VAAPIEncodeRTFormat *rt_format;
1074 const char *profile_string, *entrypoint_string;
1075 int i, j, n, depth, err;
1078 if (ctx->low_power) {
1079 #if VA_CHECK_VERSION(0, 39, 2)
1080 usable_entrypoints = vaapi_encode_entrypoints_low_power;
1082 av_log(avctx, AV_LOG_ERROR, "Low-power encoding is not "
1083 "supported with this VAAPI version.\n");
1084 return AVERROR(EINVAL);
1087 usable_entrypoints = vaapi_encode_entrypoints_normal;
1090 desc = av_pix_fmt_desc_get(ctx->input_frames->sw_format);
1092 av_log(avctx, AV_LOG_ERROR, "Invalid input pixfmt (%d).\n",
1093 ctx->input_frames->sw_format);
1094 return AVERROR(EINVAL);
1096 depth = desc->comp[0].depth;
1097 for (i = 1; i < desc->nb_components; i++) {
1098 if (desc->comp[i].depth != depth) {
1099 av_log(avctx, AV_LOG_ERROR, "Invalid input pixfmt (%s).\n",
1101 return AVERROR(EINVAL);
1104 av_log(avctx, AV_LOG_VERBOSE, "Input surface format is %s.\n",
1107 n = vaMaxNumProfiles(ctx->hwctx->display);
1108 va_profiles = av_malloc_array(n, sizeof(VAProfile));
1110 err = AVERROR(ENOMEM);
1113 vas = vaQueryConfigProfiles(ctx->hwctx->display, va_profiles, &n);
1114 if (vas != VA_STATUS_SUCCESS) {
1115 av_log(avctx, AV_LOG_ERROR, "Failed to query profiles: %d (%s).\n",
1116 vas, vaErrorStr(vas));
1117 err = AVERROR_EXTERNAL;
1121 av_assert0(ctx->codec->profiles);
1122 for (i = 0; (ctx->codec->profiles[i].av_profile !=
1123 FF_PROFILE_UNKNOWN); i++) {
1124 profile = &ctx->codec->profiles[i];
1125 if (depth != profile->depth ||
1126 desc->nb_components != profile->nb_components)
1128 if (desc->nb_components > 1 &&
1129 (desc->log2_chroma_w != profile->log2_chroma_w ||
1130 desc->log2_chroma_h != profile->log2_chroma_h))
1132 if (avctx->profile != profile->av_profile &&
1133 avctx->profile != FF_PROFILE_UNKNOWN)
1136 #if VA_CHECK_VERSION(1, 0, 0)
1137 profile_string = vaProfileStr(profile->va_profile);
1139 profile_string = "(no profile names)";
1142 for (j = 0; j < n; j++) {
1143 if (va_profiles[j] == profile->va_profile)
1147 av_log(avctx, AV_LOG_VERBOSE, "Compatible profile %s (%d) "
1148 "is not supported by driver.\n", profile_string,
1149 profile->va_profile);
1153 ctx->profile = profile;
1156 if (!ctx->profile) {
1157 av_log(avctx, AV_LOG_ERROR, "No usable encoding profile found.\n");
1158 err = AVERROR(ENOSYS);
1162 avctx->profile = profile->av_profile;
1163 ctx->va_profile = profile->va_profile;
1164 av_log(avctx, AV_LOG_VERBOSE, "Using VAAPI profile %s (%d).\n",
1165 profile_string, ctx->va_profile);
1167 n = vaMaxNumEntrypoints(ctx->hwctx->display);
1168 va_entrypoints = av_malloc_array(n, sizeof(VAEntrypoint));
1169 if (!va_entrypoints) {
1170 err = AVERROR(ENOMEM);
1173 vas = vaQueryConfigEntrypoints(ctx->hwctx->display, ctx->va_profile,
1174 va_entrypoints, &n);
1175 if (vas != VA_STATUS_SUCCESS) {
1176 av_log(avctx, AV_LOG_ERROR, "Failed to query entrypoints for "
1177 "profile %s (%d): %d (%s).\n", profile_string,
1178 ctx->va_profile, vas, vaErrorStr(vas));
1179 err = AVERROR_EXTERNAL;
1183 for (i = 0; i < n; i++) {
1184 for (j = 0; usable_entrypoints[j]; j++) {
1185 if (va_entrypoints[i] == usable_entrypoints[j])
1188 if (usable_entrypoints[j])
1192 av_log(avctx, AV_LOG_ERROR, "No usable encoding entrypoint found "
1193 "for profile %s (%d).\n", profile_string, ctx->va_profile);
1194 err = AVERROR(ENOSYS);
1198 ctx->va_entrypoint = va_entrypoints[i];
1199 #if VA_CHECK_VERSION(1, 0, 0)
1200 entrypoint_string = vaEntrypointStr(ctx->va_entrypoint);
1202 entrypoint_string = "(no entrypoint names)";
1204 av_log(avctx, AV_LOG_VERBOSE, "Using VAAPI entrypoint %s (%d).\n",
1205 entrypoint_string, ctx->va_entrypoint);
1207 for (i = 0; i < FF_ARRAY_ELEMS(vaapi_encode_rt_formats); i++) {
1208 rt_format = &vaapi_encode_rt_formats[i];
1209 if (rt_format->depth == depth &&
1210 rt_format->nb_components == profile->nb_components &&
1211 rt_format->log2_chroma_w == profile->log2_chroma_w &&
1212 rt_format->log2_chroma_h == profile->log2_chroma_h)
1215 if (i >= FF_ARRAY_ELEMS(vaapi_encode_rt_formats)) {
1216 av_log(avctx, AV_LOG_ERROR, "No usable render target format "
1217 "found for profile %s (%d) entrypoint %s (%d).\n",
1218 profile_string, ctx->va_profile,
1219 entrypoint_string, ctx->va_entrypoint);
1220 err = AVERROR(ENOSYS);
1224 rt_format_attr = (VAConfigAttrib) { VAConfigAttribRTFormat };
1225 vas = vaGetConfigAttributes(ctx->hwctx->display,
1226 ctx->va_profile, ctx->va_entrypoint,
1227 &rt_format_attr, 1);
1228 if (vas != VA_STATUS_SUCCESS) {
1229 av_log(avctx, AV_LOG_ERROR, "Failed to query RT format "
1230 "config attribute: %d (%s).\n", vas, vaErrorStr(vas));
1231 err = AVERROR_EXTERNAL;
1235 if (rt_format_attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1236 av_log(avctx, AV_LOG_VERBOSE, "RT format config attribute not "
1237 "supported by driver: assuming surface RT format %s "
1238 "is valid.\n", rt_format->name);
1239 } else if (!(rt_format_attr.value & rt_format->value)) {
1240 av_log(avctx, AV_LOG_ERROR, "Surface RT format %s not supported "
1241 "by driver for encoding profile %s (%d) entrypoint %s (%d).\n",
1242 rt_format->name, profile_string, ctx->va_profile,
1243 entrypoint_string, ctx->va_entrypoint);
1244 err = AVERROR(ENOSYS);
1247 av_log(avctx, AV_LOG_VERBOSE, "Using VAAPI render target "
1248 "format %s (%#x).\n", rt_format->name, rt_format->value);
1249 ctx->config_attributes[ctx->nb_config_attributes++] =
1251 .type = VAConfigAttribRTFormat,
1252 .value = rt_format->value,
1258 av_freep(&va_profiles);
1259 av_freep(&va_entrypoints);
1263 static av_cold int vaapi_encode_init_rate_control(AVCodecContext *avctx)
1265 VAAPIEncodeContext *ctx = avctx->priv_data;
1266 int64_t rc_bits_per_second;
1267 int rc_target_percentage;
1269 int64_t hrd_buffer_size;
1270 int64_t hrd_initial_buffer_fullness;
1272 VAConfigAttrib rc_attr = { VAConfigAttribRateControl };
1275 vas = vaGetConfigAttributes(ctx->hwctx->display,
1276 ctx->va_profile, ctx->va_entrypoint,
1278 if (vas != VA_STATUS_SUCCESS) {
1279 av_log(avctx, AV_LOG_ERROR, "Failed to query rate control "
1280 "config attribute: %d (%s).\n", vas, vaErrorStr(vas));
1281 return AVERROR_EXTERNAL;
1284 if (rc_attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1285 av_log(avctx, AV_LOG_VERBOSE, "Driver does not report any "
1286 "supported rate control modes: assuming constant-quality.\n");
1287 ctx->va_rc_mode = VA_RC_CQP;
1290 if (ctx->codec->flags & FLAG_CONSTANT_QUALITY_ONLY ||
1291 avctx->flags & AV_CODEC_FLAG_QSCALE ||
1292 avctx->bit_rate <= 0) {
1293 if (rc_attr.value & VA_RC_CQP) {
1294 av_log(avctx, AV_LOG_VERBOSE, "Using constant-quality mode.\n");
1295 ctx->va_rc_mode = VA_RC_CQP;
1296 if (avctx->bit_rate > 0 || avctx->rc_max_rate > 0) {
1297 av_log(avctx, AV_LOG_WARNING, "Bitrate target parameters "
1298 "ignored in constant-quality mode.\n");
1302 av_log(avctx, AV_LOG_ERROR, "Driver does not support "
1303 "constant-quality mode (%#x).\n", rc_attr.value);
1304 return AVERROR(EINVAL);
1308 if (!(rc_attr.value & (VA_RC_CBR | VA_RC_VBR))) {
1309 av_log(avctx, AV_LOG_ERROR, "Driver does not support any "
1310 "bitrate-targetted rate control modes.\n");
1311 return AVERROR(EINVAL);
1314 if (avctx->rc_buffer_size)
1315 hrd_buffer_size = avctx->rc_buffer_size;
1316 else if (avctx->rc_max_rate > 0)
1317 hrd_buffer_size = avctx->rc_max_rate;
1319 hrd_buffer_size = avctx->bit_rate;
1320 if (avctx->rc_initial_buffer_occupancy) {
1321 if (avctx->rc_initial_buffer_occupancy > hrd_buffer_size) {
1322 av_log(avctx, AV_LOG_ERROR, "Invalid RC buffer settings: "
1323 "must have initial buffer size (%d) < "
1324 "buffer size (%"PRId64").\n",
1325 avctx->rc_initial_buffer_occupancy, hrd_buffer_size);
1326 return AVERROR(EINVAL);
1328 hrd_initial_buffer_fullness = avctx->rc_initial_buffer_occupancy;
1330 hrd_initial_buffer_fullness = hrd_buffer_size * 3 / 4;
1333 if (avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate) {
1334 av_log(avctx, AV_LOG_ERROR, "Invalid bitrate settings: must have "
1335 "bitrate (%"PRId64") <= maxrate (%"PRId64").\n",
1336 avctx->bit_rate, avctx->rc_max_rate);
1337 return AVERROR(EINVAL);
1340 if (avctx->rc_max_rate > avctx->bit_rate) {
1341 if (!(rc_attr.value & VA_RC_VBR)) {
1342 av_log(avctx, AV_LOG_WARNING, "Driver does not support "
1343 "VBR mode (%#x), using CBR mode instead.\n",
1345 ctx->va_rc_mode = VA_RC_CBR;
1347 rc_bits_per_second = avctx->bit_rate;
1348 rc_target_percentage = 100;
1350 ctx->va_rc_mode = VA_RC_VBR;
1352 rc_bits_per_second = avctx->rc_max_rate;
1353 rc_target_percentage = (avctx->bit_rate * 100) /
1357 } else if (avctx->rc_max_rate == avctx->bit_rate) {
1358 if (!(rc_attr.value & VA_RC_CBR)) {
1359 av_log(avctx, AV_LOG_WARNING, "Driver does not support "
1360 "CBR mode (%#x), using VBR mode instead.\n",
1362 ctx->va_rc_mode = VA_RC_VBR;
1364 ctx->va_rc_mode = VA_RC_CBR;
1367 rc_bits_per_second = avctx->bit_rate;
1368 rc_target_percentage = 100;
1371 if (rc_attr.value & VA_RC_VBR) {
1372 ctx->va_rc_mode = VA_RC_VBR;
1374 // We only have a target bitrate, but VAAPI requires that a
1375 // maximum rate be supplied as well. Since the user has
1376 // offered no particular constraint, arbitrarily pick a
1377 // maximum rate of double the target rate.
1378 rc_bits_per_second = 2 * avctx->bit_rate;
1379 rc_target_percentage = 50;
1381 ctx->va_rc_mode = VA_RC_CBR;
1383 rc_bits_per_second = avctx->bit_rate;
1384 rc_target_percentage = 100;
1388 rc_window_size = (hrd_buffer_size * 1000) / rc_bits_per_second;
1390 av_log(avctx, AV_LOG_VERBOSE, "RC mode: %s, %d%% of %"PRId64" bps "
1391 "over %d ms.\n", ctx->va_rc_mode == VA_RC_VBR ? "VBR" : "CBR",
1392 rc_target_percentage, rc_bits_per_second, rc_window_size);
1393 av_log(avctx, AV_LOG_VERBOSE, "RC buffer: %"PRId64" bits, "
1394 "initial fullness %"PRId64" bits.\n",
1395 hrd_buffer_size, hrd_initial_buffer_fullness);
1397 if (rc_bits_per_second > UINT32_MAX ||
1398 hrd_buffer_size > UINT32_MAX ||
1399 hrd_initial_buffer_fullness > UINT32_MAX) {
1400 av_log(avctx, AV_LOG_ERROR, "RC parameters of 2^32 or "
1401 "greater are not supported by VAAPI.\n");
1402 return AVERROR(EINVAL);
1405 ctx->va_bit_rate = rc_bits_per_second;
1407 ctx->config_attributes[ctx->nb_config_attributes++] =
1409 .type = VAConfigAttribRateControl,
1410 .value = ctx->va_rc_mode,
1413 ctx->rc_params.misc.type = VAEncMiscParameterTypeRateControl;
1414 ctx->rc_params.rc = (VAEncMiscParameterRateControl) {
1415 .bits_per_second = rc_bits_per_second,
1416 .target_percentage = rc_target_percentage,
1417 .window_size = rc_window_size,
1419 .min_qp = (avctx->qmin > 0 ? avctx->qmin : 0),
1420 .basic_unit_size = 0,
1421 #if VA_CHECK_VERSION(1, 1, 0)
1422 .max_qp = (avctx->qmax > 0 ? avctx->qmax : 0),
1425 vaapi_encode_add_global_param(avctx, &ctx->rc_params.misc,
1426 sizeof(ctx->rc_params));
1428 ctx->hrd_params.misc.type = VAEncMiscParameterTypeHRD;
1429 ctx->hrd_params.hrd = (VAEncMiscParameterHRD) {
1430 .initial_buffer_fullness = hrd_initial_buffer_fullness,
1431 .buffer_size = hrd_buffer_size,
1433 vaapi_encode_add_global_param(avctx, &ctx->hrd_params.misc,
1434 sizeof(ctx->hrd_params));
1436 if (avctx->framerate.num > 0 && avctx->framerate.den > 0)
1437 av_reduce(&fr_num, &fr_den,
1438 avctx->framerate.num, avctx->framerate.den, 65535);
1440 av_reduce(&fr_num, &fr_den,
1441 avctx->time_base.den, avctx->time_base.num, 65535);
1443 ctx->fr_params.misc.type = VAEncMiscParameterTypeFrameRate;
1444 ctx->fr_params.fr.framerate = (unsigned int)fr_den << 16 | fr_num;
1446 #if VA_CHECK_VERSION(0, 40, 0)
1447 vaapi_encode_add_global_param(avctx, &ctx->fr_params.misc,
1448 sizeof(ctx->fr_params));
1454 static av_cold int vaapi_encode_init_gop_structure(AVCodecContext *avctx)
1456 VAAPIEncodeContext *ctx = avctx->priv_data;
1458 VAConfigAttrib attr = { VAConfigAttribEncMaxRefFrames };
1459 uint32_t ref_l0, ref_l1;
1461 vas = vaGetConfigAttributes(ctx->hwctx->display,
1465 if (vas != VA_STATUS_SUCCESS) {
1466 av_log(avctx, AV_LOG_ERROR, "Failed to query reference frames "
1467 "attribute: %d (%s).\n", vas, vaErrorStr(vas));
1468 return AVERROR_EXTERNAL;
1471 if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1472 ref_l0 = ref_l1 = 0;
1474 ref_l0 = attr.value & 0xffff;
1475 ref_l1 = attr.value >> 16 & 0xffff;
1478 if (avctx->gop_size <= 1) {
1479 av_log(avctx, AV_LOG_VERBOSE, "Using intra frames only.\n");
1481 } else if (ref_l0 < 1) {
1482 av_log(avctx, AV_LOG_ERROR, "Driver does not support any "
1483 "reference frames.\n");
1484 return AVERROR(EINVAL);
1485 } else if (ref_l1 < 1 || avctx->max_b_frames < 1) {
1486 av_log(avctx, AV_LOG_VERBOSE, "Using intra and P-frames "
1487 "(supported references: %d / %d).\n", ref_l0, ref_l1);
1488 ctx->gop_size = avctx->gop_size;
1489 ctx->p_per_i = INT_MAX;
1492 av_log(avctx, AV_LOG_VERBOSE, "Using intra, P- and B-frames "
1493 "(supported references: %d / %d).\n", ref_l0, ref_l1);
1494 ctx->gop_size = avctx->gop_size;
1495 ctx->p_per_i = INT_MAX;
1496 ctx->b_per_p = avctx->max_b_frames;
1502 static av_cold int vaapi_encode_init_slice_structure(AVCodecContext *avctx)
1504 VAAPIEncodeContext *ctx = avctx->priv_data;
1505 VAConfigAttrib attr[2] = { { VAConfigAttribEncMaxSlices },
1506 { VAConfigAttribEncSliceStructure } };
1508 uint32_t max_slices, slice_structure;
1511 if (!(ctx->codec->flags & FLAG_SLICE_CONTROL)) {
1512 if (avctx->slices > 0) {
1513 av_log(avctx, AV_LOG_WARNING, "Multiple slices were requested "
1514 "but this codec does not support controlling slices.\n");
1519 ctx->slice_block_rows = (avctx->height + ctx->slice_block_height - 1) /
1520 ctx->slice_block_height;
1521 ctx->slice_block_cols = (avctx->width + ctx->slice_block_width - 1) /
1522 ctx->slice_block_width;
1524 if (avctx->slices <= 1) {
1526 ctx->slice_size = ctx->slice_block_rows;
1530 vas = vaGetConfigAttributes(ctx->hwctx->display,
1533 attr, FF_ARRAY_ELEMS(attr));
1534 if (vas != VA_STATUS_SUCCESS) {
1535 av_log(avctx, AV_LOG_ERROR, "Failed to query slice "
1536 "attributes: %d (%s).\n", vas, vaErrorStr(vas));
1537 return AVERROR_EXTERNAL;
1539 max_slices = attr[0].value;
1540 slice_structure = attr[1].value;
1541 if (max_slices == VA_ATTRIB_NOT_SUPPORTED ||
1542 slice_structure == VA_ATTRIB_NOT_SUPPORTED) {
1543 av_log(avctx, AV_LOG_ERROR, "Driver does not support encoding "
1544 "pictures as multiple slices.\n.");
1545 return AVERROR(EINVAL);
1548 // For fixed-size slices currently we only support whole rows, making
1549 // rectangular slices. This could be extended to arbitrary runs of
1550 // blocks, but since slices tend to be a conformance requirement and
1551 // most cases (such as broadcast or bluray) want rectangular slices
1552 // only it would need to be gated behind another option.
1553 if (avctx->slices > ctx->slice_block_rows) {
1554 av_log(avctx, AV_LOG_WARNING, "Not enough rows to use "
1555 "configured number of slices (%d < %d); using "
1556 "maximum.\n", ctx->slice_block_rows, avctx->slices);
1557 req_slices = ctx->slice_block_rows;
1559 req_slices = avctx->slices;
1561 if (slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_ROWS ||
1562 slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_MACROBLOCKS) {
1563 ctx->nb_slices = req_slices;
1564 ctx->slice_size = ctx->slice_block_rows / ctx->nb_slices;
1565 } else if (slice_structure & VA_ENC_SLICE_STRUCTURE_POWER_OF_TWO_ROWS) {
1567 for (k = 1;; k *= 2) {
1568 if (2 * k * (req_slices - 1) + 1 >= ctx->slice_block_rows)
1571 ctx->nb_slices = (ctx->slice_block_rows + k - 1) / k;
1572 ctx->slice_size = k;
1573 #if VA_CHECK_VERSION(1, 0, 0)
1574 } else if (slice_structure & VA_ENC_SLICE_STRUCTURE_EQUAL_ROWS) {
1575 ctx->nb_slices = ctx->slice_block_rows;
1576 ctx->slice_size = 1;
1579 av_log(avctx, AV_LOG_ERROR, "Driver does not support any usable "
1580 "slice structure modes (%#x).\n", slice_structure);
1581 return AVERROR(EINVAL);
1584 if (ctx->nb_slices > avctx->slices) {
1585 av_log(avctx, AV_LOG_WARNING, "Slice count rounded up to "
1586 "%d (from %d) due to driver constraints on slice "
1587 "structure.\n", ctx->nb_slices, avctx->slices);
1589 if (ctx->nb_slices > max_slices) {
1590 av_log(avctx, AV_LOG_ERROR, "Driver does not support "
1591 "encoding with %d slices (max %"PRIu32").\n",
1592 ctx->nb_slices, max_slices);
1593 return AVERROR(EINVAL);
1596 av_log(avctx, AV_LOG_VERBOSE, "Encoding pictures with %d slices "
1597 "(default size %d block rows).\n",
1598 ctx->nb_slices, ctx->slice_size);
1602 static av_cold int vaapi_encode_init_packed_headers(AVCodecContext *avctx)
1604 VAAPIEncodeContext *ctx = avctx->priv_data;
1606 VAConfigAttrib attr = { VAConfigAttribEncPackedHeaders };
1608 vas = vaGetConfigAttributes(ctx->hwctx->display,
1612 if (vas != VA_STATUS_SUCCESS) {
1613 av_log(avctx, AV_LOG_ERROR, "Failed to query packed headers "
1614 "attribute: %d (%s).\n", vas, vaErrorStr(vas));
1615 return AVERROR_EXTERNAL;
1618 if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1619 if (ctx->desired_packed_headers) {
1620 av_log(avctx, AV_LOG_WARNING, "Driver does not support any "
1621 "packed headers (wanted %#x).\n",
1622 ctx->desired_packed_headers);
1624 av_log(avctx, AV_LOG_VERBOSE, "Driver does not support any "
1625 "packed headers (none wanted).\n");
1627 ctx->va_packed_headers = 0;
1629 if (ctx->desired_packed_headers & ~attr.value) {
1630 av_log(avctx, AV_LOG_WARNING, "Driver does not support some "
1631 "wanted packed headers (wanted %#x, found %#x).\n",
1632 ctx->desired_packed_headers, attr.value);
1634 av_log(avctx, AV_LOG_VERBOSE, "All wanted packed headers "
1635 "available (wanted %#x, found %#x).\n",
1636 ctx->desired_packed_headers, attr.value);
1638 ctx->va_packed_headers = ctx->desired_packed_headers & attr.value;
1641 if (ctx->va_packed_headers) {
1642 ctx->config_attributes[ctx->nb_config_attributes++] =
1644 .type = VAConfigAttribEncPackedHeaders,
1645 .value = ctx->va_packed_headers,
1649 if ( (ctx->desired_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE) &&
1650 !(ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE) &&
1651 (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER)) {
1652 av_log(avctx, AV_LOG_WARNING, "Driver does not support packed "
1653 "sequence headers, but a global header is requested.\n");
1654 av_log(avctx, AV_LOG_WARNING, "No global header will be written: "
1655 "this may result in a stream which is not usable for some "
1656 "purposes (e.g. not muxable to some containers).\n");
1662 static av_cold int vaapi_encode_init_quality(AVCodecContext *avctx)
1664 #if VA_CHECK_VERSION(0, 36, 0)
1665 VAAPIEncodeContext *ctx = avctx->priv_data;
1667 VAConfigAttrib attr = { VAConfigAttribEncQualityRange };
1668 int quality = avctx->compression_level;
1670 vas = vaGetConfigAttributes(ctx->hwctx->display,
1674 if (vas != VA_STATUS_SUCCESS) {
1675 av_log(avctx, AV_LOG_ERROR, "Failed to query quality "
1676 "config attribute: %d (%s).\n", vas, vaErrorStr(vas));
1677 return AVERROR_EXTERNAL;
1680 if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1682 av_log(avctx, AV_LOG_WARNING, "Quality attribute is not "
1683 "supported: will use default quality level.\n");
1686 if (quality > attr.value) {
1687 av_log(avctx, AV_LOG_WARNING, "Invalid quality level: "
1688 "valid range is 0-%d, using %d.\n",
1689 attr.value, attr.value);
1690 quality = attr.value;
1693 ctx->quality_params.misc.type = VAEncMiscParameterTypeQualityLevel;
1694 ctx->quality_params.quality.quality_level = quality;
1696 vaapi_encode_add_global_param(avctx, &ctx->quality_params.misc,
1697 sizeof(ctx->quality_params));
1700 av_log(avctx, AV_LOG_WARNING, "The encode quality option is "
1701 "not supported with this VAAPI version.\n");
1707 static void vaapi_encode_free_output_buffer(void *opaque,
1710 AVCodecContext *avctx = opaque;
1711 VAAPIEncodeContext *ctx = avctx->priv_data;
1712 VABufferID buffer_id;
1714 buffer_id = (VABufferID)(uintptr_t)data;
1716 vaDestroyBuffer(ctx->hwctx->display, buffer_id);
1718 av_log(avctx, AV_LOG_DEBUG, "Freed output buffer %#x\n", buffer_id);
1721 static AVBufferRef *vaapi_encode_alloc_output_buffer(void *opaque,
1724 AVCodecContext *avctx = opaque;
1725 VAAPIEncodeContext *ctx = avctx->priv_data;
1726 VABufferID buffer_id;
1730 // The output buffer size is fixed, so it needs to be large enough
1731 // to hold the largest possible compressed frame. We assume here
1732 // that the uncompressed frame plus some header data is an upper
1734 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
1735 VAEncCodedBufferType,
1736 3 * ctx->surface_width * ctx->surface_height +
1737 (1 << 16), 1, 0, &buffer_id);
1738 if (vas != VA_STATUS_SUCCESS) {
1739 av_log(avctx, AV_LOG_ERROR, "Failed to create bitstream "
1740 "output buffer: %d (%s).\n", vas, vaErrorStr(vas));
1744 av_log(avctx, AV_LOG_DEBUG, "Allocated output buffer %#x\n", buffer_id);
1746 ref = av_buffer_create((uint8_t*)(uintptr_t)buffer_id,
1748 &vaapi_encode_free_output_buffer,
1749 avctx, AV_BUFFER_FLAG_READONLY);
1751 vaDestroyBuffer(ctx->hwctx->display, buffer_id);
1758 static av_cold int vaapi_encode_create_recon_frames(AVCodecContext *avctx)
1760 VAAPIEncodeContext *ctx = avctx->priv_data;
1761 AVVAAPIHWConfig *hwconfig = NULL;
1762 AVHWFramesConstraints *constraints = NULL;
1763 enum AVPixelFormat recon_format;
1766 hwconfig = av_hwdevice_hwconfig_alloc(ctx->device_ref);
1768 err = AVERROR(ENOMEM);
1771 hwconfig->config_id = ctx->va_config;
1773 constraints = av_hwdevice_get_hwframe_constraints(ctx->device_ref,
1776 err = AVERROR(ENOMEM);
1780 // Probably we can use the input surface format as the surface format
1781 // of the reconstructed frames. If not, we just pick the first (only?)
1782 // format in the valid list and hope that it all works.
1783 recon_format = AV_PIX_FMT_NONE;
1784 if (constraints->valid_sw_formats) {
1785 for (i = 0; constraints->valid_sw_formats[i] != AV_PIX_FMT_NONE; i++) {
1786 if (ctx->input_frames->sw_format ==
1787 constraints->valid_sw_formats[i]) {
1788 recon_format = ctx->input_frames->sw_format;
1792 if (recon_format == AV_PIX_FMT_NONE) {
1793 // No match. Just use the first in the supported list and
1794 // hope for the best.
1795 recon_format = constraints->valid_sw_formats[0];
1798 // No idea what to use; copy input format.
1799 recon_format = ctx->input_frames->sw_format;
1801 av_log(avctx, AV_LOG_DEBUG, "Using %s as format of "
1802 "reconstructed frames.\n", av_get_pix_fmt_name(recon_format));
1804 if (ctx->surface_width < constraints->min_width ||
1805 ctx->surface_height < constraints->min_height ||
1806 ctx->surface_width > constraints->max_width ||
1807 ctx->surface_height > constraints->max_height) {
1808 av_log(avctx, AV_LOG_ERROR, "Hardware does not support encoding at "
1809 "size %dx%d (constraints: width %d-%d height %d-%d).\n",
1810 ctx->surface_width, ctx->surface_height,
1811 constraints->min_width, constraints->max_width,
1812 constraints->min_height, constraints->max_height);
1813 err = AVERROR(EINVAL);
1817 av_freep(&hwconfig);
1818 av_hwframe_constraints_free(&constraints);
1820 ctx->recon_frames_ref = av_hwframe_ctx_alloc(ctx->device_ref);
1821 if (!ctx->recon_frames_ref) {
1822 err = AVERROR(ENOMEM);
1825 ctx->recon_frames = (AVHWFramesContext*)ctx->recon_frames_ref->data;
1827 ctx->recon_frames->format = AV_PIX_FMT_VAAPI;
1828 ctx->recon_frames->sw_format = recon_format;
1829 ctx->recon_frames->width = ctx->surface_width;
1830 ctx->recon_frames->height = ctx->surface_height;
1831 // At most three IDR/I/P frames and two runs of B frames can be in
1832 // flight at any one time.
1833 ctx->recon_frames->initial_pool_size = 3 + 2 * ctx->b_per_p;
1835 err = av_hwframe_ctx_init(ctx->recon_frames_ref);
1837 av_log(avctx, AV_LOG_ERROR, "Failed to initialise reconstructed "
1838 "frame context: %d.\n", err);
1844 av_freep(&hwconfig);
1845 av_hwframe_constraints_free(&constraints);
1849 av_cold int ff_vaapi_encode_init(AVCodecContext *avctx)
1851 VAAPIEncodeContext *ctx = avctx->priv_data;
1852 AVVAAPIFramesContext *recon_hwctx = NULL;
1856 if (!avctx->hw_frames_ctx) {
1857 av_log(avctx, AV_LOG_ERROR, "A hardware frames reference is "
1858 "required to associate the encoding device.\n");
1859 return AVERROR(EINVAL);
1862 ctx->va_config = VA_INVALID_ID;
1863 ctx->va_context = VA_INVALID_ID;
1865 ctx->input_frames_ref = av_buffer_ref(avctx->hw_frames_ctx);
1866 if (!ctx->input_frames_ref) {
1867 err = AVERROR(ENOMEM);
1870 ctx->input_frames = (AVHWFramesContext*)ctx->input_frames_ref->data;
1872 ctx->device_ref = av_buffer_ref(ctx->input_frames->device_ref);
1873 if (!ctx->device_ref) {
1874 err = AVERROR(ENOMEM);
1877 ctx->device = (AVHWDeviceContext*)ctx->device_ref->data;
1878 ctx->hwctx = ctx->device->hwctx;
1880 err = vaapi_encode_profile_entrypoint(avctx);
1884 err = vaapi_encode_init_rate_control(avctx);
1888 err = vaapi_encode_init_gop_structure(avctx);
1892 err = vaapi_encode_init_slice_structure(avctx);
1896 err = vaapi_encode_init_packed_headers(avctx);
1900 if (avctx->compression_level >= 0) {
1901 err = vaapi_encode_init_quality(avctx);
1906 vas = vaCreateConfig(ctx->hwctx->display,
1907 ctx->va_profile, ctx->va_entrypoint,
1908 ctx->config_attributes, ctx->nb_config_attributes,
1910 if (vas != VA_STATUS_SUCCESS) {
1911 av_log(avctx, AV_LOG_ERROR, "Failed to create encode pipeline "
1912 "configuration: %d (%s).\n", vas, vaErrorStr(vas));
1917 err = vaapi_encode_create_recon_frames(avctx);
1921 recon_hwctx = ctx->recon_frames->hwctx;
1922 vas = vaCreateContext(ctx->hwctx->display, ctx->va_config,
1923 ctx->surface_width, ctx->surface_height,
1925 recon_hwctx->surface_ids,
1926 recon_hwctx->nb_surfaces,
1928 if (vas != VA_STATUS_SUCCESS) {
1929 av_log(avctx, AV_LOG_ERROR, "Failed to create encode pipeline "
1930 "context: %d (%s).\n", vas, vaErrorStr(vas));
1935 ctx->output_buffer_pool =
1936 av_buffer_pool_init2(sizeof(VABufferID), avctx,
1937 &vaapi_encode_alloc_output_buffer, NULL);
1938 if (!ctx->output_buffer_pool) {
1939 err = AVERROR(ENOMEM);
1943 if (ctx->codec->configure) {
1944 err = ctx->codec->configure(avctx);
1949 ctx->input_order = 0;
1950 ctx->output_delay = ctx->b_per_p;
1951 ctx->decode_delay = 1;
1952 ctx->output_order = - ctx->output_delay - 1;
1954 if (ctx->codec->sequence_params_size > 0) {
1955 ctx->codec_sequence_params =
1956 av_mallocz(ctx->codec->sequence_params_size);
1957 if (!ctx->codec_sequence_params) {
1958 err = AVERROR(ENOMEM);
1962 if (ctx->codec->picture_params_size > 0) {
1963 ctx->codec_picture_params =
1964 av_mallocz(ctx->codec->picture_params_size);
1965 if (!ctx->codec_picture_params) {
1966 err = AVERROR(ENOMEM);
1971 if (ctx->codec->init_sequence_params) {
1972 err = ctx->codec->init_sequence_params(avctx);
1974 av_log(avctx, AV_LOG_ERROR, "Codec sequence initialisation "
1975 "failed: %d.\n", err);
1980 // This should be configurable somehow. (Needs testing on a machine
1981 // where it actually overlaps properly, though.)
1982 ctx->issue_mode = ISSUE_MODE_MAXIMISE_THROUGHPUT;
1984 if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE &&
1985 ctx->codec->write_sequence_header &&
1986 avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
1987 char data[MAX_PARAM_BUFFER_SIZE];
1988 size_t bit_len = 8 * sizeof(data);
1990 err = ctx->codec->write_sequence_header(avctx, data, &bit_len);
1992 av_log(avctx, AV_LOG_ERROR, "Failed to write sequence header "
1993 "for extradata: %d.\n", err);
1996 avctx->extradata_size = (bit_len + 7) / 8;
1997 avctx->extradata = av_mallocz(avctx->extradata_size +
1998 AV_INPUT_BUFFER_PADDING_SIZE);
1999 if (!avctx->extradata) {
2000 err = AVERROR(ENOMEM);
2003 memcpy(avctx->extradata, data, avctx->extradata_size);
2010 ff_vaapi_encode_close(avctx);
2014 av_cold int ff_vaapi_encode_close(AVCodecContext *avctx)
2016 VAAPIEncodeContext *ctx = avctx->priv_data;
2017 VAAPIEncodePicture *pic, *next;
2019 for (pic = ctx->pic_start; pic; pic = next) {
2021 vaapi_encode_free(avctx, pic);
2024 av_buffer_pool_uninit(&ctx->output_buffer_pool);
2026 if (ctx->va_context != VA_INVALID_ID) {
2027 vaDestroyContext(ctx->hwctx->display, ctx->va_context);
2028 ctx->va_context = VA_INVALID_ID;
2031 if (ctx->va_config != VA_INVALID_ID) {
2032 vaDestroyConfig(ctx->hwctx->display, ctx->va_config);
2033 ctx->va_config = VA_INVALID_ID;
2036 av_freep(&ctx->codec_sequence_params);
2037 av_freep(&ctx->codec_picture_params);
2039 av_buffer_unref(&ctx->recon_frames_ref);
2040 av_buffer_unref(&ctx->input_frames_ref);
2041 av_buffer_unref(&ctx->device_ref);