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 "config_components.h"
20
21 #include <inttypes.h>
22 #include <string.h>
23
24 #include "libavutil/avassert.h"
25 #include "libavutil/common.h"
26 #include "libavutil/internal.h"
27 #include "libavutil/log.h"
28 #include "libavutil/pixdesc.h"
29
30 #include "vaapi_encode.h"
31 #include "encode.h"
32 #include "avcodec.h"
33
34 const AVCodecHWConfigInternal *const ff_vaapi_encode_hw_configs[] = {
35 HW_CONFIG_ENCODER_FRAMES(VAAPI, VAAPI),
36 NULL,
37 };
38
39 static const char * const picture_type_name[] = { "IDR", "I", "P", "B" };
40
vaapi_encode_make_packed_header(AVCodecContext * avctx,VAAPIEncodePicture * pic,int type,char * data,size_t bit_len)41 static int vaapi_encode_make_packed_header(AVCodecContext *avctx,
42 VAAPIEncodePicture *pic,
43 int type, char *data, size_t bit_len)
44 {
45 VAAPIEncodeContext *ctx = avctx->priv_data;
46 VAStatus vas;
47 VABufferID param_buffer, data_buffer;
48 VABufferID *tmp;
49 VAEncPackedHeaderParameterBuffer params = {
50 .type = type,
51 .bit_length = bit_len,
52 .has_emulation_bytes = 1,
53 };
54
55 tmp = av_realloc_array(pic->param_buffers, sizeof(*tmp), pic->nb_param_buffers + 2);
56 if (!tmp)
57 return AVERROR(ENOMEM);
58 pic->param_buffers = tmp;
59
60 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
61 VAEncPackedHeaderParameterBufferType,
62 sizeof(params), 1, ¶ms, ¶m_buffer);
63 if (vas != VA_STATUS_SUCCESS) {
64 av_log(avctx, AV_LOG_ERROR, "Failed to create parameter buffer "
65 "for packed header (type %d): %d (%s).\n",
66 type, vas, vaErrorStr(vas));
67 return AVERROR(EIO);
68 }
69 pic->param_buffers[pic->nb_param_buffers++] = param_buffer;
70
71 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
72 VAEncPackedHeaderDataBufferType,
73 (bit_len + 7) / 8, 1, data, &data_buffer);
74 if (vas != VA_STATUS_SUCCESS) {
75 av_log(avctx, AV_LOG_ERROR, "Failed to create data buffer "
76 "for packed header (type %d): %d (%s).\n",
77 type, vas, vaErrorStr(vas));
78 return AVERROR(EIO);
79 }
80 pic->param_buffers[pic->nb_param_buffers++] = data_buffer;
81
82 av_log(avctx, AV_LOG_DEBUG, "Packed header buffer (%d) is %#x/%#x "
83 "(%zu bits).\n", type, param_buffer, data_buffer, bit_len);
84 return 0;
85 }
86
vaapi_encode_make_param_buffer(AVCodecContext * avctx,VAAPIEncodePicture * pic,int type,char * data,size_t len)87 static int vaapi_encode_make_param_buffer(AVCodecContext *avctx,
88 VAAPIEncodePicture *pic,
89 int type, char *data, size_t len)
90 {
91 VAAPIEncodeContext *ctx = avctx->priv_data;
92 VAStatus vas;
93 VABufferID *tmp;
94 VABufferID buffer;
95
96 tmp = av_realloc_array(pic->param_buffers, sizeof(*tmp), pic->nb_param_buffers + 1);
97 if (!tmp)
98 return AVERROR(ENOMEM);
99 pic->param_buffers = tmp;
100
101 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
102 type, len, 1, data, &buffer);
103 if (vas != VA_STATUS_SUCCESS) {
104 av_log(avctx, AV_LOG_ERROR, "Failed to create parameter buffer "
105 "(type %d): %d (%s).\n", type, vas, vaErrorStr(vas));
106 return AVERROR(EIO);
107 }
108 pic->param_buffers[pic->nb_param_buffers++] = buffer;
109
110 av_log(avctx, AV_LOG_DEBUG, "Param buffer (%d) is %#x.\n",
111 type, buffer);
112 return 0;
113 }
114
vaapi_encode_make_misc_param_buffer(AVCodecContext * avctx,VAAPIEncodePicture * pic,int type,const void * data,size_t len)115 static int vaapi_encode_make_misc_param_buffer(AVCodecContext *avctx,
116 VAAPIEncodePicture *pic,
117 int type,
118 const void *data, size_t len)
119 {
120 // Construct the buffer on the stack - 1KB is much larger than any
121 // current misc parameter buffer type (the largest is EncQuality at
122 // 224 bytes).
123 uint8_t buffer[1024];
124 VAEncMiscParameterBuffer header = {
125 .type = type,
126 };
127 size_t buffer_size = sizeof(header) + len;
128 av_assert0(buffer_size <= sizeof(buffer));
129
130 memcpy(buffer, &header, sizeof(header));
131 memcpy(buffer + sizeof(header), data, len);
132
133 return vaapi_encode_make_param_buffer(avctx, pic,
134 VAEncMiscParameterBufferType,
135 buffer, buffer_size);
136 }
137
vaapi_encode_wait(AVCodecContext * avctx,VAAPIEncodePicture * pic)138 static int vaapi_encode_wait(AVCodecContext *avctx,
139 VAAPIEncodePicture *pic)
140 {
141 VAAPIEncodeContext *ctx = avctx->priv_data;
142 VAStatus vas;
143
144 av_assert0(pic->encode_issued);
145
146 if (pic->encode_complete) {
147 // Already waited for this picture.
148 return 0;
149 }
150
151 av_log(avctx, AV_LOG_DEBUG, "Sync to pic %"PRId64"/%"PRId64" "
152 "(input surface %#x).\n", pic->display_order,
153 pic->encode_order, pic->input_surface);
154
155 #if VA_CHECK_VERSION(1, 9, 0)
156 if (ctx->has_sync_buffer_func) {
157 vas = vaSyncBuffer(ctx->hwctx->display,
158 pic->output_buffer,
159 VA_TIMEOUT_INFINITE);
160 if (vas != VA_STATUS_SUCCESS) {
161 av_log(avctx, AV_LOG_ERROR, "Failed to sync to output buffer completion: "
162 "%d (%s).\n", vas, vaErrorStr(vas));
163 return AVERROR(EIO);
164 }
165 } else
166 #endif
167 { // If vaSyncBuffer is not implemented, try old version API.
168 vas = vaSyncSurface(ctx->hwctx->display, pic->input_surface);
169 if (vas != VA_STATUS_SUCCESS) {
170 av_log(avctx, AV_LOG_ERROR, "Failed to sync to picture completion: "
171 "%d (%s).\n", vas, vaErrorStr(vas));
172 return AVERROR(EIO);
173 }
174 }
175
176 // Input is definitely finished with now.
177 av_frame_free(&pic->input_image);
178
179 pic->encode_complete = 1;
180 return 0;
181 }
182
vaapi_encode_make_row_slice(AVCodecContext * avctx,VAAPIEncodePicture * pic)183 static int vaapi_encode_make_row_slice(AVCodecContext *avctx,
184 VAAPIEncodePicture *pic)
185 {
186 VAAPIEncodeContext *ctx = avctx->priv_data;
187 VAAPIEncodeSlice *slice;
188 int i, rounding;
189
190 for (i = 0; i < pic->nb_slices; i++)
191 pic->slices[i].row_size = ctx->slice_size;
192
193 rounding = ctx->slice_block_rows - ctx->nb_slices * ctx->slice_size;
194 if (rounding > 0) {
195 // Place rounding error at top and bottom of frame.
196 av_assert0(rounding < pic->nb_slices);
197 // Some Intel drivers contain a bug where the encoder will fail
198 // if the last slice is smaller than the one before it. Since
199 // that's straightforward to avoid here, just do so.
200 if (rounding <= 2) {
201 for (i = 0; i < rounding; i++)
202 ++pic->slices[i].row_size;
203 } else {
204 for (i = 0; i < (rounding + 1) / 2; i++)
205 ++pic->slices[pic->nb_slices - i - 1].row_size;
206 for (i = 0; i < rounding / 2; i++)
207 ++pic->slices[i].row_size;
208 }
209 } else if (rounding < 0) {
210 // Remove rounding error from last slice only.
211 av_assert0(rounding < ctx->slice_size);
212 pic->slices[pic->nb_slices - 1].row_size += rounding;
213 }
214
215 for (i = 0; i < pic->nb_slices; i++) {
216 slice = &pic->slices[i];
217 slice->index = i;
218 if (i == 0) {
219 slice->row_start = 0;
220 slice->block_start = 0;
221 } else {
222 const VAAPIEncodeSlice *prev = &pic->slices[i - 1];
223 slice->row_start = prev->row_start + prev->row_size;
224 slice->block_start = prev->block_start + prev->block_size;
225 }
226 slice->block_size = slice->row_size * ctx->slice_block_cols;
227
228 av_log(avctx, AV_LOG_DEBUG, "Slice %d: %d-%d (%d rows), "
229 "%d-%d (%d blocks).\n", i, slice->row_start,
230 slice->row_start + slice->row_size - 1, slice->row_size,
231 slice->block_start, slice->block_start + slice->block_size - 1,
232 slice->block_size);
233 }
234
235 return 0;
236 }
237
vaapi_encode_make_tile_slice(AVCodecContext * avctx,VAAPIEncodePicture * pic)238 static int vaapi_encode_make_tile_slice(AVCodecContext *avctx,
239 VAAPIEncodePicture *pic)
240 {
241 VAAPIEncodeContext *ctx = avctx->priv_data;
242 VAAPIEncodeSlice *slice;
243 int i, j, index;
244
245 for (i = 0; i < ctx->tile_cols; i++) {
246 for (j = 0; j < ctx->tile_rows; j++) {
247 index = j * ctx->tile_cols + i;
248 slice = &pic->slices[index];
249 slice->index = index;
250
251 pic->slices[index].block_start = ctx->col_bd[i] +
252 ctx->row_bd[j] * ctx->slice_block_cols;
253 pic->slices[index].block_size = ctx->row_height[j] * ctx->col_width[i];
254
255 av_log(avctx, AV_LOG_DEBUG, "Slice %2d: (%2d, %2d) start at: %4d "
256 "width:%2d height:%2d (%d blocks).\n", index, ctx->col_bd[i],
257 ctx->row_bd[j], slice->block_start, ctx->col_width[i],
258 ctx->row_height[j], slice->block_size);
259 }
260 }
261
262 return 0;
263 }
264
vaapi_encode_issue(AVCodecContext * avctx,VAAPIEncodePicture * pic)265 static int vaapi_encode_issue(AVCodecContext *avctx,
266 VAAPIEncodePicture *pic)
267 {
268 VAAPIEncodeContext *ctx = avctx->priv_data;
269 VAAPIEncodeSlice *slice;
270 VAStatus vas;
271 int err, i;
272 char data[MAX_PARAM_BUFFER_SIZE];
273 size_t bit_len;
274 av_unused AVFrameSideData *sd;
275
276 av_log(avctx, AV_LOG_DEBUG, "Issuing encode for pic %"PRId64"/%"PRId64" "
277 "as type %s.\n", pic->display_order, pic->encode_order,
278 picture_type_name[pic->type]);
279 if (pic->nb_refs == 0) {
280 av_log(avctx, AV_LOG_DEBUG, "No reference pictures.\n");
281 } else {
282 av_log(avctx, AV_LOG_DEBUG, "Refers to:");
283 for (i = 0; i < pic->nb_refs; i++) {
284 av_log(avctx, AV_LOG_DEBUG, " %"PRId64"/%"PRId64,
285 pic->refs[i]->display_order, pic->refs[i]->encode_order);
286 }
287 av_log(avctx, AV_LOG_DEBUG, ".\n");
288 }
289
290 av_assert0(!pic->encode_issued);
291 for (i = 0; i < pic->nb_refs; i++) {
292 av_assert0(pic->refs[i]);
293 av_assert0(pic->refs[i]->encode_issued);
294 }
295
296 av_log(avctx, AV_LOG_DEBUG, "Input surface is %#x.\n", pic->input_surface);
297
298 pic->recon_image = av_frame_alloc();
299 if (!pic->recon_image) {
300 err = AVERROR(ENOMEM);
301 goto fail;
302 }
303
304 err = av_hwframe_get_buffer(ctx->recon_frames_ref, pic->recon_image, 0);
305 if (err < 0) {
306 err = AVERROR(ENOMEM);
307 goto fail;
308 }
309 pic->recon_surface = (VASurfaceID)(uintptr_t)pic->recon_image->data[3];
310 av_log(avctx, AV_LOG_DEBUG, "Recon surface is %#x.\n", pic->recon_surface);
311
312 pic->output_buffer_ref = av_buffer_pool_get(ctx->output_buffer_pool);
313 if (!pic->output_buffer_ref) {
314 err = AVERROR(ENOMEM);
315 goto fail;
316 }
317 pic->output_buffer = (VABufferID)(uintptr_t)pic->output_buffer_ref->data;
318 av_log(avctx, AV_LOG_DEBUG, "Output buffer is %#x.\n",
319 pic->output_buffer);
320
321 if (ctx->codec->picture_params_size > 0) {
322 pic->codec_picture_params = av_malloc(ctx->codec->picture_params_size);
323 if (!pic->codec_picture_params)
324 goto fail;
325 memcpy(pic->codec_picture_params, ctx->codec_picture_params,
326 ctx->codec->picture_params_size);
327 } else {
328 av_assert0(!ctx->codec_picture_params);
329 }
330
331 pic->nb_param_buffers = 0;
332
333 if (pic->type == PICTURE_TYPE_IDR && ctx->codec->init_sequence_params) {
334 err = vaapi_encode_make_param_buffer(avctx, pic,
335 VAEncSequenceParameterBufferType,
336 ctx->codec_sequence_params,
337 ctx->codec->sequence_params_size);
338 if (err < 0)
339 goto fail;
340 }
341
342 if (pic->type == PICTURE_TYPE_IDR) {
343 for (i = 0; i < ctx->nb_global_params; i++) {
344 err = vaapi_encode_make_misc_param_buffer(avctx, pic,
345 ctx->global_params_type[i],
346 ctx->global_params[i],
347 ctx->global_params_size[i]);
348 if (err < 0)
349 goto fail;
350 }
351 }
352
353 if (ctx->codec->init_picture_params) {
354 err = ctx->codec->init_picture_params(avctx, pic);
355 if (err < 0) {
356 av_log(avctx, AV_LOG_ERROR, "Failed to initialise picture "
357 "parameters: %d.\n", err);
358 goto fail;
359 }
360 err = vaapi_encode_make_param_buffer(avctx, pic,
361 VAEncPictureParameterBufferType,
362 pic->codec_picture_params,
363 ctx->codec->picture_params_size);
364 if (err < 0)
365 goto fail;
366 }
367
368 #if VA_CHECK_VERSION(1, 5, 0)
369 if (ctx->max_frame_size) {
370 err = vaapi_encode_make_misc_param_buffer(avctx, pic,
371 VAEncMiscParameterTypeMaxFrameSize,
372 &ctx->mfs_params,
373 sizeof(ctx->mfs_params));
374 if (err < 0)
375 goto fail;
376 }
377 #endif
378
379 if (pic->type == PICTURE_TYPE_IDR) {
380 if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE &&
381 ctx->codec->write_sequence_header) {
382 bit_len = 8 * sizeof(data);
383 err = ctx->codec->write_sequence_header(avctx, data, &bit_len);
384 if (err < 0) {
385 av_log(avctx, AV_LOG_ERROR, "Failed to write per-sequence "
386 "header: %d.\n", err);
387 goto fail;
388 }
389 err = vaapi_encode_make_packed_header(avctx, pic,
390 ctx->codec->sequence_header_type,
391 data, bit_len);
392 if (err < 0)
393 goto fail;
394 }
395 }
396
397 if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_PICTURE &&
398 ctx->codec->write_picture_header) {
399 bit_len = 8 * sizeof(data);
400 err = ctx->codec->write_picture_header(avctx, pic, data, &bit_len);
401 if (err < 0) {
402 av_log(avctx, AV_LOG_ERROR, "Failed to write per-picture "
403 "header: %d.\n", err);
404 goto fail;
405 }
406 err = vaapi_encode_make_packed_header(avctx, pic,
407 ctx->codec->picture_header_type,
408 data, bit_len);
409 if (err < 0)
410 goto fail;
411 }
412
413 if (ctx->codec->write_extra_buffer) {
414 for (i = 0;; i++) {
415 size_t len = sizeof(data);
416 int type;
417 err = ctx->codec->write_extra_buffer(avctx, pic, i, &type,
418 data, &len);
419 if (err == AVERROR_EOF)
420 break;
421 if (err < 0) {
422 av_log(avctx, AV_LOG_ERROR, "Failed to write extra "
423 "buffer %d: %d.\n", i, err);
424 goto fail;
425 }
426
427 err = vaapi_encode_make_param_buffer(avctx, pic, type,
428 data, len);
429 if (err < 0)
430 goto fail;
431 }
432 }
433
434 if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_MISC &&
435 ctx->codec->write_extra_header) {
436 for (i = 0;; i++) {
437 int type;
438 bit_len = 8 * sizeof(data);
439 err = ctx->codec->write_extra_header(avctx, pic, i, &type,
440 data, &bit_len);
441 if (err == AVERROR_EOF)
442 break;
443 if (err < 0) {
444 av_log(avctx, AV_LOG_ERROR, "Failed to write extra "
445 "header %d: %d.\n", i, err);
446 goto fail;
447 }
448
449 err = vaapi_encode_make_packed_header(avctx, pic, type,
450 data, bit_len);
451 if (err < 0)
452 goto fail;
453 }
454 }
455
456 if (pic->nb_slices == 0)
457 pic->nb_slices = ctx->nb_slices;
458 if (pic->nb_slices > 0) {
459 pic->slices = av_calloc(pic->nb_slices, sizeof(*pic->slices));
460 if (!pic->slices) {
461 err = AVERROR(ENOMEM);
462 goto fail;
463 }
464
465 if (ctx->tile_rows && ctx->tile_cols)
466 vaapi_encode_make_tile_slice(avctx, pic);
467 else
468 vaapi_encode_make_row_slice(avctx, pic);
469 }
470
471 for (i = 0; i < pic->nb_slices; i++) {
472 slice = &pic->slices[i];
473
474 if (ctx->codec->slice_params_size > 0) {
475 slice->codec_slice_params = av_mallocz(ctx->codec->slice_params_size);
476 if (!slice->codec_slice_params) {
477 err = AVERROR(ENOMEM);
478 goto fail;
479 }
480 }
481
482 if (ctx->codec->init_slice_params) {
483 err = ctx->codec->init_slice_params(avctx, pic, slice);
484 if (err < 0) {
485 av_log(avctx, AV_LOG_ERROR, "Failed to initialise slice "
486 "parameters: %d.\n", err);
487 goto fail;
488 }
489 }
490
491 if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SLICE &&
492 ctx->codec->write_slice_header) {
493 bit_len = 8 * sizeof(data);
494 err = ctx->codec->write_slice_header(avctx, pic, slice,
495 data, &bit_len);
496 if (err < 0) {
497 av_log(avctx, AV_LOG_ERROR, "Failed to write per-slice "
498 "header: %d.\n", err);
499 goto fail;
500 }
501 err = vaapi_encode_make_packed_header(avctx, pic,
502 ctx->codec->slice_header_type,
503 data, bit_len);
504 if (err < 0)
505 goto fail;
506 }
507
508 if (ctx->codec->init_slice_params) {
509 err = vaapi_encode_make_param_buffer(avctx, pic,
510 VAEncSliceParameterBufferType,
511 slice->codec_slice_params,
512 ctx->codec->slice_params_size);
513 if (err < 0)
514 goto fail;
515 }
516 }
517
518 #if VA_CHECK_VERSION(1, 0, 0)
519 sd = av_frame_get_side_data(pic->input_image,
520 AV_FRAME_DATA_REGIONS_OF_INTEREST);
521 if (sd && ctx->roi_allowed) {
522 const AVRegionOfInterest *roi;
523 uint32_t roi_size;
524 VAEncMiscParameterBufferROI param_roi;
525 int nb_roi, i, v;
526
527 roi = (const AVRegionOfInterest*)sd->data;
528 roi_size = roi->self_size;
529 av_assert0(roi_size && sd->size % roi_size == 0);
530 nb_roi = sd->size / roi_size;
531 if (nb_roi > ctx->roi_max_regions) {
532 if (!ctx->roi_warned) {
533 av_log(avctx, AV_LOG_WARNING, "More ROIs set than "
534 "supported by driver (%d > %d).\n",
535 nb_roi, ctx->roi_max_regions);
536 ctx->roi_warned = 1;
537 }
538 nb_roi = ctx->roi_max_regions;
539 }
540
541 pic->roi = av_calloc(nb_roi, sizeof(*pic->roi));
542 if (!pic->roi) {
543 err = AVERROR(ENOMEM);
544 goto fail;
545 }
546 // For overlapping regions, the first in the array takes priority.
547 for (i = 0; i < nb_roi; i++) {
548 roi = (const AVRegionOfInterest*)(sd->data + roi_size * i);
549
550 av_assert0(roi->qoffset.den != 0);
551 v = roi->qoffset.num * ctx->roi_quant_range / roi->qoffset.den;
552 av_log(avctx, AV_LOG_DEBUG, "ROI: (%d,%d)-(%d,%d) -> %+d.\n",
553 roi->top, roi->left, roi->bottom, roi->right, v);
554
555 pic->roi[i] = (VAEncROI) {
556 .roi_rectangle = {
557 .x = roi->left,
558 .y = roi->top,
559 .width = roi->right - roi->left,
560 .height = roi->bottom - roi->top,
561 },
562 .roi_value = av_clip_int8(v),
563 };
564 }
565
566 param_roi = (VAEncMiscParameterBufferROI) {
567 .num_roi = nb_roi,
568 .max_delta_qp = INT8_MAX,
569 .min_delta_qp = INT8_MIN,
570 .roi = pic->roi,
571 .roi_flags.bits.roi_value_is_qp_delta = 1,
572 };
573
574 err = vaapi_encode_make_misc_param_buffer(avctx, pic,
575 VAEncMiscParameterTypeROI,
576 ¶m_roi,
577 sizeof(param_roi));
578 if (err < 0)
579 goto fail;
580 }
581 #endif
582
583 vas = vaBeginPicture(ctx->hwctx->display, ctx->va_context,
584 pic->input_surface);
585 if (vas != VA_STATUS_SUCCESS) {
586 av_log(avctx, AV_LOG_ERROR, "Failed to begin picture encode issue: "
587 "%d (%s).\n", vas, vaErrorStr(vas));
588 err = AVERROR(EIO);
589 goto fail_with_picture;
590 }
591
592 vas = vaRenderPicture(ctx->hwctx->display, ctx->va_context,
593 pic->param_buffers, pic->nb_param_buffers);
594 if (vas != VA_STATUS_SUCCESS) {
595 av_log(avctx, AV_LOG_ERROR, "Failed to upload encode parameters: "
596 "%d (%s).\n", vas, vaErrorStr(vas));
597 err = AVERROR(EIO);
598 goto fail_with_picture;
599 }
600
601 vas = vaEndPicture(ctx->hwctx->display, ctx->va_context);
602 if (vas != VA_STATUS_SUCCESS) {
603 av_log(avctx, AV_LOG_ERROR, "Failed to end picture encode issue: "
604 "%d (%s).\n", vas, vaErrorStr(vas));
605 err = AVERROR(EIO);
606 // vaRenderPicture() has been called here, so we should not destroy
607 // the parameter buffers unless separate destruction is required.
608 if (CONFIG_VAAPI_1 || ctx->hwctx->driver_quirks &
609 AV_VAAPI_DRIVER_QUIRK_RENDER_PARAM_BUFFERS)
610 goto fail;
611 else
612 goto fail_at_end;
613 }
614
615 if (CONFIG_VAAPI_1 || ctx->hwctx->driver_quirks &
616 AV_VAAPI_DRIVER_QUIRK_RENDER_PARAM_BUFFERS) {
617 for (i = 0; i < pic->nb_param_buffers; i++) {
618 vas = vaDestroyBuffer(ctx->hwctx->display,
619 pic->param_buffers[i]);
620 if (vas != VA_STATUS_SUCCESS) {
621 av_log(avctx, AV_LOG_ERROR, "Failed to destroy "
622 "param buffer %#x: %d (%s).\n",
623 pic->param_buffers[i], vas, vaErrorStr(vas));
624 // And ignore.
625 }
626 }
627 }
628
629 pic->encode_issued = 1;
630
631 return 0;
632
633 fail_with_picture:
634 vaEndPicture(ctx->hwctx->display, ctx->va_context);
635 fail:
636 for(i = 0; i < pic->nb_param_buffers; i++)
637 vaDestroyBuffer(ctx->hwctx->display, pic->param_buffers[i]);
638 if (pic->slices) {
639 for (i = 0; i < pic->nb_slices; i++)
640 av_freep(&pic->slices[i].codec_slice_params);
641 }
642 fail_at_end:
643 av_freep(&pic->codec_picture_params);
644 av_freep(&pic->param_buffers);
645 av_freep(&pic->slices);
646 av_freep(&pic->roi);
647 av_frame_free(&pic->recon_image);
648 av_buffer_unref(&pic->output_buffer_ref);
649 pic->output_buffer = VA_INVALID_ID;
650 return err;
651 }
652
vaapi_encode_output(AVCodecContext * avctx,VAAPIEncodePicture * pic,AVPacket * pkt)653 static int vaapi_encode_output(AVCodecContext *avctx,
654 VAAPIEncodePicture *pic, AVPacket *pkt)
655 {
656 VAAPIEncodeContext *ctx = avctx->priv_data;
657 VACodedBufferSegment *buf_list, *buf;
658 VAStatus vas;
659 int total_size = 0;
660 uint8_t *ptr;
661 int err;
662
663 err = vaapi_encode_wait(avctx, pic);
664 if (err < 0)
665 return err;
666
667 buf_list = NULL;
668 vas = vaMapBuffer(ctx->hwctx->display, pic->output_buffer,
669 (void**)&buf_list);
670 if (vas != VA_STATUS_SUCCESS) {
671 av_log(avctx, AV_LOG_ERROR, "Failed to map output buffers: "
672 "%d (%s).\n", vas, vaErrorStr(vas));
673 err = AVERROR(EIO);
674 goto fail;
675 }
676
677 for (buf = buf_list; buf; buf = buf->next)
678 total_size += buf->size;
679
680 err = ff_get_encode_buffer(avctx, pkt, total_size, 0);
681 ptr = pkt->data;
682
683 if (err < 0)
684 goto fail_mapped;
685
686 for (buf = buf_list; buf; buf = buf->next) {
687 av_log(avctx, AV_LOG_DEBUG, "Output buffer: %u bytes "
688 "(status %08x).\n", buf->size, buf->status);
689
690 memcpy(ptr, buf->buf, buf->size);
691 ptr += buf->size;
692 }
693
694 if (pic->type == PICTURE_TYPE_IDR)
695 pkt->flags |= AV_PKT_FLAG_KEY;
696
697 pkt->pts = pic->pts;
698
699 vas = vaUnmapBuffer(ctx->hwctx->display, pic->output_buffer);
700 if (vas != VA_STATUS_SUCCESS) {
701 av_log(avctx, AV_LOG_ERROR, "Failed to unmap output buffers: "
702 "%d (%s).\n", vas, vaErrorStr(vas));
703 err = AVERROR(EIO);
704 goto fail;
705 }
706
707 av_buffer_unref(&pic->output_buffer_ref);
708 pic->output_buffer = VA_INVALID_ID;
709
710 av_log(avctx, AV_LOG_DEBUG, "Output read for pic %"PRId64"/%"PRId64".\n",
711 pic->display_order, pic->encode_order);
712 return 0;
713
714 fail_mapped:
715 vaUnmapBuffer(ctx->hwctx->display, pic->output_buffer);
716 fail:
717 av_buffer_unref(&pic->output_buffer_ref);
718 pic->output_buffer = VA_INVALID_ID;
719 return err;
720 }
721
vaapi_encode_discard(AVCodecContext * avctx,VAAPIEncodePicture * pic)722 static int vaapi_encode_discard(AVCodecContext *avctx,
723 VAAPIEncodePicture *pic)
724 {
725 vaapi_encode_wait(avctx, pic);
726
727 if (pic->output_buffer_ref) {
728 av_log(avctx, AV_LOG_DEBUG, "Discard output for pic "
729 "%"PRId64"/%"PRId64".\n",
730 pic->display_order, pic->encode_order);
731
732 av_buffer_unref(&pic->output_buffer_ref);
733 pic->output_buffer = VA_INVALID_ID;
734 }
735
736 return 0;
737 }
738
vaapi_encode_alloc(AVCodecContext * avctx)739 static VAAPIEncodePicture *vaapi_encode_alloc(AVCodecContext *avctx)
740 {
741 VAAPIEncodeContext *ctx = avctx->priv_data;
742 VAAPIEncodePicture *pic;
743
744 pic = av_mallocz(sizeof(*pic));
745 if (!pic)
746 return NULL;
747
748 if (ctx->codec->picture_priv_data_size > 0) {
749 pic->priv_data = av_mallocz(ctx->codec->picture_priv_data_size);
750 if (!pic->priv_data) {
751 av_freep(&pic);
752 return NULL;
753 }
754 }
755
756 pic->input_surface = VA_INVALID_ID;
757 pic->recon_surface = VA_INVALID_ID;
758 pic->output_buffer = VA_INVALID_ID;
759
760 return pic;
761 }
762
vaapi_encode_free(AVCodecContext * avctx,VAAPIEncodePicture * pic)763 static int vaapi_encode_free(AVCodecContext *avctx,
764 VAAPIEncodePicture *pic)
765 {
766 int i;
767
768 if (pic->encode_issued)
769 vaapi_encode_discard(avctx, pic);
770
771 if (pic->slices) {
772 for (i = 0; i < pic->nb_slices; i++)
773 av_freep(&pic->slices[i].codec_slice_params);
774 }
775 av_freep(&pic->codec_picture_params);
776
777 av_frame_free(&pic->input_image);
778 av_frame_free(&pic->recon_image);
779
780 av_freep(&pic->param_buffers);
781 av_freep(&pic->slices);
782 // Output buffer should already be destroyed.
783 av_assert0(pic->output_buffer == VA_INVALID_ID);
784
785 av_freep(&pic->priv_data);
786 av_freep(&pic->codec_picture_params);
787 av_freep(&pic->roi);
788
789 av_free(pic);
790
791 return 0;
792 }
793
vaapi_encode_add_ref(AVCodecContext * avctx,VAAPIEncodePicture * pic,VAAPIEncodePicture * target,int is_ref,int in_dpb,int prev)794 static void vaapi_encode_add_ref(AVCodecContext *avctx,
795 VAAPIEncodePicture *pic,
796 VAAPIEncodePicture *target,
797 int is_ref, int in_dpb, int prev)
798 {
799 int refs = 0;
800
801 if (is_ref) {
802 av_assert0(pic != target);
803 av_assert0(pic->nb_refs < MAX_PICTURE_REFERENCES);
804 pic->refs[pic->nb_refs++] = target;
805 ++refs;
806 }
807
808 if (in_dpb) {
809 av_assert0(pic->nb_dpb_pics < MAX_DPB_SIZE);
810 pic->dpb[pic->nb_dpb_pics++] = target;
811 ++refs;
812 }
813
814 if (prev) {
815 av_assert0(!pic->prev);
816 pic->prev = target;
817 ++refs;
818 }
819
820 target->ref_count[0] += refs;
821 target->ref_count[1] += refs;
822 }
823
vaapi_encode_remove_refs(AVCodecContext * avctx,VAAPIEncodePicture * pic,int level)824 static void vaapi_encode_remove_refs(AVCodecContext *avctx,
825 VAAPIEncodePicture *pic,
826 int level)
827 {
828 int i;
829
830 if (pic->ref_removed[level])
831 return;
832
833 for (i = 0; i < pic->nb_refs; i++) {
834 av_assert0(pic->refs[i]);
835 --pic->refs[i]->ref_count[level];
836 av_assert0(pic->refs[i]->ref_count[level] >= 0);
837 }
838
839 for (i = 0; i < pic->nb_dpb_pics; i++) {
840 av_assert0(pic->dpb[i]);
841 --pic->dpb[i]->ref_count[level];
842 av_assert0(pic->dpb[i]->ref_count[level] >= 0);
843 }
844
845 av_assert0(pic->prev || pic->type == PICTURE_TYPE_IDR);
846 if (pic->prev) {
847 --pic->prev->ref_count[level];
848 av_assert0(pic->prev->ref_count[level] >= 0);
849 }
850
851 pic->ref_removed[level] = 1;
852 }
853
vaapi_encode_set_b_pictures(AVCodecContext * avctx,VAAPIEncodePicture * start,VAAPIEncodePicture * end,VAAPIEncodePicture * prev,int current_depth,VAAPIEncodePicture ** last)854 static void vaapi_encode_set_b_pictures(AVCodecContext *avctx,
855 VAAPIEncodePicture *start,
856 VAAPIEncodePicture *end,
857 VAAPIEncodePicture *prev,
858 int current_depth,
859 VAAPIEncodePicture **last)
860 {
861 VAAPIEncodeContext *ctx = avctx->priv_data;
862 VAAPIEncodePicture *pic, *next, *ref;
863 int i, len;
864
865 av_assert0(start && end && start != end && start->next != end);
866
867 // If we are at the maximum depth then encode all pictures as
868 // non-referenced B-pictures. Also do this if there is exactly one
869 // picture left, since there will be nothing to reference it.
870 if (current_depth == ctx->max_b_depth || start->next->next == end) {
871 for (pic = start->next; pic; pic = pic->next) {
872 if (pic == end)
873 break;
874 pic->type = PICTURE_TYPE_B;
875 pic->b_depth = current_depth;
876
877 vaapi_encode_add_ref(avctx, pic, start, 1, 1, 0);
878 vaapi_encode_add_ref(avctx, pic, end, 1, 1, 0);
879 vaapi_encode_add_ref(avctx, pic, prev, 0, 0, 1);
880
881 for (ref = end->refs[1]; ref; ref = ref->refs[1])
882 vaapi_encode_add_ref(avctx, pic, ref, 0, 1, 0);
883 }
884 *last = prev;
885
886 } else {
887 // Split the current list at the midpoint with a referenced
888 // B-picture, then descend into each side separately.
889 len = 0;
890 for (pic = start->next; pic != end; pic = pic->next)
891 ++len;
892 for (pic = start->next, i = 1; 2 * i < len; pic = pic->next, i++);
893
894 pic->type = PICTURE_TYPE_B;
895 pic->b_depth = current_depth;
896
897 pic->is_reference = 1;
898
899 vaapi_encode_add_ref(avctx, pic, pic, 0, 1, 0);
900 vaapi_encode_add_ref(avctx, pic, start, 1, 1, 0);
901 vaapi_encode_add_ref(avctx, pic, end, 1, 1, 0);
902 vaapi_encode_add_ref(avctx, pic, prev, 0, 0, 1);
903
904 for (ref = end->refs[1]; ref; ref = ref->refs[1])
905 vaapi_encode_add_ref(avctx, pic, ref, 0, 1, 0);
906
907 if (i > 1)
908 vaapi_encode_set_b_pictures(avctx, start, pic, pic,
909 current_depth + 1, &next);
910 else
911 next = pic;
912
913 vaapi_encode_set_b_pictures(avctx, pic, end, next,
914 current_depth + 1, last);
915 }
916 }
917
vaapi_encode_pick_next(AVCodecContext * avctx,VAAPIEncodePicture ** pic_out)918 static int vaapi_encode_pick_next(AVCodecContext *avctx,
919 VAAPIEncodePicture **pic_out)
920 {
921 VAAPIEncodeContext *ctx = avctx->priv_data;
922 VAAPIEncodePicture *pic = NULL, *next, *start;
923 int i, b_counter, closed_gop_end;
924
925 // If there are any B-frames already queued, the next one to encode
926 // is the earliest not-yet-issued frame for which all references are
927 // available.
928 for (pic = ctx->pic_start; pic; pic = pic->next) {
929 if (pic->encode_issued)
930 continue;
931 if (pic->type != PICTURE_TYPE_B)
932 continue;
933 for (i = 0; i < pic->nb_refs; i++) {
934 if (!pic->refs[i]->encode_issued)
935 break;
936 }
937 if (i == pic->nb_refs)
938 break;
939 }
940
941 if (pic) {
942 av_log(avctx, AV_LOG_DEBUG, "Pick B-picture at depth %d to "
943 "encode next.\n", pic->b_depth);
944 *pic_out = pic;
945 return 0;
946 }
947
948 // Find the B-per-Pth available picture to become the next picture
949 // on the top layer.
950 start = NULL;
951 b_counter = 0;
952 closed_gop_end = ctx->closed_gop ||
953 ctx->idr_counter == ctx->gop_per_idr;
954 for (pic = ctx->pic_start; pic; pic = next) {
955 next = pic->next;
956 if (pic->encode_issued) {
957 start = pic;
958 continue;
959 }
960 // If the next available picture is force-IDR, encode it to start
961 // a new GOP immediately.
962 if (pic->force_idr)
963 break;
964 if (b_counter == ctx->b_per_p)
965 break;
966 // If this picture ends a closed GOP or starts a new GOP then it
967 // needs to be in the top layer.
968 if (ctx->gop_counter + b_counter + closed_gop_end >= ctx->gop_size)
969 break;
970 // If the picture after this one is force-IDR, we need to encode
971 // this one in the top layer.
972 if (next && next->force_idr)
973 break;
974 ++b_counter;
975 }
976
977 // At the end of the stream the last picture must be in the top layer.
978 if (!pic && ctx->end_of_stream) {
979 --b_counter;
980 pic = ctx->pic_end;
981 if (pic->encode_complete)
982 return AVERROR_EOF;
983 else if (pic->encode_issued)
984 return AVERROR(EAGAIN);
985 }
986
987 if (!pic) {
988 av_log(avctx, AV_LOG_DEBUG, "Pick nothing to encode next - "
989 "need more input for reference pictures.\n");
990 return AVERROR(EAGAIN);
991 }
992 if (ctx->input_order <= ctx->decode_delay && !ctx->end_of_stream) {
993 av_log(avctx, AV_LOG_DEBUG, "Pick nothing to encode next - "
994 "need more input for timestamps.\n");
995 return AVERROR(EAGAIN);
996 }
997
998 if (pic->force_idr) {
999 av_log(avctx, AV_LOG_DEBUG, "Pick forced IDR-picture to "
1000 "encode next.\n");
1001 pic->type = PICTURE_TYPE_IDR;
1002 ctx->idr_counter = 1;
1003 ctx->gop_counter = 1;
1004
1005 } else if (ctx->gop_counter + b_counter >= ctx->gop_size) {
1006 if (ctx->idr_counter == ctx->gop_per_idr) {
1007 av_log(avctx, AV_LOG_DEBUG, "Pick new-GOP IDR-picture to "
1008 "encode next.\n");
1009 pic->type = PICTURE_TYPE_IDR;
1010 ctx->idr_counter = 1;
1011 } else {
1012 av_log(avctx, AV_LOG_DEBUG, "Pick new-GOP I-picture to "
1013 "encode next.\n");
1014 pic->type = PICTURE_TYPE_I;
1015 ++ctx->idr_counter;
1016 }
1017 ctx->gop_counter = 1;
1018
1019 } else {
1020 if (ctx->gop_counter + b_counter + closed_gop_end == ctx->gop_size) {
1021 av_log(avctx, AV_LOG_DEBUG, "Pick group-end P-picture to "
1022 "encode next.\n");
1023 } else {
1024 av_log(avctx, AV_LOG_DEBUG, "Pick normal P-picture to "
1025 "encode next.\n");
1026 }
1027 pic->type = PICTURE_TYPE_P;
1028 av_assert0(start);
1029 ctx->gop_counter += 1 + b_counter;
1030 }
1031 pic->is_reference = 1;
1032 *pic_out = pic;
1033
1034 vaapi_encode_add_ref(avctx, pic, pic, 0, 1, 0);
1035 if (pic->type != PICTURE_TYPE_IDR) {
1036 vaapi_encode_add_ref(avctx, pic, start,
1037 pic->type == PICTURE_TYPE_P,
1038 b_counter > 0, 0);
1039 vaapi_encode_add_ref(avctx, pic, ctx->next_prev, 0, 0, 1);
1040 }
1041 if (ctx->next_prev)
1042 --ctx->next_prev->ref_count[0];
1043
1044 if (b_counter > 0) {
1045 vaapi_encode_set_b_pictures(avctx, start, pic, pic, 1,
1046 &ctx->next_prev);
1047 } else {
1048 ctx->next_prev = pic;
1049 }
1050 ++ctx->next_prev->ref_count[0];
1051 return 0;
1052 }
1053
vaapi_encode_clear_old(AVCodecContext * avctx)1054 static int vaapi_encode_clear_old(AVCodecContext *avctx)
1055 {
1056 VAAPIEncodeContext *ctx = avctx->priv_data;
1057 VAAPIEncodePicture *pic, *prev, *next;
1058
1059 av_assert0(ctx->pic_start);
1060
1061 // Remove direct references once each picture is complete.
1062 for (pic = ctx->pic_start; pic; pic = pic->next) {
1063 if (pic->encode_complete && pic->next)
1064 vaapi_encode_remove_refs(avctx, pic, 0);
1065 }
1066
1067 // Remove indirect references once a picture has no direct references.
1068 for (pic = ctx->pic_start; pic; pic = pic->next) {
1069 if (pic->encode_complete && pic->ref_count[0] == 0)
1070 vaapi_encode_remove_refs(avctx, pic, 1);
1071 }
1072
1073 // Clear out all complete pictures with no remaining references.
1074 prev = NULL;
1075 for (pic = ctx->pic_start; pic; pic = next) {
1076 next = pic->next;
1077 if (pic->encode_complete && pic->ref_count[1] == 0) {
1078 av_assert0(pic->ref_removed[0] && pic->ref_removed[1]);
1079 if (prev)
1080 prev->next = next;
1081 else
1082 ctx->pic_start = next;
1083 vaapi_encode_free(avctx, pic);
1084 } else {
1085 prev = pic;
1086 }
1087 }
1088
1089 return 0;
1090 }
1091
vaapi_encode_check_frame(AVCodecContext * avctx,const AVFrame * frame)1092 static int vaapi_encode_check_frame(AVCodecContext *avctx,
1093 const AVFrame *frame)
1094 {
1095 VAAPIEncodeContext *ctx = avctx->priv_data;
1096
1097 if ((frame->crop_top || frame->crop_bottom ||
1098 frame->crop_left || frame->crop_right) && !ctx->crop_warned) {
1099 av_log(avctx, AV_LOG_WARNING, "Cropping information on input "
1100 "frames ignored due to lack of API support.\n");
1101 ctx->crop_warned = 1;
1102 }
1103
1104 if (!ctx->roi_allowed) {
1105 AVFrameSideData *sd =
1106 av_frame_get_side_data(frame, AV_FRAME_DATA_REGIONS_OF_INTEREST);
1107
1108 if (sd && !ctx->roi_warned) {
1109 av_log(avctx, AV_LOG_WARNING, "ROI side data on input "
1110 "frames ignored due to lack of driver support.\n");
1111 ctx->roi_warned = 1;
1112 }
1113 }
1114
1115 return 0;
1116 }
1117
vaapi_encode_send_frame(AVCodecContext * avctx,AVFrame * frame)1118 static int vaapi_encode_send_frame(AVCodecContext *avctx, AVFrame *frame)
1119 {
1120 VAAPIEncodeContext *ctx = avctx->priv_data;
1121 VAAPIEncodePicture *pic;
1122 int err;
1123
1124 if (frame) {
1125 av_log(avctx, AV_LOG_DEBUG, "Input frame: %ux%u (%"PRId64").\n",
1126 frame->width, frame->height, frame->pts);
1127
1128 err = vaapi_encode_check_frame(avctx, frame);
1129 if (err < 0)
1130 return err;
1131
1132 pic = vaapi_encode_alloc(avctx);
1133 if (!pic)
1134 return AVERROR(ENOMEM);
1135
1136 pic->input_image = av_frame_alloc();
1137 if (!pic->input_image) {
1138 err = AVERROR(ENOMEM);
1139 goto fail;
1140 }
1141
1142 if (ctx->input_order == 0 || frame->pict_type == AV_PICTURE_TYPE_I)
1143 pic->force_idr = 1;
1144
1145 pic->input_surface = (VASurfaceID)(uintptr_t)frame->data[3];
1146 pic->pts = frame->pts;
1147
1148 av_frame_move_ref(pic->input_image, frame);
1149
1150 if (ctx->input_order == 0)
1151 ctx->first_pts = pic->pts;
1152 if (ctx->input_order == ctx->decode_delay)
1153 ctx->dts_pts_diff = pic->pts - ctx->first_pts;
1154 if (ctx->output_delay > 0)
1155 ctx->ts_ring[ctx->input_order %
1156 (3 * ctx->output_delay + ctx->async_depth)] = pic->pts;
1157
1158 pic->display_order = ctx->input_order;
1159 ++ctx->input_order;
1160
1161 if (ctx->pic_start) {
1162 ctx->pic_end->next = pic;
1163 ctx->pic_end = pic;
1164 } else {
1165 ctx->pic_start = pic;
1166 ctx->pic_end = pic;
1167 }
1168
1169 } else {
1170 ctx->end_of_stream = 1;
1171
1172 // Fix timestamps if we hit end-of-stream before the initial decode
1173 // delay has elapsed.
1174 if (ctx->input_order < ctx->decode_delay)
1175 ctx->dts_pts_diff = ctx->pic_end->pts - ctx->first_pts;
1176 }
1177
1178 return 0;
1179
1180 fail:
1181 vaapi_encode_free(avctx, pic);
1182 return err;
1183 }
1184
ff_vaapi_encode_receive_packet(AVCodecContext * avctx,AVPacket * pkt)1185 int ff_vaapi_encode_receive_packet(AVCodecContext *avctx, AVPacket *pkt)
1186 {
1187 VAAPIEncodeContext *ctx = avctx->priv_data;
1188 VAAPIEncodePicture *pic;
1189 AVFrame *frame = ctx->frame;
1190 int err;
1191
1192 err = ff_encode_get_frame(avctx, frame);
1193 if (err < 0 && err != AVERROR_EOF)
1194 return err;
1195
1196 if (err == AVERROR_EOF)
1197 frame = NULL;
1198
1199 err = vaapi_encode_send_frame(avctx, frame);
1200 if (err < 0)
1201 return err;
1202
1203 if (!ctx->pic_start) {
1204 if (ctx->end_of_stream)
1205 return AVERROR_EOF;
1206 else
1207 return AVERROR(EAGAIN);
1208 }
1209
1210 if (ctx->has_sync_buffer_func) {
1211 pic = NULL;
1212
1213 if (av_fifo_can_write(ctx->encode_fifo)) {
1214 err = vaapi_encode_pick_next(avctx, &pic);
1215 if (!err) {
1216 av_assert0(pic);
1217 pic->encode_order = ctx->encode_order +
1218 av_fifo_can_read(ctx->encode_fifo);
1219 err = vaapi_encode_issue(avctx, pic);
1220 if (err < 0) {
1221 av_log(avctx, AV_LOG_ERROR, "Encode failed: %d.\n", err);
1222 return err;
1223 }
1224 av_fifo_write(ctx->encode_fifo, &pic, 1);
1225 }
1226 }
1227
1228 if (!av_fifo_can_read(ctx->encode_fifo))
1229 return err;
1230
1231 // More frames can be buffered
1232 if (av_fifo_can_write(ctx->encode_fifo) && !ctx->end_of_stream)
1233 return AVERROR(EAGAIN);
1234
1235 av_fifo_read(ctx->encode_fifo, &pic, 1);
1236 ctx->encode_order = pic->encode_order + 1;
1237 } else {
1238 pic = NULL;
1239 err = vaapi_encode_pick_next(avctx, &pic);
1240 if (err < 0)
1241 return err;
1242 av_assert0(pic);
1243
1244 pic->encode_order = ctx->encode_order++;
1245
1246 err = vaapi_encode_issue(avctx, pic);
1247 if (err < 0) {
1248 av_log(avctx, AV_LOG_ERROR, "Encode failed: %d.\n", err);
1249 return err;
1250 }
1251 }
1252
1253 err = vaapi_encode_output(avctx, pic, pkt);
1254 if (err < 0) {
1255 av_log(avctx, AV_LOG_ERROR, "Output failed: %d.\n", err);
1256 return err;
1257 }
1258
1259 if (ctx->output_delay == 0) {
1260 pkt->dts = pkt->pts;
1261 } else if (pic->encode_order < ctx->decode_delay) {
1262 if (ctx->ts_ring[pic->encode_order] < INT64_MIN + ctx->dts_pts_diff)
1263 pkt->dts = INT64_MIN;
1264 else
1265 pkt->dts = ctx->ts_ring[pic->encode_order] - ctx->dts_pts_diff;
1266 } else {
1267 pkt->dts = ctx->ts_ring[(pic->encode_order - ctx->decode_delay) %
1268 (3 * ctx->output_delay + ctx->async_depth)];
1269 }
1270 av_log(avctx, AV_LOG_DEBUG, "Output packet: pts %"PRId64" dts %"PRId64".\n",
1271 pkt->pts, pkt->dts);
1272
1273 ctx->output_order = pic->encode_order;
1274 vaapi_encode_clear_old(avctx);
1275
1276 return 0;
1277 }
1278
1279
vaapi_encode_add_global_param(AVCodecContext * avctx,int type,void * buffer,size_t size)1280 static av_cold void vaapi_encode_add_global_param(AVCodecContext *avctx, int type,
1281 void *buffer, size_t size)
1282 {
1283 VAAPIEncodeContext *ctx = avctx->priv_data;
1284
1285 av_assert0(ctx->nb_global_params < MAX_GLOBAL_PARAMS);
1286
1287 ctx->global_params_type[ctx->nb_global_params] = type;
1288 ctx->global_params [ctx->nb_global_params] = buffer;
1289 ctx->global_params_size[ctx->nb_global_params] = size;
1290
1291 ++ctx->nb_global_params;
1292 }
1293
1294 typedef struct VAAPIEncodeRTFormat {
1295 const char *name;
1296 unsigned int value;
1297 int depth;
1298 int nb_components;
1299 int log2_chroma_w;
1300 int log2_chroma_h;
1301 } VAAPIEncodeRTFormat;
1302
1303 static const VAAPIEncodeRTFormat vaapi_encode_rt_formats[] = {
1304 { "YUV400", VA_RT_FORMAT_YUV400, 8, 1, },
1305 { "YUV420", VA_RT_FORMAT_YUV420, 8, 3, 1, 1 },
1306 { "YUV422", VA_RT_FORMAT_YUV422, 8, 3, 1, 0 },
1307 #if VA_CHECK_VERSION(1, 2, 0)
1308 { "YUV422_10", VA_RT_FORMAT_YUV422_10, 10, 3, 1, 0 },
1309 #endif
1310 { "YUV444", VA_RT_FORMAT_YUV444, 8, 3, 0, 0 },
1311 { "YUV411", VA_RT_FORMAT_YUV411, 8, 3, 2, 0 },
1312 #if VA_CHECK_VERSION(0, 38, 1)
1313 { "YUV420_10", VA_RT_FORMAT_YUV420_10BPP, 10, 3, 1, 1 },
1314 #endif
1315 };
1316
1317 static const VAEntrypoint vaapi_encode_entrypoints_normal[] = {
1318 VAEntrypointEncSlice,
1319 VAEntrypointEncPicture,
1320 #if VA_CHECK_VERSION(0, 39, 2)
1321 VAEntrypointEncSliceLP,
1322 #endif
1323 0
1324 };
1325 #if VA_CHECK_VERSION(0, 39, 2)
1326 static const VAEntrypoint vaapi_encode_entrypoints_low_power[] = {
1327 VAEntrypointEncSliceLP,
1328 0
1329 };
1330 #endif
1331
vaapi_encode_profile_entrypoint(AVCodecContext * avctx)1332 static av_cold int vaapi_encode_profile_entrypoint(AVCodecContext *avctx)
1333 {
1334 VAAPIEncodeContext *ctx = avctx->priv_data;
1335 VAProfile *va_profiles = NULL;
1336 VAEntrypoint *va_entrypoints = NULL;
1337 VAStatus vas;
1338 const VAEntrypoint *usable_entrypoints;
1339 const VAAPIEncodeProfile *profile;
1340 const AVPixFmtDescriptor *desc;
1341 VAConfigAttrib rt_format_attr;
1342 const VAAPIEncodeRTFormat *rt_format;
1343 const char *profile_string, *entrypoint_string;
1344 int i, j, n, depth, err;
1345
1346
1347 if (ctx->low_power) {
1348 #if VA_CHECK_VERSION(0, 39, 2)
1349 usable_entrypoints = vaapi_encode_entrypoints_low_power;
1350 #else
1351 av_log(avctx, AV_LOG_ERROR, "Low-power encoding is not "
1352 "supported with this VAAPI version.\n");
1353 return AVERROR(EINVAL);
1354 #endif
1355 } else {
1356 usable_entrypoints = vaapi_encode_entrypoints_normal;
1357 }
1358
1359 desc = av_pix_fmt_desc_get(ctx->input_frames->sw_format);
1360 if (!desc) {
1361 av_log(avctx, AV_LOG_ERROR, "Invalid input pixfmt (%d).\n",
1362 ctx->input_frames->sw_format);
1363 return AVERROR(EINVAL);
1364 }
1365 depth = desc->comp[0].depth;
1366 for (i = 1; i < desc->nb_components; i++) {
1367 if (desc->comp[i].depth != depth) {
1368 av_log(avctx, AV_LOG_ERROR, "Invalid input pixfmt (%s).\n",
1369 desc->name);
1370 return AVERROR(EINVAL);
1371 }
1372 }
1373 av_log(avctx, AV_LOG_VERBOSE, "Input surface format is %s.\n",
1374 desc->name);
1375
1376 n = vaMaxNumProfiles(ctx->hwctx->display);
1377 va_profiles = av_malloc_array(n, sizeof(VAProfile));
1378 if (!va_profiles) {
1379 err = AVERROR(ENOMEM);
1380 goto fail;
1381 }
1382 vas = vaQueryConfigProfiles(ctx->hwctx->display, va_profiles, &n);
1383 if (vas != VA_STATUS_SUCCESS) {
1384 av_log(avctx, AV_LOG_ERROR, "Failed to query profiles: %d (%s).\n",
1385 vas, vaErrorStr(vas));
1386 err = AVERROR_EXTERNAL;
1387 goto fail;
1388 }
1389
1390 av_assert0(ctx->codec->profiles);
1391 for (i = 0; (ctx->codec->profiles[i].av_profile !=
1392 FF_PROFILE_UNKNOWN); i++) {
1393 profile = &ctx->codec->profiles[i];
1394 if (depth != profile->depth ||
1395 desc->nb_components != profile->nb_components)
1396 continue;
1397 if (desc->nb_components > 1 &&
1398 (desc->log2_chroma_w != profile->log2_chroma_w ||
1399 desc->log2_chroma_h != profile->log2_chroma_h))
1400 continue;
1401 if (avctx->profile != profile->av_profile &&
1402 avctx->profile != FF_PROFILE_UNKNOWN)
1403 continue;
1404
1405 #if VA_CHECK_VERSION(1, 0, 0)
1406 profile_string = vaProfileStr(profile->va_profile);
1407 #else
1408 profile_string = "(no profile names)";
1409 #endif
1410
1411 for (j = 0; j < n; j++) {
1412 if (va_profiles[j] == profile->va_profile)
1413 break;
1414 }
1415 if (j >= n) {
1416 av_log(avctx, AV_LOG_VERBOSE, "Compatible profile %s (%d) "
1417 "is not supported by driver.\n", profile_string,
1418 profile->va_profile);
1419 continue;
1420 }
1421
1422 ctx->profile = profile;
1423 break;
1424 }
1425 if (!ctx->profile) {
1426 av_log(avctx, AV_LOG_ERROR, "No usable encoding profile found.\n");
1427 err = AVERROR(ENOSYS);
1428 goto fail;
1429 }
1430
1431 avctx->profile = profile->av_profile;
1432 ctx->va_profile = profile->va_profile;
1433 av_log(avctx, AV_LOG_VERBOSE, "Using VAAPI profile %s (%d).\n",
1434 profile_string, ctx->va_profile);
1435
1436 n = vaMaxNumEntrypoints(ctx->hwctx->display);
1437 va_entrypoints = av_malloc_array(n, sizeof(VAEntrypoint));
1438 if (!va_entrypoints) {
1439 err = AVERROR(ENOMEM);
1440 goto fail;
1441 }
1442 vas = vaQueryConfigEntrypoints(ctx->hwctx->display, ctx->va_profile,
1443 va_entrypoints, &n);
1444 if (vas != VA_STATUS_SUCCESS) {
1445 av_log(avctx, AV_LOG_ERROR, "Failed to query entrypoints for "
1446 "profile %s (%d): %d (%s).\n", profile_string,
1447 ctx->va_profile, vas, vaErrorStr(vas));
1448 err = AVERROR_EXTERNAL;
1449 goto fail;
1450 }
1451
1452 for (i = 0; i < n; i++) {
1453 for (j = 0; usable_entrypoints[j]; j++) {
1454 if (va_entrypoints[i] == usable_entrypoints[j])
1455 break;
1456 }
1457 if (usable_entrypoints[j])
1458 break;
1459 }
1460 if (i >= n) {
1461 av_log(avctx, AV_LOG_ERROR, "No usable encoding entrypoint found "
1462 "for profile %s (%d).\n", profile_string, ctx->va_profile);
1463 err = AVERROR(ENOSYS);
1464 goto fail;
1465 }
1466
1467 ctx->va_entrypoint = va_entrypoints[i];
1468 #if VA_CHECK_VERSION(1, 0, 0)
1469 entrypoint_string = vaEntrypointStr(ctx->va_entrypoint);
1470 #else
1471 entrypoint_string = "(no entrypoint names)";
1472 #endif
1473 av_log(avctx, AV_LOG_VERBOSE, "Using VAAPI entrypoint %s (%d).\n",
1474 entrypoint_string, ctx->va_entrypoint);
1475
1476 for (i = 0; i < FF_ARRAY_ELEMS(vaapi_encode_rt_formats); i++) {
1477 rt_format = &vaapi_encode_rt_formats[i];
1478 if (rt_format->depth == depth &&
1479 rt_format->nb_components == profile->nb_components &&
1480 rt_format->log2_chroma_w == profile->log2_chroma_w &&
1481 rt_format->log2_chroma_h == profile->log2_chroma_h)
1482 break;
1483 }
1484 if (i >= FF_ARRAY_ELEMS(vaapi_encode_rt_formats)) {
1485 av_log(avctx, AV_LOG_ERROR, "No usable render target format "
1486 "found for profile %s (%d) entrypoint %s (%d).\n",
1487 profile_string, ctx->va_profile,
1488 entrypoint_string, ctx->va_entrypoint);
1489 err = AVERROR(ENOSYS);
1490 goto fail;
1491 }
1492
1493 rt_format_attr = (VAConfigAttrib) { VAConfigAttribRTFormat };
1494 vas = vaGetConfigAttributes(ctx->hwctx->display,
1495 ctx->va_profile, ctx->va_entrypoint,
1496 &rt_format_attr, 1);
1497 if (vas != VA_STATUS_SUCCESS) {
1498 av_log(avctx, AV_LOG_ERROR, "Failed to query RT format "
1499 "config attribute: %d (%s).\n", vas, vaErrorStr(vas));
1500 err = AVERROR_EXTERNAL;
1501 goto fail;
1502 }
1503
1504 if (rt_format_attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1505 av_log(avctx, AV_LOG_VERBOSE, "RT format config attribute not "
1506 "supported by driver: assuming surface RT format %s "
1507 "is valid.\n", rt_format->name);
1508 } else if (!(rt_format_attr.value & rt_format->value)) {
1509 av_log(avctx, AV_LOG_ERROR, "Surface RT format %s not supported "
1510 "by driver for encoding profile %s (%d) entrypoint %s (%d).\n",
1511 rt_format->name, profile_string, ctx->va_profile,
1512 entrypoint_string, ctx->va_entrypoint);
1513 err = AVERROR(ENOSYS);
1514 goto fail;
1515 } else {
1516 av_log(avctx, AV_LOG_VERBOSE, "Using VAAPI render target "
1517 "format %s (%#x).\n", rt_format->name, rt_format->value);
1518 ctx->config_attributes[ctx->nb_config_attributes++] =
1519 (VAConfigAttrib) {
1520 .type = VAConfigAttribRTFormat,
1521 .value = rt_format->value,
1522 };
1523 }
1524
1525 err = 0;
1526 fail:
1527 av_freep(&va_profiles);
1528 av_freep(&va_entrypoints);
1529 return err;
1530 }
1531
1532 static const VAAPIEncodeRCMode vaapi_encode_rc_modes[] = {
1533 // Bitrate Quality
1534 // | Maxrate | HRD/VBV
1535 { 0 }, // | | | |
1536 { RC_MODE_CQP, "CQP", 1, VA_RC_CQP, 0, 0, 1, 0 },
1537 { RC_MODE_CBR, "CBR", 1, VA_RC_CBR, 1, 0, 0, 1 },
1538 { RC_MODE_VBR, "VBR", 1, VA_RC_VBR, 1, 1, 0, 1 },
1539 #if VA_CHECK_VERSION(1, 1, 0)
1540 { RC_MODE_ICQ, "ICQ", 1, VA_RC_ICQ, 0, 0, 1, 0 },
1541 #else
1542 { RC_MODE_ICQ, "ICQ", 0 },
1543 #endif
1544 #if VA_CHECK_VERSION(1, 3, 0)
1545 { RC_MODE_QVBR, "QVBR", 1, VA_RC_QVBR, 1, 1, 1, 1 },
1546 { RC_MODE_AVBR, "AVBR", 0, VA_RC_AVBR, 1, 0, 0, 0 },
1547 #else
1548 { RC_MODE_QVBR, "QVBR", 0 },
1549 { RC_MODE_AVBR, "AVBR", 0 },
1550 #endif
1551 };
1552
vaapi_encode_init_rate_control(AVCodecContext * avctx)1553 static av_cold int vaapi_encode_init_rate_control(AVCodecContext *avctx)
1554 {
1555 VAAPIEncodeContext *ctx = avctx->priv_data;
1556 uint32_t supported_va_rc_modes;
1557 const VAAPIEncodeRCMode *rc_mode;
1558 int64_t rc_bits_per_second;
1559 int rc_target_percentage;
1560 int rc_window_size;
1561 int rc_quality;
1562 int64_t hrd_buffer_size;
1563 int64_t hrd_initial_buffer_fullness;
1564 int fr_num, fr_den;
1565 VAConfigAttrib rc_attr = { VAConfigAttribRateControl };
1566 VAStatus vas;
1567 char supported_rc_modes_string[64];
1568
1569 vas = vaGetConfigAttributes(ctx->hwctx->display,
1570 ctx->va_profile, ctx->va_entrypoint,
1571 &rc_attr, 1);
1572 if (vas != VA_STATUS_SUCCESS) {
1573 av_log(avctx, AV_LOG_ERROR, "Failed to query rate control "
1574 "config attribute: %d (%s).\n", vas, vaErrorStr(vas));
1575 return AVERROR_EXTERNAL;
1576 }
1577 if (rc_attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1578 av_log(avctx, AV_LOG_VERBOSE, "Driver does not report any "
1579 "supported rate control modes: assuming CQP only.\n");
1580 supported_va_rc_modes = VA_RC_CQP;
1581 strcpy(supported_rc_modes_string, "unknown");
1582 } else {
1583 char *str = supported_rc_modes_string;
1584 size_t len = sizeof(supported_rc_modes_string);
1585 int i, first = 1, res;
1586
1587 supported_va_rc_modes = rc_attr.value;
1588 for (i = 0; i < FF_ARRAY_ELEMS(vaapi_encode_rc_modes); i++) {
1589 rc_mode = &vaapi_encode_rc_modes[i];
1590 if (supported_va_rc_modes & rc_mode->va_mode) {
1591 res = snprintf(str, len, "%s%s",
1592 first ? "" : ", ", rc_mode->name);
1593 first = 0;
1594 if (res < 0) {
1595 *str = 0;
1596 break;
1597 }
1598 len -= res;
1599 str += res;
1600 if (len == 0)
1601 break;
1602 }
1603 }
1604
1605 av_log(avctx, AV_LOG_DEBUG, "Driver supports RC modes %s.\n",
1606 supported_rc_modes_string);
1607 }
1608
1609 // Rate control mode selection:
1610 // * If the user has set a mode explicitly with the rc_mode option,
1611 // use it and fail if it is not available.
1612 // * If an explicit QP option has been set, use CQP.
1613 // * If the codec is CQ-only, use CQP.
1614 // * If the QSCALE avcodec option is set, use CQP.
1615 // * If bitrate and quality are both set, try QVBR.
1616 // * If quality is set, try ICQ, then CQP.
1617 // * If bitrate and maxrate are set and have the same value, try CBR.
1618 // * If a bitrate is set, try AVBR, then VBR, then CBR.
1619 // * If no bitrate is set, try ICQ, then CQP.
1620
1621 #define TRY_RC_MODE(mode, fail) do { \
1622 rc_mode = &vaapi_encode_rc_modes[mode]; \
1623 if (!(rc_mode->va_mode & supported_va_rc_modes)) { \
1624 if (fail) { \
1625 av_log(avctx, AV_LOG_ERROR, "Driver does not support %s " \
1626 "RC mode (supported modes: %s).\n", rc_mode->name, \
1627 supported_rc_modes_string); \
1628 return AVERROR(EINVAL); \
1629 } \
1630 av_log(avctx, AV_LOG_DEBUG, "Driver does not support %s " \
1631 "RC mode.\n", rc_mode->name); \
1632 rc_mode = NULL; \
1633 } else { \
1634 goto rc_mode_found; \
1635 } \
1636 } while (0)
1637
1638 if (ctx->explicit_rc_mode)
1639 TRY_RC_MODE(ctx->explicit_rc_mode, 1);
1640
1641 if (ctx->explicit_qp)
1642 TRY_RC_MODE(RC_MODE_CQP, 1);
1643
1644 if (ctx->codec->flags & FLAG_CONSTANT_QUALITY_ONLY)
1645 TRY_RC_MODE(RC_MODE_CQP, 1);
1646
1647 if (avctx->flags & AV_CODEC_FLAG_QSCALE)
1648 TRY_RC_MODE(RC_MODE_CQP, 1);
1649
1650 if (avctx->bit_rate > 0 && avctx->global_quality > 0)
1651 TRY_RC_MODE(RC_MODE_QVBR, 0);
1652
1653 if (avctx->global_quality > 0) {
1654 TRY_RC_MODE(RC_MODE_ICQ, 0);
1655 TRY_RC_MODE(RC_MODE_CQP, 0);
1656 }
1657
1658 if (avctx->bit_rate > 0 && avctx->rc_max_rate == avctx->bit_rate)
1659 TRY_RC_MODE(RC_MODE_CBR, 0);
1660
1661 if (avctx->bit_rate > 0) {
1662 TRY_RC_MODE(RC_MODE_AVBR, 0);
1663 TRY_RC_MODE(RC_MODE_VBR, 0);
1664 TRY_RC_MODE(RC_MODE_CBR, 0);
1665 } else {
1666 TRY_RC_MODE(RC_MODE_ICQ, 0);
1667 TRY_RC_MODE(RC_MODE_CQP, 0);
1668 }
1669
1670 av_log(avctx, AV_LOG_ERROR, "Driver does not support any "
1671 "RC mode compatible with selected options "
1672 "(supported modes: %s).\n", supported_rc_modes_string);
1673 return AVERROR(EINVAL);
1674
1675 rc_mode_found:
1676 if (rc_mode->bitrate) {
1677 if (avctx->bit_rate <= 0) {
1678 av_log(avctx, AV_LOG_ERROR, "Bitrate must be set for %s "
1679 "RC mode.\n", rc_mode->name);
1680 return AVERROR(EINVAL);
1681 }
1682
1683 if (rc_mode->mode == RC_MODE_AVBR) {
1684 // For maximum confusion AVBR is hacked into the existing API
1685 // by overloading some of the fields with completely different
1686 // meanings.
1687
1688 // Target percentage does not apply in AVBR mode.
1689 rc_bits_per_second = avctx->bit_rate;
1690
1691 // Accuracy tolerance range for meeting the specified target
1692 // bitrate. It's very unclear how this is actually intended
1693 // to work - since we do want to get the specified bitrate,
1694 // set the accuracy to 100% for now.
1695 rc_target_percentage = 100;
1696
1697 // Convergence period in frames. The GOP size reflects the
1698 // user's intended block size for cutting, so reusing that
1699 // as the convergence period seems a reasonable default.
1700 rc_window_size = avctx->gop_size > 0 ? avctx->gop_size : 60;
1701
1702 } else if (rc_mode->maxrate) {
1703 if (avctx->rc_max_rate > 0) {
1704 if (avctx->rc_max_rate < avctx->bit_rate) {
1705 av_log(avctx, AV_LOG_ERROR, "Invalid bitrate settings: "
1706 "bitrate (%"PRId64") must not be greater than "
1707 "maxrate (%"PRId64").\n", avctx->bit_rate,
1708 avctx->rc_max_rate);
1709 return AVERROR(EINVAL);
1710 }
1711 rc_bits_per_second = avctx->rc_max_rate;
1712 rc_target_percentage = (avctx->bit_rate * 100) /
1713 avctx->rc_max_rate;
1714 } else {
1715 // We only have a target bitrate, but this mode requires
1716 // that a maximum rate be supplied as well. Since the
1717 // user does not want this to be a constraint, arbitrarily
1718 // pick a maximum rate of double the target rate.
1719 rc_bits_per_second = 2 * avctx->bit_rate;
1720 rc_target_percentage = 50;
1721 }
1722 } else {
1723 if (avctx->rc_max_rate > avctx->bit_rate) {
1724 av_log(avctx, AV_LOG_WARNING, "Max bitrate is ignored "
1725 "in %s RC mode.\n", rc_mode->name);
1726 }
1727 rc_bits_per_second = avctx->bit_rate;
1728 rc_target_percentage = 100;
1729 }
1730 } else {
1731 rc_bits_per_second = 0;
1732 rc_target_percentage = 100;
1733 }
1734
1735 if (rc_mode->quality) {
1736 if (ctx->explicit_qp) {
1737 rc_quality = ctx->explicit_qp;
1738 } else if (avctx->global_quality > 0) {
1739 rc_quality = avctx->global_quality;
1740 } else {
1741 rc_quality = ctx->codec->default_quality;
1742 av_log(avctx, AV_LOG_WARNING, "No quality level set; "
1743 "using default (%d).\n", rc_quality);
1744 }
1745 } else {
1746 rc_quality = 0;
1747 }
1748
1749 if (rc_mode->hrd) {
1750 if (avctx->rc_buffer_size)
1751 hrd_buffer_size = avctx->rc_buffer_size;
1752 else if (avctx->rc_max_rate > 0)
1753 hrd_buffer_size = avctx->rc_max_rate;
1754 else
1755 hrd_buffer_size = avctx->bit_rate;
1756 if (avctx->rc_initial_buffer_occupancy) {
1757 if (avctx->rc_initial_buffer_occupancy > hrd_buffer_size) {
1758 av_log(avctx, AV_LOG_ERROR, "Invalid RC buffer settings: "
1759 "must have initial buffer size (%d) <= "
1760 "buffer size (%"PRId64").\n",
1761 avctx->rc_initial_buffer_occupancy, hrd_buffer_size);
1762 return AVERROR(EINVAL);
1763 }
1764 hrd_initial_buffer_fullness = avctx->rc_initial_buffer_occupancy;
1765 } else {
1766 hrd_initial_buffer_fullness = hrd_buffer_size * 3 / 4;
1767 }
1768
1769 rc_window_size = (hrd_buffer_size * 1000) / rc_bits_per_second;
1770 } else {
1771 if (avctx->rc_buffer_size || avctx->rc_initial_buffer_occupancy) {
1772 av_log(avctx, AV_LOG_WARNING, "Buffering settings are ignored "
1773 "in %s RC mode.\n", rc_mode->name);
1774 }
1775
1776 hrd_buffer_size = 0;
1777 hrd_initial_buffer_fullness = 0;
1778
1779 if (rc_mode->mode != RC_MODE_AVBR) {
1780 // Already set (with completely different meaning) for AVBR.
1781 rc_window_size = 1000;
1782 }
1783 }
1784
1785 if (rc_bits_per_second > UINT32_MAX ||
1786 hrd_buffer_size > UINT32_MAX ||
1787 hrd_initial_buffer_fullness > UINT32_MAX) {
1788 av_log(avctx, AV_LOG_ERROR, "RC parameters of 2^32 or "
1789 "greater are not supported by VAAPI.\n");
1790 return AVERROR(EINVAL);
1791 }
1792
1793 ctx->rc_mode = rc_mode;
1794 ctx->rc_quality = rc_quality;
1795 ctx->va_rc_mode = rc_mode->va_mode;
1796 ctx->va_bit_rate = rc_bits_per_second;
1797
1798 av_log(avctx, AV_LOG_VERBOSE, "RC mode: %s.\n", rc_mode->name);
1799 if (rc_attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1800 // This driver does not want the RC mode attribute to be set.
1801 } else {
1802 ctx->config_attributes[ctx->nb_config_attributes++] =
1803 (VAConfigAttrib) {
1804 .type = VAConfigAttribRateControl,
1805 .value = ctx->va_rc_mode,
1806 };
1807 }
1808
1809 if (rc_mode->quality)
1810 av_log(avctx, AV_LOG_VERBOSE, "RC quality: %d.\n", rc_quality);
1811
1812 if (rc_mode->va_mode != VA_RC_CQP) {
1813 if (rc_mode->mode == RC_MODE_AVBR) {
1814 av_log(avctx, AV_LOG_VERBOSE, "RC target: %"PRId64" bps "
1815 "converging in %d frames with %d%% accuracy.\n",
1816 rc_bits_per_second, rc_window_size,
1817 rc_target_percentage);
1818 } else if (rc_mode->bitrate) {
1819 av_log(avctx, AV_LOG_VERBOSE, "RC target: %d%% of "
1820 "%"PRId64" bps over %d ms.\n", rc_target_percentage,
1821 rc_bits_per_second, rc_window_size);
1822 }
1823
1824 ctx->rc_params = (VAEncMiscParameterRateControl) {
1825 .bits_per_second = rc_bits_per_second,
1826 .target_percentage = rc_target_percentage,
1827 .window_size = rc_window_size,
1828 .initial_qp = 0,
1829 .min_qp = (avctx->qmin > 0 ? avctx->qmin : 0),
1830 .basic_unit_size = 0,
1831 #if VA_CHECK_VERSION(1, 1, 0)
1832 .ICQ_quality_factor = av_clip(rc_quality, 1, 51),
1833 .max_qp = (avctx->qmax > 0 ? avctx->qmax : 0),
1834 #endif
1835 #if VA_CHECK_VERSION(1, 3, 0)
1836 .quality_factor = rc_quality,
1837 #endif
1838 };
1839 vaapi_encode_add_global_param(avctx,
1840 VAEncMiscParameterTypeRateControl,
1841 &ctx->rc_params,
1842 sizeof(ctx->rc_params));
1843 }
1844
1845 if (rc_mode->hrd) {
1846 av_log(avctx, AV_LOG_VERBOSE, "RC buffer: %"PRId64" bits, "
1847 "initial fullness %"PRId64" bits.\n",
1848 hrd_buffer_size, hrd_initial_buffer_fullness);
1849
1850 ctx->hrd_params = (VAEncMiscParameterHRD) {
1851 .initial_buffer_fullness = hrd_initial_buffer_fullness,
1852 .buffer_size = hrd_buffer_size,
1853 };
1854 vaapi_encode_add_global_param(avctx,
1855 VAEncMiscParameterTypeHRD,
1856 &ctx->hrd_params,
1857 sizeof(ctx->hrd_params));
1858 }
1859
1860 if (avctx->framerate.num > 0 && avctx->framerate.den > 0)
1861 av_reduce(&fr_num, &fr_den,
1862 avctx->framerate.num, avctx->framerate.den, 65535);
1863 else
1864 av_reduce(&fr_num, &fr_den,
1865 avctx->time_base.den, avctx->time_base.num, 65535);
1866
1867 av_log(avctx, AV_LOG_VERBOSE, "RC framerate: %d/%d (%.2f fps).\n",
1868 fr_num, fr_den, (double)fr_num / fr_den);
1869
1870 ctx->fr_params = (VAEncMiscParameterFrameRate) {
1871 .framerate = (unsigned int)fr_den << 16 | fr_num,
1872 };
1873 #if VA_CHECK_VERSION(0, 40, 0)
1874 vaapi_encode_add_global_param(avctx,
1875 VAEncMiscParameterTypeFrameRate,
1876 &ctx->fr_params,
1877 sizeof(ctx->fr_params));
1878 #endif
1879
1880 return 0;
1881 }
1882
vaapi_encode_init_max_frame_size(AVCodecContext * avctx)1883 static av_cold int vaapi_encode_init_max_frame_size(AVCodecContext *avctx)
1884 {
1885 #if VA_CHECK_VERSION(1, 5, 0)
1886 VAAPIEncodeContext *ctx = avctx->priv_data;
1887 VAConfigAttrib attr = { VAConfigAttribMaxFrameSize };
1888 VAStatus vas;
1889
1890 if (ctx->va_rc_mode == VA_RC_CQP) {
1891 ctx->max_frame_size = 0;
1892 av_log(avctx, AV_LOG_ERROR, "Max frame size is invalid in CQP rate "
1893 "control mode.\n");
1894 return AVERROR(EINVAL);
1895 }
1896
1897 vas = vaGetConfigAttributes(ctx->hwctx->display,
1898 ctx->va_profile,
1899 ctx->va_entrypoint,
1900 &attr, 1);
1901 if (vas != VA_STATUS_SUCCESS) {
1902 ctx->max_frame_size = 0;
1903 av_log(avctx, AV_LOG_ERROR, "Failed to query max frame size "
1904 "config attribute: %d (%s).\n", vas, vaErrorStr(vas));
1905 return AVERROR_EXTERNAL;
1906 }
1907
1908 if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1909 ctx->max_frame_size = 0;
1910 av_log(avctx, AV_LOG_ERROR, "Max frame size attribute "
1911 "is not supported.\n");
1912 return AVERROR(EINVAL);
1913 } else {
1914 VAConfigAttribValMaxFrameSize attr_mfs;
1915 attr_mfs.value = attr.value;
1916 // Prefer to use VAEncMiscParameterTypeMaxFrameSize for max frame size.
1917 if (!attr_mfs.bits.max_frame_size && attr_mfs.bits.multiple_pass) {
1918 ctx->max_frame_size = 0;
1919 av_log(avctx, AV_LOG_ERROR, "Driver only supports multiple pass "
1920 "max frame size which has not been implemented in FFmpeg.\n");
1921 return AVERROR(EINVAL);
1922 }
1923
1924 ctx->mfs_params = (VAEncMiscParameterBufferMaxFrameSize){
1925 .max_frame_size = ctx->max_frame_size * 8,
1926 };
1927
1928 av_log(avctx, AV_LOG_VERBOSE, "Set max frame size: %d bytes.\n",
1929 ctx->max_frame_size);
1930 }
1931 #else
1932 av_log(avctx, AV_LOG_ERROR, "The max frame size option is not supported with "
1933 "this VAAPI version.\n");
1934 return AVERROR(EINVAL);
1935 #endif
1936
1937 return 0;
1938 }
1939
vaapi_encode_init_gop_structure(AVCodecContext * avctx)1940 static av_cold int vaapi_encode_init_gop_structure(AVCodecContext *avctx)
1941 {
1942 VAAPIEncodeContext *ctx = avctx->priv_data;
1943 VAStatus vas;
1944 VAConfigAttrib attr = { VAConfigAttribEncMaxRefFrames };
1945 uint32_t ref_l0, ref_l1;
1946 int prediction_pre_only;
1947
1948 vas = vaGetConfigAttributes(ctx->hwctx->display,
1949 ctx->va_profile,
1950 ctx->va_entrypoint,
1951 &attr, 1);
1952 if (vas != VA_STATUS_SUCCESS) {
1953 av_log(avctx, AV_LOG_ERROR, "Failed to query reference frames "
1954 "attribute: %d (%s).\n", vas, vaErrorStr(vas));
1955 return AVERROR_EXTERNAL;
1956 }
1957
1958 if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1959 ref_l0 = ref_l1 = 0;
1960 } else {
1961 ref_l0 = attr.value & 0xffff;
1962 ref_l1 = attr.value >> 16 & 0xffff;
1963 }
1964
1965 ctx->p_to_gpb = 0;
1966 prediction_pre_only = 0;
1967
1968 #if VA_CHECK_VERSION(1, 9, 0)
1969 if (!(ctx->codec->flags & FLAG_INTRA_ONLY ||
1970 avctx->gop_size <= 1)) {
1971 attr = (VAConfigAttrib) { VAConfigAttribPredictionDirection };
1972 vas = vaGetConfigAttributes(ctx->hwctx->display,
1973 ctx->va_profile,
1974 ctx->va_entrypoint,
1975 &attr, 1);
1976 if (vas != VA_STATUS_SUCCESS) {
1977 av_log(avctx, AV_LOG_WARNING, "Failed to query prediction direction "
1978 "attribute: %d (%s).\n", vas, vaErrorStr(vas));
1979 return AVERROR_EXTERNAL;
1980 } else if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
1981 av_log(avctx, AV_LOG_VERBOSE, "Driver does not report any additional "
1982 "prediction constraints.\n");
1983 } else {
1984 if (((ref_l0 > 0 || ref_l1 > 0) && !(attr.value & VA_PREDICTION_DIRECTION_PREVIOUS)) ||
1985 ((ref_l1 == 0) && (attr.value & (VA_PREDICTION_DIRECTION_FUTURE | VA_PREDICTION_DIRECTION_BI_NOT_EMPTY)))) {
1986 av_log(avctx, AV_LOG_ERROR, "Driver report incorrect prediction "
1987 "direction attribute.\n");
1988 return AVERROR_EXTERNAL;
1989 }
1990
1991 if (!(attr.value & VA_PREDICTION_DIRECTION_FUTURE)) {
1992 if (ref_l0 > 0 && ref_l1 > 0) {
1993 prediction_pre_only = 1;
1994 av_log(avctx, AV_LOG_VERBOSE, "Driver only support same reference "
1995 "lists for B-frames.\n");
1996 }
1997 }
1998
1999 if (attr.value & VA_PREDICTION_DIRECTION_BI_NOT_EMPTY) {
2000 if (ref_l0 > 0 && ref_l1 > 0) {
2001 ctx->p_to_gpb = 1;
2002 av_log(avctx, AV_LOG_VERBOSE, "Driver does not support P-frames, "
2003 "replacing them with B-frames.\n");
2004 }
2005 }
2006 }
2007 }
2008 #endif
2009
2010 if (ctx->codec->flags & FLAG_INTRA_ONLY ||
2011 avctx->gop_size <= 1) {
2012 av_log(avctx, AV_LOG_VERBOSE, "Using intra frames only.\n");
2013 ctx->gop_size = 1;
2014 } else if (ref_l0 < 1) {
2015 av_log(avctx, AV_LOG_ERROR, "Driver does not support any "
2016 "reference frames.\n");
2017 return AVERROR(EINVAL);
2018 } else if (!(ctx->codec->flags & FLAG_B_PICTURES) ||
2019 ref_l1 < 1 || avctx->max_b_frames < 1 ||
2020 prediction_pre_only) {
2021 if (ctx->p_to_gpb)
2022 av_log(avctx, AV_LOG_VERBOSE, "Using intra and B-frames "
2023 "(supported references: %d / %d).\n",
2024 ref_l0, ref_l1);
2025 else
2026 av_log(avctx, AV_LOG_VERBOSE, "Using intra and P-frames "
2027 "(supported references: %d / %d).\n", ref_l0, ref_l1);
2028 ctx->gop_size = avctx->gop_size;
2029 ctx->p_per_i = INT_MAX;
2030 ctx->b_per_p = 0;
2031 } else {
2032 if (ctx->p_to_gpb)
2033 av_log(avctx, AV_LOG_VERBOSE, "Using intra and B-frames "
2034 "(supported references: %d / %d).\n",
2035 ref_l0, ref_l1);
2036 else
2037 av_log(avctx, AV_LOG_VERBOSE, "Using intra, P- and B-frames "
2038 "(supported references: %d / %d).\n", ref_l0, ref_l1);
2039 ctx->gop_size = avctx->gop_size;
2040 ctx->p_per_i = INT_MAX;
2041 ctx->b_per_p = avctx->max_b_frames;
2042 if (ctx->codec->flags & FLAG_B_PICTURE_REFERENCES) {
2043 ctx->max_b_depth = FFMIN(ctx->desired_b_depth,
2044 av_log2(ctx->b_per_p) + 1);
2045 } else {
2046 ctx->max_b_depth = 1;
2047 }
2048 }
2049
2050 if (ctx->codec->flags & FLAG_NON_IDR_KEY_PICTURES) {
2051 ctx->closed_gop = !!(avctx->flags & AV_CODEC_FLAG_CLOSED_GOP);
2052 ctx->gop_per_idr = ctx->idr_interval + 1;
2053 } else {
2054 ctx->closed_gop = 1;
2055 ctx->gop_per_idr = 1;
2056 }
2057
2058 return 0;
2059 }
2060
vaapi_encode_init_row_slice_structure(AVCodecContext * avctx,uint32_t slice_structure)2061 static av_cold int vaapi_encode_init_row_slice_structure(AVCodecContext *avctx,
2062 uint32_t slice_structure)
2063 {
2064 VAAPIEncodeContext *ctx = avctx->priv_data;
2065 int req_slices;
2066
2067 // For fixed-size slices currently we only support whole rows, making
2068 // rectangular slices. This could be extended to arbitrary runs of
2069 // blocks, but since slices tend to be a conformance requirement and
2070 // most cases (such as broadcast or bluray) want rectangular slices
2071 // only it would need to be gated behind another option.
2072 if (avctx->slices > ctx->slice_block_rows) {
2073 av_log(avctx, AV_LOG_WARNING, "Not enough rows to use "
2074 "configured number of slices (%d < %d); using "
2075 "maximum.\n", ctx->slice_block_rows, avctx->slices);
2076 req_slices = ctx->slice_block_rows;
2077 } else {
2078 req_slices = avctx->slices;
2079 }
2080 if (slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_ROWS ||
2081 slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_MACROBLOCKS) {
2082 ctx->nb_slices = req_slices;
2083 ctx->slice_size = ctx->slice_block_rows / ctx->nb_slices;
2084 } else if (slice_structure & VA_ENC_SLICE_STRUCTURE_POWER_OF_TWO_ROWS) {
2085 int k;
2086 for (k = 1;; k *= 2) {
2087 if (2 * k * (req_slices - 1) + 1 >= ctx->slice_block_rows)
2088 break;
2089 }
2090 ctx->nb_slices = (ctx->slice_block_rows + k - 1) / k;
2091 ctx->slice_size = k;
2092 #if VA_CHECK_VERSION(1, 0, 0)
2093 } else if (slice_structure & VA_ENC_SLICE_STRUCTURE_EQUAL_ROWS) {
2094 ctx->nb_slices = ctx->slice_block_rows;
2095 ctx->slice_size = 1;
2096 #endif
2097 } else {
2098 av_log(avctx, AV_LOG_ERROR, "Driver does not support any usable "
2099 "slice structure modes (%#x).\n", slice_structure);
2100 return AVERROR(EINVAL);
2101 }
2102
2103 return 0;
2104 }
2105
vaapi_encode_init_tile_slice_structure(AVCodecContext * avctx,uint32_t slice_structure)2106 static av_cold int vaapi_encode_init_tile_slice_structure(AVCodecContext *avctx,
2107 uint32_t slice_structure)
2108 {
2109 VAAPIEncodeContext *ctx = avctx->priv_data;
2110 int i, req_tiles;
2111
2112 if (!(slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_MACROBLOCKS ||
2113 (slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_ROWS &&
2114 ctx->tile_cols == 1))) {
2115 av_log(avctx, AV_LOG_ERROR, "Supported slice structure (%#x) doesn't work for "
2116 "current tile requirement.\n", slice_structure);
2117 return AVERROR(EINVAL);
2118 }
2119
2120 if (ctx->tile_rows > ctx->slice_block_rows ||
2121 ctx->tile_cols > ctx->slice_block_cols) {
2122 av_log(avctx, AV_LOG_WARNING, "Not enough block rows/cols (%d x %d) "
2123 "for configured number of tile (%d x %d); ",
2124 ctx->slice_block_rows, ctx->slice_block_cols,
2125 ctx->tile_rows, ctx->tile_cols);
2126 ctx->tile_rows = ctx->tile_rows > ctx->slice_block_rows ?
2127 ctx->slice_block_rows : ctx->tile_rows;
2128 ctx->tile_cols = ctx->tile_cols > ctx->slice_block_cols ?
2129 ctx->slice_block_cols : ctx->tile_cols;
2130 av_log(avctx, AV_LOG_WARNING, "using allowed maximum (%d x %d).\n",
2131 ctx->tile_rows, ctx->tile_cols);
2132 }
2133
2134 req_tiles = ctx->tile_rows * ctx->tile_cols;
2135
2136 // Tile slice is not allowed to cross the boundary of a tile due to
2137 // the constraints of media-driver. Currently we support one slice
2138 // per tile. This could be extended to multiple slices per tile.
2139 if (avctx->slices != req_tiles)
2140 av_log(avctx, AV_LOG_WARNING, "The number of requested slices "
2141 "mismatches with configured number of tile (%d != %d); "
2142 "using requested tile number for slice.\n",
2143 avctx->slices, req_tiles);
2144
2145 ctx->nb_slices = req_tiles;
2146
2147 // Default in uniform spacing
2148 // 6-3, 6-5
2149 for (i = 0; i < ctx->tile_cols; i++) {
2150 ctx->col_width[i] = ( i + 1 ) * ctx->slice_block_cols / ctx->tile_cols -
2151 i * ctx->slice_block_cols / ctx->tile_cols;
2152 ctx->col_bd[i + 1] = ctx->col_bd[i] + ctx->col_width[i];
2153 }
2154 // 6-4, 6-6
2155 for (i = 0; i < ctx->tile_rows; i++) {
2156 ctx->row_height[i] = ( i + 1 ) * ctx->slice_block_rows / ctx->tile_rows -
2157 i * ctx->slice_block_rows / ctx->tile_rows;
2158 ctx->row_bd[i + 1] = ctx->row_bd[i] + ctx->row_height[i];
2159 }
2160
2161 av_log(avctx, AV_LOG_VERBOSE, "Encoding pictures with %d x %d tile.\n",
2162 ctx->tile_rows, ctx->tile_cols);
2163
2164 return 0;
2165 }
2166
vaapi_encode_init_slice_structure(AVCodecContext * avctx)2167 static av_cold int vaapi_encode_init_slice_structure(AVCodecContext *avctx)
2168 {
2169 VAAPIEncodeContext *ctx = avctx->priv_data;
2170 VAConfigAttrib attr[3] = { { VAConfigAttribEncMaxSlices },
2171 { VAConfigAttribEncSliceStructure },
2172 #if VA_CHECK_VERSION(1, 1, 0)
2173 { VAConfigAttribEncTileSupport },
2174 #endif
2175 };
2176 VAStatus vas;
2177 uint32_t max_slices, slice_structure;
2178 int ret;
2179
2180 if (!(ctx->codec->flags & FLAG_SLICE_CONTROL)) {
2181 if (avctx->slices > 0) {
2182 av_log(avctx, AV_LOG_WARNING, "Multiple slices were requested "
2183 "but this codec does not support controlling slices.\n");
2184 }
2185 return 0;
2186 }
2187
2188 av_assert0(ctx->slice_block_height > 0 && ctx->slice_block_width > 0);
2189
2190 ctx->slice_block_rows = (avctx->height + ctx->slice_block_height - 1) /
2191 ctx->slice_block_height;
2192 ctx->slice_block_cols = (avctx->width + ctx->slice_block_width - 1) /
2193 ctx->slice_block_width;
2194
2195 if (avctx->slices <= 1 && !ctx->tile_rows && !ctx->tile_cols) {
2196 ctx->nb_slices = 1;
2197 ctx->slice_size = ctx->slice_block_rows;
2198 return 0;
2199 }
2200
2201 vas = vaGetConfigAttributes(ctx->hwctx->display,
2202 ctx->va_profile,
2203 ctx->va_entrypoint,
2204 attr, FF_ARRAY_ELEMS(attr));
2205 if (vas != VA_STATUS_SUCCESS) {
2206 av_log(avctx, AV_LOG_ERROR, "Failed to query slice "
2207 "attributes: %d (%s).\n", vas, vaErrorStr(vas));
2208 return AVERROR_EXTERNAL;
2209 }
2210 max_slices = attr[0].value;
2211 slice_structure = attr[1].value;
2212 if (max_slices == VA_ATTRIB_NOT_SUPPORTED ||
2213 slice_structure == VA_ATTRIB_NOT_SUPPORTED) {
2214 av_log(avctx, AV_LOG_ERROR, "Driver does not support encoding "
2215 "pictures as multiple slices.\n.");
2216 return AVERROR(EINVAL);
2217 }
2218
2219 if (ctx->tile_rows && ctx->tile_cols) {
2220 #if VA_CHECK_VERSION(1, 1, 0)
2221 uint32_t tile_support = attr[2].value;
2222 if (tile_support == VA_ATTRIB_NOT_SUPPORTED) {
2223 av_log(avctx, AV_LOG_ERROR, "Driver does not support encoding "
2224 "pictures as multiple tiles.\n.");
2225 return AVERROR(EINVAL);
2226 }
2227 #else
2228 av_log(avctx, AV_LOG_ERROR, "Tile encoding option is "
2229 "not supported with this VAAPI version.\n");
2230 return AVERROR(EINVAL);
2231 #endif
2232 }
2233
2234 if (ctx->tile_rows && ctx->tile_cols)
2235 ret = vaapi_encode_init_tile_slice_structure(avctx, slice_structure);
2236 else
2237 ret = vaapi_encode_init_row_slice_structure(avctx, slice_structure);
2238 if (ret < 0)
2239 return ret;
2240
2241 if (ctx->nb_slices > avctx->slices) {
2242 av_log(avctx, AV_LOG_WARNING, "Slice count rounded up to "
2243 "%d (from %d) due to driver constraints on slice "
2244 "structure.\n", ctx->nb_slices, avctx->slices);
2245 }
2246 if (ctx->nb_slices > max_slices) {
2247 av_log(avctx, AV_LOG_ERROR, "Driver does not support "
2248 "encoding with %d slices (max %"PRIu32").\n",
2249 ctx->nb_slices, max_slices);
2250 return AVERROR(EINVAL);
2251 }
2252
2253 av_log(avctx, AV_LOG_VERBOSE, "Encoding pictures with %d slices.\n",
2254 ctx->nb_slices);
2255 return 0;
2256 }
2257
vaapi_encode_init_packed_headers(AVCodecContext * avctx)2258 static av_cold int vaapi_encode_init_packed_headers(AVCodecContext *avctx)
2259 {
2260 VAAPIEncodeContext *ctx = avctx->priv_data;
2261 VAStatus vas;
2262 VAConfigAttrib attr = { VAConfigAttribEncPackedHeaders };
2263
2264 vas = vaGetConfigAttributes(ctx->hwctx->display,
2265 ctx->va_profile,
2266 ctx->va_entrypoint,
2267 &attr, 1);
2268 if (vas != VA_STATUS_SUCCESS) {
2269 av_log(avctx, AV_LOG_ERROR, "Failed to query packed headers "
2270 "attribute: %d (%s).\n", vas, vaErrorStr(vas));
2271 return AVERROR_EXTERNAL;
2272 }
2273
2274 if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
2275 if (ctx->desired_packed_headers) {
2276 av_log(avctx, AV_LOG_WARNING, "Driver does not support any "
2277 "packed headers (wanted %#x).\n",
2278 ctx->desired_packed_headers);
2279 } else {
2280 av_log(avctx, AV_LOG_VERBOSE, "Driver does not support any "
2281 "packed headers (none wanted).\n");
2282 }
2283 ctx->va_packed_headers = 0;
2284 } else {
2285 if (ctx->desired_packed_headers & ~attr.value) {
2286 av_log(avctx, AV_LOG_WARNING, "Driver does not support some "
2287 "wanted packed headers (wanted %#x, found %#x).\n",
2288 ctx->desired_packed_headers, attr.value);
2289 } else {
2290 av_log(avctx, AV_LOG_VERBOSE, "All wanted packed headers "
2291 "available (wanted %#x, found %#x).\n",
2292 ctx->desired_packed_headers, attr.value);
2293 }
2294 ctx->va_packed_headers = ctx->desired_packed_headers & attr.value;
2295 }
2296
2297 if (ctx->va_packed_headers) {
2298 ctx->config_attributes[ctx->nb_config_attributes++] =
2299 (VAConfigAttrib) {
2300 .type = VAConfigAttribEncPackedHeaders,
2301 .value = ctx->va_packed_headers,
2302 };
2303 }
2304
2305 if ( (ctx->desired_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE) &&
2306 !(ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE) &&
2307 (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER)) {
2308 av_log(avctx, AV_LOG_WARNING, "Driver does not support packed "
2309 "sequence headers, but a global header is requested.\n");
2310 av_log(avctx, AV_LOG_WARNING, "No global header will be written: "
2311 "this may result in a stream which is not usable for some "
2312 "purposes (e.g. not muxable to some containers).\n");
2313 }
2314
2315 return 0;
2316 }
2317
vaapi_encode_init_quality(AVCodecContext * avctx)2318 static av_cold int vaapi_encode_init_quality(AVCodecContext *avctx)
2319 {
2320 #if VA_CHECK_VERSION(0, 36, 0)
2321 VAAPIEncodeContext *ctx = avctx->priv_data;
2322 VAStatus vas;
2323 VAConfigAttrib attr = { VAConfigAttribEncQualityRange };
2324 int quality = avctx->compression_level;
2325
2326 vas = vaGetConfigAttributes(ctx->hwctx->display,
2327 ctx->va_profile,
2328 ctx->va_entrypoint,
2329 &attr, 1);
2330 if (vas != VA_STATUS_SUCCESS) {
2331 av_log(avctx, AV_LOG_ERROR, "Failed to query quality "
2332 "config attribute: %d (%s).\n", vas, vaErrorStr(vas));
2333 return AVERROR_EXTERNAL;
2334 }
2335
2336 if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
2337 if (quality != 0) {
2338 av_log(avctx, AV_LOG_WARNING, "Quality attribute is not "
2339 "supported: will use default quality level.\n");
2340 }
2341 } else {
2342 if (quality > attr.value) {
2343 av_log(avctx, AV_LOG_WARNING, "Invalid quality level: "
2344 "valid range is 0-%d, using %d.\n",
2345 attr.value, attr.value);
2346 quality = attr.value;
2347 }
2348
2349 ctx->quality_params = (VAEncMiscParameterBufferQualityLevel) {
2350 .quality_level = quality,
2351 };
2352 vaapi_encode_add_global_param(avctx,
2353 VAEncMiscParameterTypeQualityLevel,
2354 &ctx->quality_params,
2355 sizeof(ctx->quality_params));
2356 }
2357 #else
2358 av_log(avctx, AV_LOG_WARNING, "The encode quality option is "
2359 "not supported with this VAAPI version.\n");
2360 #endif
2361
2362 return 0;
2363 }
2364
vaapi_encode_init_roi(AVCodecContext * avctx)2365 static av_cold int vaapi_encode_init_roi(AVCodecContext *avctx)
2366 {
2367 #if VA_CHECK_VERSION(1, 0, 0)
2368 VAAPIEncodeContext *ctx = avctx->priv_data;
2369 VAStatus vas;
2370 VAConfigAttrib attr = { VAConfigAttribEncROI };
2371
2372 vas = vaGetConfigAttributes(ctx->hwctx->display,
2373 ctx->va_profile,
2374 ctx->va_entrypoint,
2375 &attr, 1);
2376 if (vas != VA_STATUS_SUCCESS) {
2377 av_log(avctx, AV_LOG_ERROR, "Failed to query ROI "
2378 "config attribute: %d (%s).\n", vas, vaErrorStr(vas));
2379 return AVERROR_EXTERNAL;
2380 }
2381
2382 if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
2383 ctx->roi_allowed = 0;
2384 } else {
2385 VAConfigAttribValEncROI roi = {
2386 .value = attr.value,
2387 };
2388
2389 ctx->roi_max_regions = roi.bits.num_roi_regions;
2390 ctx->roi_allowed = ctx->roi_max_regions > 0 &&
2391 (ctx->va_rc_mode == VA_RC_CQP ||
2392 roi.bits.roi_rc_qp_delta_support);
2393 }
2394 #endif
2395 return 0;
2396 }
2397
vaapi_encode_free_output_buffer(void * opaque,uint8_t * data)2398 static void vaapi_encode_free_output_buffer(void *opaque,
2399 uint8_t *data)
2400 {
2401 AVCodecContext *avctx = opaque;
2402 VAAPIEncodeContext *ctx = avctx->priv_data;
2403 VABufferID buffer_id;
2404
2405 buffer_id = (VABufferID)(uintptr_t)data;
2406
2407 vaDestroyBuffer(ctx->hwctx->display, buffer_id);
2408
2409 av_log(avctx, AV_LOG_DEBUG, "Freed output buffer %#x\n", buffer_id);
2410 }
2411
vaapi_encode_alloc_output_buffer(void * opaque,size_t size)2412 static AVBufferRef *vaapi_encode_alloc_output_buffer(void *opaque,
2413 size_t size)
2414 {
2415 AVCodecContext *avctx = opaque;
2416 VAAPIEncodeContext *ctx = avctx->priv_data;
2417 VABufferID buffer_id;
2418 VAStatus vas;
2419 AVBufferRef *ref;
2420
2421 // The output buffer size is fixed, so it needs to be large enough
2422 // to hold the largest possible compressed frame. We assume here
2423 // that the uncompressed frame plus some header data is an upper
2424 // bound on that.
2425 vas = vaCreateBuffer(ctx->hwctx->display, ctx->va_context,
2426 VAEncCodedBufferType,
2427 3 * ctx->surface_width * ctx->surface_height +
2428 (1 << 16), 1, 0, &buffer_id);
2429 if (vas != VA_STATUS_SUCCESS) {
2430 av_log(avctx, AV_LOG_ERROR, "Failed to create bitstream "
2431 "output buffer: %d (%s).\n", vas, vaErrorStr(vas));
2432 return NULL;
2433 }
2434
2435 av_log(avctx, AV_LOG_DEBUG, "Allocated output buffer %#x\n", buffer_id);
2436
2437 ref = av_buffer_create((uint8_t*)(uintptr_t)buffer_id,
2438 sizeof(buffer_id),
2439 &vaapi_encode_free_output_buffer,
2440 avctx, AV_BUFFER_FLAG_READONLY);
2441 if (!ref) {
2442 vaDestroyBuffer(ctx->hwctx->display, buffer_id);
2443 return NULL;
2444 }
2445
2446 return ref;
2447 }
2448
vaapi_encode_create_recon_frames(AVCodecContext * avctx)2449 static av_cold int vaapi_encode_create_recon_frames(AVCodecContext *avctx)
2450 {
2451 VAAPIEncodeContext *ctx = avctx->priv_data;
2452 AVVAAPIHWConfig *hwconfig = NULL;
2453 AVHWFramesConstraints *constraints = NULL;
2454 enum AVPixelFormat recon_format;
2455 int err, i;
2456
2457 hwconfig = av_hwdevice_hwconfig_alloc(ctx->device_ref);
2458 if (!hwconfig) {
2459 err = AVERROR(ENOMEM);
2460 goto fail;
2461 }
2462 hwconfig->config_id = ctx->va_config;
2463
2464 constraints = av_hwdevice_get_hwframe_constraints(ctx->device_ref,
2465 hwconfig);
2466 if (!constraints) {
2467 err = AVERROR(ENOMEM);
2468 goto fail;
2469 }
2470
2471 // Probably we can use the input surface format as the surface format
2472 // of the reconstructed frames. If not, we just pick the first (only?)
2473 // format in the valid list and hope that it all works.
2474 recon_format = AV_PIX_FMT_NONE;
2475 if (constraints->valid_sw_formats) {
2476 for (i = 0; constraints->valid_sw_formats[i] != AV_PIX_FMT_NONE; i++) {
2477 if (ctx->input_frames->sw_format ==
2478 constraints->valid_sw_formats[i]) {
2479 recon_format = ctx->input_frames->sw_format;
2480 break;
2481 }
2482 }
2483 if (recon_format == AV_PIX_FMT_NONE) {
2484 // No match. Just use the first in the supported list and
2485 // hope for the best.
2486 recon_format = constraints->valid_sw_formats[0];
2487 }
2488 } else {
2489 // No idea what to use; copy input format.
2490 recon_format = ctx->input_frames->sw_format;
2491 }
2492 av_log(avctx, AV_LOG_DEBUG, "Using %s as format of "
2493 "reconstructed frames.\n", av_get_pix_fmt_name(recon_format));
2494
2495 if (ctx->surface_width < constraints->min_width ||
2496 ctx->surface_height < constraints->min_height ||
2497 ctx->surface_width > constraints->max_width ||
2498 ctx->surface_height > constraints->max_height) {
2499 av_log(avctx, AV_LOG_ERROR, "Hardware does not support encoding at "
2500 "size %dx%d (constraints: width %d-%d height %d-%d).\n",
2501 ctx->surface_width, ctx->surface_height,
2502 constraints->min_width, constraints->max_width,
2503 constraints->min_height, constraints->max_height);
2504 err = AVERROR(EINVAL);
2505 goto fail;
2506 }
2507
2508 av_freep(&hwconfig);
2509 av_hwframe_constraints_free(&constraints);
2510
2511 ctx->recon_frames_ref = av_hwframe_ctx_alloc(ctx->device_ref);
2512 if (!ctx->recon_frames_ref) {
2513 err = AVERROR(ENOMEM);
2514 goto fail;
2515 }
2516 ctx->recon_frames = (AVHWFramesContext*)ctx->recon_frames_ref->data;
2517
2518 ctx->recon_frames->format = AV_PIX_FMT_VAAPI;
2519 ctx->recon_frames->sw_format = recon_format;
2520 ctx->recon_frames->width = ctx->surface_width;
2521 ctx->recon_frames->height = ctx->surface_height;
2522
2523 err = av_hwframe_ctx_init(ctx->recon_frames_ref);
2524 if (err < 0) {
2525 av_log(avctx, AV_LOG_ERROR, "Failed to initialise reconstructed "
2526 "frame context: %d.\n", err);
2527 goto fail;
2528 }
2529
2530 err = 0;
2531 fail:
2532 av_freep(&hwconfig);
2533 av_hwframe_constraints_free(&constraints);
2534 return err;
2535 }
2536
ff_vaapi_encode_init(AVCodecContext * avctx)2537 av_cold int ff_vaapi_encode_init(AVCodecContext *avctx)
2538 {
2539 VAAPIEncodeContext *ctx = avctx->priv_data;
2540 AVVAAPIFramesContext *recon_hwctx = NULL;
2541 VAStatus vas;
2542 int err;
2543
2544 ctx->va_config = VA_INVALID_ID;
2545 ctx->va_context = VA_INVALID_ID;
2546
2547 /* If you add something that can fail above this av_frame_alloc(),
2548 * modify ff_vaapi_encode_close() accordingly. */
2549 ctx->frame = av_frame_alloc();
2550 if (!ctx->frame) {
2551 return AVERROR(ENOMEM);
2552 }
2553
2554 if (!avctx->hw_frames_ctx) {
2555 av_log(avctx, AV_LOG_ERROR, "A hardware frames reference is "
2556 "required to associate the encoding device.\n");
2557 return AVERROR(EINVAL);
2558 }
2559
2560 ctx->input_frames_ref = av_buffer_ref(avctx->hw_frames_ctx);
2561 if (!ctx->input_frames_ref) {
2562 err = AVERROR(ENOMEM);
2563 goto fail;
2564 }
2565 ctx->input_frames = (AVHWFramesContext*)ctx->input_frames_ref->data;
2566
2567 ctx->device_ref = av_buffer_ref(ctx->input_frames->device_ref);
2568 if (!ctx->device_ref) {
2569 err = AVERROR(ENOMEM);
2570 goto fail;
2571 }
2572 ctx->device = (AVHWDeviceContext*)ctx->device_ref->data;
2573 ctx->hwctx = ctx->device->hwctx;
2574
2575 err = vaapi_encode_profile_entrypoint(avctx);
2576 if (err < 0)
2577 goto fail;
2578
2579 if (ctx->codec->get_encoder_caps) {
2580 err = ctx->codec->get_encoder_caps(avctx);
2581 if (err < 0)
2582 goto fail;
2583 } else {
2584 // Assume 16x16 blocks.
2585 ctx->surface_width = FFALIGN(avctx->width, 16);
2586 ctx->surface_height = FFALIGN(avctx->height, 16);
2587 if (ctx->codec->flags & FLAG_SLICE_CONTROL) {
2588 ctx->slice_block_width = 16;
2589 ctx->slice_block_height = 16;
2590 }
2591 }
2592
2593 err = vaapi_encode_init_rate_control(avctx);
2594 if (err < 0)
2595 goto fail;
2596
2597 err = vaapi_encode_init_gop_structure(avctx);
2598 if (err < 0)
2599 goto fail;
2600
2601 err = vaapi_encode_init_slice_structure(avctx);
2602 if (err < 0)
2603 goto fail;
2604
2605 err = vaapi_encode_init_packed_headers(avctx);
2606 if (err < 0)
2607 goto fail;
2608
2609 err = vaapi_encode_init_roi(avctx);
2610 if (err < 0)
2611 goto fail;
2612
2613 if (avctx->compression_level >= 0) {
2614 err = vaapi_encode_init_quality(avctx);
2615 if (err < 0)
2616 goto fail;
2617 }
2618
2619 if (ctx->max_frame_size) {
2620 err = vaapi_encode_init_max_frame_size(avctx);
2621 if (err < 0)
2622 goto fail;
2623 }
2624
2625 vas = vaCreateConfig(ctx->hwctx->display,
2626 ctx->va_profile, ctx->va_entrypoint,
2627 ctx->config_attributes, ctx->nb_config_attributes,
2628 &ctx->va_config);
2629 if (vas != VA_STATUS_SUCCESS) {
2630 av_log(avctx, AV_LOG_ERROR, "Failed to create encode pipeline "
2631 "configuration: %d (%s).\n", vas, vaErrorStr(vas));
2632 err = AVERROR(EIO);
2633 goto fail;
2634 }
2635
2636 err = vaapi_encode_create_recon_frames(avctx);
2637 if (err < 0)
2638 goto fail;
2639
2640 recon_hwctx = ctx->recon_frames->hwctx;
2641 vas = vaCreateContext(ctx->hwctx->display, ctx->va_config,
2642 ctx->surface_width, ctx->surface_height,
2643 VA_PROGRESSIVE,
2644 recon_hwctx->surface_ids,
2645 recon_hwctx->nb_surfaces,
2646 &ctx->va_context);
2647 if (vas != VA_STATUS_SUCCESS) {
2648 av_log(avctx, AV_LOG_ERROR, "Failed to create encode pipeline "
2649 "context: %d (%s).\n", vas, vaErrorStr(vas));
2650 err = AVERROR(EIO);
2651 goto fail;
2652 }
2653
2654 ctx->output_buffer_pool =
2655 av_buffer_pool_init2(sizeof(VABufferID), avctx,
2656 &vaapi_encode_alloc_output_buffer, NULL);
2657 if (!ctx->output_buffer_pool) {
2658 err = AVERROR(ENOMEM);
2659 goto fail;
2660 }
2661
2662 if (ctx->codec->configure) {
2663 err = ctx->codec->configure(avctx);
2664 if (err < 0)
2665 goto fail;
2666 }
2667
2668 ctx->output_delay = ctx->b_per_p;
2669 ctx->decode_delay = ctx->max_b_depth;
2670
2671 if (ctx->codec->sequence_params_size > 0) {
2672 ctx->codec_sequence_params =
2673 av_mallocz(ctx->codec->sequence_params_size);
2674 if (!ctx->codec_sequence_params) {
2675 err = AVERROR(ENOMEM);
2676 goto fail;
2677 }
2678 }
2679 if (ctx->codec->picture_params_size > 0) {
2680 ctx->codec_picture_params =
2681 av_mallocz(ctx->codec->picture_params_size);
2682 if (!ctx->codec_picture_params) {
2683 err = AVERROR(ENOMEM);
2684 goto fail;
2685 }
2686 }
2687
2688 if (ctx->codec->init_sequence_params) {
2689 err = ctx->codec->init_sequence_params(avctx);
2690 if (err < 0) {
2691 av_log(avctx, AV_LOG_ERROR, "Codec sequence initialisation "
2692 "failed: %d.\n", err);
2693 goto fail;
2694 }
2695 }
2696
2697 if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE &&
2698 ctx->codec->write_sequence_header &&
2699 avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) {
2700 char data[MAX_PARAM_BUFFER_SIZE];
2701 size_t bit_len = 8 * sizeof(data);
2702
2703 err = ctx->codec->write_sequence_header(avctx, data, &bit_len);
2704 if (err < 0) {
2705 av_log(avctx, AV_LOG_ERROR, "Failed to write sequence header "
2706 "for extradata: %d.\n", err);
2707 goto fail;
2708 } else {
2709 avctx->extradata_size = (bit_len + 7) / 8;
2710 avctx->extradata = av_mallocz(avctx->extradata_size +
2711 AV_INPUT_BUFFER_PADDING_SIZE);
2712 if (!avctx->extradata) {
2713 err = AVERROR(ENOMEM);
2714 goto fail;
2715 }
2716 memcpy(avctx->extradata, data, avctx->extradata_size);
2717 }
2718 }
2719
2720 #if VA_CHECK_VERSION(1, 9, 0)
2721 // check vaSyncBuffer function
2722 vas = vaSyncBuffer(ctx->hwctx->display, VA_INVALID_ID, 0);
2723 if (vas != VA_STATUS_ERROR_UNIMPLEMENTED) {
2724 ctx->has_sync_buffer_func = 1;
2725 ctx->encode_fifo = av_fifo_alloc2(ctx->async_depth,
2726 sizeof(VAAPIEncodePicture *),
2727 0);
2728 if (!ctx->encode_fifo)
2729 return AVERROR(ENOMEM);
2730 }
2731 #endif
2732
2733 return 0;
2734
2735 fail:
2736 return err;
2737 }
2738
ff_vaapi_encode_close(AVCodecContext * avctx)2739 av_cold int ff_vaapi_encode_close(AVCodecContext *avctx)
2740 {
2741 VAAPIEncodeContext *ctx = avctx->priv_data;
2742 VAAPIEncodePicture *pic, *next;
2743
2744 /* We check ctx->frame to know whether ff_vaapi_encode_init()
2745 * has been called and va_config/va_context initialized. */
2746 if (!ctx->frame)
2747 return 0;
2748
2749 for (pic = ctx->pic_start; pic; pic = next) {
2750 next = pic->next;
2751 vaapi_encode_free(avctx, pic);
2752 }
2753
2754 av_buffer_pool_uninit(&ctx->output_buffer_pool);
2755
2756 if (ctx->va_context != VA_INVALID_ID) {
2757 if (ctx->hwctx)
2758 vaDestroyContext(ctx->hwctx->display, ctx->va_context);
2759 ctx->va_context = VA_INVALID_ID;
2760 }
2761
2762 if (ctx->va_config != VA_INVALID_ID) {
2763 if (ctx->hwctx)
2764 vaDestroyConfig(ctx->hwctx->display, ctx->va_config);
2765 ctx->va_config = VA_INVALID_ID;
2766 }
2767
2768 av_frame_free(&ctx->frame);
2769
2770 av_freep(&ctx->codec_sequence_params);
2771 av_freep(&ctx->codec_picture_params);
2772 av_fifo_freep2(&ctx->encode_fifo);
2773
2774 av_buffer_unref(&ctx->recon_frames_ref);
2775 av_buffer_unref(&ctx->input_frames_ref);
2776 av_buffer_unref(&ctx->device_ref);
2777
2778 return 0;
2779 }
2780