2 * Copyright (c) 2018 Ronald S. Bultje <rsbultje gmail com>
3 * Copyright (c) 2018 James Almer <jamrial gmail com>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include <dav1d/dav1d.h>
24 #include "libavutil/avassert.h"
25 #include "libavutil/film_grain_params.h"
26 #include "libavutil/mastering_display_metadata.h"
27 #include "libavutil/imgutils.h"
28 #include "libavutil/opt.h"
32 #include "bytestream.h"
36 typedef struct Libdav1dContext {
50 static const enum AVPixelFormat pix_fmt[][3] = {
51 [DAV1D_PIXEL_LAYOUT_I400] = { AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12 },
52 [DAV1D_PIXEL_LAYOUT_I420] = { AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV420P12 },
53 [DAV1D_PIXEL_LAYOUT_I422] = { AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV422P12 },
54 [DAV1D_PIXEL_LAYOUT_I444] = { AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV444P12 },
57 static const enum AVPixelFormat pix_fmt_rgb[3] = {
58 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRP12,
61 static void libdav1d_log_callback(void *opaque, const char *fmt, va_list vl)
63 AVCodecContext *c = opaque;
65 av_vlog(c, AV_LOG_ERROR, fmt, vl);
68 static int libdav1d_picture_allocator(Dav1dPicture *p, void *cookie)
70 Libdav1dContext *dav1d = cookie;
71 enum AVPixelFormat format = pix_fmt[p->p.layout][p->seq_hdr->hbd];
72 int ret, linesize[4], h = FFALIGN(p->p.h, 128), w = FFALIGN(p->p.w, 128);
73 uint8_t *aligned_ptr, *data[4];
76 ret = av_image_get_buffer_size(format, w, h, DAV1D_PICTURE_ALIGNMENT);
80 if (ret != dav1d->pool_size) {
81 av_buffer_pool_uninit(&dav1d->pool);
82 // Use twice the amount of required padding bytes for aligned_ptr below.
83 dav1d->pool = av_buffer_pool_init(ret + DAV1D_PICTURE_ALIGNMENT * 2, NULL);
86 return AVERROR(ENOMEM);
88 dav1d->pool_size = ret;
90 buf = av_buffer_pool_get(dav1d->pool);
92 return AVERROR(ENOMEM);
94 // libdav1d requires DAV1D_PICTURE_ALIGNMENT aligned buffers, which av_malloc()
95 // doesn't guarantee for example when AVX is disabled at configure time.
96 // Use the extra DAV1D_PICTURE_ALIGNMENT padding bytes in the buffer to align it
98 aligned_ptr = (uint8_t *)FFALIGN((uintptr_t)buf->data, DAV1D_PICTURE_ALIGNMENT);
99 ret = av_image_fill_arrays(data, linesize, aligned_ptr, format, w, h,
100 DAV1D_PICTURE_ALIGNMENT);
102 av_buffer_unref(&buf);
106 p->data[0] = data[0];
107 p->data[1] = data[1];
108 p->data[2] = data[2];
109 p->stride[0] = linesize[0];
110 p->stride[1] = linesize[1];
111 p->allocator_data = buf;
116 static void libdav1d_picture_release(Dav1dPicture *p, void *cookie)
118 AVBufferRef *buf = p->allocator_data;
120 av_buffer_unref(&buf);
123 static av_cold int libdav1d_init(AVCodecContext *c)
125 Libdav1dContext *dav1d = c->priv_data;
127 int threads = (c->thread_count ? c->thread_count : av_cpu_count()) * 3 / 2;
130 av_log(c, AV_LOG_INFO, "libdav1d %s\n", dav1d_version());
132 dav1d_default_settings(&s);
134 s.logger.callback = libdav1d_log_callback;
135 s.allocator.cookie = dav1d;
136 s.allocator.alloc_picture_callback = libdav1d_picture_allocator;
137 s.allocator.release_picture_callback = libdav1d_picture_release;
138 s.frame_size_limit = c->max_pixels;
139 if (dav1d->apply_grain >= 0)
140 s.apply_grain = dav1d->apply_grain;
141 else if (c->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN)
144 s.all_layers = dav1d->all_layers;
145 if (dav1d->operating_point >= 0)
146 s.operating_point = dav1d->operating_point;
148 s.n_tile_threads = dav1d->tile_threads
149 ? dav1d->tile_threads
150 : FFMIN(floor(sqrt(threads)), DAV1D_MAX_TILE_THREADS);
151 s.n_frame_threads = dav1d->frame_threads
152 ? dav1d->frame_threads
153 : FFMIN(ceil(threads / s.n_tile_threads), DAV1D_MAX_FRAME_THREADS);
154 av_log(c, AV_LOG_DEBUG, "Using %d frame threads, %d tile threads\n",
155 s.n_frame_threads, s.n_tile_threads);
157 res = dav1d_open(&dav1d->c, &s);
159 return AVERROR(ENOMEM);
164 static void libdav1d_flush(AVCodecContext *c)
166 Libdav1dContext *dav1d = c->priv_data;
168 dav1d_data_unref(&dav1d->data);
169 dav1d_flush(dav1d->c);
172 static void libdav1d_data_free(const uint8_t *data, void *opaque) {
173 AVBufferRef *buf = opaque;
175 av_buffer_unref(&buf);
178 static void libdav1d_user_data_free(const uint8_t *data, void *opaque) {
179 av_assert0(data == opaque);
183 static int libdav1d_receive_frame(AVCodecContext *c, AVFrame *frame)
185 Libdav1dContext *dav1d = c->priv_data;
186 Dav1dData *data = &dav1d->data;
187 Dav1dPicture pic = { 0 }, *p = &pic;
191 AVPacket pkt = { 0 };
193 res = ff_decode_get_packet(c, &pkt);
194 if (res < 0 && res != AVERROR_EOF)
198 res = dav1d_data_wrap(data, pkt.data, pkt.size, libdav1d_data_free, pkt.buf);
200 av_packet_unref(&pkt);
204 data->m.timestamp = pkt.pts;
205 data->m.offset = pkt.pos;
206 data->m.duration = pkt.duration;
209 av_packet_unref(&pkt);
211 if (c->reordered_opaque != AV_NOPTS_VALUE) {
212 uint8_t *reordered_opaque = av_malloc(sizeof(c->reordered_opaque));
213 if (!reordered_opaque) {
214 dav1d_data_unref(data);
215 return AVERROR(ENOMEM);
218 memcpy(reordered_opaque, &c->reordered_opaque, sizeof(c->reordered_opaque));
219 res = dav1d_data_wrap_user_data(data, reordered_opaque,
220 libdav1d_user_data_free, reordered_opaque);
222 av_free(reordered_opaque);
223 dav1d_data_unref(data);
230 res = dav1d_send_data(dav1d->c, data);
232 if (res == AVERROR(EINVAL))
233 res = AVERROR_INVALIDDATA;
234 if (res != AVERROR(EAGAIN))
238 res = dav1d_get_picture(dav1d->c, p);
240 if (res == AVERROR(EINVAL))
241 res = AVERROR_INVALIDDATA;
242 else if (res == AVERROR(EAGAIN) && c->internal->draining)
248 av_assert0(p->data[0] && p->allocator_data);
250 // This requires the custom allocator above
251 frame->buf[0] = av_buffer_ref(p->allocator_data);
252 if (!frame->buf[0]) {
253 dav1d_picture_unref(p);
254 return AVERROR(ENOMEM);
257 frame->data[0] = p->data[0];
258 frame->data[1] = p->data[1];
259 frame->data[2] = p->data[2];
260 frame->linesize[0] = p->stride[0];
261 frame->linesize[1] = p->stride[1];
262 frame->linesize[2] = p->stride[1];
264 c->profile = p->seq_hdr->profile;
265 c->level = ((p->seq_hdr->operating_points[0].major_level - 2) << 2)
266 | p->seq_hdr->operating_points[0].minor_level;
267 frame->width = p->p.w;
268 frame->height = p->p.h;
269 if (c->width != p->p.w || c->height != p->p.h) {
270 res = ff_set_dimensions(c, p->p.w, p->p.h);
275 av_reduce(&frame->sample_aspect_ratio.num,
276 &frame->sample_aspect_ratio.den,
277 frame->height * (int64_t)p->frame_hdr->render_width,
278 frame->width * (int64_t)p->frame_hdr->render_height,
280 ff_set_sar(c, frame->sample_aspect_ratio);
282 switch (p->seq_hdr->chr) {
283 case DAV1D_CHR_VERTICAL:
284 frame->chroma_location = c->chroma_sample_location = AVCHROMA_LOC_LEFT;
286 case DAV1D_CHR_COLOCATED:
287 frame->chroma_location = c->chroma_sample_location = AVCHROMA_LOC_TOPLEFT;
290 frame->colorspace = c->colorspace = (enum AVColorSpace) p->seq_hdr->mtrx;
291 frame->color_primaries = c->color_primaries = (enum AVColorPrimaries) p->seq_hdr->pri;
292 frame->color_trc = c->color_trc = (enum AVColorTransferCharacteristic) p->seq_hdr->trc;
293 frame->color_range = c->color_range = p->seq_hdr->color_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
295 if (p->p.layout == DAV1D_PIXEL_LAYOUT_I444 &&
296 p->seq_hdr->mtrx == DAV1D_MC_IDENTITY &&
297 p->seq_hdr->pri == DAV1D_COLOR_PRI_BT709 &&
298 p->seq_hdr->trc == DAV1D_TRC_SRGB)
299 frame->format = c->pix_fmt = pix_fmt_rgb[p->seq_hdr->hbd];
301 frame->format = c->pix_fmt = pix_fmt[p->p.layout][p->seq_hdr->hbd];
303 if (p->m.user_data.data)
304 memcpy(&frame->reordered_opaque, p->m.user_data.data, sizeof(frame->reordered_opaque));
306 frame->reordered_opaque = AV_NOPTS_VALUE;
308 if (p->seq_hdr->num_units_in_tick && p->seq_hdr->time_scale) {
309 av_reduce(&c->framerate.den, &c->framerate.num,
310 p->seq_hdr->num_units_in_tick, p->seq_hdr->time_scale, INT_MAX);
311 if (p->seq_hdr->equal_picture_interval)
312 c->ticks_per_frame = p->seq_hdr->num_ticks_per_picture;
315 // match timestamps and packet size
316 frame->pts = p->m.timestamp;
317 frame->pkt_dts = p->m.timestamp;
318 frame->pkt_pos = p->m.offset;
319 frame->pkt_size = p->m.size;
320 frame->pkt_duration = p->m.duration;
321 frame->key_frame = p->frame_hdr->frame_type == DAV1D_FRAME_TYPE_KEY;
323 switch (p->frame_hdr->frame_type) {
324 case DAV1D_FRAME_TYPE_KEY:
325 case DAV1D_FRAME_TYPE_INTRA:
326 frame->pict_type = AV_PICTURE_TYPE_I;
328 case DAV1D_FRAME_TYPE_INTER:
329 frame->pict_type = AV_PICTURE_TYPE_P;
331 case DAV1D_FRAME_TYPE_SWITCH:
332 frame->pict_type = AV_PICTURE_TYPE_SP;
335 res = AVERROR_INVALIDDATA;
339 if (p->mastering_display) {
340 AVMasteringDisplayMetadata *mastering = av_mastering_display_metadata_create_side_data(frame);
342 res = AVERROR(ENOMEM);
346 for (int i = 0; i < 3; i++) {
347 mastering->display_primaries[i][0] = av_make_q(p->mastering_display->primaries[i][0], 1 << 16);
348 mastering->display_primaries[i][1] = av_make_q(p->mastering_display->primaries[i][1], 1 << 16);
350 mastering->white_point[0] = av_make_q(p->mastering_display->white_point[0], 1 << 16);
351 mastering->white_point[1] = av_make_q(p->mastering_display->white_point[1], 1 << 16);
353 mastering->max_luminance = av_make_q(p->mastering_display->max_luminance, 1 << 8);
354 mastering->min_luminance = av_make_q(p->mastering_display->min_luminance, 1 << 14);
356 mastering->has_primaries = 1;
357 mastering->has_luminance = 1;
359 if (p->content_light) {
360 AVContentLightMetadata *light = av_content_light_metadata_create_side_data(frame);
362 res = AVERROR(ENOMEM);
365 light->MaxCLL = p->content_light->max_content_light_level;
366 light->MaxFALL = p->content_light->max_frame_average_light_level;
370 unsigned int user_identifier;
372 bytestream2_init(&gb, p->itut_t35->payload, p->itut_t35->payload_size);
373 bytestream2_skip(&gb, 1); // terminal provider code
374 bytestream2_skip(&gb, 1); // terminal provider oriented code
375 user_identifier = bytestream2_get_be32(&gb);
376 switch (user_identifier) {
377 case MKBETAG('G', 'A', '9', '4'): { // closed captions
378 AVBufferRef *buf = NULL;
380 res = ff_parse_a53_cc(&buf, gb.buffer, bytestream2_get_bytes_left(&gb));
386 if (!av_frame_new_side_data_from_buf(frame, AV_FRAME_DATA_A53_CC, buf))
387 av_buffer_unref(&buf);
389 c->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
392 default: // ignore unsupported identifiers
396 if (p->frame_hdr->film_grain.present && (!dav1d->apply_grain ||
397 (c->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN))) {
398 AVFilmGrainParams *fgp = av_film_grain_params_create_side_data(frame);
400 res = AVERROR(ENOMEM);
404 fgp->type = AV_FILM_GRAIN_PARAMS_AV1;
405 fgp->seed = p->frame_hdr->film_grain.data.seed;
406 fgp->codec.aom.num_y_points = p->frame_hdr->film_grain.data.num_y_points;
407 fgp->codec.aom.chroma_scaling_from_luma = p->frame_hdr->film_grain.data.chroma_scaling_from_luma;
408 fgp->codec.aom.scaling_shift = p->frame_hdr->film_grain.data.scaling_shift;
409 fgp->codec.aom.ar_coeff_lag = p->frame_hdr->film_grain.data.ar_coeff_lag;
410 fgp->codec.aom.ar_coeff_shift = p->frame_hdr->film_grain.data.ar_coeff_shift;
411 fgp->codec.aom.grain_scale_shift = p->frame_hdr->film_grain.data.grain_scale_shift;
412 fgp->codec.aom.overlap_flag = p->frame_hdr->film_grain.data.overlap_flag;
413 fgp->codec.aom.limit_output_range = p->frame_hdr->film_grain.data.clip_to_restricted_range;
415 memcpy(&fgp->codec.aom.y_points, &p->frame_hdr->film_grain.data.y_points,
416 sizeof(fgp->codec.aom.y_points));
417 memcpy(&fgp->codec.aom.num_uv_points, &p->frame_hdr->film_grain.data.num_uv_points,
418 sizeof(fgp->codec.aom.num_uv_points));
419 memcpy(&fgp->codec.aom.uv_points, &p->frame_hdr->film_grain.data.uv_points,
420 sizeof(fgp->codec.aom.uv_points));
421 memcpy(&fgp->codec.aom.ar_coeffs_y, &p->frame_hdr->film_grain.data.ar_coeffs_y,
422 sizeof(fgp->codec.aom.ar_coeffs_y));
423 memcpy(&fgp->codec.aom.ar_coeffs_uv[0], &p->frame_hdr->film_grain.data.ar_coeffs_uv[0],
424 sizeof(fgp->codec.aom.ar_coeffs_uv[0]));
425 memcpy(&fgp->codec.aom.ar_coeffs_uv[1], &p->frame_hdr->film_grain.data.ar_coeffs_uv[1],
426 sizeof(fgp->codec.aom.ar_coeffs_uv[1]));
427 memcpy(&fgp->codec.aom.uv_mult, &p->frame_hdr->film_grain.data.uv_mult,
428 sizeof(fgp->codec.aom.uv_mult));
429 memcpy(&fgp->codec.aom.uv_mult_luma, &p->frame_hdr->film_grain.data.uv_luma_mult,
430 sizeof(fgp->codec.aom.uv_mult_luma));
431 memcpy(&fgp->codec.aom.uv_offset, &p->frame_hdr->film_grain.data.uv_offset,
432 sizeof(fgp->codec.aom.uv_offset));
437 dav1d_picture_unref(p);
439 av_frame_unref(frame);
443 static av_cold int libdav1d_close(AVCodecContext *c)
445 Libdav1dContext *dav1d = c->priv_data;
447 av_buffer_pool_uninit(&dav1d->pool);
448 dav1d_data_unref(&dav1d->data);
449 dav1d_close(&dav1d->c);
454 #define OFFSET(x) offsetof(Libdav1dContext, x)
455 #define VD AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM
456 static const AVOption libdav1d_options[] = {
457 { "tilethreads", "Tile threads", OFFSET(tile_threads), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, DAV1D_MAX_TILE_THREADS, VD },
458 { "framethreads", "Frame threads", OFFSET(frame_threads), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, DAV1D_MAX_FRAME_THREADS, VD },
459 { "filmgrain", "Apply Film Grain", OFFSET(apply_grain), AV_OPT_TYPE_BOOL, { .i64 = -1 }, -1, 1, VD | AV_OPT_FLAG_DEPRECATED },
460 { "oppoint", "Select an operating point of the scalable bitstream", OFFSET(operating_point), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 31, VD },
461 { "alllayers", "Output all spatial layers", OFFSET(all_layers), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, VD },
465 static const AVClass libdav1d_class = {
466 .class_name = "libdav1d decoder",
467 .item_name = av_default_item_name,
468 .option = libdav1d_options,
469 .version = LIBAVUTIL_VERSION_INT,
472 const AVCodec ff_libdav1d_decoder = {
474 .long_name = NULL_IF_CONFIG_SMALL("dav1d AV1 decoder by VideoLAN"),
475 .type = AVMEDIA_TYPE_VIDEO,
476 .id = AV_CODEC_ID_AV1,
477 .priv_data_size = sizeof(Libdav1dContext),
478 .init = libdav1d_init,
479 .close = libdav1d_close,
480 .flush = libdav1d_flush,
481 .receive_frame = libdav1d_receive_frame,
482 .capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_OTHER_THREADS,
483 .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_SETS_PKT_DTS |
484 FF_CODEC_CAP_AUTO_THREADS,
485 .priv_class = &libdav1d_class,
486 .wrapper_name = "libdav1d",