| @@ -0,0 +1,852 @@ |
| 1 | +/* |
| 2 | + * AV1 video decoder |
| 3 | + * |
| 4 | + * This file is part of FFmpeg. |
| 5 | + * |
| 6 | + * FFmpeg is free software; you can redistribute it and/or |
| 7 | + * modify it under the terms of the GNU Lesser General Public |
| 8 | + * License as published by the Free Software Foundation; either |
| 9 | + * version 2.1 of the License, or (at your option) any later version. |
| 10 | + * |
| 11 | + * FFmpeg is distributed in the hope that it will be useful, |
| 12 | + * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | + * Lesser General Public License for more details. |
| 15 | + * |
| 16 | + * You should have received a copy of the GNU Lesser General Public |
| 17 | + * License along with FFmpeg; if not, write to the Free Software |
| 18 | + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 19 | + */ |
| 20 | + |
| 21 | +#include "libavutil/pixdesc.h" |
| 22 | +#include "avcodec.h" |
| 23 | +#include "av1dec.h" |
| 24 | +#include "get_bits.h" |
| 25 | +#include "hwconfig.h" |
| 26 | +#include "internal.h" |
| 27 | +#include "profiles.h" |
| 28 | + |
| 29 | +static void setup_past_independence(AV1Frame *f) |
| 30 | +{ |
| 31 | + f->loop_filter_delta_enabled = 1; |
| 32 | + |
| 33 | + f->loop_filter_ref_deltas[AV1_REF_FRAME_INTRA] = 1; |
| 34 | + f->loop_filter_ref_deltas[AV1_REF_FRAME_LAST] = 0; |
| 35 | + f->loop_filter_ref_deltas[AV1_REF_FRAME_LAST2] = 0; |
| 36 | + f->loop_filter_ref_deltas[AV1_REF_FRAME_LAST3] = 0; |
| 37 | + f->loop_filter_ref_deltas[AV1_REF_FRAME_GOLDEN] = -1; |
| 38 | + f->loop_filter_ref_deltas[AV1_REF_FRAME_BWDREF] = 0; |
| 39 | + f->loop_filter_ref_deltas[AV1_REF_FRAME_ALTREF2] = -1; |
| 40 | + f->loop_filter_ref_deltas[AV1_REF_FRAME_ALTREF] = -1; |
| 41 | + |
| 42 | + f->loop_filter_mode_deltas[0] = 0; |
| 43 | + f->loop_filter_mode_deltas[1] = 0; |
| 44 | +} |
| 45 | + |
| 46 | +static void load_previous_and_update(AV1DecContext *s) |
| 47 | +{ |
| 48 | + uint8_t primary_frame, prev_frame; |
| 49 | + |
| 50 | + primary_frame = s->raw_frame_header->primary_ref_frame; |
| 51 | + prev_frame = s->raw_frame_header->ref_frame_idx[primary_frame]; |
| 52 | + memcpy(s->cur_frame.loop_filter_ref_deltas, |
| 53 | + s->ref[prev_frame].loop_filter_ref_deltas, |
| 54 | + AV1_NUM_REF_FRAMES * sizeof(int8_t)); |
| 55 | + memcpy(s->cur_frame.loop_filter_mode_deltas, |
| 56 | + s->ref[prev_frame].loop_filter_mode_deltas, |
| 57 | + 2 * sizeof(int8_t)); |
| 58 | + |
| 59 | + if (s->raw_frame_header->loop_filter_delta_update) { |
| 60 | + for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) { |
| 61 | + if (s->raw_frame_header->update_ref_delta[i]) |
| 62 | + s->cur_frame.loop_filter_ref_deltas[i] = |
| 63 | + s->raw_frame_header->loop_filter_ref_deltas[i]; |
| 64 | + } |
| 65 | + |
| 66 | + for (int i = 0; i < 2; i++) { |
| 67 | + if (s->raw_frame_header->update_mode_delta[i]) |
| 68 | + s->cur_frame.loop_filter_mode_deltas[i] = |
| 69 | + s->raw_frame_header->loop_filter_mode_deltas[i]; |
| 70 | + } |
| 71 | + } |
| 72 | + |
| 73 | + s->cur_frame.loop_filter_delta_enabled = |
| 74 | + s->raw_frame_header->loop_filter_delta_enabled; |
| 75 | +} |
| 76 | + |
| 77 | +static uint32_t inverse_recenter(int r, uint32_t v) |
| 78 | +{ |
| 79 | + if (v > 2 * r) |
| 80 | + return v; |
| 81 | + else if (v & 1) |
| 82 | + return r - ((v + 1) >> 1); |
| 83 | + else |
| 84 | + return r + (v >> 1); |
| 85 | +} |
| 86 | + |
| 87 | +static uint32_t decode_unsigned_subexp_with_ref(uint32_t sub_exp, |
| 88 | + int mx, int r) |
| 89 | +{ |
| 90 | + if ((r << 1) <= mx) { |
| 91 | + return inverse_recenter(r, sub_exp); |
| 92 | + } else { |
| 93 | + return mx - 1 - inverse_recenter(mx - 1 - r, sub_exp); |
| 94 | + } |
| 95 | +} |
| 96 | + |
| 97 | +static int32_t decode_signed_subexp_with_ref(uint32_t sub_exp, int low, |
| 98 | + int high, int r) |
| 99 | +{ |
| 100 | + int32_t x = decode_unsigned_subexp_with_ref(sub_exp, high - low, r - low); |
| 101 | + return x + low; |
| 102 | +} |
| 103 | + |
| 104 | +static void read_global_param(AV1DecContext *s, int type, int ref, int idx) |
| 105 | +{ |
| 106 | + uint8_t primary_frame, prev_frame; |
| 107 | + uint32_t abs_bits, prec_bits, round, prec_diff, sub, mx; |
| 108 | + int32_t r; |
| 109 | + |
| 110 | + primary_frame = s->raw_frame_header->primary_ref_frame; |
| 111 | + prev_frame = s->raw_frame_header->ref_frame_idx[primary_frame]; |
| 112 | + abs_bits = AV1_GM_ABS_ALPHA_BITS; |
| 113 | + prec_bits = AV1_GM_ALPHA_PREC_BITS; |
| 114 | + |
| 115 | + if (idx < 2) { |
| 116 | + if (type == AV1_WARP_MODEL_TRANSLATION) { |
| 117 | + abs_bits = AV1_GM_ABS_TRANS_ONLY_BITS - |
| 118 | + !s->raw_frame_header->allow_high_precision_mv; |
| 119 | + prec_bits = AV1_GM_TRANS_ONLY_PREC_BITS - |
| 120 | + !s->raw_frame_header->allow_high_precision_mv; |
| 121 | + } else { |
| 122 | + abs_bits = AV1_GM_ABS_TRANS_BITS; |
| 123 | + prec_bits = AV1_GM_TRANS_PREC_BITS; |
| 124 | + } |
| 125 | + } |
| 126 | + round = (idx % 3) == 2 ? (1 << AV1_WARPEDMODEL_PREC_BITS) : 0; |
| 127 | + prec_diff = AV1_WARPEDMODEL_PREC_BITS - prec_bits; |
| 128 | + sub = (idx % 3) == 2 ? (1 << prec_bits) : 0; |
| 129 | + mx = 1 << abs_bits; |
| 130 | + r = (s->ref[prev_frame].gm_params[ref][idx] >> prec_diff) - sub; |
| 131 | + |
| 132 | + s->cur_frame.gm_params[ref][idx] = |
| 133 | + (decode_signed_subexp_with_ref(s->raw_frame_header->gm_params[ref][idx], |
| 134 | + -mx, mx + 1, r) << prec_diff) + round; |
| 135 | +} |
| 136 | + |
| 137 | +/** |
| 138 | +* update gm type/params, since cbs already implemented part of this funcation, |
| 139 | +* so we don't need to full implement spec. |
| 140 | +*/ |
| 141 | +static void global_motion_params(AV1DecContext *s) |
| 142 | +{ |
| 143 | + const AV1RawFrameHeader *header = s->raw_frame_header; |
| 144 | + int type, ref; |
| 145 | + |
| 146 | + for (ref = AV1_REF_FRAME_LAST; ref <= AV1_REF_FRAME_ALTREF; ref++) { |
| 147 | + s->cur_frame.gm_type[ref] = AV1_WARP_MODEL_IDENTITY; |
| 148 | + for (int i = 0; i < 6; i++) |
| 149 | + s->cur_frame.gm_params[ref][i] = (i % 3 == 2) ? |
| 150 | + 1 << AV1_WARPEDMODEL_PREC_BITS : 0; |
| 151 | + } |
| 152 | + if (header->frame_type == AV1_FRAME_KEY || |
| 153 | + header->frame_type == AV1_FRAME_INTRA_ONLY) |
| 154 | + return; |
| 155 | + |
| 156 | + for (ref = AV1_REF_FRAME_LAST; ref <= AV1_REF_FRAME_ALTREF; ref++) { |
| 157 | + if (header->is_global[ref]) { |
| 158 | + if (header->is_rot_zoom[ref]) { |
| 159 | + type = AV1_WARP_MODEL_ROTZOOM; |
| 160 | + } else { |
| 161 | + type = header->is_translation[ref] ? AV1_WARP_MODEL_TRANSLATION |
| 162 | + : AV1_WARP_MODEL_AFFINE; |
| 163 | + } |
| 164 | + } else { |
| 165 | + type = AV1_WARP_MODEL_IDENTITY; |
| 166 | + } |
| 167 | + s->cur_frame.gm_type[ref] = type; |
| 168 | + |
| 169 | + if (type >= AV1_WARP_MODEL_ROTZOOM) { |
| 170 | + read_global_param(s, type, ref, 2); |
| 171 | + read_global_param(s, type, ref, 3); |
| 172 | + if (type == AV1_WARP_MODEL_AFFINE) { |
| 173 | + read_global_param(s, type, ref, 4); |
| 174 | + read_global_param(s, type, ref, 5); |
| 175 | + } else { |
| 176 | + s->cur_frame.gm_params[ref][4] = -s->cur_frame.gm_params[ref][3]; |
| 177 | + s->cur_frame.gm_params[ref][5] = s->cur_frame.gm_params[ref][2]; |
| 178 | + } |
| 179 | + } |
| 180 | + if (type >= AV1_WARP_MODEL_TRANSLATION) { |
| 181 | + read_global_param(s, type, ref, 0); |
| 182 | + read_global_param(s, type, ref, 1); |
| 183 | + } |
| 184 | + } |
| 185 | +} |
| 186 | + |
| 187 | +static int init_tile_data(AV1DecContext *s) |
| 188 | + |
| 189 | +{ |
| 190 | + int cur_tile_num = |
| 191 | + s->raw_frame_header->tile_cols * s->raw_frame_header->tile_rows; |
| 192 | + if (s->tile_num < cur_tile_num) { |
| 193 | + int ret = av_reallocp_array(&s->tile_group_info, cur_tile_num, |
| 194 | + sizeof(TileGroupInfo)); |
| 195 | + if (ret < 0) { |
| 196 | + s->tile_num = 0; |
| 197 | + return ret; |
| 198 | + } |
| 199 | + } |
| 200 | + s->tile_num = cur_tile_num; |
| 201 | + |
| 202 | + return 0; |
| 203 | +} |
| 204 | + |
| 205 | +static int get_tiles_info(AVCodecContext *avctx, const AV1RawTileGroup *tile_group) |
| 206 | +{ |
| 207 | + AV1DecContext *s = avctx->priv_data; |
| 208 | + GetBitContext gb; |
| 209 | + uint16_t tile_num, tile_row, tile_col; |
| 210 | + uint32_t size = 0, size_bytes = 0, offset = 0; |
| 211 | + int ret; |
| 212 | + |
| 213 | + if ((ret = init_get_bits8(&gb, |
| 214 | + tile_group->tile_data.data, |
| 215 | + tile_group->tile_data.data_size)) < 0) { |
| 216 | + av_log(avctx, AV_LOG_ERROR, "Failed to initialize bitstream reader.\n"); |
| 217 | + return ret; |
| 218 | + } |
| 219 | + |
| 220 | + s->tg_start = tile_group->tg_start; |
| 221 | + s->tg_end = tile_group->tg_end; |
| 222 | + |
| 223 | + for (tile_num = tile_group->tg_start; tile_num <= tile_group->tg_end; tile_num++) { |
| 224 | + tile_row = tile_num / s->raw_frame_header->tile_cols; |
| 225 | + tile_col = tile_num % s->raw_frame_header->tile_cols; |
| 226 | + |
| 227 | + if (tile_num == tile_group->tg_end) { |
| 228 | + s->tile_group_info[tile_num].tile_size = get_bits_left(&gb) / 8; |
| 229 | + s->tile_group_info[tile_num].tile_offset = offset; |
| 230 | + s->tile_group_info[tile_num].tile_row = tile_row; |
| 231 | + s->tile_group_info[tile_num].tile_column = tile_col; |
| 232 | + return 0; |
| 233 | + } |
| 234 | + size_bytes = s->raw_frame_header->tile_size_bytes_minus1 + 1; |
| 235 | + size = get_bits_le(&gb, size_bytes * 8) + 1; |
| 236 | + skip_bits(&gb, size * 8); |
| 237 | + |
| 238 | + offset += size_bytes; |
| 239 | + |
| 240 | + s->tile_group_info[tile_num].tile_size = size; |
| 241 | + s->tile_group_info[tile_num].tile_offset = offset; |
| 242 | + s->tile_group_info[tile_num].tile_row = tile_row; |
| 243 | + s->tile_group_info[tile_num].tile_column = tile_col; |
| 244 | + |
| 245 | + offset += size; |
| 246 | + } |
| 247 | + |
| 248 | + return 0; |
| 249 | + |
| 250 | +} |
| 251 | + |
| 252 | +static int get_pixel_format(AVCodecContext *avctx) |
| 253 | +{ |
| 254 | + AV1DecContext *s = avctx->priv_data; |
| 255 | + const AV1RawSequenceHeader *seq = s->raw_seq; |
| 256 | + uint8_t bit_depth; |
| 257 | + int ret; |
| 258 | + enum AVPixelFormat pix_fmt = AV_PIX_FMT_NONE; |
| 259 | +#define HWACCEL_MAX (0) |
| 260 | + enum AVPixelFormat pix_fmts[HWACCEL_MAX + 1], *fmtp = pix_fmts; |
| 261 | + |
| 262 | + /** |
| 263 | + * check if the HW accel is inited correctly. If not, return un-implemented. |
| 264 | + * Since now the av1 decoder doesn't support native decode, if it will be |
| 265 | + * implemented in the future, need remove this check. |
| 266 | + */ |
| 267 | + if (!avctx->hwaccel) { |
| 268 | + av_log(avctx, AV_LOG_ERROR, "Your platform doesn't suppport" |
| 269 | + " hardware accelerated AV1 decoding.\n"); |
| 270 | + return AVERROR(ENOSYS); |
| 271 | + } |
| 272 | + |
| 273 | + if (seq->seq_profile == 2 && seq->color_config.high_bitdepth) |
| 274 | + bit_depth = seq->color_config.twelve_bit ? 12 : 10; |
| 275 | + else if (seq->seq_profile <= 2) |
| 276 | + bit_depth = seq->color_config.high_bitdepth ? 10 : 8; |
| 277 | + else { |
| 278 | + av_log(avctx, AV_LOG_ERROR, |
| 279 | + "Unknown AV1 profile %d.\n", seq->seq_profile); |
| 280 | + return -1; |
| 281 | + } |
| 282 | + |
| 283 | + if (!seq->color_config.mono_chrome) { |
| 284 | + // 4:4:4 x:0 y:0, 4:2:2 x:1 y:0, 4:2:0 x:1 y:1 |
| 285 | + if (seq->color_config.subsampling_x == 0 && |
| 286 | + seq->color_config.subsampling_y == 0) { |
| 287 | + if (bit_depth == 8) |
| 288 | + pix_fmt = AV_PIX_FMT_YUV444P; |
| 289 | + else if (bit_depth == 10) |
| 290 | + pix_fmt = AV_PIX_FMT_YUV444P10; |
| 291 | + else if (bit_depth == 12) |
| 292 | + pix_fmt = AV_PIX_FMT_YUV444P12; |
| 293 | + else |
| 294 | + av_log(avctx, AV_LOG_WARNING, "Unknown AV1 pixel format.\n"); |
| 295 | + } else if (seq->color_config.subsampling_x == 1 && |
| 296 | + seq->color_config.subsampling_y == 0) { |
| 297 | + if (bit_depth == 8) |
| 298 | + pix_fmt = AV_PIX_FMT_YUV422P; |
| 299 | + else if (bit_depth == 10) |
| 300 | + pix_fmt = AV_PIX_FMT_YUV422P10; |
| 301 | + else if (bit_depth == 12) |
| 302 | + pix_fmt = AV_PIX_FMT_YUV422P12; |
| 303 | + else |
| 304 | + av_log(avctx, AV_LOG_WARNING, "Unknown AV1 pixel format.\n"); |
| 305 | + } else if (seq->color_config.subsampling_x == 1 && |
| 306 | + seq->color_config.subsampling_y == 1) { |
| 307 | + if (bit_depth == 8) |
| 308 | + pix_fmt = AV_PIX_FMT_YUV420P; |
| 309 | + else if (bit_depth == 10) |
| 310 | + pix_fmt = AV_PIX_FMT_YUV420P10; |
| 311 | + else if (bit_depth == 12) |
| 312 | + pix_fmt = AV_PIX_FMT_YUV420P12; |
| 313 | + else |
| 314 | + av_log(avctx, AV_LOG_WARNING, "Unknown AV1 pixel format.\n"); |
| 315 | + } |
| 316 | + } else { |
| 317 | + if (seq->color_config.subsampling_x == 1 && |
| 318 | + seq->color_config.subsampling_y == 1) |
| 319 | + pix_fmt = AV_PIX_FMT_YUV440P; |
| 320 | + else |
| 321 | + av_log(avctx, AV_LOG_WARNING, "Unknown AV1 pixel format.\n"); |
| 322 | + } |
| 323 | + |
| 324 | + av_log(avctx, AV_LOG_DEBUG, "AV1 decode get format: %s.\n", |
| 325 | + av_get_pix_fmt_name(pix_fmt)); |
| 326 | + |
| 327 | + if (pix_fmt == AV_PIX_FMT_NONE) |
| 328 | + return -1; |
| 329 | + s->pix_fmt = pix_fmt; |
| 330 | + |
| 331 | + *fmtp = AV_PIX_FMT_NONE; |
| 332 | + avctx->sw_pix_fmt = s->pix_fmt; |
| 333 | + ret = ff_thread_get_format(avctx, pix_fmts); |
| 334 | + if (ret < 0) |
| 335 | + return ret; |
| 336 | + |
| 337 | + avctx->pix_fmt = ret; |
| 338 | + |
| 339 | + return 0; |
| 340 | +} |
| 341 | + |
| 342 | +static void av1_frame_unref(AVCodecContext *avctx, AV1Frame *f) |
| 343 | +{ |
| 344 | + ff_thread_release_buffer(avctx, &f->tf); |
| 345 | + av_buffer_unref(&f->hwaccel_priv_buf); |
| 346 | + f->hwaccel_picture_private = NULL; |
| 347 | +} |
| 348 | + |
| 349 | +static int av1_frame_ref(AVCodecContext *avctx, AV1Frame *dst, const AV1Frame *src) |
| 350 | +{ |
| 351 | + int ret; |
| 352 | + |
| 353 | + ret = ff_thread_ref_frame(&dst->tf, &src->tf); |
| 354 | + if (ret < 0) |
| 355 | + return ret; |
| 356 | + |
| 357 | + if (src->hwaccel_picture_private) { |
| 358 | + dst->hwaccel_priv_buf = av_buffer_ref(src->hwaccel_priv_buf); |
| 359 | + if (!dst->hwaccel_priv_buf) |
| 360 | + goto fail; |
| 361 | + dst->hwaccel_picture_private = dst->hwaccel_priv_buf->data; |
| 362 | + } |
| 363 | + |
| 364 | + dst->loop_filter_delta_enabled = src->loop_filter_delta_enabled; |
| 365 | + memcpy(dst->loop_filter_ref_deltas, |
| 366 | + src->loop_filter_ref_deltas, |
| 367 | + AV1_NUM_REF_FRAMES * sizeof(int8_t)); |
| 368 | + memcpy(dst->loop_filter_mode_deltas, |
| 369 | + src->loop_filter_mode_deltas, |
| 370 | + 2 * sizeof(int8_t)); |
| 371 | + memcpy(dst->gm_type, |
| 372 | + src->gm_type, |
| 373 | + AV1_NUM_REF_FRAMES * sizeof(uint8_t)); |
| 374 | + memcpy(dst->gm_params, |
| 375 | + src->gm_params, |
| 376 | + AV1_NUM_REF_FRAMES * 6 * sizeof(int32_t)); |
| 377 | + |
| 378 | + return 0; |
| 379 | + |
| 380 | +fail: |
| 381 | + av1_frame_unref(avctx, dst); |
| 382 | + return AVERROR(ENOMEM); |
| 383 | +} |
| 384 | + |
| 385 | +static av_cold int av1_decode_free(AVCodecContext *avctx) |
| 386 | +{ |
| 387 | + AV1DecContext *s = avctx->priv_data; |
| 388 | + |
| 389 | + for (int i = 0; i < FF_ARRAY_ELEMS(s->ref); i++) { |
| 390 | + if (s->ref[i].tf.f->buf[0]) |
| 391 | + av1_frame_unref(avctx, &s->ref[i]); |
| 392 | + av_frame_free(&s->ref[i].tf.f); |
| 393 | + } |
| 394 | + if (s->cur_frame.tf.f->buf[0]) |
| 395 | + av1_frame_unref(avctx, &s->cur_frame); |
| 396 | + av_frame_free(&s->cur_frame.tf.f); |
| 397 | + |
| 398 | + av_buffer_unref(&s->seq_ref); |
| 399 | + av_buffer_unref(&s->header_ref); |
| 400 | + av_freep(&s->tile_group_info); |
| 401 | + |
| 402 | + ff_cbs_fragment_free(&s->current_obu); |
| 403 | + ff_cbs_close(&s->cbc); |
| 404 | + |
| 405 | + return 0; |
| 406 | +} |
| 407 | + |
| 408 | +static int set_context_with_sequence(AVCodecContext *avctx, |
| 409 | + const AV1RawSequenceHeader *seq) |
| 410 | +{ |
| 411 | + avctx->profile = seq->seq_profile; |
| 412 | + avctx->level = seq->seq_level_idx[0]; |
| 413 | + |
| 414 | + avctx->color_range = |
| 415 | + seq->color_config.color_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG; |
| 416 | + avctx->color_primaries = seq->color_config.color_primaries; |
| 417 | + avctx->colorspace = seq->color_config.color_primaries; |
| 418 | + avctx->color_trc = seq->color_config.transfer_characteristics; |
| 419 | + |
| 420 | + if (seq->timing_info.num_units_in_display_tick && |
| 421 | + seq->timing_info.time_scale) { |
| 422 | + av_reduce(&avctx->framerate.den, &avctx->framerate.num, |
| 423 | + seq->timing_info.num_units_in_display_tick, |
| 424 | + seq->timing_info.time_scale, |
| 425 | + INT_MAX); |
| 426 | + if (seq->timing_info.equal_picture_interval) |
| 427 | + avctx->ticks_per_frame = seq->timing_info.num_ticks_per_picture_minus_1 + 1; |
| 428 | + } |
| 429 | + |
| 430 | + return 0; |
| 431 | +} |
| 432 | + |
| 433 | +static int update_context_with_frame_header(AVCodecContext *avctx, |
| 434 | + const AV1RawFrameHeader *header) |
| 435 | +{ |
| 436 | + AVRational aspect_ratio; |
| 437 | + int width = header->frame_width_minus_1 + 1; |
| 438 | + int height = header->frame_height_minus_1 + 1; |
| 439 | + int r_width = header->render_width_minus_1 + 1; |
| 440 | + int r_height = header->render_height_minus_1 + 1; |
| 441 | + int ret; |
| 442 | + |
| 443 | + if (avctx->width != width || avctx->height != height) { |
| 444 | + ret = ff_set_dimensions(avctx, width, height); |
| 445 | + if (ret < 0) |
| 446 | + return ret; |
| 447 | + } |
| 448 | + |
| 449 | + av_reduce(&aspect_ratio.num, &aspect_ratio.den, |
| 450 | + (int64_t)height * r_width, |
| 451 | + (int64_t)width * r_height, |
| 452 | + INT_MAX); |
| 453 | + |
| 454 | + if (avctx->sample_aspect_ratio.num != aspect_ratio.num || |
| 455 | + avctx->sample_aspect_ratio.den != aspect_ratio.den) { |
| 456 | + ret = ff_set_sar(avctx, aspect_ratio); |
| 457 | + if (ret < 0) |
| 458 | + return ret; |
| 459 | + } |
| 460 | + |
| 461 | + return 0; |
| 462 | +} |
| 463 | + |
| 464 | +static av_cold int av1_decode_init(AVCodecContext *avctx) |
| 465 | +{ |
| 466 | + AV1DecContext *s = avctx->priv_data; |
| 467 | + AV1RawSequenceHeader *seq; |
| 468 | + int ret; |
| 469 | + |
| 470 | + s->avctx = avctx; |
| 471 | + s->pix_fmt = AV_PIX_FMT_NONE; |
| 472 | + |
| 473 | + for (int i = 0; i < FF_ARRAY_ELEMS(s->ref); i++) { |
| 474 | + s->ref[i].tf.f = av_frame_alloc(); |
| 475 | + if (!s->ref[i].tf.f) { |
| 476 | + av1_decode_free(avctx); |
| 477 | + av_log(avctx, AV_LOG_ERROR, |
| 478 | + "Failed to allocate reference frame buffer %d.\n", i); |
| 479 | + return AVERROR(ENOMEM); |
| 480 | + } |
| 481 | + } |
| 482 | + |
| 483 | + s->cur_frame.tf.f = av_frame_alloc(); |
| 484 | + if (!s->cur_frame.tf.f) { |
| 485 | + av1_decode_free(avctx); |
| 486 | + av_log(avctx, AV_LOG_ERROR, |
| 487 | + "Failed to allocate current frame buffer.\n"); |
| 488 | + return AVERROR(ENOMEM); |
| 489 | + } |
| 490 | + |
| 491 | + ret = ff_cbs_init(&s->cbc, AV_CODEC_ID_AV1, avctx); |
| 492 | + if (ret < 0) |
| 493 | + return ret; |
| 494 | + |
| 495 | + if (avctx->extradata && avctx->extradata_size) { |
| 496 | + ret = ff_cbs_read(s->cbc, &s->current_obu, avctx->extradata, |
| 497 | + avctx->extradata_size); |
| 498 | + if (ret < 0) { |
| 499 | + av_log(avctx, AV_LOG_WARNING, "Failed to read extradata.\n"); |
| 500 | + goto end; |
| 501 | + } |
| 502 | + |
| 503 | + seq = ((CodedBitstreamAV1Context *)(s->cbc->priv_data))->sequence_header; |
| 504 | + if (!seq) { |
| 505 | + av_log(avctx, AV_LOG_WARNING, "No sequence header available.\n"); |
| 506 | + goto end; |
| 507 | + } |
| 508 | + |
| 509 | + ret = set_context_with_sequence(avctx, seq); |
| 510 | + if (ret < 0) { |
| 511 | + av_log(avctx, AV_LOG_WARNING, "Failed to set decoder context.\n"); |
| 512 | + goto end; |
| 513 | + } |
| 514 | + |
| 515 | + end: |
| 516 | + ff_cbs_fragment_reset(&s->current_obu); |
| 517 | + } |
| 518 | + |
| 519 | + return ret; |
| 520 | +} |
| 521 | + |
| 522 | +static int av1_frame_alloc(AVCodecContext *avctx, AV1Frame *f) |
| 523 | +{ |
| 524 | + AV1DecContext *s = avctx->priv_data; |
| 525 | + AV1RawFrameHeader *header= s->raw_frame_header; |
| 526 | + AVFrame *frame; |
| 527 | + int ret; |
| 528 | + |
| 529 | + ret = update_context_with_frame_header(avctx, header); |
| 530 | + if (ret < 0) { |
| 531 | + av_log(avctx, AV_LOG_ERROR, "Failed to update context with frame header\n"); |
| 532 | + return ret; |
| 533 | + } |
| 534 | + |
| 535 | + if ((ret = ff_thread_get_buffer(avctx, &f->tf, AV_GET_BUFFER_FLAG_REF)) < 0) |
| 536 | + return ret; |
| 537 | + |
| 538 | + frame = f->tf.f; |
| 539 | + frame->key_frame = header->frame_type == AV1_FRAME_KEY; |
| 540 | + |
| 541 | + switch (header->frame_type) { |
| 542 | + case AV1_FRAME_KEY: |
| 543 | + case AV1_FRAME_INTRA_ONLY: |
| 544 | + frame->pict_type = AV_PICTURE_TYPE_I; |
| 545 | + break; |
| 546 | + case AV1_FRAME_INTER: |
| 547 | + frame->pict_type = AV_PICTURE_TYPE_P; |
| 548 | + break; |
| 549 | + case AV1_FRAME_SWITCH: |
| 550 | + frame->pict_type = AV_PICTURE_TYPE_SP; |
| 551 | + break; |
| 552 | + } |
| 553 | + |
| 554 | + if (avctx->hwaccel) { |
| 555 | + const AVHWAccel *hwaccel = avctx->hwaccel; |
| 556 | + if (hwaccel->frame_priv_data_size) { |
| 557 | + f->hwaccel_priv_buf = |
| 558 | + av_buffer_allocz(hwaccel->frame_priv_data_size); |
| 559 | + if (!f->hwaccel_priv_buf) |
| 560 | + goto fail; |
| 561 | + f->hwaccel_picture_private = f->hwaccel_priv_buf->data; |
| 562 | + } |
| 563 | + } |
| 564 | + return 0; |
| 565 | + |
| 566 | +fail: |
| 567 | + av1_frame_unref(avctx, f); |
| 568 | + return AVERROR(ENOMEM); |
| 569 | +} |
| 570 | + |
| 571 | +static int set_output_frame(AVCodecContext *avctx, AVFrame *frame, |
| 572 | + const AVPacket *pkt, int *got_frame) |
| 573 | +{ |
| 574 | + AV1DecContext *s = avctx->priv_data; |
| 575 | + const AV1RawFrameHeader *header = s->raw_frame_header; |
| 576 | + const AVFrame *srcframe; |
| 577 | + int ret; |
| 578 | + |
| 579 | + if (header->show_existing_frame) |
| 580 | + srcframe = s->ref[header->frame_to_show_map_idx].tf.f; |
| 581 | + else |
| 582 | + srcframe = s->cur_frame.tf.f; |
| 583 | + |
| 584 | + ret = av_frame_ref(frame, srcframe); |
| 585 | + if (ret < 0) |
| 586 | + return ret; |
| 587 | + |
| 588 | + frame->pts = pkt->pts; |
| 589 | + frame->pkt_dts = pkt->dts; |
| 590 | + frame->pkt_size = pkt->size; |
| 591 | + |
| 592 | + *got_frame = 1; |
| 593 | + |
| 594 | + return 0; |
| 595 | +} |
| 596 | + |
| 597 | +static int update_reference_list(AVCodecContext *avctx) |
| 598 | +{ |
| 599 | + AV1DecContext *s = avctx->priv_data; |
| 600 | + const AV1RawFrameHeader *header = s->raw_frame_header; |
| 601 | + int ret; |
| 602 | + |
| 603 | + for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) { |
| 604 | + if (header->refresh_frame_flags & (1 << i)) { |
| 605 | + if (s->ref[i].tf.f->buf[0]) |
| 606 | + av1_frame_unref(avctx, &s->ref[i]); |
| 607 | + if ((ret = av1_frame_ref(avctx, &s->ref[i], &s->cur_frame)) < 0) { |
| 608 | + av_log(avctx, AV_LOG_ERROR, |
| 609 | + "Failed to update frame %d in reference list\n", i); |
| 610 | + return ret; |
| 611 | + } |
| 612 | + } |
| 613 | + } |
| 614 | + return 0; |
| 615 | +} |
| 616 | + |
| 617 | +static int get_current_frame(AVCodecContext *avctx) |
| 618 | +{ |
| 619 | + AV1DecContext *s = avctx->priv_data; |
| 620 | + int ret; |
| 621 | + |
| 622 | + if (s->cur_frame.tf.f->buf[0]) |
| 623 | + av1_frame_unref(avctx, &s->cur_frame); |
| 624 | + |
| 625 | + ret = av1_frame_alloc(avctx, &s->cur_frame); |
| 626 | + if (ret < 0) { |
| 627 | + av_log(avctx, AV_LOG_ERROR, |
| 628 | + "Failed to allocate space for current frame.\n"); |
| 629 | + return ret; |
| 630 | + } |
| 631 | + |
| 632 | + ret = init_tile_data(s); |
| 633 | + if (ret < 0) { |
| 634 | + av_log(avctx, AV_LOG_ERROR, "Failed to init tile data.\n"); |
| 635 | + return ret; |
| 636 | + } |
| 637 | + |
| 638 | + if (s->raw_frame_header->primary_ref_frame == AV1_PRIMARY_REF_NONE) |
| 639 | + setup_past_independence(&s->cur_frame); |
| 640 | + else |
| 641 | + load_previous_and_update(s); |
| 642 | + |
| 643 | + global_motion_params(s); |
| 644 | + |
| 645 | + return ret; |
| 646 | +} |
| 647 | + |
| 648 | +static int av1_decode_frame(AVCodecContext *avctx, void *frame, |
| 649 | + int *got_frame, AVPacket *pkt) |
| 650 | +{ |
| 651 | + AV1DecContext *s = avctx->priv_data; |
| 652 | + AV1RawTileGroup *raw_tile_group = NULL; |
| 653 | + int ret; |
| 654 | + |
| 655 | + ret = ff_cbs_read_packet(s->cbc, &s->current_obu, pkt); |
| 656 | + if (ret < 0) { |
| 657 | + av_log(avctx, AV_LOG_ERROR, "Failed to read packet.\n"); |
| 658 | + goto end; |
| 659 | + } |
| 660 | + av_log(avctx, AV_LOG_DEBUG, "Total obu for this frame:%d.\n", |
| 661 | + s->current_obu.nb_units); |
| 662 | + |
| 663 | + for (int i = 0; i < s->current_obu.nb_units; i++) { |
| 664 | + CodedBitstreamUnit *unit = &s->current_obu.units[i]; |
| 665 | + AV1RawOBU *obu = unit->content; |
| 666 | + av_log(avctx, AV_LOG_DEBUG, "Obu idx:%d, obu type:%d.\n", i, unit->type); |
| 667 | + |
| 668 | + switch (unit->type) { |
| 669 | + case AV1_OBU_SEQUENCE_HEADER: |
| 670 | + av_buffer_unref(&s->seq_ref); |
| 671 | + s->seq_ref = av_buffer_ref(unit->content_ref); |
| 672 | + if (!s->seq_ref) { |
| 673 | + ret = AVERROR(ENOMEM); |
| 674 | + goto end; |
| 675 | + } |
| 676 | + |
| 677 | + s->raw_seq = &obu->obu.sequence_header; |
| 678 | + |
| 679 | + ret = set_context_with_sequence(avctx, s->raw_seq); |
| 680 | + if (ret < 0) { |
| 681 | + av_log(avctx, AV_LOG_ERROR, "Failed to set context.\n"); |
| 682 | + goto end; |
| 683 | + } |
| 684 | + |
| 685 | + if (s->pix_fmt == AV_PIX_FMT_NONE) { |
| 686 | + ret = get_pixel_format(avctx); |
| 687 | + if (ret < 0) { |
| 688 | + av_log(avctx, AV_LOG_ERROR, |
| 689 | + "Failed to get pixel format.\n"); |
| 690 | + goto end; |
| 691 | + } |
| 692 | + } |
| 693 | + |
| 694 | + if (avctx->hwaccel && avctx->hwaccel->decode_params) { |
| 695 | + ret = avctx->hwaccel->decode_params(avctx, unit->type, unit->data, |
| 696 | + unit->data_size); |
| 697 | + if (ret < 0) { |
| 698 | + av_log(avctx, AV_LOG_ERROR, "HW accel decode params fail.\n"); |
| 699 | + goto end; |
| 700 | + } |
| 701 | + } |
| 702 | + break; |
| 703 | + case AV1_OBU_REDUNDANT_FRAME_HEADER: |
| 704 | + if (s->raw_frame_header) |
| 705 | + break; |
| 706 | + // fall-through |
| 707 | + case AV1_OBU_FRAME: |
| 708 | + case AV1_OBU_FRAME_HEADER: |
| 709 | + if (!s->seq_ref) { |
| 710 | + av_log(avctx, AV_LOG_ERROR, "Missing Sequence Header.\n"); |
| 711 | + ret = AVERROR_INVALIDDATA; |
| 712 | + goto end; |
| 713 | + } |
| 714 | + |
| 715 | + av_buffer_unref(&s->header_ref); |
| 716 | + s->header_ref = av_buffer_ref(unit->content_ref); |
| 717 | + if (!s->header_ref) { |
| 718 | + ret = AVERROR(ENOMEM); |
| 719 | + goto end; |
| 720 | + } |
| 721 | + |
| 722 | + if (unit->type == AV1_OBU_FRAME) |
| 723 | + s->raw_frame_header = &obu->obu.frame.header; |
| 724 | + else |
| 725 | + s->raw_frame_header = &obu->obu.frame_header; |
| 726 | + |
| 727 | + if (s->raw_frame_header->show_existing_frame) { |
| 728 | + ret = set_output_frame(avctx, frame, pkt, got_frame); |
| 729 | + if (ret < 0) |
| 730 | + av_log(avctx, AV_LOG_ERROR, "Set output frame error.\n"); |
| 731 | + |
| 732 | + s->raw_frame_header = NULL; |
| 733 | + |
| 734 | + goto end; |
| 735 | + } |
| 736 | + |
| 737 | + ret = get_current_frame(avctx); |
| 738 | + if (ret < 0) { |
| 739 | + av_log(avctx, AV_LOG_ERROR, "Get current frame error\n"); |
| 740 | + goto end; |
| 741 | + } |
| 742 | + |
| 743 | + if (avctx->hwaccel) { |
| 744 | + ret = avctx->hwaccel->start_frame(avctx, unit->data, |
| 745 | + unit->data_size); |
| 746 | + if (ret < 0) { |
| 747 | + av_log(avctx, AV_LOG_ERROR, "HW accel start frame fail.\n"); |
| 748 | + goto end; |
| 749 | + } |
| 750 | + } |
| 751 | + if (unit->type != AV1_OBU_FRAME) |
| 752 | + break; |
| 753 | + // fall-through |
| 754 | + case AV1_OBU_TILE_GROUP: |
| 755 | + if (!s->raw_frame_header) { |
| 756 | + av_log(avctx, AV_LOG_ERROR, "Missing Frame Header.\n"); |
| 757 | + ret = AVERROR_INVALIDDATA; |
| 758 | + goto end; |
| 759 | + } |
| 760 | + |
| 761 | + if (unit->type == AV1_OBU_FRAME) |
| 762 | + raw_tile_group = &obu->obu.frame.tile_group; |
| 763 | + else |
| 764 | + raw_tile_group = &obu->obu.tile_group; |
| 765 | + |
| 766 | + get_tiles_info(avctx, raw_tile_group); |
| 767 | + |
| 768 | + if (avctx->hwaccel) { |
| 769 | + ret = avctx->hwaccel->decode_slice(avctx, |
| 770 | + raw_tile_group->tile_data.data, |
| 771 | + raw_tile_group->tile_data.data_size); |
| 772 | + if (ret < 0) { |
| 773 | + av_log(avctx, AV_LOG_ERROR, |
| 774 | + "HW accel decode slice fail.\n"); |
| 775 | + goto end; |
| 776 | + } |
| 777 | + } |
| 778 | + break; |
| 779 | + case AV1_OBU_TILE_LIST: |
| 780 | + case AV1_OBU_TEMPORAL_DELIMITER: |
| 781 | + case AV1_OBU_PADDING: |
| 782 | + case AV1_OBU_METADATA: |
| 783 | + break; |
| 784 | + default: |
| 785 | + av_log(avctx, AV_LOG_DEBUG, |
| 786 | + "Unknown obu type: %d (%"SIZE_SPECIFIER" bits).\n", |
| 787 | + unit->type, unit->data_size); |
| 788 | + } |
| 789 | + |
| 790 | + if (raw_tile_group && (s->tile_num == raw_tile_group->tg_end + 1)) { |
| 791 | + if (avctx->hwaccel) { |
| 792 | + ret = avctx->hwaccel->end_frame(avctx); |
| 793 | + if (ret < 0) { |
| 794 | + av_log(avctx, AV_LOG_ERROR, "HW accel end frame fail.\n"); |
| 795 | + goto end; |
| 796 | + } |
| 797 | + } |
| 798 | + |
| 799 | + ret = update_reference_list(avctx); |
| 800 | + if (ret < 0) { |
| 801 | + av_log(avctx, AV_LOG_ERROR, "Failed to update reference list.\n"); |
| 802 | + goto end; |
| 803 | + } |
| 804 | + |
| 805 | + if (s->raw_frame_header->show_frame) { |
| 806 | + ret = set_output_frame(avctx, frame, pkt, got_frame); |
| 807 | + if (ret < 0) { |
| 808 | + av_log(avctx, AV_LOG_ERROR, "Set output frame error\n"); |
| 809 | + goto end; |
| 810 | + } |
| 811 | + } |
| 812 | + raw_tile_group = NULL; |
| 813 | + s->raw_frame_header = NULL; |
| 814 | + } |
| 815 | + } |
| 816 | + |
| 817 | +end: |
| 818 | + ff_cbs_fragment_reset(&s->current_obu); |
| 819 | + return ret; |
| 820 | +} |
| 821 | + |
| 822 | +static void av1_decode_flush(AVCodecContext *avctx) |
| 823 | +{ |
| 824 | + AV1DecContext *s = avctx->priv_data; |
| 825 | + |
| 826 | + for (int i = 0; i < FF_ARRAY_ELEMS(s->ref); i++) |
| 827 | + av1_frame_unref(avctx, &s->ref[i]); |
| 828 | + |
| 829 | + av1_frame_unref(avctx, &s->cur_frame); |
| 830 | + s->raw_frame_header = NULL; |
| 831 | + s->raw_seq = NULL; |
| 832 | +} |
| 833 | + |
| 834 | +AVCodec ff_av1_decoder = { |
| 835 | + .name = "av1", |
| 836 | + .long_name = NULL_IF_CONFIG_SMALL("Alliance for Open Media AV1"), |
| 837 | + .type = AVMEDIA_TYPE_VIDEO, |
| 838 | + .id = AV_CODEC_ID_AV1, |
| 839 | + .priv_data_size = sizeof(AV1DecContext), |
| 840 | + .init = av1_decode_init, |
| 841 | + .close = av1_decode_free, |
| 842 | + .decode = av1_decode_frame, |
| 843 | + .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_AVOID_PROBING, |
| 844 | + .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE | |
| 845 | + FF_CODEC_CAP_INIT_CLEANUP | |
| 846 | + FF_CODEC_CAP_SETS_PKT_DTS, |
| 847 | + .flush = av1_decode_flush, |
| 848 | + .profiles = NULL_IF_CONFIG_SMALL(ff_av1_profiles), |
| 849 | + .hw_configs = (const AVCodecHWConfigInternal * []) { |
| 850 | + NULL |
| 851 | + }, |
| 852 | +}; |