vaapi_av1.c (20455B)
1 /* 2 * AV1 HW decode acceleration through VA API 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/frame.h" 22 #include "libavutil/mem.h" 23 #include "hwaccel_internal.h" 24 #include "vaapi_decode.h" 25 #include "internal.h" 26 #include "av1dec.h" 27 #include "thread.h" 28 29 typedef struct VAAPIAV1FrameRef { 30 AVFrame *frame; 31 int valid; 32 } VAAPIAV1FrameRef; 33 34 typedef struct VAAPIAV1DecContext { 35 VAAPIDecodeContext base; 36 37 /** 38 * For film grain case, VAAPI generate 2 output for each frame, 39 * current_frame will not apply film grain, and will be used for 40 * references for next frames. Maintain the reference list without 41 * applying film grain here. And current_display_picture will be 42 * used to apply film grain and push to downstream. 43 */ 44 VAAPIAV1FrameRef ref_tab[AV1_NUM_REF_FRAMES]; 45 AVFrame *tmp_frame; 46 47 int nb_slice_params; 48 VASliceParameterBufferAV1 *slice_params; 49 } VAAPIAV1DecContext; 50 51 static VASurfaceID vaapi_av1_surface_id(AV1Frame *vf) 52 { 53 if (vf->f) 54 return ff_vaapi_get_surface_id(vf->f); 55 else 56 return VA_INVALID_SURFACE; 57 } 58 59 static int8_t vaapi_av1_get_bit_depth_idx(AVCodecContext *avctx) 60 { 61 AV1DecContext *s = avctx->priv_data; 62 const AV1RawSequenceHeader *seq = s->raw_seq; 63 int8_t bit_depth = 8; 64 65 if (seq->seq_profile == 2 && seq->color_config.high_bitdepth) 66 bit_depth = seq->color_config.twelve_bit ? 12 : 10; 67 else if (seq->seq_profile <= 2) 68 bit_depth = seq->color_config.high_bitdepth ? 10 : 8; 69 else { 70 av_log(avctx, AV_LOG_ERROR, 71 "Couldn't get bit depth from profile:%d.\n", seq->seq_profile); 72 return -1; 73 } 74 return bit_depth == 8 ? 0 : bit_depth == 10 ? 1 : 2; 75 } 76 77 static int vaapi_av1_decode_init(AVCodecContext *avctx) 78 { 79 VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data; 80 81 ctx->tmp_frame = av_frame_alloc(); 82 if (!ctx->tmp_frame) 83 return AVERROR(ENOMEM); 84 85 for (int i = 0; i < FF_ARRAY_ELEMS(ctx->ref_tab); i++) { 86 ctx->ref_tab[i].frame = av_frame_alloc(); 87 if (!ctx->ref_tab[i].frame) 88 return AVERROR(ENOMEM); 89 ctx->ref_tab[i].valid = 0; 90 } 91 92 return ff_vaapi_decode_init(avctx); 93 } 94 95 static int vaapi_av1_decode_uninit(AVCodecContext *avctx) 96 { 97 VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data; 98 99 av_frame_free(&ctx->tmp_frame); 100 101 for (int i = 0; i < FF_ARRAY_ELEMS(ctx->ref_tab); i++) 102 av_frame_free(&ctx->ref_tab[i].frame); 103 104 av_freep(&ctx->slice_params); 105 106 return ff_vaapi_decode_uninit(avctx); 107 } 108 109 110 static int vaapi_av1_start_frame(AVCodecContext *avctx, 111 av_unused const uint8_t *buffer, 112 av_unused uint32_t size) 113 { 114 AV1DecContext *s = avctx->priv_data; 115 const AV1RawSequenceHeader *seq = s->raw_seq; 116 const AV1RawFrameHeader *frame_header = s->raw_frame_header; 117 const AV1RawFilmGrainParams *film_grain = &s->cur_frame.film_grain; 118 VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private; 119 VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data; 120 VADecPictureParameterBufferAV1 pic_param; 121 int8_t bit_depth_idx; 122 int err = 0; 123 int apply_grain = !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN) && film_grain->apply_grain; 124 uint8_t remap_lr_type[4] = {AV1_RESTORE_NONE, AV1_RESTORE_SWITCHABLE, AV1_RESTORE_WIENER, AV1_RESTORE_SGRPROJ}; 125 uint8_t segmentation_feature_signed[AV1_SEG_LVL_MAX] = {1, 1, 1, 1, 1, 0, 0, 0}; 126 uint8_t segmentation_feature_max[AV1_SEG_LVL_MAX] = {255, AV1_MAX_LOOP_FILTER, 127 AV1_MAX_LOOP_FILTER, AV1_MAX_LOOP_FILTER, AV1_MAX_LOOP_FILTER, 7 , 0 , 0 }; 128 129 bit_depth_idx = vaapi_av1_get_bit_depth_idx(avctx); 130 if (bit_depth_idx < 0) 131 goto fail; 132 133 if (apply_grain) { 134 if (ctx->tmp_frame->buf[0]) 135 av_frame_unref(ctx->tmp_frame); 136 err = ff_thread_get_buffer(avctx, ctx->tmp_frame, AV_GET_BUFFER_FLAG_REF); 137 if (err < 0) 138 goto fail; 139 pic->output_surface = ff_vaapi_get_surface_id(ctx->tmp_frame); 140 } else { 141 pic->output_surface = ff_vaapi_get_surface_id(s->cur_frame.f); 142 } 143 144 memset(&pic_param, 0, sizeof(VADecPictureParameterBufferAV1)); 145 pic_param = (VADecPictureParameterBufferAV1) { 146 .profile = seq->seq_profile, 147 .order_hint_bits_minus_1 = seq->order_hint_bits_minus_1, 148 .bit_depth_idx = bit_depth_idx, 149 .matrix_coefficients = seq->color_config.matrix_coefficients, 150 .current_frame = pic->output_surface, 151 .current_display_picture = ff_vaapi_get_surface_id(s->cur_frame.f), 152 .frame_width_minus1 = frame_header->frame_width_minus_1, 153 .frame_height_minus1 = frame_header->frame_height_minus_1, 154 .primary_ref_frame = frame_header->primary_ref_frame, 155 .order_hint = frame_header->order_hint, 156 .tile_cols = frame_header->tile_cols, 157 .tile_rows = frame_header->tile_rows, 158 .context_update_tile_id = frame_header->context_update_tile_id, 159 .superres_scale_denominator = frame_header->use_superres ? 160 frame_header->coded_denom + AV1_SUPERRES_DENOM_MIN : 161 AV1_SUPERRES_NUM, 162 .interp_filter = frame_header->interpolation_filter, 163 .filter_level[0] = frame_header->loop_filter_level[0], 164 .filter_level[1] = frame_header->loop_filter_level[1], 165 .filter_level_u = frame_header->loop_filter_level[2], 166 .filter_level_v = frame_header->loop_filter_level[3], 167 .base_qindex = frame_header->base_q_idx, 168 .y_dc_delta_q = frame_header->delta_q_y_dc, 169 .u_dc_delta_q = frame_header->delta_q_u_dc, 170 .u_ac_delta_q = frame_header->delta_q_u_ac, 171 .v_dc_delta_q = frame_header->delta_q_v_dc, 172 .v_ac_delta_q = frame_header->delta_q_v_ac, 173 .cdef_damping_minus_3 = frame_header->cdef_damping_minus_3, 174 .cdef_bits = frame_header->cdef_bits, 175 .seq_info_fields.fields = { 176 .still_picture = seq->still_picture, 177 .use_128x128_superblock = seq->use_128x128_superblock, 178 .enable_filter_intra = seq->enable_filter_intra, 179 .enable_intra_edge_filter = seq->enable_intra_edge_filter, 180 .enable_interintra_compound = seq->enable_interintra_compound, 181 .enable_masked_compound = seq->enable_masked_compound, 182 .enable_dual_filter = seq->enable_dual_filter, 183 .enable_order_hint = seq->enable_order_hint, 184 .enable_jnt_comp = seq->enable_jnt_comp, 185 .enable_cdef = seq->enable_cdef, 186 .mono_chrome = seq->color_config.mono_chrome, 187 .color_range = seq->color_config.color_range, 188 .subsampling_x = seq->color_config.subsampling_x, 189 .subsampling_y = seq->color_config.subsampling_y, 190 .chroma_sample_position = seq->color_config.chroma_sample_position, 191 .film_grain_params_present = seq->film_grain_params_present && 192 !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN), 193 }, 194 .seg_info.segment_info_fields.bits = { 195 .enabled = frame_header->segmentation_enabled, 196 .update_map = frame_header->segmentation_update_map, 197 .temporal_update = frame_header->segmentation_temporal_update, 198 .update_data = frame_header->segmentation_update_data, 199 }, 200 .film_grain_info = { 201 .film_grain_info_fields.bits = { 202 .apply_grain = apply_grain, 203 .chroma_scaling_from_luma = film_grain->chroma_scaling_from_luma, 204 .grain_scaling_minus_8 = film_grain->grain_scaling_minus_8, 205 .ar_coeff_lag = film_grain->ar_coeff_lag, 206 .ar_coeff_shift_minus_6 = film_grain->ar_coeff_shift_minus_6, 207 .grain_scale_shift = film_grain->grain_scale_shift, 208 .overlap_flag = film_grain->overlap_flag, 209 .clip_to_restricted_range = film_grain->clip_to_restricted_range, 210 }, 211 .grain_seed = film_grain->grain_seed, 212 .num_y_points = film_grain->num_y_points, 213 .num_cb_points = film_grain->num_cb_points, 214 .num_cr_points = film_grain->num_cr_points, 215 .cb_mult = film_grain->cb_mult, 216 .cb_luma_mult = film_grain->cb_luma_mult, 217 .cb_offset = film_grain->cb_offset, 218 .cr_mult = film_grain->cr_mult, 219 .cr_luma_mult = film_grain->cr_luma_mult, 220 .cr_offset = film_grain->cr_offset, 221 }, 222 .pic_info_fields.bits = { 223 .frame_type = frame_header->frame_type, 224 .show_frame = frame_header->show_frame, 225 .showable_frame = frame_header->showable_frame, 226 .error_resilient_mode = frame_header->error_resilient_mode, 227 .disable_cdf_update = frame_header->disable_cdf_update, 228 .allow_screen_content_tools = frame_header->allow_screen_content_tools, 229 .force_integer_mv = s->cur_frame.force_integer_mv, 230 .allow_intrabc = frame_header->allow_intrabc, 231 .use_superres = frame_header->use_superres, 232 .allow_high_precision_mv = frame_header->allow_high_precision_mv, 233 .is_motion_mode_switchable = frame_header->is_motion_mode_switchable, 234 .use_ref_frame_mvs = frame_header->use_ref_frame_mvs, 235 .disable_frame_end_update_cdf = frame_header->disable_frame_end_update_cdf, 236 .uniform_tile_spacing_flag = frame_header->uniform_tile_spacing_flag, 237 .allow_warped_motion = frame_header->allow_warped_motion, 238 }, 239 .loop_filter_info_fields.bits = { 240 .sharpness_level = frame_header->loop_filter_sharpness, 241 .mode_ref_delta_enabled = frame_header->loop_filter_delta_enabled, 242 .mode_ref_delta_update = frame_header->loop_filter_delta_update, 243 }, 244 .mode_control_fields.bits = { 245 .delta_q_present_flag = frame_header->delta_q_present, 246 .log2_delta_q_res = frame_header->delta_q_res, 247 .delta_lf_present_flag = frame_header->delta_lf_present, 248 .log2_delta_lf_res = frame_header->delta_lf_res, 249 .delta_lf_multi = frame_header->delta_lf_multi, 250 .tx_mode = frame_header->tx_mode, 251 .reference_select = frame_header->reference_select, 252 .reduced_tx_set_used = frame_header->reduced_tx_set, 253 .skip_mode_present = frame_header->skip_mode_present, 254 }, 255 .loop_restoration_fields.bits = { 256 .yframe_restoration_type = remap_lr_type[frame_header->lr_type[0]], 257 .cbframe_restoration_type = remap_lr_type[frame_header->lr_type[1]], 258 .crframe_restoration_type = remap_lr_type[frame_header->lr_type[2]], 259 .lr_unit_shift = frame_header->lr_unit_shift, 260 .lr_uv_shift = frame_header->lr_uv_shift, 261 }, 262 .qmatrix_fields.bits = { 263 .using_qmatrix = frame_header->using_qmatrix, 264 .qm_y = frame_header->qm_y, 265 .qm_u = frame_header->qm_u, 266 .qm_v = frame_header->qm_v, 267 } 268 }; 269 270 for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) { 271 if (pic_param.pic_info_fields.bits.frame_type == AV1_FRAME_KEY && frame_header->show_frame) 272 pic_param.ref_frame_map[i] = VA_INVALID_ID; 273 else 274 pic_param.ref_frame_map[i] = ctx->ref_tab[i].valid ? 275 ff_vaapi_get_surface_id(ctx->ref_tab[i].frame) : 276 vaapi_av1_surface_id(&s->ref[i]); 277 } 278 for (int i = 0; i < AV1_REFS_PER_FRAME; i++) { 279 pic_param.ref_frame_idx[i] = frame_header->ref_frame_idx[i]; 280 } 281 for (int i = 0; i < AV1_TOTAL_REFS_PER_FRAME; i++) { 282 pic_param.ref_deltas[i] = frame_header->loop_filter_ref_deltas[i]; 283 } 284 for (int i = 0; i < 2; i++) { 285 pic_param.mode_deltas[i] = frame_header->loop_filter_mode_deltas[i]; 286 } 287 for (int i = 0; i < (1 << frame_header->cdef_bits); i++) { 288 pic_param.cdef_y_strengths[i] = 289 (frame_header->cdef_y_pri_strength[i] << 2) + 290 frame_header->cdef_y_sec_strength[i]; 291 pic_param.cdef_uv_strengths[i] = 292 (frame_header->cdef_uv_pri_strength[i] << 2) + 293 frame_header->cdef_uv_sec_strength[i]; 294 } 295 for (int i = 0; i < frame_header->tile_cols; i++) { 296 pic_param.width_in_sbs_minus_1[i] = 297 frame_header->width_in_sbs_minus_1[i]; 298 } 299 for (int i = 0; i < frame_header->tile_rows; i++) { 300 pic_param.height_in_sbs_minus_1[i] = 301 frame_header->height_in_sbs_minus_1[i]; 302 } 303 for (int i = AV1_REF_FRAME_LAST; i <= AV1_REF_FRAME_ALTREF; i++) { 304 pic_param.wm[i - 1].invalid = s->cur_frame.gm_invalid[i]; 305 pic_param.wm[i - 1].wmtype = s->cur_frame.gm_type[i]; 306 for (int j = 0; j < 6; j++) 307 pic_param.wm[i - 1].wmmat[j] = s->cur_frame.gm_params[i][j]; 308 } 309 for (int i = 0; i < AV1_MAX_SEGMENTS; i++) { 310 for (int j = 0; j < AV1_SEG_LVL_MAX; j++) { 311 pic_param.seg_info.feature_mask[i] |= (frame_header->feature_enabled[i][j] << j); 312 if (segmentation_feature_signed[j]) 313 pic_param.seg_info.feature_data[i][j] = av_clip(frame_header->feature_value[i][j], 314 -segmentation_feature_max[j], segmentation_feature_max[j]); 315 else 316 pic_param.seg_info.feature_data[i][j] = av_clip(frame_header->feature_value[i][j], 317 0, segmentation_feature_max[j]); 318 } 319 } 320 if (apply_grain) { 321 for (int i = 0; i < film_grain->num_y_points; i++) { 322 pic_param.film_grain_info.point_y_value[i] = 323 film_grain->point_y_value[i]; 324 pic_param.film_grain_info.point_y_scaling[i] = 325 film_grain->point_y_scaling[i]; 326 } 327 for (int i = 0; i < film_grain->num_cb_points; i++) { 328 pic_param.film_grain_info.point_cb_value[i] = 329 film_grain->point_cb_value[i]; 330 pic_param.film_grain_info.point_cb_scaling[i] = 331 film_grain->point_cb_scaling[i]; 332 } 333 for (int i = 0; i < film_grain->num_cr_points; i++) { 334 pic_param.film_grain_info.point_cr_value[i] = 335 film_grain->point_cr_value[i]; 336 pic_param.film_grain_info.point_cr_scaling[i] = 337 film_grain->point_cr_scaling[i]; 338 } 339 for (int i = 0; i < 24; i++) { 340 pic_param.film_grain_info.ar_coeffs_y[i] = 341 film_grain->ar_coeffs_y_plus_128[i] - 128; 342 } 343 for (int i = 0; i < 25; i++) { 344 pic_param.film_grain_info.ar_coeffs_cb[i] = 345 film_grain->ar_coeffs_cb_plus_128[i] - 128; 346 pic_param.film_grain_info.ar_coeffs_cr[i] = 347 film_grain->ar_coeffs_cr_plus_128[i] - 128; 348 } 349 } 350 err = ff_vaapi_decode_make_param_buffer(avctx, pic, 351 VAPictureParameterBufferType, 352 &pic_param, sizeof(pic_param)); 353 if (err < 0) 354 goto fail; 355 356 return 0; 357 358 fail: 359 ff_vaapi_decode_cancel(avctx, pic); 360 return err; 361 } 362 363 static int vaapi_av1_end_frame(AVCodecContext *avctx) 364 { 365 const AV1DecContext *s = avctx->priv_data; 366 const AV1RawFrameHeader *header = s->raw_frame_header; 367 const AV1RawFilmGrainParams *film_grain = &s->cur_frame.film_grain; 368 VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private; 369 VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data; 370 371 int apply_grain = !(avctx->export_side_data & AV_CODEC_EXPORT_DATA_FILM_GRAIN) && film_grain->apply_grain; 372 int ret; 373 ret = ff_vaapi_decode_issue(avctx, pic); 374 if (ret < 0) 375 return ret; 376 377 for (int i = 0; i < AV1_NUM_REF_FRAMES; i++) { 378 if (header->refresh_frame_flags & (1 << i)) { 379 if (ctx->ref_tab[i].frame->buf[0]) 380 av_frame_unref(ctx->ref_tab[i].frame); 381 382 if (apply_grain) { 383 ret = av_frame_ref(ctx->ref_tab[i].frame, ctx->tmp_frame); 384 if (ret < 0) 385 return ret; 386 ctx->ref_tab[i].valid = 1; 387 } else { 388 ctx->ref_tab[i].valid = 0; 389 } 390 } 391 } 392 393 return 0; 394 } 395 396 static int vaapi_av1_decode_slice(AVCodecContext *avctx, 397 const uint8_t *buffer, 398 uint32_t size) 399 { 400 const AV1DecContext *s = avctx->priv_data; 401 VAAPIDecodePicture *pic = s->cur_frame.hwaccel_picture_private; 402 VAAPIAV1DecContext *ctx = avctx->internal->hwaccel_priv_data; 403 int err, nb_params; 404 405 nb_params = s->tg_end - s->tg_start + 1; 406 if (ctx->nb_slice_params < nb_params) { 407 VASliceParameterBufferAV1 *tmp = av_realloc_array(ctx->slice_params, 408 nb_params, 409 sizeof(*ctx->slice_params)); 410 if (!tmp) { 411 ctx->nb_slice_params = 0; 412 err = AVERROR(ENOMEM); 413 goto fail; 414 } 415 ctx->slice_params = tmp; 416 ctx->nb_slice_params = nb_params; 417 } 418 419 for (int i = s->tg_start; i <= s->tg_end; i++) { 420 ctx->slice_params[i - s->tg_start] = (VASliceParameterBufferAV1) { 421 .slice_data_size = s->tile_group_info[i].tile_size, 422 .slice_data_offset = s->tile_group_info[i].tile_offset, 423 .slice_data_flag = VA_SLICE_DATA_FLAG_ALL, 424 .tile_row = s->tile_group_info[i].tile_row, 425 .tile_column = s->tile_group_info[i].tile_column, 426 .tg_start = s->tg_start, 427 .tg_end = s->tg_end, 428 }; 429 } 430 431 err = ff_vaapi_decode_make_slice_buffer(avctx, pic, ctx->slice_params, nb_params, 432 sizeof(VASliceParameterBufferAV1), 433 buffer, 434 size); 435 if (err) 436 goto fail; 437 438 return 0; 439 440 fail: 441 ff_vaapi_decode_cancel(avctx, pic); 442 return err; 443 } 444 445 const FFHWAccel ff_av1_vaapi_hwaccel = { 446 .p.name = "av1_vaapi", 447 .p.type = AVMEDIA_TYPE_VIDEO, 448 .p.id = AV_CODEC_ID_AV1, 449 .p.pix_fmt = AV_PIX_FMT_VAAPI, 450 .start_frame = vaapi_av1_start_frame, 451 .end_frame = vaapi_av1_end_frame, 452 .decode_slice = vaapi_av1_decode_slice, 453 .frame_priv_data_size = sizeof(VAAPIDecodePicture), 454 .init = vaapi_av1_decode_init, 455 .uninit = vaapi_av1_decode_uninit, 456 .frame_params = ff_vaapi_common_frame_params, 457 .priv_data_size = sizeof(VAAPIAV1DecContext), 458 .caps_internal = HWACCEL_CAP_ASYNC_SAFE, 459 };