| @@ -0,0 +1,847 @@ |
| 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 <va/va.h> |
| 20 | +#include <va/va_enc_h264.h> |
| 21 | + |
| 22 | +#include "libavutil/avassert.h" |
| 23 | +#include "libavutil/internal.h" |
| 24 | +#include "libavutil/opt.h" |
| 25 | +#include "libavutil/pixfmt.h" |
| 26 | + |
| 27 | +#include "avcodec.h" |
| 28 | +#include "h264.h" |
| 29 | +#include "internal.h" |
| 30 | +#include "vaapi_encode.h" |
| 31 | +#include "vaapi_encode_h26x.h" |
| 32 | + |
| 33 | +enum { |
| 34 | + SLICE_TYPE_P = 0, |
| 35 | + SLICE_TYPE_B = 1, |
| 36 | + SLICE_TYPE_I = 2, |
| 37 | + SLICE_TYPE_SP = 3, |
| 38 | + SLICE_TYPE_SI = 4, |
| 39 | +}; |
| 40 | + |
| 41 | +// This structure contains all possibly-useful per-sequence syntax elements |
| 42 | +// which are not already contained in the various VAAPI structures. |
| 43 | +typedef struct VAAPIEncodeH264MiscSequenceParams { |
| 44 | + unsigned int profile_idc; |
| 45 | + char constraint_set0_flag; |
| 46 | + char constraint_set1_flag; |
| 47 | + char constraint_set2_flag; |
| 48 | + char constraint_set3_flag; |
| 49 | + char constraint_set4_flag; |
| 50 | + char constraint_set5_flag; |
| 51 | + |
| 52 | + char separate_colour_plane_flag; |
| 53 | + char qpprime_y_zero_transform_bypass_flag; |
| 54 | + |
| 55 | + char gaps_in_frame_num_allowed_flag; |
| 56 | + char delta_pic_order_always_zero_flag; |
| 57 | + char bottom_field_pic_order_in_frame_present_flag; |
| 58 | + |
| 59 | + unsigned int num_slice_groups_minus1; |
| 60 | + unsigned int slice_group_map_type; |
| 61 | + |
| 62 | + int pic_init_qs_minus26; |
| 63 | + |
| 64 | + char vui_parameters_present_flag; |
| 65 | +} VAAPIEncodeH264MiscSequenceParams; |
| 66 | + |
| 67 | +// This structure contains all possibly-useful per-slice syntax elements |
| 68 | +// which are not already contained in the various VAAPI structures. |
| 69 | +typedef struct VAAPIEncodeH264MiscSliceParams { |
| 70 | + unsigned int nal_unit_type; |
| 71 | + unsigned int nal_ref_idc; |
| 72 | + |
| 73 | + unsigned int colour_plane_id; |
| 74 | + char field_pic_flag; |
| 75 | + char bottom_field_flag; |
| 76 | + |
| 77 | + unsigned int redundant_pic_cnt; |
| 78 | + |
| 79 | + char sp_for_switch_flag; |
| 80 | + int slice_qs_delta; |
| 81 | + |
| 82 | + char ref_pic_list_modification_flag_l0; |
| 83 | + char ref_pic_list_modification_flag_l1; |
| 84 | + |
| 85 | + char no_output_of_prior_pics_flag; |
| 86 | + char long_term_reference_flag; |
| 87 | + char adaptive_ref_pic_marking_mode_flag; |
| 88 | +} VAAPIEncodeH264MiscSliceParams; |
| 89 | + |
| 90 | +typedef struct VAAPIEncodeH264Slice { |
| 91 | + VAAPIEncodeH264MiscSliceParams misc_slice_params; |
| 92 | +} VAAPIEncodeH264Slice; |
| 93 | + |
| 94 | +typedef struct VAAPIEncodeH264Context { |
| 95 | + VAAPIEncodeH264MiscSequenceParams misc_sequence_params; |
| 96 | + |
| 97 | + int mb_width; |
| 98 | + int mb_height; |
| 99 | + |
| 100 | + int fixed_qp_idr; |
| 101 | + int fixed_qp_p; |
| 102 | + int fixed_qp_b; |
| 103 | + |
| 104 | + int64_t idr_pic_count; |
| 105 | + int64_t last_idr_frame; |
| 106 | + |
| 107 | + // RefPicList management. |
| 108 | +} VAAPIEncodeH264Context; |
| 109 | + |
| 110 | + |
| 111 | +#define vseq_var(name) vseq->name, name |
| 112 | +#define vseq_field(name) vseq->seq_fields.bits.name, name |
| 113 | +#define vpic_var(name) vpic->name, name |
| 114 | +#define vpic_field(name) vpic->pic_fields.bits.name, name |
| 115 | +#define vslice_var(name) vslice->name, name |
| 116 | +#define vslice_field(name) vslice->slice_fields.bits.name, name |
| 117 | +#define mseq_var(name) mseq->name, name |
| 118 | +#define mslice_var(name) mslice->name, name |
| 119 | + |
| 120 | +static void vaapi_encode_h264_write_nal_header(PutBitContext *pbc, |
| 121 | + int nal_unit_type, int nal_ref_idc) |
| 122 | +{ |
| 123 | + u(1, 0, forbidden_zero_bit); |
| 124 | + u(2, nal_ref_idc, nal_ref_idc); |
| 125 | + u(5, nal_unit_type, nal_unit_type); |
| 126 | +} |
| 127 | + |
| 128 | +static void vaapi_encode_h264_write_trailing_rbsp(PutBitContext *pbc) |
| 129 | +{ |
| 130 | + u(1, 1, rbsp_stop_one_bit); |
| 131 | + while (put_bits_count(pbc) & 7) |
| 132 | + u(1, 0, rbsp_alignment_zero_bit); |
| 133 | +} |
| 134 | + |
| 135 | +static void vaapi_encode_h264_write_sps(PutBitContext *pbc, |
| 136 | + VAAPIEncodeContext *ctx) |
| 137 | +{ |
| 138 | + VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
| 139 | + VAAPIEncodeH264Context *priv = ctx->priv_data; |
| 140 | + VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
| 141 | + int i; |
| 142 | + |
| 143 | + vaapi_encode_h264_write_nal_header(pbc, NAL_SPS, 3); |
| 144 | + |
| 145 | + u(8, mseq_var(profile_idc)); |
| 146 | + u(1, mseq_var(constraint_set0_flag)); |
| 147 | + u(1, mseq_var(constraint_set1_flag)); |
| 148 | + u(1, mseq_var(constraint_set2_flag)); |
| 149 | + u(1, mseq_var(constraint_set3_flag)); |
| 150 | + u(1, mseq_var(constraint_set4_flag)); |
| 151 | + u(1, mseq_var(constraint_set5_flag)); |
| 152 | + u(2, 0, reserved_zero_2bits); |
| 153 | + |
| 154 | + u(8, vseq_var(level_idc)); |
| 155 | + |
| 156 | + ue(vseq_var(seq_parameter_set_id)); |
| 157 | + |
| 158 | + if (mseq->profile_idc == 100 || mseq->profile_idc == 110 || |
| 159 | + mseq->profile_idc == 122 || mseq->profile_idc == 244 || |
| 160 | + mseq->profile_idc == 44 || mseq->profile_idc == 83 || |
| 161 | + mseq->profile_idc == 86 || mseq->profile_idc == 118 || |
| 162 | + mseq->profile_idc == 128 || mseq->profile_idc == 138) { |
| 163 | + ue(vseq_field(chroma_format_idc)); |
| 164 | + |
| 165 | + if (vseq->seq_fields.bits.chroma_format_idc == 3) |
| 166 | + u(1, mseq_var(separate_colour_plane_flag)); |
| 167 | + |
| 168 | + ue(vseq_var(bit_depth_luma_minus8)); |
| 169 | + ue(vseq_var(bit_depth_chroma_minus8)); |
| 170 | + |
| 171 | + u(1, mseq_var(qpprime_y_zero_transform_bypass_flag)); |
| 172 | + |
| 173 | + u(1, vseq_field(seq_scaling_matrix_present_flag)); |
| 174 | + if (vseq->seq_fields.bits.seq_scaling_matrix_present_flag) { |
| 175 | + av_assert0(0 && "scaling matrices not supported"); |
| 176 | + } |
| 177 | + } |
| 178 | + |
| 179 | + ue(vseq_field(log2_max_frame_num_minus4)); |
| 180 | + ue(vseq_field(pic_order_cnt_type)); |
| 181 | + |
| 182 | + if (vseq->seq_fields.bits.pic_order_cnt_type == 0) { |
| 183 | + ue(vseq_field(log2_max_pic_order_cnt_lsb_minus4)); |
| 184 | + } else if (vseq->seq_fields.bits.pic_order_cnt_type == 1) { |
| 185 | + u(1, mseq_var(delta_pic_order_always_zero_flag)); |
| 186 | + se(vseq_var(offset_for_non_ref_pic)); |
| 187 | + se(vseq_var(offset_for_top_to_bottom_field)); |
| 188 | + ue(vseq_var(num_ref_frames_in_pic_order_cnt_cycle)); |
| 189 | + |
| 190 | + for (i = 0; i < vseq->num_ref_frames_in_pic_order_cnt_cycle; i++) |
| 191 | + se(vseq_var(offset_for_ref_frame[i])); |
| 192 | + } |
| 193 | + |
| 194 | + ue(vseq_var(max_num_ref_frames)); |
| 195 | + u(1, mseq_var(gaps_in_frame_num_allowed_flag)); |
| 196 | + |
| 197 | + ue(vseq->picture_width_in_mbs - 1, pic_width_in_mbs_minus1); |
| 198 | + ue(vseq->picture_height_in_mbs - 1, pic_height_in_mbs_minus1); |
| 199 | + |
| 200 | + u(1, vseq_field(frame_mbs_only_flag)); |
| 201 | + if (!vseq->seq_fields.bits.frame_mbs_only_flag) |
| 202 | + u(1, vseq_field(mb_adaptive_frame_field_flag)); |
| 203 | + |
| 204 | + u(1, vseq_field(direct_8x8_inference_flag)); |
| 205 | + |
| 206 | + u(1, vseq_var(frame_cropping_flag)); |
| 207 | + if (vseq->frame_cropping_flag) { |
| 208 | + ue(vseq_var(frame_crop_left_offset)); |
| 209 | + ue(vseq_var(frame_crop_right_offset)); |
| 210 | + ue(vseq_var(frame_crop_top_offset)); |
| 211 | + ue(vseq_var(frame_crop_bottom_offset)); |
| 212 | + } |
| 213 | + |
| 214 | + u(1, mseq_var(vui_parameters_present_flag)); |
| 215 | + |
| 216 | + vaapi_encode_h264_write_trailing_rbsp(pbc); |
| 217 | +} |
| 218 | + |
| 219 | +static void vaapi_encode_h264_write_pps(PutBitContext *pbc, |
| 220 | + VAAPIEncodeContext *ctx) |
| 221 | +{ |
| 222 | + VAEncPictureParameterBufferH264 *vpic = ctx->codec_picture_params; |
| 223 | + VAAPIEncodeH264Context *priv = ctx->priv_data; |
| 224 | + VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
| 225 | + |
| 226 | + vaapi_encode_h264_write_nal_header(pbc, NAL_PPS, 3); |
| 227 | + |
| 228 | + ue(vpic_var(pic_parameter_set_id)); |
| 229 | + ue(vpic_var(seq_parameter_set_id)); |
| 230 | + |
| 231 | + u(1, vpic_field(entropy_coding_mode_flag)); |
| 232 | + u(1, mseq_var(bottom_field_pic_order_in_frame_present_flag)); |
| 233 | + |
| 234 | + ue(mseq_var(num_slice_groups_minus1)); |
| 235 | + if (mseq->num_slice_groups_minus1 > 0) { |
| 236 | + ue(mseq_var(slice_group_map_type)); |
| 237 | + av_assert0(0 && "slice groups not supported"); |
| 238 | + } |
| 239 | + |
| 240 | + ue(vpic_var(num_ref_idx_l0_active_minus1)); |
| 241 | + ue(vpic_var(num_ref_idx_l1_active_minus1)); |
| 242 | + |
| 243 | + u(1, vpic_field(weighted_pred_flag)); |
| 244 | + u(2, vpic_field(weighted_bipred_idc)); |
| 245 | + |
| 246 | + se(vpic->pic_init_qp - 26, pic_init_qp_minus26); |
| 247 | + se(mseq_var(pic_init_qs_minus26)); |
| 248 | + se(vpic_var(chroma_qp_index_offset)); |
| 249 | + |
| 250 | + u(1, vpic_field(deblocking_filter_control_present_flag)); |
| 251 | + u(1, vpic_field(constrained_intra_pred_flag)); |
| 252 | + u(1, vpic_field(redundant_pic_cnt_present_flag)); |
| 253 | + u(1, vpic_field(transform_8x8_mode_flag)); |
| 254 | + |
| 255 | + u(1, vpic_field(pic_scaling_matrix_present_flag)); |
| 256 | + if (vpic->pic_fields.bits.pic_scaling_matrix_present_flag) { |
| 257 | + av_assert0(0 && "scaling matrices not supported"); |
| 258 | + } |
| 259 | + |
| 260 | + se(vpic_var(second_chroma_qp_index_offset)); |
| 261 | + |
| 262 | + vaapi_encode_h264_write_trailing_rbsp(pbc); |
| 263 | +} |
| 264 | + |
| 265 | +static void vaapi_encode_h264_write_slice_header2(PutBitContext *pbc, |
| 266 | + VAAPIEncodeContext *ctx, |
| 267 | + VAAPIEncodePicture *pic, |
| 268 | + VAAPIEncodeSlice *slice) |
| 269 | +{ |
| 270 | + VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
| 271 | + VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; |
| 272 | + VAEncSliceParameterBufferH264 *vslice = slice->codec_slice_params; |
| 273 | + VAAPIEncodeH264Context *priv = ctx->priv_data; |
| 274 | + VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
| 275 | + VAAPIEncodeH264Slice *pslice = slice->priv_data; |
| 276 | + VAAPIEncodeH264MiscSliceParams *mslice = &pslice->misc_slice_params; |
| 277 | + |
| 278 | + vaapi_encode_h264_write_nal_header(pbc, mslice->nal_unit_type, |
| 279 | + mslice->nal_ref_idc); |
| 280 | + |
| 281 | + ue(vslice->macroblock_address, first_mb_in_slice); |
| 282 | + ue(vslice_var(slice_type)); |
| 283 | + ue(vpic_var(pic_parameter_set_id)); |
| 284 | + |
| 285 | + if (mseq->separate_colour_plane_flag) { |
| 286 | + u(2, mslice_var(colour_plane_id)); |
| 287 | + } |
| 288 | + |
| 289 | + u(4 + vseq->seq_fields.bits.log2_max_frame_num_minus4, |
| 290 | + (vpic->frame_num & |
| 291 | + ((1 << (4 + vseq->seq_fields.bits.log2_max_frame_num_minus4)) - 1)), |
| 292 | + frame_num); |
| 293 | + |
| 294 | + if (!vseq->seq_fields.bits.frame_mbs_only_flag) { |
| 295 | + u(1, mslice_var(field_pic_flag)); |
| 296 | + if (mslice->field_pic_flag) |
| 297 | + u(1, mslice_var(bottom_field_flag)); |
| 298 | + } |
| 299 | + |
| 300 | + if (vpic->pic_fields.bits.idr_pic_flag) { |
| 301 | + ue(vslice_var(idr_pic_id)); |
| 302 | + } |
| 303 | + |
| 304 | + if (vseq->seq_fields.bits.pic_order_cnt_type == 0) { |
| 305 | + u(4 + vseq->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4, |
| 306 | + vslice_var(pic_order_cnt_lsb)); |
| 307 | + if (mseq->bottom_field_pic_order_in_frame_present_flag && |
| 308 | + !mslice->field_pic_flag) { |
| 309 | + se(vslice_var(delta_pic_order_cnt_bottom)); |
| 310 | + } |
| 311 | + } |
| 312 | + |
| 313 | + if (vseq->seq_fields.bits.pic_order_cnt_type == 1 && |
| 314 | + !vseq->seq_fields.bits.delta_pic_order_always_zero_flag) { |
| 315 | + se(vslice_var(delta_pic_order_cnt[0])); |
| 316 | + if (mseq->bottom_field_pic_order_in_frame_present_flag && |
| 317 | + !mslice->field_pic_flag) { |
| 318 | + se(vslice_var(delta_pic_order_cnt[1])); |
| 319 | + } |
| 320 | + } |
| 321 | + |
| 322 | + if (vpic->pic_fields.bits.redundant_pic_cnt_present_flag) { |
| 323 | + ue(mslice_var(redundant_pic_cnt)); |
| 324 | + } |
| 325 | + |
| 326 | + if (vslice->slice_type == SLICE_TYPE_B) { |
| 327 | + u(1, vslice_var(direct_spatial_mv_pred_flag)); |
| 328 | + } |
| 329 | + |
| 330 | + if (vslice->slice_type == SLICE_TYPE_P || |
| 331 | + vslice->slice_type == SLICE_TYPE_SP || |
| 332 | + vslice->slice_type == SLICE_TYPE_B) { |
| 333 | + u(1, vslice_var(num_ref_idx_active_override_flag)); |
| 334 | + if (vslice->num_ref_idx_active_override_flag) { |
| 335 | + ue(vslice_var(num_ref_idx_l0_active_minus1)); |
| 336 | + if (vslice->slice_type == SLICE_TYPE_B) |
| 337 | + ue(vslice_var(num_ref_idx_l1_active_minus1)); |
| 338 | + } |
| 339 | + } |
| 340 | + |
| 341 | + if (mslice->nal_unit_type == 20 || mslice->nal_unit_type == 21) { |
| 342 | + av_assert0(0 && "no MVC support"); |
| 343 | + } else { |
| 344 | + if (vslice->slice_type % 5 != 2 && vslice->slice_type % 5 != 4) { |
| 345 | + u(1, mslice_var(ref_pic_list_modification_flag_l0)); |
| 346 | + if (mslice->ref_pic_list_modification_flag_l0) { |
| 347 | + av_assert0(0 && "ref pic list modification"); |
| 348 | + } |
| 349 | + } |
| 350 | + if (vslice->slice_type % 5 == 1) { |
| 351 | + u(1, mslice_var(ref_pic_list_modification_flag_l1)); |
| 352 | + if (mslice->ref_pic_list_modification_flag_l1) { |
| 353 | + av_assert0(0 && "ref pic list modification"); |
| 354 | + } |
| 355 | + } |
| 356 | + } |
| 357 | + |
| 358 | + if ((vpic->pic_fields.bits.weighted_pred_flag && |
| 359 | + (vslice->slice_type == SLICE_TYPE_P || |
| 360 | + vslice->slice_type == SLICE_TYPE_SP)) || |
| 361 | + (vpic->pic_fields.bits.weighted_bipred_idc == 1 && |
| 362 | + vslice->slice_type == SLICE_TYPE_B)) { |
| 363 | + av_assert0(0 && "prediction weights not supported"); |
| 364 | + } |
| 365 | + |
| 366 | + av_assert0(mslice->nal_ref_idc > 0 == |
| 367 | + vpic->pic_fields.bits.reference_pic_flag); |
| 368 | + if (mslice->nal_ref_idc != 0) { |
| 369 | + if (vpic->pic_fields.bits.idr_pic_flag) { |
| 370 | + u(1, mslice_var(no_output_of_prior_pics_flag)); |
| 371 | + u(1, mslice_var(long_term_reference_flag)); |
| 372 | + } else { |
| 373 | + u(1, mslice_var(adaptive_ref_pic_marking_mode_flag)); |
| 374 | + if (mslice->adaptive_ref_pic_marking_mode_flag) { |
| 375 | + av_assert0(0 && "MMCOs not supported"); |
| 376 | + } |
| 377 | + } |
| 378 | + } |
| 379 | + |
| 380 | + if (vpic->pic_fields.bits.entropy_coding_mode_flag && |
| 381 | + vslice->slice_type != SLICE_TYPE_I && |
| 382 | + vslice->slice_type != SLICE_TYPE_SI) { |
| 383 | + ue(vslice_var(cabac_init_idc)); |
| 384 | + } |
| 385 | + |
| 386 | + se(vslice_var(slice_qp_delta)); |
| 387 | + if (vslice->slice_type == SLICE_TYPE_SP || |
| 388 | + vslice->slice_type == SLICE_TYPE_SI) { |
| 389 | + if (vslice->slice_type == SLICE_TYPE_SP) |
| 390 | + u(1, mslice_var(sp_for_switch_flag)); |
| 391 | + se(mslice_var(slice_qs_delta)); |
| 392 | + } |
| 393 | + |
| 394 | + if (vpic->pic_fields.bits.deblocking_filter_control_present_flag) { |
| 395 | + ue(vslice_var(disable_deblocking_filter_idc)); |
| 396 | + if (vslice->disable_deblocking_filter_idc != 1) { |
| 397 | + se(vslice_var(slice_alpha_c0_offset_div2)); |
| 398 | + se(vslice_var(slice_beta_offset_div2)); |
| 399 | + } |
| 400 | + } |
| 401 | + |
| 402 | + if (mseq->num_slice_groups_minus1 > 0 && |
| 403 | + mseq->slice_group_map_type >= 3 && mseq->slice_group_map_type <= 5) { |
| 404 | + av_assert0(0 && "slice groups not supported"); |
| 405 | + } |
| 406 | + |
| 407 | + // No alignment - this need not be a byte boundary. |
| 408 | +} |
| 409 | + |
| 410 | +static int vaapi_encode_h264_write_sequence_header(AVCodecContext *avctx, |
| 411 | + char *data, size_t *data_len) |
| 412 | +{ |
| 413 | + VAAPIEncodeContext *ctx = avctx->priv_data; |
| 414 | + PutBitContext pbc; |
| 415 | + char tmp[256]; |
| 416 | + int err; |
| 417 | + size_t nal_len, bit_len, bit_pos, next_len; |
| 418 | + |
| 419 | + bit_len = *data_len; |
| 420 | + bit_pos = 0; |
| 421 | + |
| 422 | + init_put_bits(&pbc, tmp, sizeof(tmp)); |
| 423 | + vaapi_encode_h264_write_sps(&pbc, ctx); |
| 424 | + nal_len = put_bits_count(&pbc); |
| 425 | + flush_put_bits(&pbc); |
| 426 | + |
| 427 | + next_len = bit_len - bit_pos; |
| 428 | + err = ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data + bit_pos / 8, |
| 429 | + &next_len, |
| 430 | + tmp, nal_len); |
| 431 | + if (err < 0) |
| 432 | + return err; |
| 433 | + bit_pos += next_len; |
| 434 | + |
| 435 | + init_put_bits(&pbc, tmp, sizeof(tmp)); |
| 436 | + vaapi_encode_h264_write_pps(&pbc, ctx); |
| 437 | + nal_len = put_bits_count(&pbc); |
| 438 | + flush_put_bits(&pbc); |
| 439 | + |
| 440 | + next_len = bit_len - bit_pos; |
| 441 | + err = ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data + bit_pos / 8, |
| 442 | + &next_len, |
| 443 | + tmp, nal_len); |
| 444 | + if (err < 0) |
| 445 | + return err; |
| 446 | + bit_pos += next_len; |
| 447 | + |
| 448 | + *data_len = bit_pos; |
| 449 | + return 0; |
| 450 | +} |
| 451 | + |
| 452 | +static int vaapi_encode_h264_write_slice_header(AVCodecContext *avctx, |
| 453 | + VAAPIEncodePicture *pic, |
| 454 | + VAAPIEncodeSlice *slice, |
| 455 | + char *data, size_t *data_len) |
| 456 | +{ |
| 457 | + VAAPIEncodeContext *ctx = avctx->priv_data; |
| 458 | + PutBitContext pbc; |
| 459 | + char tmp[256]; |
| 460 | + size_t header_len; |
| 461 | + |
| 462 | + init_put_bits(&pbc, tmp, sizeof(tmp)); |
| 463 | + vaapi_encode_h264_write_slice_header2(&pbc, ctx, pic, slice); |
| 464 | + header_len = put_bits_count(&pbc); |
| 465 | + flush_put_bits(&pbc); |
| 466 | + |
| 467 | + return ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data, data_len, |
| 468 | + tmp, header_len); |
| 469 | +} |
| 470 | + |
| 471 | +static int vaapi_encode_h264_init_sequence_params(AVCodecContext *avctx) |
| 472 | +{ |
| 473 | + VAAPIEncodeContext *ctx = avctx->priv_data; |
| 474 | + VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
| 475 | + VAEncPictureParameterBufferH264 *vpic = ctx->codec_picture_params; |
| 476 | + VAAPIEncodeH264Context *priv = ctx->priv_data; |
| 477 | + VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
| 478 | + int i; |
| 479 | + |
| 480 | + { |
| 481 | + vseq->seq_parameter_set_id = 0; |
| 482 | + |
| 483 | + vseq->level_idc = avctx->level; |
| 484 | + |
| 485 | + vseq->max_num_ref_frames = 2; |
| 486 | + |
| 487 | + vseq->picture_width_in_mbs = priv->mb_width; |
| 488 | + vseq->picture_height_in_mbs = priv->mb_height; |
| 489 | + |
| 490 | + vseq->seq_fields.bits.chroma_format_idc = 1; |
| 491 | + vseq->seq_fields.bits.frame_mbs_only_flag = 1; |
| 492 | + vseq->seq_fields.bits.direct_8x8_inference_flag = 1; |
| 493 | + vseq->seq_fields.bits.log2_max_frame_num_minus4 = 4; |
| 494 | + vseq->seq_fields.bits.pic_order_cnt_type = 0; |
| 495 | + |
| 496 | + if (ctx->input_width != ctx->aligned_width || |
| 497 | + ctx->input_height != ctx->aligned_height) { |
| 498 | + vseq->frame_cropping_flag = 1; |
| 499 | + |
| 500 | + vseq->frame_crop_left_offset = 0; |
| 501 | + vseq->frame_crop_right_offset = |
| 502 | + (ctx->aligned_width - ctx->input_width) / 2; |
| 503 | + vseq->frame_crop_top_offset = 0; |
| 504 | + vseq->frame_crop_bottom_offset = |
| 505 | + (ctx->aligned_height - ctx->input_height) / 2; |
| 506 | + } else { |
| 507 | + vseq->frame_cropping_flag = 0; |
| 508 | + } |
| 509 | + } |
| 510 | + |
| 511 | + { |
| 512 | + vpic->CurrPic.picture_id = VA_INVALID_ID; |
| 513 | + vpic->CurrPic.flags = VA_PICTURE_H264_INVALID; |
| 514 | + |
| 515 | + for (i = 0; i < FF_ARRAY_ELEMS(vpic->ReferenceFrames); i++) { |
| 516 | + vpic->ReferenceFrames[i].picture_id = VA_INVALID_ID; |
| 517 | + vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_INVALID; |
| 518 | + } |
| 519 | + |
| 520 | + vpic->coded_buf = VA_INVALID_ID; |
| 521 | + |
| 522 | + vpic->pic_parameter_set_id = 0; |
| 523 | + vpic->seq_parameter_set_id = 0; |
| 524 | + |
| 525 | + vpic->num_ref_idx_l0_active_minus1 = 0; |
| 526 | + vpic->num_ref_idx_l1_active_minus1 = 0; |
| 527 | + |
| 528 | + vpic->pic_fields.bits.entropy_coding_mode_flag = |
| 529 | + ((avctx->profile & 0xff) != 66); |
| 530 | + vpic->pic_fields.bits.weighted_pred_flag = 0; |
| 531 | + vpic->pic_fields.bits.weighted_bipred_idc = 0; |
| 532 | + vpic->pic_fields.bits.transform_8x8_mode_flag = |
| 533 | + ((avctx->profile & 0xff) >= 100); |
| 534 | + |
| 535 | + vpic->pic_init_qp = priv->fixed_qp_idr; |
| 536 | + } |
| 537 | + |
| 538 | + { |
| 539 | + mseq->profile_idc = avctx->profile & 0xff; |
| 540 | + |
| 541 | + if (avctx->profile & FF_PROFILE_H264_CONSTRAINED) |
| 542 | + mseq->constraint_set1_flag = 1; |
| 543 | + if (avctx->profile & FF_PROFILE_H264_INTRA) |
| 544 | + mseq->constraint_set3_flag = 1; |
| 545 | + } |
| 546 | + |
| 547 | + return 0; |
| 548 | +} |
| 549 | + |
| 550 | +static int vaapi_encode_h264_init_picture_params(AVCodecContext *avctx, |
| 551 | + VAAPIEncodePicture *pic) |
| 552 | +{ |
| 553 | + VAAPIEncodeContext *ctx = avctx->priv_data; |
| 554 | + VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
| 555 | + VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; |
| 556 | + VAAPIEncodeH264Context *priv = ctx->priv_data; |
| 557 | + int i; |
| 558 | + |
| 559 | + if (pic->type == PICTURE_TYPE_IDR) { |
| 560 | + av_assert0(pic->display_order == pic->encode_order); |
| 561 | + priv->last_idr_frame = pic->display_order; |
| 562 | + } else { |
| 563 | + av_assert0(pic->display_order > priv->last_idr_frame); |
| 564 | + } |
| 565 | + |
| 566 | + vpic->frame_num = (pic->encode_order - priv->last_idr_frame) & |
| 567 | + ((1 << (4 + vseq->seq_fields.bits.log2_max_frame_num_minus4)) - 1); |
| 568 | + |
| 569 | + vpic->CurrPic.picture_id = pic->recon_surface; |
| 570 | + vpic->CurrPic.frame_idx = vpic->frame_num; |
| 571 | + vpic->CurrPic.flags = 0; |
| 572 | + vpic->CurrPic.TopFieldOrderCnt = pic->display_order; |
| 573 | + vpic->CurrPic.BottomFieldOrderCnt = pic->display_order; |
| 574 | + |
| 575 | + for (i = 0; i < pic->nb_refs; i++) { |
| 576 | + VAAPIEncodePicture *ref = pic->refs[i]; |
| 577 | + av_assert0(ref && ref->encode_order >= priv->last_idr_frame); |
| 578 | + vpic->ReferenceFrames[i].picture_id = ref->recon_surface; |
| 579 | + vpic->ReferenceFrames[i].frame_idx = |
| 580 | + ref->encode_order - priv->last_idr_frame; |
| 581 | + vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_SHORT_TERM_REFERENCE; |
| 582 | + vpic->ReferenceFrames[i].TopFieldOrderCnt = ref->display_order; |
| 583 | + vpic->ReferenceFrames[i].BottomFieldOrderCnt = ref->display_order; |
| 584 | + } |
| 585 | + for (; i < FF_ARRAY_ELEMS(vpic->ReferenceFrames); i++) { |
| 586 | + vpic->ReferenceFrames[i].picture_id = VA_INVALID_ID; |
| 587 | + vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_INVALID; |
| 588 | + } |
| 589 | + |
| 590 | + vpic->coded_buf = pic->output_buffer; |
| 591 | + |
| 592 | + vpic->pic_fields.bits.idr_pic_flag = (pic->type == PICTURE_TYPE_IDR); |
| 593 | + vpic->pic_fields.bits.reference_pic_flag = (pic->type != PICTURE_TYPE_B); |
| 594 | + |
| 595 | + pic->nb_slices = 1; |
| 596 | + |
| 597 | + return 0; |
| 598 | +} |
| 599 | + |
| 600 | +static int vaapi_encode_h264_init_slice_params(AVCodecContext *avctx, |
| 601 | + VAAPIEncodePicture *pic, |
| 602 | + VAAPIEncodeSlice *slice) |
| 603 | +{ |
| 604 | + VAAPIEncodeContext *ctx = avctx->priv_data; |
| 605 | + VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
| 606 | + VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; |
| 607 | + VAEncSliceParameterBufferH264 *vslice = slice->codec_slice_params; |
| 608 | + VAAPIEncodeH264Context *priv = ctx->priv_data; |
| 609 | + VAAPIEncodeH264Slice *pslice; |
| 610 | + VAAPIEncodeH264MiscSliceParams *mslice; |
| 611 | + int i; |
| 612 | + |
| 613 | + slice->priv_data = av_mallocz(sizeof(*pslice)); |
| 614 | + if (!slice->priv_data) |
| 615 | + return AVERROR(ENOMEM); |
| 616 | + pslice = slice->priv_data; |
| 617 | + mslice = &pslice->misc_slice_params; |
| 618 | + |
| 619 | + if (pic->type == PICTURE_TYPE_IDR) |
| 620 | + mslice->nal_unit_type = NAL_IDR_SLICE; |
| 621 | + else |
| 622 | + mslice->nal_unit_type = NAL_SLICE; |
| 623 | + |
| 624 | + switch (pic->type) { |
| 625 | + case PICTURE_TYPE_IDR: |
| 626 | + vslice->slice_type = SLICE_TYPE_I; |
| 627 | + mslice->nal_ref_idc = 3; |
| 628 | + break; |
| 629 | + case PICTURE_TYPE_I: |
| 630 | + vslice->slice_type = SLICE_TYPE_I; |
| 631 | + mslice->nal_ref_idc = 2; |
| 632 | + break; |
| 633 | + case PICTURE_TYPE_P: |
| 634 | + vslice->slice_type = SLICE_TYPE_P; |
| 635 | + mslice->nal_ref_idc = 1; |
| 636 | + break; |
| 637 | + case PICTURE_TYPE_B: |
| 638 | + vslice->slice_type = SLICE_TYPE_B; |
| 639 | + mslice->nal_ref_idc = 0; |
| 640 | + break; |
| 641 | + default: |
| 642 | + av_assert0(0 && "invalid picture type"); |
| 643 | + } |
| 644 | + |
| 645 | + // Only one slice per frame. |
| 646 | + vslice->macroblock_address = 0; |
| 647 | + vslice->num_macroblocks = priv->mb_width * priv->mb_height; |
| 648 | + |
| 649 | + vslice->macroblock_info = VA_INVALID_ID; |
| 650 | + |
| 651 | + vslice->pic_parameter_set_id = vpic->pic_parameter_set_id; |
| 652 | + vslice->idr_pic_id = priv->idr_pic_count++; |
| 653 | + |
| 654 | + vslice->pic_order_cnt_lsb = pic->display_order & |
| 655 | + ((1 << (4 + vseq->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4)) - 1); |
| 656 | + |
| 657 | + for (i = 0; i < FF_ARRAY_ELEMS(vslice->RefPicList0); i++) { |
| 658 | + vslice->RefPicList0[i].picture_id = VA_INVALID_ID; |
| 659 | + vslice->RefPicList0[i].flags = VA_PICTURE_H264_INVALID; |
| 660 | + vslice->RefPicList1[i].picture_id = VA_INVALID_ID; |
| 661 | + vslice->RefPicList1[i].flags = VA_PICTURE_H264_INVALID; |
| 662 | + } |
| 663 | + |
| 664 | + av_assert0(pic->nb_refs <= 2); |
| 665 | + if (pic->nb_refs >= 1) { |
| 666 | + // Backward reference for P or B frame. |
| 667 | + av_assert0(pic->type == PICTURE_TYPE_P || |
| 668 | + pic->type == PICTURE_TYPE_B); |
| 669 | + |
| 670 | + vslice->num_ref_idx_l0_active_minus1 = 0; |
| 671 | + vslice->RefPicList0[0] = vpic->ReferenceFrames[0]; |
| 672 | + } |
| 673 | + if (pic->nb_refs >= 2) { |
| 674 | + // Forward reference for B frame. |
| 675 | + av_assert0(pic->type == PICTURE_TYPE_B); |
| 676 | + |
| 677 | + vslice->num_ref_idx_l1_active_minus1 = 0; |
| 678 | + vslice->RefPicList1[0] = vpic->ReferenceFrames[1]; |
| 679 | + } |
| 680 | + |
| 681 | + if (pic->type == PICTURE_TYPE_B) |
| 682 | + vslice->slice_qp_delta = priv->fixed_qp_b - vpic->pic_init_qp; |
| 683 | + else if (pic->type == PICTURE_TYPE_P) |
| 684 | + vslice->slice_qp_delta = priv->fixed_qp_p - vpic->pic_init_qp; |
| 685 | + else |
| 686 | + vslice->slice_qp_delta = priv->fixed_qp_idr - vpic->pic_init_qp; |
| 687 | + |
| 688 | + vslice->direct_spatial_mv_pred_flag = 1; |
| 689 | + |
| 690 | + return 0; |
| 691 | +} |
| 692 | + |
| 693 | +static VAConfigAttrib vaapi_encode_h264_config_attributes[] = { |
| 694 | + { .type = VAConfigAttribRTFormat, |
| 695 | + .value = VA_RT_FORMAT_YUV420 }, |
| 696 | + { .type = VAConfigAttribRateControl, |
| 697 | + .value = VA_RC_CQP }, |
| 698 | + { .type = VAConfigAttribEncPackedHeaders, |
| 699 | + .value = (VA_ENC_PACKED_HEADER_SEQUENCE | |
| 700 | + VA_ENC_PACKED_HEADER_SLICE) }, |
| 701 | +}; |
| 702 | + |
| 703 | +static av_cold int vaapi_encode_h264_init_internal(AVCodecContext *avctx) |
| 704 | +{ |
| 705 | + VAAPIEncodeContext *ctx = avctx->priv_data; |
| 706 | + VAAPIEncodeH264Context *priv = ctx->priv_data; |
| 707 | + |
| 708 | + switch (avctx->profile) { |
| 709 | + case FF_PROFILE_H264_CONSTRAINED_BASELINE: |
| 710 | + ctx->va_profile = VAProfileH264ConstrainedBaseline; |
| 711 | + break; |
| 712 | + case FF_PROFILE_H264_BASELINE: |
| 713 | + ctx->va_profile = VAProfileH264Baseline; |
| 714 | + break; |
| 715 | + case FF_PROFILE_H264_MAIN: |
| 716 | + ctx->va_profile = VAProfileH264Main; |
| 717 | + break; |
| 718 | + case FF_PROFILE_H264_EXTENDED: |
| 719 | + av_log(avctx, AV_LOG_ERROR, "H.264 extended profile " |
| 720 | + "is not supported.\n"); |
| 721 | + return AVERROR_PATCHWELCOME; |
| 722 | + case FF_PROFILE_UNKNOWN: |
| 723 | + case FF_PROFILE_H264_HIGH: |
| 724 | + ctx->va_profile = VAProfileH264High; |
| 725 | + break; |
| 726 | + case FF_PROFILE_H264_HIGH_10: |
| 727 | + case FF_PROFILE_H264_HIGH_10_INTRA: |
| 728 | + av_log(avctx, AV_LOG_ERROR, "H.264 10-bit profiles " |
| 729 | + "are not supported.\n"); |
| 730 | + return AVERROR_PATCHWELCOME; |
| 731 | + case FF_PROFILE_H264_HIGH_422: |
| 732 | + case FF_PROFILE_H264_HIGH_422_INTRA: |
| 733 | + case FF_PROFILE_H264_HIGH_444: |
| 734 | + case FF_PROFILE_H264_HIGH_444_PREDICTIVE: |
| 735 | + case FF_PROFILE_H264_HIGH_444_INTRA: |
| 736 | + case FF_PROFILE_H264_CAVLC_444: |
| 737 | + av_log(avctx, AV_LOG_ERROR, "H.264 non-4:2:0 profiles " |
| 738 | + "are not supported.\n"); |
| 739 | + return AVERROR_PATCHWELCOME; |
| 740 | + default: |
| 741 | + av_log(avctx, AV_LOG_ERROR, "Unknown H.264 profile %d.\n", |
| 742 | + avctx->profile); |
| 743 | + return AVERROR(EINVAL); |
| 744 | + } |
| 745 | + ctx->va_entrypoint = VAEntrypointEncSlice; |
| 746 | + |
| 747 | + ctx->va_rc_mode = VA_RC_CQP; |
| 748 | + |
| 749 | + ctx->input_width = avctx->width; |
| 750 | + ctx->input_height = avctx->height; |
| 751 | + ctx->aligned_width = FFALIGN(ctx->input_width, 16); |
| 752 | + ctx->aligned_height = FFALIGN(ctx->input_height, 16); |
| 753 | + priv->mb_width = ctx->aligned_width / 16; |
| 754 | + priv->mb_height = ctx->aligned_height / 16; |
| 755 | + |
| 756 | + if (avctx->bit_rate > 0) { |
| 757 | + av_log(avctx, AV_LOG_ERROR, "Constant bitrate encoding is not " |
| 758 | + "supported!\n"); |
| 759 | + return AVERROR_PATCHWELCOME; |
| 760 | + } |
| 761 | + |
| 762 | + priv->fixed_qp_p = avctx->global_quality; |
| 763 | + if (avctx->i_quant_factor > 0.0) |
| 764 | + priv->fixed_qp_idr = (int)((priv->fixed_qp_p * avctx->i_quant_factor + |
| 765 | + avctx->i_quant_offset) + 0.5); |
| 766 | + else |
| 767 | + priv->fixed_qp_idr = priv->fixed_qp_p; |
| 768 | + if (avctx->b_quant_factor > 0.0) |
| 769 | + priv->fixed_qp_b = (int)((priv->fixed_qp_p * avctx->b_quant_factor + |
| 770 | + avctx->b_quant_offset) + 0.5); |
| 771 | + else |
| 772 | + priv->fixed_qp_b = priv->fixed_qp_p; |
| 773 | + av_log(avctx, AV_LOG_DEBUG, "QP = %d / %d / %d for IDR / P / B frames.\n", |
| 774 | + priv->fixed_qp_idr, priv->fixed_qp_p, priv->fixed_qp_b); |
| 775 | + |
| 776 | + ctx->config_attributes = vaapi_encode_h264_config_attributes; |
| 777 | + ctx->nb_config_attributes = |
| 778 | + FF_ARRAY_ELEMS(vaapi_encode_h264_config_attributes); |
| 779 | + |
| 780 | + ctx->nb_recon_frames = 20; |
| 781 | + |
| 782 | + return 0; |
| 783 | +} |
| 784 | + |
| 785 | +static VAAPIEncodeType vaapi_encode_type_h264 = { |
| 786 | + .priv_data_size = sizeof(VAAPIEncodeH264Context), |
| 787 | + |
| 788 | + .init = &vaapi_encode_h264_init_internal, |
| 789 | + |
| 790 | + .sequence_params_size = sizeof(VAEncSequenceParameterBufferH264), |
| 791 | + .init_sequence_params = &vaapi_encode_h264_init_sequence_params, |
| 792 | + |
| 793 | + .picture_params_size = sizeof(VAEncPictureParameterBufferH264), |
| 794 | + .init_picture_params = &vaapi_encode_h264_init_picture_params, |
| 795 | + |
| 796 | + .slice_params_size = sizeof(VAEncSliceParameterBufferH264), |
| 797 | + .init_slice_params = &vaapi_encode_h264_init_slice_params, |
| 798 | + |
| 799 | + .sequence_header_type = VAEncPackedHeaderSequence, |
| 800 | + .write_sequence_header = &vaapi_encode_h264_write_sequence_header, |
| 801 | + |
| 802 | + .slice_header_type = VAEncPackedHeaderH264_Slice, |
| 803 | + .write_slice_header = &vaapi_encode_h264_write_slice_header, |
| 804 | +}; |
| 805 | + |
| 806 | +static av_cold int vaapi_encode_h264_init(AVCodecContext *avctx) |
| 807 | +{ |
| 808 | + return ff_vaapi_encode_init(avctx, &vaapi_encode_type_h264); |
| 809 | +} |
| 810 | + |
| 811 | +static const AVCodecDefault vaapi_encode_h264_defaults[] = { |
| 812 | + { "profile", "100" }, |
| 813 | + { "level", "51" }, |
| 814 | + { "b", "0" }, |
| 815 | + { "bf", "2" }, |
| 816 | + { "g", "120" }, |
| 817 | + { "global_quality", "20" }, |
| 818 | + { "i_qfactor", "1.0" }, |
| 819 | + { "i_qoffset", "0.0" }, |
| 820 | + { "b_qfactor", "1.2" }, |
| 821 | + { "b_qoffset", "0.0" }, |
| 822 | + { NULL }, |
| 823 | +}; |
| 824 | + |
| 825 | +static const AVClass vaapi_encode_h264_class = { |
| 826 | + .class_name = "h264_vaapi", |
| 827 | + .item_name = av_default_item_name, |
| 828 | + .version = LIBAVUTIL_VERSION_INT, |
| 829 | +}; |
| 830 | + |
| 831 | +AVCodec ff_h264_vaapi_encoder = { |
| 832 | + .name = "h264_vaapi", |
| 833 | + .long_name = NULL_IF_CONFIG_SMALL("H.264/AVC (VAAPI)"), |
| 834 | + .type = AVMEDIA_TYPE_VIDEO, |
| 835 | + .id = AV_CODEC_ID_H264, |
| 836 | + .priv_data_size = sizeof(VAAPIEncodeContext), |
| 837 | + .init = &vaapi_encode_h264_init, |
| 838 | + .encode2 = &ff_vaapi_encode2, |
| 839 | + .close = &ff_vaapi_encode_close, |
| 840 | + .priv_class = &vaapi_encode_h264_class, |
| 841 | + .capabilities = AV_CODEC_CAP_DELAY, |
| 842 | + .defaults = vaapi_encode_h264_defaults, |
| 843 | + .pix_fmts = (const enum AVPixelFormat[]) { |
| 844 | + AV_PIX_FMT_VAAPI, |
| 845 | + AV_PIX_FMT_NONE, |
| 846 | + }, |
| 847 | +}; |