blob: 92e29554edca58bc5b86e2e734c34132fd42e653
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 "h264_sei.h" |
30 | #include "internal.h" |
31 | #include "vaapi_encode.h" |
32 | #include "vaapi_encode_h26x.h" |
33 | |
34 | enum { |
35 | SLICE_TYPE_P = 0, |
36 | SLICE_TYPE_B = 1, |
37 | SLICE_TYPE_I = 2, |
38 | SLICE_TYPE_SP = 3, |
39 | SLICE_TYPE_SI = 4, |
40 | }; |
41 | |
42 | // This structure contains all possibly-useful per-sequence syntax elements |
43 | // which are not already contained in the various VAAPI structures. |
44 | typedef struct VAAPIEncodeH264MiscSequenceParams { |
45 | unsigned int profile_idc; |
46 | char constraint_set0_flag; |
47 | char constraint_set1_flag; |
48 | char constraint_set2_flag; |
49 | char constraint_set3_flag; |
50 | char constraint_set4_flag; |
51 | char constraint_set5_flag; |
52 | |
53 | char separate_colour_plane_flag; |
54 | char qpprime_y_zero_transform_bypass_flag; |
55 | |
56 | char gaps_in_frame_num_allowed_flag; |
57 | char delta_pic_order_always_zero_flag; |
58 | char bottom_field_pic_order_in_frame_present_flag; |
59 | |
60 | unsigned int num_slice_groups_minus1; |
61 | unsigned int slice_group_map_type; |
62 | |
63 | int pic_init_qs_minus26; |
64 | |
65 | char overscan_info_present_flag; |
66 | char overscan_appropriate_flag; |
67 | |
68 | char video_signal_type_present_flag; |
69 | unsigned int video_format; |
70 | char video_full_range_flag; |
71 | char colour_description_present_flag; |
72 | unsigned int colour_primaries; |
73 | unsigned int transfer_characteristics; |
74 | unsigned int matrix_coefficients; |
75 | |
76 | char chroma_loc_info_present_flag; |
77 | unsigned int chroma_sample_loc_type_top_field; |
78 | unsigned int chroma_sample_loc_type_bottom_field; |
79 | |
80 | // Some timing elements are in VAEncSequenceParameterBufferH264. |
81 | char fixed_frame_rate_flag; |
82 | |
83 | char nal_hrd_parameters_present_flag; |
84 | char vcl_hrd_parameters_present_flag; |
85 | char low_delay_hrd_flag; |
86 | char pic_struct_present_flag; |
87 | |
88 | char motion_vectors_over_pic_boundaries_flag; |
89 | unsigned int max_bytes_per_pic_denom; |
90 | unsigned int max_bits_per_mb_denom; |
91 | unsigned int max_num_reorder_frames; |
92 | unsigned int max_dec_pic_buffering; |
93 | |
94 | unsigned int cpb_cnt_minus1; |
95 | unsigned int bit_rate_scale; |
96 | unsigned int cpb_size_scale; |
97 | unsigned int bit_rate_value_minus1[32]; |
98 | unsigned int cpb_size_value_minus1[32]; |
99 | char cbr_flag[32]; |
100 | unsigned int initial_cpb_removal_delay_length_minus1; |
101 | unsigned int cpb_removal_delay_length_minus1; |
102 | unsigned int dpb_output_delay_length_minus1; |
103 | unsigned int time_offset_length; |
104 | |
105 | unsigned int initial_cpb_removal_delay; |
106 | unsigned int initial_cpb_removal_delay_offset; |
107 | |
108 | unsigned int pic_struct; |
109 | } VAAPIEncodeH264MiscSequenceParams; |
110 | |
111 | // This structure contains all possibly-useful per-slice syntax elements |
112 | // which are not already contained in the various VAAPI structures. |
113 | typedef struct VAAPIEncodeH264MiscSliceParams { |
114 | unsigned int nal_unit_type; |
115 | unsigned int nal_ref_idc; |
116 | |
117 | unsigned int colour_plane_id; |
118 | char field_pic_flag; |
119 | char bottom_field_flag; |
120 | |
121 | unsigned int redundant_pic_cnt; |
122 | |
123 | char sp_for_switch_flag; |
124 | int slice_qs_delta; |
125 | |
126 | char ref_pic_list_modification_flag_l0; |
127 | char ref_pic_list_modification_flag_l1; |
128 | |
129 | char no_output_of_prior_pics_flag; |
130 | char long_term_reference_flag; |
131 | char adaptive_ref_pic_marking_mode_flag; |
132 | } VAAPIEncodeH264MiscSliceParams; |
133 | |
134 | typedef struct VAAPIEncodeH264Slice { |
135 | VAAPIEncodeH264MiscSliceParams misc_slice_params; |
136 | } VAAPIEncodeH264Slice; |
137 | |
138 | typedef struct VAAPIEncodeH264Context { |
139 | VAAPIEncodeH264MiscSequenceParams misc_sequence_params; |
140 | |
141 | int mb_width; |
142 | int mb_height; |
143 | |
144 | int fixed_qp_idr; |
145 | int fixed_qp_p; |
146 | int fixed_qp_b; |
147 | |
148 | int next_frame_num; |
149 | int64_t last_idr_frame; |
150 | int64_t idr_pic_count; |
151 | |
152 | int cpb_delay; |
153 | int dpb_delay; |
154 | |
155 | // Rate control configuration. |
156 | int send_timing_sei; |
157 | |
158 | #if VA_CHECK_VERSION(0, 36, 0) |
159 | // Speed-quality tradeoff setting. |
160 | struct { |
161 | VAEncMiscParameterBuffer misc; |
162 | VAEncMiscParameterBufferQualityLevel quality; |
163 | } quality_params; |
164 | #endif |
165 | } VAAPIEncodeH264Context; |
166 | |
167 | typedef struct VAAPIEncodeH264Options { |
168 | int qp; |
169 | int quality; |
170 | int low_power; |
171 | } VAAPIEncodeH264Options; |
172 | |
173 | |
174 | #define vseq_var(name) vseq->name, name |
175 | #define vseq_field(name) vseq->seq_fields.bits.name, name |
176 | #define vvui_field(name) vseq->vui_fields.bits.name, name |
177 | #define vpic_var(name) vpic->name, name |
178 | #define vpic_field(name) vpic->pic_fields.bits.name, name |
179 | #define vslice_var(name) vslice->name, name |
180 | #define vslice_field(name) vslice->slice_fields.bits.name, name |
181 | #define mseq_var(name) mseq->name, name |
182 | #define mslice_var(name) mslice->name, name |
183 | |
184 | static void vaapi_encode_h264_write_nal_header(PutBitContext *pbc, |
185 | int nal_unit_type, int nal_ref_idc) |
186 | { |
187 | u(1, 0, forbidden_zero_bit); |
188 | u(2, nal_ref_idc, nal_ref_idc); |
189 | u(5, nal_unit_type, nal_unit_type); |
190 | } |
191 | |
192 | static void vaapi_encode_h264_write_trailing_rbsp(PutBitContext *pbc) |
193 | { |
194 | u(1, 1, rbsp_stop_one_bit); |
195 | while (put_bits_count(pbc) & 7) |
196 | u(1, 0, rbsp_alignment_zero_bit); |
197 | } |
198 | |
199 | static void vaapi_encode_h264_write_vui(PutBitContext *pbc, |
200 | VAAPIEncodeContext *ctx) |
201 | { |
202 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
203 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
204 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
205 | int i; |
206 | |
207 | u(1, vvui_field(aspect_ratio_info_present_flag)); |
208 | if (vseq->vui_fields.bits.aspect_ratio_info_present_flag) { |
209 | u(8, vseq_var(aspect_ratio_idc)); |
210 | if (vseq->aspect_ratio_idc == 255) { |
211 | u(16, vseq_var(sar_width)); |
212 | u(16, vseq_var(sar_height)); |
213 | } |
214 | } |
215 | |
216 | u(1, mseq_var(overscan_info_present_flag)); |
217 | if (mseq->overscan_info_present_flag) |
218 | u(1, mseq_var(overscan_appropriate_flag)); |
219 | |
220 | u(1, mseq_var(video_signal_type_present_flag)); |
221 | if (mseq->video_signal_type_present_flag) { |
222 | u(3, mseq_var(video_format)); |
223 | u(1, mseq_var(video_full_range_flag)); |
224 | u(1, mseq_var(colour_description_present_flag)); |
225 | if (mseq->colour_description_present_flag) { |
226 | u(8, mseq_var(colour_primaries)); |
227 | u(8, mseq_var(transfer_characteristics)); |
228 | u(8, mseq_var(matrix_coefficients)); |
229 | } |
230 | } |
231 | |
232 | u(1, mseq_var(chroma_loc_info_present_flag)); |
233 | if (mseq->chroma_loc_info_present_flag) { |
234 | ue(mseq_var(chroma_sample_loc_type_top_field)); |
235 | ue(mseq_var(chroma_sample_loc_type_bottom_field)); |
236 | } |
237 | |
238 | u(1, vvui_field(timing_info_present_flag)); |
239 | if (vseq->vui_fields.bits.timing_info_present_flag) { |
240 | u(32, vseq_var(num_units_in_tick)); |
241 | u(32, vseq_var(time_scale)); |
242 | u(1, mseq_var(fixed_frame_rate_flag)); |
243 | } |
244 | |
245 | u(1, mseq_var(nal_hrd_parameters_present_flag)); |
246 | if (mseq->nal_hrd_parameters_present_flag) { |
247 | ue(mseq_var(cpb_cnt_minus1)); |
248 | u(4, mseq_var(bit_rate_scale)); |
249 | u(4, mseq_var(cpb_size_scale)); |
250 | for (i = 0; i <= mseq->cpb_cnt_minus1; i++) { |
251 | ue(mseq_var(bit_rate_value_minus1[i])); |
252 | ue(mseq_var(cpb_size_value_minus1[i])); |
253 | u(1, mseq_var(cbr_flag[i])); |
254 | } |
255 | u(5, mseq_var(initial_cpb_removal_delay_length_minus1)); |
256 | u(5, mseq_var(cpb_removal_delay_length_minus1)); |
257 | u(5, mseq_var(dpb_output_delay_length_minus1)); |
258 | u(5, mseq_var(time_offset_length)); |
259 | } |
260 | u(1, mseq_var(vcl_hrd_parameters_present_flag)); |
261 | if (mseq->vcl_hrd_parameters_present_flag) { |
262 | av_assert0(0 && "vcl hrd parameters not supported"); |
263 | } |
264 | |
265 | if (mseq->nal_hrd_parameters_present_flag || |
266 | mseq->vcl_hrd_parameters_present_flag) |
267 | u(1, mseq_var(low_delay_hrd_flag)); |
268 | u(1, mseq_var(pic_struct_present_flag)); |
269 | |
270 | u(1, vvui_field(bitstream_restriction_flag)); |
271 | if (vseq->vui_fields.bits.bitstream_restriction_flag) { |
272 | u(1, mseq_var(motion_vectors_over_pic_boundaries_flag)); |
273 | ue(mseq_var(max_bytes_per_pic_denom)); |
274 | ue(mseq_var(max_bits_per_mb_denom)); |
275 | ue(vvui_field(log2_max_mv_length_horizontal)); |
276 | ue(vvui_field(log2_max_mv_length_vertical)); |
277 | ue(mseq_var(max_num_reorder_frames)); |
278 | ue(mseq_var(max_dec_pic_buffering)); |
279 | } |
280 | } |
281 | |
282 | static void vaapi_encode_h264_write_sps(PutBitContext *pbc, |
283 | VAAPIEncodeContext *ctx) |
284 | { |
285 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
286 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
287 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
288 | int i; |
289 | |
290 | vaapi_encode_h264_write_nal_header(pbc, H264_NAL_SPS, 3); |
291 | |
292 | u(8, mseq_var(profile_idc)); |
293 | u(1, mseq_var(constraint_set0_flag)); |
294 | u(1, mseq_var(constraint_set1_flag)); |
295 | u(1, mseq_var(constraint_set2_flag)); |
296 | u(1, mseq_var(constraint_set3_flag)); |
297 | u(1, mseq_var(constraint_set4_flag)); |
298 | u(1, mseq_var(constraint_set5_flag)); |
299 | u(2, 0, reserved_zero_2bits); |
300 | |
301 | u(8, vseq_var(level_idc)); |
302 | |
303 | ue(vseq_var(seq_parameter_set_id)); |
304 | |
305 | if (mseq->profile_idc == 100 || mseq->profile_idc == 110 || |
306 | mseq->profile_idc == 122 || mseq->profile_idc == 244 || |
307 | mseq->profile_idc == 44 || mseq->profile_idc == 83 || |
308 | mseq->profile_idc == 86 || mseq->profile_idc == 118 || |
309 | mseq->profile_idc == 128 || mseq->profile_idc == 138) { |
310 | ue(vseq_field(chroma_format_idc)); |
311 | |
312 | if (vseq->seq_fields.bits.chroma_format_idc == 3) |
313 | u(1, mseq_var(separate_colour_plane_flag)); |
314 | |
315 | ue(vseq_var(bit_depth_luma_minus8)); |
316 | ue(vseq_var(bit_depth_chroma_minus8)); |
317 | |
318 | u(1, mseq_var(qpprime_y_zero_transform_bypass_flag)); |
319 | |
320 | u(1, vseq_field(seq_scaling_matrix_present_flag)); |
321 | if (vseq->seq_fields.bits.seq_scaling_matrix_present_flag) { |
322 | av_assert0(0 && "scaling matrices not supported"); |
323 | } |
324 | } |
325 | |
326 | ue(vseq_field(log2_max_frame_num_minus4)); |
327 | ue(vseq_field(pic_order_cnt_type)); |
328 | |
329 | if (vseq->seq_fields.bits.pic_order_cnt_type == 0) { |
330 | ue(vseq_field(log2_max_pic_order_cnt_lsb_minus4)); |
331 | } else if (vseq->seq_fields.bits.pic_order_cnt_type == 1) { |
332 | u(1, mseq_var(delta_pic_order_always_zero_flag)); |
333 | se(vseq_var(offset_for_non_ref_pic)); |
334 | se(vseq_var(offset_for_top_to_bottom_field)); |
335 | ue(vseq_var(num_ref_frames_in_pic_order_cnt_cycle)); |
336 | |
337 | for (i = 0; i < vseq->num_ref_frames_in_pic_order_cnt_cycle; i++) |
338 | se(vseq_var(offset_for_ref_frame[i])); |
339 | } |
340 | |
341 | ue(vseq_var(max_num_ref_frames)); |
342 | u(1, mseq_var(gaps_in_frame_num_allowed_flag)); |
343 | |
344 | ue(vseq->picture_width_in_mbs - 1, pic_width_in_mbs_minus1); |
345 | ue(vseq->picture_height_in_mbs - 1, pic_height_in_mbs_minus1); |
346 | |
347 | u(1, vseq_field(frame_mbs_only_flag)); |
348 | if (!vseq->seq_fields.bits.frame_mbs_only_flag) |
349 | u(1, vseq_field(mb_adaptive_frame_field_flag)); |
350 | |
351 | u(1, vseq_field(direct_8x8_inference_flag)); |
352 | |
353 | u(1, vseq_var(frame_cropping_flag)); |
354 | if (vseq->frame_cropping_flag) { |
355 | ue(vseq_var(frame_crop_left_offset)); |
356 | ue(vseq_var(frame_crop_right_offset)); |
357 | ue(vseq_var(frame_crop_top_offset)); |
358 | ue(vseq_var(frame_crop_bottom_offset)); |
359 | } |
360 | |
361 | u(1, vseq_var(vui_parameters_present_flag)); |
362 | if (vseq->vui_parameters_present_flag) |
363 | vaapi_encode_h264_write_vui(pbc, ctx); |
364 | |
365 | vaapi_encode_h264_write_trailing_rbsp(pbc); |
366 | } |
367 | |
368 | static void vaapi_encode_h264_write_pps(PutBitContext *pbc, |
369 | VAAPIEncodeContext *ctx) |
370 | { |
371 | VAEncPictureParameterBufferH264 *vpic = ctx->codec_picture_params; |
372 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
373 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
374 | |
375 | vaapi_encode_h264_write_nal_header(pbc, H264_NAL_PPS, 3); |
376 | |
377 | ue(vpic_var(pic_parameter_set_id)); |
378 | ue(vpic_var(seq_parameter_set_id)); |
379 | |
380 | u(1, vpic_field(entropy_coding_mode_flag)); |
381 | u(1, mseq_var(bottom_field_pic_order_in_frame_present_flag)); |
382 | |
383 | ue(mseq_var(num_slice_groups_minus1)); |
384 | if (mseq->num_slice_groups_minus1 > 0) { |
385 | ue(mseq_var(slice_group_map_type)); |
386 | av_assert0(0 && "slice groups not supported"); |
387 | } |
388 | |
389 | ue(vpic_var(num_ref_idx_l0_active_minus1)); |
390 | ue(vpic_var(num_ref_idx_l1_active_minus1)); |
391 | |
392 | u(1, vpic_field(weighted_pred_flag)); |
393 | u(2, vpic_field(weighted_bipred_idc)); |
394 | |
395 | se(vpic->pic_init_qp - 26, pic_init_qp_minus26); |
396 | se(mseq_var(pic_init_qs_minus26)); |
397 | se(vpic_var(chroma_qp_index_offset)); |
398 | |
399 | u(1, vpic_field(deblocking_filter_control_present_flag)); |
400 | u(1, vpic_field(constrained_intra_pred_flag)); |
401 | u(1, vpic_field(redundant_pic_cnt_present_flag)); |
402 | u(1, vpic_field(transform_8x8_mode_flag)); |
403 | |
404 | u(1, vpic_field(pic_scaling_matrix_present_flag)); |
405 | if (vpic->pic_fields.bits.pic_scaling_matrix_present_flag) { |
406 | av_assert0(0 && "scaling matrices not supported"); |
407 | } |
408 | |
409 | se(vpic_var(second_chroma_qp_index_offset)); |
410 | |
411 | vaapi_encode_h264_write_trailing_rbsp(pbc); |
412 | } |
413 | |
414 | static void vaapi_encode_h264_write_slice_header2(PutBitContext *pbc, |
415 | VAAPIEncodeContext *ctx, |
416 | VAAPIEncodePicture *pic, |
417 | VAAPIEncodeSlice *slice) |
418 | { |
419 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
420 | VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; |
421 | VAEncSliceParameterBufferH264 *vslice = slice->codec_slice_params; |
422 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
423 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
424 | VAAPIEncodeH264Slice *pslice = slice->priv_data; |
425 | VAAPIEncodeH264MiscSliceParams *mslice = &pslice->misc_slice_params; |
426 | |
427 | vaapi_encode_h264_write_nal_header(pbc, mslice->nal_unit_type, |
428 | mslice->nal_ref_idc); |
429 | |
430 | ue(vslice->macroblock_address, first_mb_in_slice); |
431 | ue(vslice_var(slice_type)); |
432 | ue(vpic_var(pic_parameter_set_id)); |
433 | |
434 | if (mseq->separate_colour_plane_flag) { |
435 | u(2, mslice_var(colour_plane_id)); |
436 | } |
437 | |
438 | u(4 + vseq->seq_fields.bits.log2_max_frame_num_minus4, |
439 | (vpic->frame_num & |
440 | ((1 << (4 + vseq->seq_fields.bits.log2_max_frame_num_minus4)) - 1)), |
441 | frame_num); |
442 | |
443 | if (!vseq->seq_fields.bits.frame_mbs_only_flag) { |
444 | u(1, mslice_var(field_pic_flag)); |
445 | if (mslice->field_pic_flag) |
446 | u(1, mslice_var(bottom_field_flag)); |
447 | } |
448 | |
449 | if (vpic->pic_fields.bits.idr_pic_flag) { |
450 | ue(vslice_var(idr_pic_id)); |
451 | } |
452 | |
453 | if (vseq->seq_fields.bits.pic_order_cnt_type == 0) { |
454 | u(4 + vseq->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4, |
455 | vslice_var(pic_order_cnt_lsb)); |
456 | if (mseq->bottom_field_pic_order_in_frame_present_flag && |
457 | !mslice->field_pic_flag) { |
458 | se(vslice_var(delta_pic_order_cnt_bottom)); |
459 | } |
460 | } |
461 | |
462 | if (vseq->seq_fields.bits.pic_order_cnt_type == 1 && |
463 | !vseq->seq_fields.bits.delta_pic_order_always_zero_flag) { |
464 | se(vslice_var(delta_pic_order_cnt[0])); |
465 | if (mseq->bottom_field_pic_order_in_frame_present_flag && |
466 | !mslice->field_pic_flag) { |
467 | se(vslice_var(delta_pic_order_cnt[1])); |
468 | } |
469 | } |
470 | |
471 | if (vpic->pic_fields.bits.redundant_pic_cnt_present_flag) { |
472 | ue(mslice_var(redundant_pic_cnt)); |
473 | } |
474 | |
475 | if (vslice->slice_type == SLICE_TYPE_B) { |
476 | u(1, vslice_var(direct_spatial_mv_pred_flag)); |
477 | } |
478 | |
479 | if (vslice->slice_type == SLICE_TYPE_P || |
480 | vslice->slice_type == SLICE_TYPE_SP || |
481 | vslice->slice_type == SLICE_TYPE_B) { |
482 | u(1, vslice_var(num_ref_idx_active_override_flag)); |
483 | if (vslice->num_ref_idx_active_override_flag) { |
484 | ue(vslice_var(num_ref_idx_l0_active_minus1)); |
485 | if (vslice->slice_type == SLICE_TYPE_B) |
486 | ue(vslice_var(num_ref_idx_l1_active_minus1)); |
487 | } |
488 | } |
489 | |
490 | if (mslice->nal_unit_type == 20 || mslice->nal_unit_type == 21) { |
491 | av_assert0(0 && "no MVC support"); |
492 | } else { |
493 | if (vslice->slice_type % 5 != 2 && vslice->slice_type % 5 != 4) { |
494 | u(1, mslice_var(ref_pic_list_modification_flag_l0)); |
495 | if (mslice->ref_pic_list_modification_flag_l0) { |
496 | av_assert0(0 && "ref pic list modification"); |
497 | } |
498 | } |
499 | if (vslice->slice_type % 5 == 1) { |
500 | u(1, mslice_var(ref_pic_list_modification_flag_l1)); |
501 | if (mslice->ref_pic_list_modification_flag_l1) { |
502 | av_assert0(0 && "ref pic list modification"); |
503 | } |
504 | } |
505 | } |
506 | |
507 | if ((vpic->pic_fields.bits.weighted_pred_flag && |
508 | (vslice->slice_type == SLICE_TYPE_P || |
509 | vslice->slice_type == SLICE_TYPE_SP)) || |
510 | (vpic->pic_fields.bits.weighted_bipred_idc == 1 && |
511 | vslice->slice_type == SLICE_TYPE_B)) { |
512 | av_assert0(0 && "prediction weights not supported"); |
513 | } |
514 | |
515 | av_assert0(mslice->nal_ref_idc > 0 == |
516 | vpic->pic_fields.bits.reference_pic_flag); |
517 | if (mslice->nal_ref_idc != 0) { |
518 | if (vpic->pic_fields.bits.idr_pic_flag) { |
519 | u(1, mslice_var(no_output_of_prior_pics_flag)); |
520 | u(1, mslice_var(long_term_reference_flag)); |
521 | } else { |
522 | u(1, mslice_var(adaptive_ref_pic_marking_mode_flag)); |
523 | if (mslice->adaptive_ref_pic_marking_mode_flag) { |
524 | av_assert0(0 && "MMCOs not supported"); |
525 | } |
526 | } |
527 | } |
528 | |
529 | if (vpic->pic_fields.bits.entropy_coding_mode_flag && |
530 | vslice->slice_type != SLICE_TYPE_I && |
531 | vslice->slice_type != SLICE_TYPE_SI) { |
532 | ue(vslice_var(cabac_init_idc)); |
533 | } |
534 | |
535 | se(vslice_var(slice_qp_delta)); |
536 | if (vslice->slice_type == SLICE_TYPE_SP || |
537 | vslice->slice_type == SLICE_TYPE_SI) { |
538 | if (vslice->slice_type == SLICE_TYPE_SP) |
539 | u(1, mslice_var(sp_for_switch_flag)); |
540 | se(mslice_var(slice_qs_delta)); |
541 | } |
542 | |
543 | if (vpic->pic_fields.bits.deblocking_filter_control_present_flag) { |
544 | ue(vslice_var(disable_deblocking_filter_idc)); |
545 | if (vslice->disable_deblocking_filter_idc != 1) { |
546 | se(vslice_var(slice_alpha_c0_offset_div2)); |
547 | se(vslice_var(slice_beta_offset_div2)); |
548 | } |
549 | } |
550 | |
551 | if (mseq->num_slice_groups_minus1 > 0 && |
552 | mseq->slice_group_map_type >= 3 && mseq->slice_group_map_type <= 5) { |
553 | av_assert0(0 && "slice groups not supported"); |
554 | } |
555 | |
556 | // No alignment - this need not be a byte boundary. |
557 | } |
558 | |
559 | static void vaapi_encode_h264_write_buffering_period(PutBitContext *pbc, |
560 | VAAPIEncodeContext *ctx, |
561 | VAAPIEncodePicture *pic) |
562 | { |
563 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
564 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
565 | VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; |
566 | int i; |
567 | |
568 | ue(vpic_var(seq_parameter_set_id)); |
569 | |
570 | if (mseq->nal_hrd_parameters_present_flag) { |
571 | for (i = 0; i <= mseq->cpb_cnt_minus1; i++) { |
572 | u(mseq->initial_cpb_removal_delay_length_minus1 + 1, |
573 | mseq_var(initial_cpb_removal_delay)); |
574 | u(mseq->initial_cpb_removal_delay_length_minus1 + 1, |
575 | mseq_var(initial_cpb_removal_delay_offset)); |
576 | } |
577 | } |
578 | if (mseq->vcl_hrd_parameters_present_flag) { |
579 | av_assert0(0 && "vcl hrd parameters not supported"); |
580 | } |
581 | } |
582 | |
583 | static void vaapi_encode_h264_write_pic_timing(PutBitContext *pbc, |
584 | VAAPIEncodeContext *ctx, |
585 | VAAPIEncodePicture *pic) |
586 | { |
587 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
588 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
589 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
590 | int i, num_clock_ts; |
591 | |
592 | if (mseq->nal_hrd_parameters_present_flag || |
593 | mseq->vcl_hrd_parameters_present_flag) { |
594 | u(mseq->cpb_removal_delay_length_minus1 + 1, |
595 | 2 * vseq->num_units_in_tick * priv->cpb_delay, |
596 | cpb_removal_delay); |
597 | u(mseq->dpb_output_delay_length_minus1 + 1, |
598 | 2 * vseq->num_units_in_tick * priv->dpb_delay, |
599 | dpb_output_delay); |
600 | } |
601 | if (mseq->pic_struct_present_flag) { |
602 | u(4, mseq_var(pic_struct)); |
603 | num_clock_ts = (mseq->pic_struct <= 2 ? 1 : |
604 | mseq->pic_struct <= 4 ? 2 : |
605 | mseq->pic_struct <= 8 ? 3 : 0); |
606 | for (i = 0; i < num_clock_ts; i++) { |
607 | u(1, 0, clock_timestamp_flag[i]); |
608 | // No full timestamp information. |
609 | } |
610 | } |
611 | } |
612 | |
613 | static void vaapi_encode_h264_write_identifier(PutBitContext *pbc, |
614 | VAAPIEncodeContext *ctx, |
615 | VAAPIEncodePicture *pic) |
616 | { |
617 | const char *lavc = LIBAVCODEC_IDENT; |
618 | const char *vaapi = VA_VERSION_S; |
619 | const char *driver = vaQueryVendorString(ctx->hwctx->display); |
620 | char tmp[256]; |
621 | int i; |
622 | |
623 | // Random (version 4) ISO 11578 UUID. |
624 | uint8_t uuid[16] = { |
625 | 0x59, 0x94, 0x8b, 0x28, 0x11, 0xec, 0x45, 0xaf, |
626 | 0x96, 0x75, 0x19, 0xd4, 0x1f, 0xea, 0xa9, 0x4d, |
627 | }; |
628 | |
629 | for (i = 0; i < 16; i++) |
630 | u(8, uuid[i], uuid_iso_iec_11578); |
631 | |
632 | snprintf(tmp, sizeof(tmp), "%s / VAAPI %s / %s", lavc, vaapi, driver); |
633 | for (i = 0; i < sizeof(tmp) && tmp[i]; i++) |
634 | u(8, tmp[i], user_data_payload_byte); |
635 | } |
636 | |
637 | static void vaapi_encode_h264_write_sei(PutBitContext *pbc, |
638 | VAAPIEncodeContext *ctx, |
639 | VAAPIEncodePicture *pic) |
640 | { |
641 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
642 | PutBitContext payload_bits; |
643 | char payload[256]; |
644 | int payload_type, payload_size, i; |
645 | void (*write_payload)(PutBitContext *pbc, |
646 | VAAPIEncodeContext *ctx, |
647 | VAAPIEncodePicture *pic) = NULL; |
648 | |
649 | vaapi_encode_h264_write_nal_header(pbc, H264_NAL_SEI, 0); |
650 | |
651 | for (payload_type = 0; payload_type < 64; payload_type++) { |
652 | switch (payload_type) { |
653 | case SEI_TYPE_BUFFERING_PERIOD: |
654 | if (!priv->send_timing_sei || |
655 | pic->type != PICTURE_TYPE_IDR) |
656 | continue; |
657 | write_payload = &vaapi_encode_h264_write_buffering_period; |
658 | break; |
659 | case SEI_TYPE_PIC_TIMING: |
660 | if (!priv->send_timing_sei) |
661 | continue; |
662 | write_payload = &vaapi_encode_h264_write_pic_timing; |
663 | break; |
664 | case SEI_TYPE_USER_DATA_UNREGISTERED: |
665 | if (pic->encode_order != 0) |
666 | continue; |
667 | write_payload = &vaapi_encode_h264_write_identifier; |
668 | break; |
669 | default: |
670 | continue; |
671 | } |
672 | |
673 | init_put_bits(&payload_bits, payload, sizeof(payload)); |
674 | write_payload(&payload_bits, ctx, pic); |
675 | if (put_bits_count(&payload_bits) & 7) { |
676 | write_u(&payload_bits, 1, 1, bit_equal_to_one); |
677 | while (put_bits_count(&payload_bits) & 7) |
678 | write_u(&payload_bits, 1, 0, bit_equal_to_zero); |
679 | } |
680 | payload_size = put_bits_count(&payload_bits) / 8; |
681 | flush_put_bits(&payload_bits); |
682 | |
683 | u(8, payload_type, last_payload_type_byte); |
684 | u(8, payload_size, last_payload_size_byte); |
685 | for (i = 0; i < payload_size; i++) |
686 | u(8, payload[i] & 0xff, sei_payload); |
687 | } |
688 | |
689 | vaapi_encode_h264_write_trailing_rbsp(pbc); |
690 | } |
691 | |
692 | static int vaapi_encode_h264_write_sequence_header(AVCodecContext *avctx, |
693 | char *data, size_t *data_len) |
694 | { |
695 | VAAPIEncodeContext *ctx = avctx->priv_data; |
696 | PutBitContext pbc; |
697 | char tmp[256]; |
698 | int err; |
699 | size_t nal_len, bit_len, bit_pos, next_len; |
700 | |
701 | bit_len = *data_len; |
702 | bit_pos = 0; |
703 | |
704 | init_put_bits(&pbc, tmp, sizeof(tmp)); |
705 | vaapi_encode_h264_write_sps(&pbc, ctx); |
706 | nal_len = put_bits_count(&pbc); |
707 | flush_put_bits(&pbc); |
708 | |
709 | next_len = bit_len - bit_pos; |
710 | err = ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data + bit_pos / 8, |
711 | &next_len, |
712 | tmp, nal_len); |
713 | if (err < 0) |
714 | return err; |
715 | bit_pos += next_len; |
716 | |
717 | init_put_bits(&pbc, tmp, sizeof(tmp)); |
718 | vaapi_encode_h264_write_pps(&pbc, ctx); |
719 | nal_len = put_bits_count(&pbc); |
720 | flush_put_bits(&pbc); |
721 | |
722 | next_len = bit_len - bit_pos; |
723 | err = ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data + bit_pos / 8, |
724 | &next_len, |
725 | tmp, nal_len); |
726 | if (err < 0) |
727 | return err; |
728 | bit_pos += next_len; |
729 | |
730 | *data_len = bit_pos; |
731 | return 0; |
732 | } |
733 | |
734 | static int vaapi_encode_h264_write_slice_header(AVCodecContext *avctx, |
735 | VAAPIEncodePicture *pic, |
736 | VAAPIEncodeSlice *slice, |
737 | char *data, size_t *data_len) |
738 | { |
739 | VAAPIEncodeContext *ctx = avctx->priv_data; |
740 | PutBitContext pbc; |
741 | char tmp[256]; |
742 | size_t header_len; |
743 | |
744 | init_put_bits(&pbc, tmp, sizeof(tmp)); |
745 | vaapi_encode_h264_write_slice_header2(&pbc, ctx, pic, slice); |
746 | header_len = put_bits_count(&pbc); |
747 | flush_put_bits(&pbc); |
748 | |
749 | return ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data, data_len, |
750 | tmp, header_len); |
751 | } |
752 | |
753 | static int vaapi_encode_h264_write_extra_header(AVCodecContext *avctx, |
754 | VAAPIEncodePicture *pic, |
755 | int index, int *type, |
756 | char *data, size_t *data_len) |
757 | { |
758 | VAAPIEncodeContext *ctx = avctx->priv_data; |
759 | PutBitContext pbc; |
760 | char tmp[256]; |
761 | size_t header_len; |
762 | |
763 | if (index == 0 && ctx->va_rc_mode == VA_RC_CBR) { |
764 | *type = VAEncPackedHeaderH264_SEI; |
765 | |
766 | init_put_bits(&pbc, tmp, sizeof(tmp)); |
767 | vaapi_encode_h264_write_sei(&pbc, ctx, pic); |
768 | header_len = put_bits_count(&pbc); |
769 | flush_put_bits(&pbc); |
770 | |
771 | return ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data, data_len, |
772 | tmp, header_len); |
773 | |
774 | } else { |
775 | return AVERROR_EOF; |
776 | } |
777 | } |
778 | |
779 | static int vaapi_encode_h264_init_sequence_params(AVCodecContext *avctx) |
780 | { |
781 | VAAPIEncodeContext *ctx = avctx->priv_data; |
782 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
783 | VAEncPictureParameterBufferH264 *vpic = ctx->codec_picture_params; |
784 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
785 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; |
786 | int i; |
787 | |
788 | { |
789 | vseq->seq_parameter_set_id = 0; |
790 | |
791 | vseq->level_idc = avctx->level; |
792 | |
793 | vseq->max_num_ref_frames = 1 + (avctx->max_b_frames > 0); |
794 | |
795 | vseq->picture_width_in_mbs = priv->mb_width; |
796 | vseq->picture_height_in_mbs = priv->mb_height; |
797 | |
798 | vseq->seq_fields.bits.chroma_format_idc = 1; |
799 | vseq->seq_fields.bits.frame_mbs_only_flag = 1; |
800 | vseq->seq_fields.bits.direct_8x8_inference_flag = 1; |
801 | vseq->seq_fields.bits.log2_max_frame_num_minus4 = 4; |
802 | vseq->seq_fields.bits.pic_order_cnt_type = 0; |
803 | vseq->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4 = |
804 | av_clip(av_log2(avctx->max_b_frames + 1) - 2, 0, 12); |
805 | |
806 | if (avctx->width != ctx->surface_width || |
807 | avctx->height != ctx->surface_height) { |
808 | vseq->frame_cropping_flag = 1; |
809 | |
810 | vseq->frame_crop_left_offset = 0; |
811 | vseq->frame_crop_right_offset = |
812 | (ctx->surface_width - avctx->width) / 2; |
813 | vseq->frame_crop_top_offset = 0; |
814 | vseq->frame_crop_bottom_offset = |
815 | (ctx->surface_height - avctx->height) / 2; |
816 | } else { |
817 | vseq->frame_cropping_flag = 0; |
818 | } |
819 | |
820 | vseq->vui_parameters_present_flag = 1; |
821 | if (avctx->sample_aspect_ratio.num != 0) { |
822 | vseq->vui_fields.bits.aspect_ratio_info_present_flag = 1; |
823 | // There is a large enum of these which we could support |
824 | // individually rather than using the generic X/Y form? |
825 | if (avctx->sample_aspect_ratio.num == |
826 | avctx->sample_aspect_ratio.den) { |
827 | vseq->aspect_ratio_idc = 1; |
828 | } else { |
829 | vseq->aspect_ratio_idc = 255; // Extended SAR. |
830 | vseq->sar_width = avctx->sample_aspect_ratio.num; |
831 | vseq->sar_height = avctx->sample_aspect_ratio.den; |
832 | } |
833 | } |
834 | if (avctx->color_primaries != AVCOL_PRI_UNSPECIFIED || |
835 | avctx->color_trc != AVCOL_TRC_UNSPECIFIED || |
836 | avctx->colorspace != AVCOL_SPC_UNSPECIFIED) { |
837 | mseq->video_signal_type_present_flag = 1; |
838 | mseq->video_format = 5; // Unspecified. |
839 | mseq->video_full_range_flag = 0; |
840 | mseq->colour_description_present_flag = 1; |
841 | // These enums are derived from the standard and hence |
842 | // we can just use the values directly. |
843 | mseq->colour_primaries = avctx->color_primaries; |
844 | mseq->transfer_characteristics = avctx->color_trc; |
845 | mseq->matrix_coefficients = avctx->colorspace; |
846 | } |
847 | |
848 | vseq->vui_fields.bits.bitstream_restriction_flag = 1; |
849 | mseq->motion_vectors_over_pic_boundaries_flag = 1; |
850 | mseq->max_bytes_per_pic_denom = 0; |
851 | mseq->max_bits_per_mb_denom = 0; |
852 | vseq->vui_fields.bits.log2_max_mv_length_horizontal = 16; |
853 | vseq->vui_fields.bits.log2_max_mv_length_vertical = 16; |
854 | |
855 | mseq->max_num_reorder_frames = (avctx->max_b_frames > 0); |
856 | mseq->max_dec_pic_buffering = vseq->max_num_ref_frames; |
857 | |
858 | vseq->bits_per_second = avctx->bit_rate; |
859 | |
860 | vseq->vui_fields.bits.timing_info_present_flag = 1; |
861 | if (avctx->framerate.num > 0 && avctx->framerate.den > 0) { |
862 | vseq->num_units_in_tick = avctx->framerate.den; |
863 | vseq->time_scale = 2 * avctx->framerate.num; |
864 | mseq->fixed_frame_rate_flag = 1; |
865 | } else { |
866 | vseq->num_units_in_tick = avctx->time_base.num; |
867 | vseq->time_scale = 2 * avctx->time_base.den; |
868 | mseq->fixed_frame_rate_flag = 0; |
869 | } |
870 | |
871 | if (ctx->va_rc_mode == VA_RC_CBR) { |
872 | priv->send_timing_sei = 1; |
873 | mseq->nal_hrd_parameters_present_flag = 1; |
874 | |
875 | mseq->cpb_cnt_minus1 = 0; |
876 | |
877 | // Try to scale these to a sensible range so that the |
878 | // golomb encode of the value is not overlong. |
879 | mseq->bit_rate_scale = |
880 | av_clip_uintp2(av_log2(avctx->bit_rate) - 15 - 6, 4); |
881 | mseq->bit_rate_value_minus1[0] = |
882 | (avctx->bit_rate >> mseq->bit_rate_scale + 6) - 1; |
883 | |
884 | mseq->cpb_size_scale = |
885 | av_clip_uintp2(av_log2(ctx->hrd_params.hrd.buffer_size) - 15 - 4, 4); |
886 | mseq->cpb_size_value_minus1[0] = |
887 | (ctx->hrd_params.hrd.buffer_size >> mseq->cpb_size_scale + 4) - 1; |
888 | |
889 | // CBR mode isn't actually available here, despite naming. |
890 | mseq->cbr_flag[0] = 0; |
891 | |
892 | mseq->initial_cpb_removal_delay_length_minus1 = 23; |
893 | mseq->cpb_removal_delay_length_minus1 = 23; |
894 | mseq->dpb_output_delay_length_minus1 = 7; |
895 | mseq->time_offset_length = 0; |
896 | |
897 | // This calculation can easily overflow 32 bits. |
898 | mseq->initial_cpb_removal_delay = 90000 * |
899 | (uint64_t)ctx->hrd_params.hrd.initial_buffer_fullness / |
900 | ctx->hrd_params.hrd.buffer_size; |
901 | |
902 | mseq->initial_cpb_removal_delay_offset = 0; |
903 | } else { |
904 | priv->send_timing_sei = 0; |
905 | mseq->nal_hrd_parameters_present_flag = 0; |
906 | } |
907 | |
908 | vseq->intra_period = ctx->p_per_i * (ctx->b_per_p + 1); |
909 | vseq->intra_idr_period = vseq->intra_period; |
910 | vseq->ip_period = ctx->b_per_p + 1; |
911 | } |
912 | |
913 | { |
914 | vpic->CurrPic.picture_id = VA_INVALID_ID; |
915 | vpic->CurrPic.flags = VA_PICTURE_H264_INVALID; |
916 | |
917 | for (i = 0; i < FF_ARRAY_ELEMS(vpic->ReferenceFrames); i++) { |
918 | vpic->ReferenceFrames[i].picture_id = VA_INVALID_ID; |
919 | vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_INVALID; |
920 | } |
921 | |
922 | vpic->coded_buf = VA_INVALID_ID; |
923 | |
924 | vpic->pic_parameter_set_id = 0; |
925 | vpic->seq_parameter_set_id = 0; |
926 | |
927 | vpic->num_ref_idx_l0_active_minus1 = 0; |
928 | vpic->num_ref_idx_l1_active_minus1 = 0; |
929 | |
930 | vpic->pic_fields.bits.entropy_coding_mode_flag = |
931 | ((avctx->profile & 0xff) != 66); |
932 | vpic->pic_fields.bits.weighted_pred_flag = 0; |
933 | vpic->pic_fields.bits.weighted_bipred_idc = 0; |
934 | vpic->pic_fields.bits.transform_8x8_mode_flag = |
935 | ((avctx->profile & 0xff) >= 100); |
936 | |
937 | vpic->pic_init_qp = priv->fixed_qp_idr; |
938 | } |
939 | |
940 | { |
941 | mseq->profile_idc = avctx->profile & 0xff; |
942 | |
943 | if (avctx->profile & FF_PROFILE_H264_CONSTRAINED) |
944 | mseq->constraint_set1_flag = 1; |
945 | if (avctx->profile & FF_PROFILE_H264_INTRA) |
946 | mseq->constraint_set3_flag = 1; |
947 | } |
948 | |
949 | return 0; |
950 | } |
951 | |
952 | static int vaapi_encode_h264_init_picture_params(AVCodecContext *avctx, |
953 | VAAPIEncodePicture *pic) |
954 | { |
955 | VAAPIEncodeContext *ctx = avctx->priv_data; |
956 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
957 | VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; |
958 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
959 | int i; |
960 | |
961 | if (pic->type == PICTURE_TYPE_IDR) { |
962 | av_assert0(pic->display_order == pic->encode_order); |
963 | vpic->frame_num = 0; |
964 | priv->next_frame_num = 1; |
965 | priv->cpb_delay = 0; |
966 | priv->last_idr_frame = pic->display_order; |
967 | } else { |
968 | vpic->frame_num = priv->next_frame_num; |
969 | if (pic->type != PICTURE_TYPE_B) { |
970 | // nal_ref_idc != 0 |
971 | ++priv->next_frame_num; |
972 | } |
973 | ++priv->cpb_delay; |
974 | } |
975 | priv->dpb_delay = pic->display_order - pic->encode_order + 1; |
976 | |
977 | vpic->frame_num = vpic->frame_num & |
978 | ((1 << (4 + vseq->seq_fields.bits.log2_max_frame_num_minus4)) - 1); |
979 | |
980 | vpic->CurrPic.picture_id = pic->recon_surface; |
981 | vpic->CurrPic.frame_idx = vpic->frame_num; |
982 | vpic->CurrPic.flags = 0; |
983 | vpic->CurrPic.TopFieldOrderCnt = pic->display_order - priv->last_idr_frame; |
984 | vpic->CurrPic.BottomFieldOrderCnt = pic->display_order - priv->last_idr_frame; |
985 | |
986 | for (i = 0; i < pic->nb_refs; i++) { |
987 | VAAPIEncodePicture *ref = pic->refs[i]; |
988 | av_assert0(ref && ref->encode_order < pic->encode_order); |
989 | vpic->ReferenceFrames[i].picture_id = ref->recon_surface; |
990 | vpic->ReferenceFrames[i].frame_idx = ref->encode_order; |
991 | vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_SHORT_TERM_REFERENCE; |
992 | vpic->ReferenceFrames[i].TopFieldOrderCnt = ref->display_order - priv->last_idr_frame; |
993 | vpic->ReferenceFrames[i].BottomFieldOrderCnt = ref->display_order - priv->last_idr_frame; |
994 | } |
995 | for (; i < FF_ARRAY_ELEMS(vpic->ReferenceFrames); i++) { |
996 | vpic->ReferenceFrames[i].picture_id = VA_INVALID_ID; |
997 | vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_INVALID; |
998 | } |
999 | |
1000 | vpic->coded_buf = pic->output_buffer; |
1001 | |
1002 | vpic->pic_fields.bits.idr_pic_flag = (pic->type == PICTURE_TYPE_IDR); |
1003 | vpic->pic_fields.bits.reference_pic_flag = (pic->type != PICTURE_TYPE_B); |
1004 | |
1005 | pic->nb_slices = 1; |
1006 | |
1007 | return 0; |
1008 | } |
1009 | |
1010 | static int vaapi_encode_h264_init_slice_params(AVCodecContext *avctx, |
1011 | VAAPIEncodePicture *pic, |
1012 | VAAPIEncodeSlice *slice) |
1013 | { |
1014 | VAAPIEncodeContext *ctx = avctx->priv_data; |
1015 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; |
1016 | VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; |
1017 | VAEncSliceParameterBufferH264 *vslice = slice->codec_slice_params; |
1018 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
1019 | VAAPIEncodeH264Slice *pslice; |
1020 | VAAPIEncodeH264MiscSliceParams *mslice; |
1021 | int i; |
1022 | |
1023 | slice->priv_data = av_mallocz(sizeof(*pslice)); |
1024 | if (!slice->priv_data) |
1025 | return AVERROR(ENOMEM); |
1026 | pslice = slice->priv_data; |
1027 | mslice = &pslice->misc_slice_params; |
1028 | |
1029 | if (pic->type == PICTURE_TYPE_IDR) |
1030 | mslice->nal_unit_type = H264_NAL_IDR_SLICE; |
1031 | else |
1032 | mslice->nal_unit_type = H264_NAL_SLICE; |
1033 | |
1034 | switch (pic->type) { |
1035 | case PICTURE_TYPE_IDR: |
1036 | vslice->slice_type = SLICE_TYPE_I; |
1037 | mslice->nal_ref_idc = 3; |
1038 | break; |
1039 | case PICTURE_TYPE_I: |
1040 | vslice->slice_type = SLICE_TYPE_I; |
1041 | mslice->nal_ref_idc = 2; |
1042 | break; |
1043 | case PICTURE_TYPE_P: |
1044 | vslice->slice_type = SLICE_TYPE_P; |
1045 | mslice->nal_ref_idc = 1; |
1046 | break; |
1047 | case PICTURE_TYPE_B: |
1048 | vslice->slice_type = SLICE_TYPE_B; |
1049 | mslice->nal_ref_idc = 0; |
1050 | break; |
1051 | default: |
1052 | av_assert0(0 && "invalid picture type"); |
1053 | } |
1054 | |
1055 | // Only one slice per frame. |
1056 | vslice->macroblock_address = 0; |
1057 | vslice->num_macroblocks = priv->mb_width * priv->mb_height; |
1058 | |
1059 | vslice->macroblock_info = VA_INVALID_ID; |
1060 | |
1061 | vslice->pic_parameter_set_id = vpic->pic_parameter_set_id; |
1062 | vslice->idr_pic_id = priv->idr_pic_count++; |
1063 | |
1064 | vslice->pic_order_cnt_lsb = (pic->display_order - priv->last_idr_frame) & |
1065 | ((1 << (4 + vseq->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4)) - 1); |
1066 | |
1067 | for (i = 0; i < FF_ARRAY_ELEMS(vslice->RefPicList0); i++) { |
1068 | vslice->RefPicList0[i].picture_id = VA_INVALID_ID; |
1069 | vslice->RefPicList0[i].flags = VA_PICTURE_H264_INVALID; |
1070 | vslice->RefPicList1[i].picture_id = VA_INVALID_ID; |
1071 | vslice->RefPicList1[i].flags = VA_PICTURE_H264_INVALID; |
1072 | } |
1073 | |
1074 | av_assert0(pic->nb_refs <= 2); |
1075 | if (pic->nb_refs >= 1) { |
1076 | // Backward reference for P- or B-frame. |
1077 | av_assert0(pic->type == PICTURE_TYPE_P || |
1078 | pic->type == PICTURE_TYPE_B); |
1079 | |
1080 | vslice->num_ref_idx_l0_active_minus1 = 0; |
1081 | vslice->RefPicList0[0] = vpic->ReferenceFrames[0]; |
1082 | } |
1083 | if (pic->nb_refs >= 2) { |
1084 | // Forward reference for B-frame. |
1085 | av_assert0(pic->type == PICTURE_TYPE_B); |
1086 | |
1087 | vslice->num_ref_idx_l1_active_minus1 = 0; |
1088 | vslice->RefPicList1[0] = vpic->ReferenceFrames[1]; |
1089 | } |
1090 | |
1091 | if (pic->type == PICTURE_TYPE_B) |
1092 | vslice->slice_qp_delta = priv->fixed_qp_b - vpic->pic_init_qp; |
1093 | else if (pic->type == PICTURE_TYPE_P) |
1094 | vslice->slice_qp_delta = priv->fixed_qp_p - vpic->pic_init_qp; |
1095 | else |
1096 | vslice->slice_qp_delta = priv->fixed_qp_idr - vpic->pic_init_qp; |
1097 | |
1098 | vslice->direct_spatial_mv_pred_flag = 1; |
1099 | |
1100 | return 0; |
1101 | } |
1102 | |
1103 | static av_cold int vaapi_encode_h264_configure(AVCodecContext *avctx) |
1104 | { |
1105 | VAAPIEncodeContext *ctx = avctx->priv_data; |
1106 | VAAPIEncodeH264Context *priv = ctx->priv_data; |
1107 | VAAPIEncodeH264Options *opt = ctx->codec_options; |
1108 | |
1109 | priv->mb_width = FFALIGN(avctx->width, 16) / 16; |
1110 | priv->mb_height = FFALIGN(avctx->height, 16) / 16; |
1111 | |
1112 | if (ctx->va_rc_mode == VA_RC_CQP) { |
1113 | priv->fixed_qp_p = opt->qp; |
1114 | if (avctx->i_quant_factor > 0.0) |
1115 | priv->fixed_qp_idr = (int)((priv->fixed_qp_p * avctx->i_quant_factor + |
1116 | avctx->i_quant_offset) + 0.5); |
1117 | else |
1118 | priv->fixed_qp_idr = priv->fixed_qp_p; |
1119 | if (avctx->b_quant_factor > 0.0) |
1120 | priv->fixed_qp_b = (int)((priv->fixed_qp_p * avctx->b_quant_factor + |
1121 | avctx->b_quant_offset) + 0.5); |
1122 | else |
1123 | priv->fixed_qp_b = priv->fixed_qp_p; |
1124 | |
1125 | av_log(avctx, AV_LOG_DEBUG, "Using fixed QP = " |
1126 | "%d / %d / %d for IDR- / P- / B-frames.\n", |
1127 | priv->fixed_qp_idr, priv->fixed_qp_p, priv->fixed_qp_b); |
1128 | |
1129 | } else if (ctx->va_rc_mode == VA_RC_CBR || |
1130 | ctx->va_rc_mode == VA_RC_VBR) { |
1131 | // These still need to be set for pic_init_qp/slice_qp_delta. |
1132 | priv->fixed_qp_idr = 26; |
1133 | priv->fixed_qp_p = 26; |
1134 | priv->fixed_qp_b = 26; |
1135 | |
1136 | av_log(avctx, AV_LOG_DEBUG, "Using %s-bitrate = %"PRId64" bps.\n", |
1137 | ctx->va_rc_mode == VA_RC_CBR ? "constant" : "variable", |
1138 | avctx->bit_rate); |
1139 | |
1140 | } else { |
1141 | av_assert0(0 && "Invalid RC mode."); |
1142 | } |
1143 | |
1144 | if (opt->quality > 0) { |
1145 | #if VA_CHECK_VERSION(0, 36, 0) |
1146 | priv->quality_params.misc.type = |
1147 | VAEncMiscParameterTypeQualityLevel; |
1148 | priv->quality_params.quality.quality_level = opt->quality; |
1149 | |
1150 | ctx->global_params[ctx->nb_global_params] = |
1151 | &priv->quality_params.misc; |
1152 | ctx->global_params_size[ctx->nb_global_params++] = |
1153 | sizeof(priv->quality_params); |
1154 | #else |
1155 | av_log(avctx, AV_LOG_WARNING, "The encode quality option is not " |
1156 | "supported with this VAAPI version.\n"); |
1157 | #endif |
1158 | } |
1159 | |
1160 | return 0; |
1161 | } |
1162 | |
1163 | static const VAAPIEncodeType vaapi_encode_type_h264 = { |
1164 | .priv_data_size = sizeof(VAAPIEncodeH264Context), |
1165 | |
1166 | .configure = &vaapi_encode_h264_configure, |
1167 | |
1168 | .sequence_params_size = sizeof(VAEncSequenceParameterBufferH264), |
1169 | .init_sequence_params = &vaapi_encode_h264_init_sequence_params, |
1170 | |
1171 | .picture_params_size = sizeof(VAEncPictureParameterBufferH264), |
1172 | .init_picture_params = &vaapi_encode_h264_init_picture_params, |
1173 | |
1174 | .slice_params_size = sizeof(VAEncSliceParameterBufferH264), |
1175 | .init_slice_params = &vaapi_encode_h264_init_slice_params, |
1176 | |
1177 | .sequence_header_type = VAEncPackedHeaderSequence, |
1178 | .write_sequence_header = &vaapi_encode_h264_write_sequence_header, |
1179 | |
1180 | .slice_header_type = VAEncPackedHeaderH264_Slice, |
1181 | .write_slice_header = &vaapi_encode_h264_write_slice_header, |
1182 | |
1183 | .write_extra_header = &vaapi_encode_h264_write_extra_header, |
1184 | }; |
1185 | |
1186 | static av_cold int vaapi_encode_h264_init(AVCodecContext *avctx) |
1187 | { |
1188 | VAAPIEncodeContext *ctx = avctx->priv_data; |
1189 | VAAPIEncodeH264Options *opt = |
1190 | (VAAPIEncodeH264Options*)ctx->codec_options_data; |
1191 | |
1192 | ctx->codec = &vaapi_encode_type_h264; |
1193 | |
1194 | switch (avctx->profile) { |
1195 | case FF_PROFILE_H264_CONSTRAINED_BASELINE: |
1196 | ctx->va_profile = VAProfileH264ConstrainedBaseline; |
1197 | if (avctx->max_b_frames != 0) { |
1198 | avctx->max_b_frames = 0; |
1199 | av_log(avctx, AV_LOG_WARNING, "H.264 constrained baseline profile " |
1200 | "doesn't support encoding with B frames, disabling them.\n"); |
1201 | } |
1202 | break; |
1203 | case FF_PROFILE_H264_BASELINE: |
1204 | ctx->va_profile = VAProfileH264Baseline; |
1205 | if (avctx->max_b_frames != 0) { |
1206 | avctx->max_b_frames = 0; |
1207 | av_log(avctx, AV_LOG_WARNING, "H.264 baseline profile " |
1208 | "doesn't support encoding with B frames, disabling them.\n"); |
1209 | } |
1210 | break; |
1211 | case FF_PROFILE_H264_MAIN: |
1212 | ctx->va_profile = VAProfileH264Main; |
1213 | break; |
1214 | case FF_PROFILE_H264_EXTENDED: |
1215 | av_log(avctx, AV_LOG_ERROR, "H.264 extended profile " |
1216 | "is not supported.\n"); |
1217 | return AVERROR_PATCHWELCOME; |
1218 | case FF_PROFILE_UNKNOWN: |
1219 | case FF_PROFILE_H264_HIGH: |
1220 | ctx->va_profile = VAProfileH264High; |
1221 | break; |
1222 | case FF_PROFILE_H264_HIGH_10: |
1223 | case FF_PROFILE_H264_HIGH_10_INTRA: |
1224 | av_log(avctx, AV_LOG_ERROR, "H.264 10-bit profiles " |
1225 | "are not supported.\n"); |
1226 | return AVERROR_PATCHWELCOME; |
1227 | case FF_PROFILE_H264_HIGH_422: |
1228 | case FF_PROFILE_H264_HIGH_422_INTRA: |
1229 | case FF_PROFILE_H264_HIGH_444: |
1230 | case FF_PROFILE_H264_HIGH_444_PREDICTIVE: |
1231 | case FF_PROFILE_H264_HIGH_444_INTRA: |
1232 | case FF_PROFILE_H264_CAVLC_444: |
1233 | av_log(avctx, AV_LOG_ERROR, "H.264 non-4:2:0 profiles " |
1234 | "are not supported.\n"); |
1235 | return AVERROR_PATCHWELCOME; |
1236 | default: |
1237 | av_log(avctx, AV_LOG_ERROR, "Unknown H.264 profile %d.\n", |
1238 | avctx->profile); |
1239 | return AVERROR(EINVAL); |
1240 | } |
1241 | if (opt->low_power) { |
1242 | #if VA_CHECK_VERSION(0, 39, 2) |
1243 | ctx->va_entrypoint = VAEntrypointEncSliceLP; |
1244 | #else |
1245 | av_log(avctx, AV_LOG_ERROR, "Low-power encoding is not " |
1246 | "supported with this VAAPI version.\n"); |
1247 | return AVERROR(EINVAL); |
1248 | #endif |
1249 | } else { |
1250 | ctx->va_entrypoint = VAEntrypointEncSlice; |
1251 | } |
1252 | |
1253 | // Only 8-bit encode is supported. |
1254 | ctx->va_rt_format = VA_RT_FORMAT_YUV420; |
1255 | |
1256 | if (avctx->bit_rate > 0) { |
1257 | if (avctx->rc_max_rate == avctx->bit_rate) |
1258 | ctx->va_rc_mode = VA_RC_CBR; |
1259 | else |
1260 | ctx->va_rc_mode = VA_RC_VBR; |
1261 | } else |
1262 | ctx->va_rc_mode = VA_RC_CQP; |
1263 | |
1264 | ctx->va_packed_headers = |
1265 | VA_ENC_PACKED_HEADER_SEQUENCE | // SPS and PPS. |
1266 | VA_ENC_PACKED_HEADER_SLICE | // Slice headers. |
1267 | VA_ENC_PACKED_HEADER_MISC; // SEI. |
1268 | |
1269 | ctx->surface_width = FFALIGN(avctx->width, 16); |
1270 | ctx->surface_height = FFALIGN(avctx->height, 16); |
1271 | |
1272 | return ff_vaapi_encode_init(avctx); |
1273 | } |
1274 | |
1275 | #define OFFSET(x) (offsetof(VAAPIEncodeContext, codec_options_data) + \ |
1276 | offsetof(VAAPIEncodeH264Options, x)) |
1277 | #define FLAGS (AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM) |
1278 | static const AVOption vaapi_encode_h264_options[] = { |
1279 | { "qp", "Constant QP (for P-frames; scaled by qfactor/qoffset for I/B)", |
1280 | OFFSET(qp), AV_OPT_TYPE_INT, { .i64 = 20 }, 0, 52, FLAGS }, |
1281 | { "quality", "Set encode quality (trades off against speed, higher is faster)", |
1282 | OFFSET(quality), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 8, FLAGS }, |
1283 | { "low_power", "Use low-power encoding mode (experimental: only supported " |
1284 | "on some platforms, does not support all features)", |
1285 | OFFSET(low_power), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS }, |
1286 | { NULL }, |
1287 | }; |
1288 | |
1289 | static const AVCodecDefault vaapi_encode_h264_defaults[] = { |
1290 | { "profile", "100" }, |
1291 | { "level", "51" }, |
1292 | { "b", "0" }, |
1293 | { "bf", "2" }, |
1294 | { "g", "120" }, |
1295 | { "i_qfactor", "1" }, |
1296 | { "i_qoffset", "0" }, |
1297 | { "b_qfactor", "6/5" }, |
1298 | { "b_qoffset", "0" }, |
1299 | { "qmin", "0" }, |
1300 | { NULL }, |
1301 | }; |
1302 | |
1303 | static const AVClass vaapi_encode_h264_class = { |
1304 | .class_name = "h264_vaapi", |
1305 | .item_name = av_default_item_name, |
1306 | .option = vaapi_encode_h264_options, |
1307 | .version = LIBAVUTIL_VERSION_INT, |
1308 | }; |
1309 | |
1310 | AVCodec ff_h264_vaapi_encoder = { |
1311 | .name = "h264_vaapi", |
1312 | .long_name = NULL_IF_CONFIG_SMALL("H.264/AVC (VAAPI)"), |
1313 | .type = AVMEDIA_TYPE_VIDEO, |
1314 | .id = AV_CODEC_ID_H264, |
1315 | .priv_data_size = (sizeof(VAAPIEncodeContext) + |
1316 | sizeof(VAAPIEncodeH264Options)), |
1317 | .init = &vaapi_encode_h264_init, |
1318 | .encode2 = &ff_vaapi_encode2, |
1319 | .close = &ff_vaapi_encode_close, |
1320 | .priv_class = &vaapi_encode_h264_class, |
1321 | .capabilities = AV_CODEC_CAP_DELAY, |
1322 | .defaults = vaapi_encode_h264_defaults, |
1323 | .pix_fmts = (const enum AVPixelFormat[]) { |
1324 | AV_PIX_FMT_VAAPI, |
1325 | AV_PIX_FMT_NONE, |
1326 | }, |
1327 | }; |
1328 |