Commit | Line | Data |
---|---|---|
2c62fcdf MT |
1 | /* |
2 | * This file is part of Libav. | |
3 | * | |
4 | * Libav 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 | * Libav 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 Libav; 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" | |
251cbb44 | 28 | #include "h264.h" |
48e2967c | 29 | #include "h264_sei.h" |
2c62fcdf MT |
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 | ||
081961f8 MT |
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 | char bitstream_restriction_flag; | |
48e2967c MT |
88 | |
89 | unsigned int cpb_cnt_minus1; | |
90 | unsigned int bit_rate_scale; | |
91 | unsigned int cpb_size_scale; | |
92 | unsigned int bit_rate_value_minus1[32]; | |
93 | unsigned int cpb_size_value_minus1[32]; | |
94 | char cbr_flag[32]; | |
95 | unsigned int initial_cpb_removal_delay_length_minus1; | |
96 | unsigned int cpb_removal_delay_length_minus1; | |
97 | unsigned int dpb_output_delay_length_minus1; | |
98 | unsigned int time_offset_length; | |
99 | ||
100 | unsigned int initial_cpb_removal_delay; | |
101 | unsigned int initial_cpb_removal_delay_offset; | |
102 | ||
103 | unsigned int pic_struct; | |
2c62fcdf MT |
104 | } VAAPIEncodeH264MiscSequenceParams; |
105 | ||
106 | // This structure contains all possibly-useful per-slice syntax elements | |
107 | // which are not already contained in the various VAAPI structures. | |
108 | typedef struct VAAPIEncodeH264MiscSliceParams { | |
109 | unsigned int nal_unit_type; | |
110 | unsigned int nal_ref_idc; | |
111 | ||
112 | unsigned int colour_plane_id; | |
113 | char field_pic_flag; | |
114 | char bottom_field_flag; | |
115 | ||
116 | unsigned int redundant_pic_cnt; | |
117 | ||
118 | char sp_for_switch_flag; | |
119 | int slice_qs_delta; | |
120 | ||
121 | char ref_pic_list_modification_flag_l0; | |
122 | char ref_pic_list_modification_flag_l1; | |
123 | ||
124 | char no_output_of_prior_pics_flag; | |
125 | char long_term_reference_flag; | |
126 | char adaptive_ref_pic_marking_mode_flag; | |
127 | } VAAPIEncodeH264MiscSliceParams; | |
128 | ||
129 | typedef struct VAAPIEncodeH264Slice { | |
130 | VAAPIEncodeH264MiscSliceParams misc_slice_params; | |
131 | } VAAPIEncodeH264Slice; | |
132 | ||
133 | typedef struct VAAPIEncodeH264Context { | |
134 | VAAPIEncodeH264MiscSequenceParams misc_sequence_params; | |
135 | ||
136 | int mb_width; | |
137 | int mb_height; | |
138 | ||
139 | int fixed_qp_idr; | |
140 | int fixed_qp_p; | |
141 | int fixed_qp_b; | |
142 | ||
b51c7c6b | 143 | int next_frame_num; |
2c62fcdf | 144 | int64_t idr_pic_count; |
2c62fcdf | 145 | |
48e2967c MT |
146 | int cpb_delay; |
147 | int dpb_delay; | |
148 | ||
6e8f66fc | 149 | // Rate control configuration. |
48e2967c | 150 | int send_timing_sei; |
6e8f66fc MT |
151 | struct { |
152 | VAEncMiscParameterBuffer misc; | |
153 | VAEncMiscParameterRateControl rc; | |
154 | } rc_params; | |
155 | struct { | |
156 | VAEncMiscParameterBuffer misc; | |
157 | VAEncMiscParameterHRD hrd; | |
158 | } hrd_params; | |
fcf536b1 MT |
159 | |
160 | #if VA_CHECK_VERSION(0, 36, 0) | |
161 | // Speed-quality tradeoff setting. | |
162 | struct { | |
163 | VAEncMiscParameterBuffer misc; | |
164 | VAEncMiscParameterBufferQualityLevel quality; | |
165 | } quality_params; | |
166 | #endif | |
2c62fcdf MT |
167 | } VAAPIEncodeH264Context; |
168 | ||
9629701c MT |
169 | typedef struct VAAPIEncodeH264Options { |
170 | int qp; | |
fcf536b1 | 171 | int quality; |
a86aa160 | 172 | int low_power; |
9629701c MT |
173 | } VAAPIEncodeH264Options; |
174 | ||
2c62fcdf MT |
175 | |
176 | #define vseq_var(name) vseq->name, name | |
177 | #define vseq_field(name) vseq->seq_fields.bits.name, name | |
081961f8 | 178 | #define vvui_field(name) vseq->vui_fields.bits.name, name |
2c62fcdf MT |
179 | #define vpic_var(name) vpic->name, name |
180 | #define vpic_field(name) vpic->pic_fields.bits.name, name | |
181 | #define vslice_var(name) vslice->name, name | |
182 | #define vslice_field(name) vslice->slice_fields.bits.name, name | |
183 | #define mseq_var(name) mseq->name, name | |
184 | #define mslice_var(name) mslice->name, name | |
185 | ||
186 | static void vaapi_encode_h264_write_nal_header(PutBitContext *pbc, | |
187 | int nal_unit_type, int nal_ref_idc) | |
188 | { | |
189 | u(1, 0, forbidden_zero_bit); | |
190 | u(2, nal_ref_idc, nal_ref_idc); | |
191 | u(5, nal_unit_type, nal_unit_type); | |
192 | } | |
193 | ||
194 | static void vaapi_encode_h264_write_trailing_rbsp(PutBitContext *pbc) | |
195 | { | |
196 | u(1, 1, rbsp_stop_one_bit); | |
197 | while (put_bits_count(pbc) & 7) | |
198 | u(1, 0, rbsp_alignment_zero_bit); | |
199 | } | |
200 | ||
081961f8 MT |
201 | static void vaapi_encode_h264_write_vui(PutBitContext *pbc, |
202 | VAAPIEncodeContext *ctx) | |
203 | { | |
204 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; | |
205 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
206 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; | |
48e2967c | 207 | int i; |
081961f8 MT |
208 | |
209 | u(1, vvui_field(aspect_ratio_info_present_flag)); | |
210 | if (vseq->vui_fields.bits.aspect_ratio_info_present_flag) { | |
211 | u(8, vseq_var(aspect_ratio_idc)); | |
212 | if (vseq->aspect_ratio_idc == 255) { | |
213 | u(16, vseq_var(sar_width)); | |
214 | u(16, vseq_var(sar_height)); | |
215 | } | |
216 | } | |
217 | ||
218 | u(1, mseq_var(overscan_info_present_flag)); | |
219 | if (mseq->overscan_info_present_flag) | |
220 | u(1, mseq_var(overscan_appropriate_flag)); | |
221 | ||
222 | u(1, mseq_var(video_signal_type_present_flag)); | |
223 | if (mseq->video_signal_type_present_flag) { | |
224 | u(3, mseq_var(video_format)); | |
225 | u(1, mseq_var(video_full_range_flag)); | |
226 | u(1, mseq_var(colour_description_present_flag)); | |
227 | if (mseq->colour_description_present_flag) { | |
228 | u(8, mseq_var(colour_primaries)); | |
229 | u(8, mseq_var(transfer_characteristics)); | |
230 | u(8, mseq_var(matrix_coefficients)); | |
231 | } | |
232 | } | |
233 | ||
234 | u(1, mseq_var(chroma_loc_info_present_flag)); | |
235 | if (mseq->chroma_loc_info_present_flag) { | |
236 | ue(mseq_var(chroma_sample_loc_type_top_field)); | |
237 | ue(mseq_var(chroma_sample_loc_type_bottom_field)); | |
238 | } | |
239 | ||
240 | u(1, vvui_field(timing_info_present_flag)); | |
241 | if (vseq->vui_fields.bits.timing_info_present_flag) { | |
242 | u(32, vseq_var(num_units_in_tick)); | |
243 | u(32, vseq_var(time_scale)); | |
244 | u(1, mseq_var(fixed_frame_rate_flag)); | |
245 | } | |
246 | ||
247 | u(1, mseq_var(nal_hrd_parameters_present_flag)); | |
248 | if (mseq->nal_hrd_parameters_present_flag) { | |
48e2967c MT |
249 | ue(mseq_var(cpb_cnt_minus1)); |
250 | u(4, mseq_var(bit_rate_scale)); | |
251 | u(4, mseq_var(cpb_size_scale)); | |
252 | for (i = 0; i <= mseq->cpb_cnt_minus1; i++) { | |
253 | ue(mseq_var(bit_rate_value_minus1[i])); | |
254 | ue(mseq_var(cpb_size_value_minus1[i])); | |
255 | u(1, mseq_var(cbr_flag[i])); | |
256 | } | |
257 | u(5, mseq_var(initial_cpb_removal_delay_length_minus1)); | |
258 | u(5, mseq_var(cpb_removal_delay_length_minus1)); | |
259 | u(5, mseq_var(dpb_output_delay_length_minus1)); | |
260 | u(5, mseq_var(time_offset_length)); | |
081961f8 MT |
261 | } |
262 | u(1, mseq_var(vcl_hrd_parameters_present_flag)); | |
263 | if (mseq->vcl_hrd_parameters_present_flag) { | |
264 | av_assert0(0 && "vcl hrd parameters not supported"); | |
265 | } | |
266 | ||
267 | if (mseq->nal_hrd_parameters_present_flag || | |
268 | mseq->vcl_hrd_parameters_present_flag) | |
269 | u(1, mseq_var(low_delay_hrd_flag)); | |
270 | u(1, mseq_var(pic_struct_present_flag)); | |
271 | ||
272 | u(1, vvui_field(bitstream_restriction_flag)); | |
273 | if (vseq->vui_fields.bits.bitstream_restriction_flag) { | |
274 | av_assert0(0 && "bitstream restrictions not supported"); | |
275 | } | |
276 | } | |
277 | ||
2c62fcdf MT |
278 | static void vaapi_encode_h264_write_sps(PutBitContext *pbc, |
279 | VAAPIEncodeContext *ctx) | |
280 | { | |
281 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; | |
282 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
283 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; | |
284 | int i; | |
285 | ||
5c2fb561 | 286 | vaapi_encode_h264_write_nal_header(pbc, H264_NAL_SPS, 3); |
2c62fcdf MT |
287 | |
288 | u(8, mseq_var(profile_idc)); | |
289 | u(1, mseq_var(constraint_set0_flag)); | |
290 | u(1, mseq_var(constraint_set1_flag)); | |
291 | u(1, mseq_var(constraint_set2_flag)); | |
292 | u(1, mseq_var(constraint_set3_flag)); | |
293 | u(1, mseq_var(constraint_set4_flag)); | |
294 | u(1, mseq_var(constraint_set5_flag)); | |
295 | u(2, 0, reserved_zero_2bits); | |
296 | ||
297 | u(8, vseq_var(level_idc)); | |
298 | ||
299 | ue(vseq_var(seq_parameter_set_id)); | |
300 | ||
301 | if (mseq->profile_idc == 100 || mseq->profile_idc == 110 || | |
302 | mseq->profile_idc == 122 || mseq->profile_idc == 244 || | |
303 | mseq->profile_idc == 44 || mseq->profile_idc == 83 || | |
304 | mseq->profile_idc == 86 || mseq->profile_idc == 118 || | |
305 | mseq->profile_idc == 128 || mseq->profile_idc == 138) { | |
306 | ue(vseq_field(chroma_format_idc)); | |
307 | ||
308 | if (vseq->seq_fields.bits.chroma_format_idc == 3) | |
309 | u(1, mseq_var(separate_colour_plane_flag)); | |
310 | ||
311 | ue(vseq_var(bit_depth_luma_minus8)); | |
312 | ue(vseq_var(bit_depth_chroma_minus8)); | |
313 | ||
314 | u(1, mseq_var(qpprime_y_zero_transform_bypass_flag)); | |
315 | ||
316 | u(1, vseq_field(seq_scaling_matrix_present_flag)); | |
317 | if (vseq->seq_fields.bits.seq_scaling_matrix_present_flag) { | |
318 | av_assert0(0 && "scaling matrices not supported"); | |
319 | } | |
320 | } | |
321 | ||
322 | ue(vseq_field(log2_max_frame_num_minus4)); | |
323 | ue(vseq_field(pic_order_cnt_type)); | |
324 | ||
325 | if (vseq->seq_fields.bits.pic_order_cnt_type == 0) { | |
326 | ue(vseq_field(log2_max_pic_order_cnt_lsb_minus4)); | |
327 | } else if (vseq->seq_fields.bits.pic_order_cnt_type == 1) { | |
328 | u(1, mseq_var(delta_pic_order_always_zero_flag)); | |
329 | se(vseq_var(offset_for_non_ref_pic)); | |
330 | se(vseq_var(offset_for_top_to_bottom_field)); | |
331 | ue(vseq_var(num_ref_frames_in_pic_order_cnt_cycle)); | |
332 | ||
333 | for (i = 0; i < vseq->num_ref_frames_in_pic_order_cnt_cycle; i++) | |
334 | se(vseq_var(offset_for_ref_frame[i])); | |
335 | } | |
336 | ||
337 | ue(vseq_var(max_num_ref_frames)); | |
338 | u(1, mseq_var(gaps_in_frame_num_allowed_flag)); | |
339 | ||
340 | ue(vseq->picture_width_in_mbs - 1, pic_width_in_mbs_minus1); | |
341 | ue(vseq->picture_height_in_mbs - 1, pic_height_in_mbs_minus1); | |
342 | ||
343 | u(1, vseq_field(frame_mbs_only_flag)); | |
344 | if (!vseq->seq_fields.bits.frame_mbs_only_flag) | |
345 | u(1, vseq_field(mb_adaptive_frame_field_flag)); | |
346 | ||
347 | u(1, vseq_field(direct_8x8_inference_flag)); | |
348 | ||
349 | u(1, vseq_var(frame_cropping_flag)); | |
350 | if (vseq->frame_cropping_flag) { | |
351 | ue(vseq_var(frame_crop_left_offset)); | |
352 | ue(vseq_var(frame_crop_right_offset)); | |
353 | ue(vseq_var(frame_crop_top_offset)); | |
354 | ue(vseq_var(frame_crop_bottom_offset)); | |
355 | } | |
356 | ||
081961f8 MT |
357 | u(1, vseq_var(vui_parameters_present_flag)); |
358 | if (vseq->vui_parameters_present_flag) | |
359 | vaapi_encode_h264_write_vui(pbc, ctx); | |
2c62fcdf MT |
360 | |
361 | vaapi_encode_h264_write_trailing_rbsp(pbc); | |
362 | } | |
363 | ||
364 | static void vaapi_encode_h264_write_pps(PutBitContext *pbc, | |
365 | VAAPIEncodeContext *ctx) | |
366 | { | |
367 | VAEncPictureParameterBufferH264 *vpic = ctx->codec_picture_params; | |
368 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
369 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; | |
370 | ||
5c2fb561 | 371 | vaapi_encode_h264_write_nal_header(pbc, H264_NAL_PPS, 3); |
2c62fcdf MT |
372 | |
373 | ue(vpic_var(pic_parameter_set_id)); | |
374 | ue(vpic_var(seq_parameter_set_id)); | |
375 | ||
376 | u(1, vpic_field(entropy_coding_mode_flag)); | |
377 | u(1, mseq_var(bottom_field_pic_order_in_frame_present_flag)); | |
378 | ||
379 | ue(mseq_var(num_slice_groups_minus1)); | |
380 | if (mseq->num_slice_groups_minus1 > 0) { | |
381 | ue(mseq_var(slice_group_map_type)); | |
382 | av_assert0(0 && "slice groups not supported"); | |
383 | } | |
384 | ||
385 | ue(vpic_var(num_ref_idx_l0_active_minus1)); | |
386 | ue(vpic_var(num_ref_idx_l1_active_minus1)); | |
387 | ||
388 | u(1, vpic_field(weighted_pred_flag)); | |
389 | u(2, vpic_field(weighted_bipred_idc)); | |
390 | ||
391 | se(vpic->pic_init_qp - 26, pic_init_qp_minus26); | |
392 | se(mseq_var(pic_init_qs_minus26)); | |
393 | se(vpic_var(chroma_qp_index_offset)); | |
394 | ||
395 | u(1, vpic_field(deblocking_filter_control_present_flag)); | |
396 | u(1, vpic_field(constrained_intra_pred_flag)); | |
397 | u(1, vpic_field(redundant_pic_cnt_present_flag)); | |
398 | u(1, vpic_field(transform_8x8_mode_flag)); | |
399 | ||
400 | u(1, vpic_field(pic_scaling_matrix_present_flag)); | |
401 | if (vpic->pic_fields.bits.pic_scaling_matrix_present_flag) { | |
402 | av_assert0(0 && "scaling matrices not supported"); | |
403 | } | |
404 | ||
405 | se(vpic_var(second_chroma_qp_index_offset)); | |
406 | ||
407 | vaapi_encode_h264_write_trailing_rbsp(pbc); | |
408 | } | |
409 | ||
410 | static void vaapi_encode_h264_write_slice_header2(PutBitContext *pbc, | |
411 | VAAPIEncodeContext *ctx, | |
412 | VAAPIEncodePicture *pic, | |
413 | VAAPIEncodeSlice *slice) | |
414 | { | |
415 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; | |
416 | VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; | |
417 | VAEncSliceParameterBufferH264 *vslice = slice->codec_slice_params; | |
418 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
419 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; | |
420 | VAAPIEncodeH264Slice *pslice = slice->priv_data; | |
421 | VAAPIEncodeH264MiscSliceParams *mslice = &pslice->misc_slice_params; | |
422 | ||
423 | vaapi_encode_h264_write_nal_header(pbc, mslice->nal_unit_type, | |
424 | mslice->nal_ref_idc); | |
425 | ||
426 | ue(vslice->macroblock_address, first_mb_in_slice); | |
427 | ue(vslice_var(slice_type)); | |
428 | ue(vpic_var(pic_parameter_set_id)); | |
429 | ||
430 | if (mseq->separate_colour_plane_flag) { | |
431 | u(2, mslice_var(colour_plane_id)); | |
432 | } | |
433 | ||
434 | u(4 + vseq->seq_fields.bits.log2_max_frame_num_minus4, | |
435 | (vpic->frame_num & | |
436 | ((1 << (4 + vseq->seq_fields.bits.log2_max_frame_num_minus4)) - 1)), | |
437 | frame_num); | |
438 | ||
439 | if (!vseq->seq_fields.bits.frame_mbs_only_flag) { | |
440 | u(1, mslice_var(field_pic_flag)); | |
441 | if (mslice->field_pic_flag) | |
442 | u(1, mslice_var(bottom_field_flag)); | |
443 | } | |
444 | ||
445 | if (vpic->pic_fields.bits.idr_pic_flag) { | |
446 | ue(vslice_var(idr_pic_id)); | |
447 | } | |
448 | ||
449 | if (vseq->seq_fields.bits.pic_order_cnt_type == 0) { | |
450 | u(4 + vseq->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4, | |
451 | vslice_var(pic_order_cnt_lsb)); | |
452 | if (mseq->bottom_field_pic_order_in_frame_present_flag && | |
453 | !mslice->field_pic_flag) { | |
454 | se(vslice_var(delta_pic_order_cnt_bottom)); | |
455 | } | |
456 | } | |
457 | ||
458 | if (vseq->seq_fields.bits.pic_order_cnt_type == 1 && | |
459 | !vseq->seq_fields.bits.delta_pic_order_always_zero_flag) { | |
460 | se(vslice_var(delta_pic_order_cnt[0])); | |
461 | if (mseq->bottom_field_pic_order_in_frame_present_flag && | |
462 | !mslice->field_pic_flag) { | |
463 | se(vslice_var(delta_pic_order_cnt[1])); | |
464 | } | |
465 | } | |
466 | ||
467 | if (vpic->pic_fields.bits.redundant_pic_cnt_present_flag) { | |
468 | ue(mslice_var(redundant_pic_cnt)); | |
469 | } | |
470 | ||
471 | if (vslice->slice_type == SLICE_TYPE_B) { | |
472 | u(1, vslice_var(direct_spatial_mv_pred_flag)); | |
473 | } | |
474 | ||
475 | if (vslice->slice_type == SLICE_TYPE_P || | |
476 | vslice->slice_type == SLICE_TYPE_SP || | |
477 | vslice->slice_type == SLICE_TYPE_B) { | |
478 | u(1, vslice_var(num_ref_idx_active_override_flag)); | |
479 | if (vslice->num_ref_idx_active_override_flag) { | |
480 | ue(vslice_var(num_ref_idx_l0_active_minus1)); | |
481 | if (vslice->slice_type == SLICE_TYPE_B) | |
482 | ue(vslice_var(num_ref_idx_l1_active_minus1)); | |
483 | } | |
484 | } | |
485 | ||
486 | if (mslice->nal_unit_type == 20 || mslice->nal_unit_type == 21) { | |
487 | av_assert0(0 && "no MVC support"); | |
488 | } else { | |
489 | if (vslice->slice_type % 5 != 2 && vslice->slice_type % 5 != 4) { | |
490 | u(1, mslice_var(ref_pic_list_modification_flag_l0)); | |
491 | if (mslice->ref_pic_list_modification_flag_l0) { | |
492 | av_assert0(0 && "ref pic list modification"); | |
493 | } | |
494 | } | |
495 | if (vslice->slice_type % 5 == 1) { | |
496 | u(1, mslice_var(ref_pic_list_modification_flag_l1)); | |
497 | if (mslice->ref_pic_list_modification_flag_l1) { | |
498 | av_assert0(0 && "ref pic list modification"); | |
499 | } | |
500 | } | |
501 | } | |
502 | ||
503 | if ((vpic->pic_fields.bits.weighted_pred_flag && | |
504 | (vslice->slice_type == SLICE_TYPE_P || | |
505 | vslice->slice_type == SLICE_TYPE_SP)) || | |
506 | (vpic->pic_fields.bits.weighted_bipred_idc == 1 && | |
507 | vslice->slice_type == SLICE_TYPE_B)) { | |
508 | av_assert0(0 && "prediction weights not supported"); | |
509 | } | |
510 | ||
511 | av_assert0(mslice->nal_ref_idc > 0 == | |
512 | vpic->pic_fields.bits.reference_pic_flag); | |
513 | if (mslice->nal_ref_idc != 0) { | |
514 | if (vpic->pic_fields.bits.idr_pic_flag) { | |
515 | u(1, mslice_var(no_output_of_prior_pics_flag)); | |
516 | u(1, mslice_var(long_term_reference_flag)); | |
517 | } else { | |
518 | u(1, mslice_var(adaptive_ref_pic_marking_mode_flag)); | |
519 | if (mslice->adaptive_ref_pic_marking_mode_flag) { | |
520 | av_assert0(0 && "MMCOs not supported"); | |
521 | } | |
522 | } | |
523 | } | |
524 | ||
525 | if (vpic->pic_fields.bits.entropy_coding_mode_flag && | |
526 | vslice->slice_type != SLICE_TYPE_I && | |
527 | vslice->slice_type != SLICE_TYPE_SI) { | |
528 | ue(vslice_var(cabac_init_idc)); | |
529 | } | |
530 | ||
531 | se(vslice_var(slice_qp_delta)); | |
532 | if (vslice->slice_type == SLICE_TYPE_SP || | |
533 | vslice->slice_type == SLICE_TYPE_SI) { | |
534 | if (vslice->slice_type == SLICE_TYPE_SP) | |
535 | u(1, mslice_var(sp_for_switch_flag)); | |
536 | se(mslice_var(slice_qs_delta)); | |
537 | } | |
538 | ||
539 | if (vpic->pic_fields.bits.deblocking_filter_control_present_flag) { | |
540 | ue(vslice_var(disable_deblocking_filter_idc)); | |
541 | if (vslice->disable_deblocking_filter_idc != 1) { | |
542 | se(vslice_var(slice_alpha_c0_offset_div2)); | |
543 | se(vslice_var(slice_beta_offset_div2)); | |
544 | } | |
545 | } | |
546 | ||
547 | if (mseq->num_slice_groups_minus1 > 0 && | |
548 | mseq->slice_group_map_type >= 3 && mseq->slice_group_map_type <= 5) { | |
549 | av_assert0(0 && "slice groups not supported"); | |
550 | } | |
551 | ||
552 | // No alignment - this need not be a byte boundary. | |
553 | } | |
554 | ||
48e2967c MT |
555 | static void vaapi_encode_h264_write_buffering_period(PutBitContext *pbc, |
556 | VAAPIEncodeContext *ctx, | |
557 | VAAPIEncodePicture *pic) | |
558 | { | |
559 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
560 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; | |
561 | VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; | |
562 | int i; | |
563 | ||
564 | ue(vpic_var(seq_parameter_set_id)); | |
565 | ||
566 | if (mseq->nal_hrd_parameters_present_flag) { | |
567 | for (i = 0; i <= mseq->cpb_cnt_minus1; i++) { | |
568 | u(mseq->initial_cpb_removal_delay_length_minus1 + 1, | |
569 | mseq_var(initial_cpb_removal_delay)); | |
570 | u(mseq->initial_cpb_removal_delay_length_minus1 + 1, | |
571 | mseq_var(initial_cpb_removal_delay_offset)); | |
572 | } | |
573 | } | |
574 | if (mseq->vcl_hrd_parameters_present_flag) { | |
575 | av_assert0(0 && "vcl hrd parameters not supported"); | |
576 | } | |
577 | } | |
578 | ||
579 | static void vaapi_encode_h264_write_pic_timing(PutBitContext *pbc, | |
580 | VAAPIEncodeContext *ctx, | |
581 | VAAPIEncodePicture *pic) | |
582 | { | |
583 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; | |
584 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
585 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; | |
586 | int i, num_clock_ts; | |
587 | ||
588 | if (mseq->nal_hrd_parameters_present_flag || | |
589 | mseq->vcl_hrd_parameters_present_flag) { | |
590 | u(mseq->cpb_removal_delay_length_minus1 + 1, | |
591 | 2 * vseq->num_units_in_tick * priv->cpb_delay, | |
592 | cpb_removal_delay); | |
593 | u(mseq->dpb_output_delay_length_minus1 + 1, | |
594 | 2 * vseq->num_units_in_tick * priv->dpb_delay, | |
595 | dpb_output_delay); | |
596 | } | |
597 | if (mseq->pic_struct_present_flag) { | |
598 | u(4, mseq_var(pic_struct)); | |
599 | num_clock_ts = (mseq->pic_struct <= 2 ? 1 : | |
600 | mseq->pic_struct <= 4 ? 2 : | |
601 | mseq->pic_struct <= 8 ? 3 : 0); | |
602 | for (i = 0; i < num_clock_ts; i++) { | |
603 | u(1, 0, clock_timestamp_flag[i]); | |
604 | // No full timestamp information. | |
605 | } | |
606 | } | |
607 | } | |
608 | ||
02fa1ad9 MT |
609 | static void vaapi_encode_h264_write_identifier(PutBitContext *pbc, |
610 | VAAPIEncodeContext *ctx, | |
611 | VAAPIEncodePicture *pic) | |
612 | { | |
613 | const char *lavc = LIBAVCODEC_IDENT; | |
614 | const char *vaapi = VA_VERSION_S; | |
615 | const char *driver = vaQueryVendorString(ctx->hwctx->display); | |
616 | char tmp[256]; | |
617 | int i; | |
618 | ||
619 | // Random (version 4) ISO 11578 UUID. | |
620 | uint8_t uuid[16] = { | |
621 | 0x59, 0x94, 0x8b, 0x28, 0x11, 0xec, 0x45, 0xaf, | |
622 | 0x96, 0x75, 0x19, 0xd4, 0x1f, 0xea, 0xa9, 0x4d, | |
623 | }; | |
624 | ||
625 | for (i = 0; i < 16; i++) | |
626 | u(8, uuid[i], uuid_iso_iec_11578); | |
627 | ||
628 | snprintf(tmp, sizeof(tmp), "%s / VAAPI %s / %s", lavc, vaapi, driver); | |
629 | for (i = 0; i < sizeof(tmp) && tmp[i]; i++) | |
630 | u(8, tmp[i], user_data_payload_byte); | |
631 | } | |
632 | ||
48e2967c MT |
633 | static void vaapi_encode_h264_write_sei(PutBitContext *pbc, |
634 | VAAPIEncodeContext *ctx, | |
635 | VAAPIEncodePicture *pic) | |
636 | { | |
637 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
638 | PutBitContext payload_bits; | |
639 | char payload[256]; | |
640 | int payload_type, payload_size, i; | |
641 | void (*write_payload)(PutBitContext *pbc, | |
642 | VAAPIEncodeContext *ctx, | |
643 | VAAPIEncodePicture *pic) = NULL; | |
644 | ||
5c2fb561 | 645 | vaapi_encode_h264_write_nal_header(pbc, H264_NAL_SEI, 0); |
48e2967c MT |
646 | |
647 | for (payload_type = 0; payload_type < 64; payload_type++) { | |
648 | switch (payload_type) { | |
649 | case SEI_TYPE_BUFFERING_PERIOD: | |
650 | if (!priv->send_timing_sei || | |
651 | pic->type != PICTURE_TYPE_IDR) | |
652 | continue; | |
653 | write_payload = &vaapi_encode_h264_write_buffering_period; | |
654 | break; | |
655 | case SEI_TYPE_PIC_TIMING: | |
656 | if (!priv->send_timing_sei) | |
657 | continue; | |
658 | write_payload = &vaapi_encode_h264_write_pic_timing; | |
659 | break; | |
02fa1ad9 MT |
660 | case SEI_TYPE_USER_DATA_UNREGISTERED: |
661 | if (pic->encode_order != 0) | |
662 | continue; | |
663 | write_payload = &vaapi_encode_h264_write_identifier; | |
664 | break; | |
48e2967c MT |
665 | default: |
666 | continue; | |
667 | } | |
668 | ||
669 | init_put_bits(&payload_bits, payload, sizeof(payload)); | |
670 | write_payload(&payload_bits, ctx, pic); | |
671 | if (put_bits_count(&payload_bits) & 7) { | |
672 | write_u(&payload_bits, 1, 1, bit_equal_to_one); | |
673 | while (put_bits_count(&payload_bits) & 7) | |
674 | write_u(&payload_bits, 1, 0, bit_equal_to_zero); | |
675 | } | |
676 | payload_size = put_bits_count(&payload_bits) / 8; | |
677 | flush_put_bits(&payload_bits); | |
678 | ||
679 | u(8, payload_type, last_payload_type_byte); | |
680 | u(8, payload_size, last_payload_size_byte); | |
681 | for (i = 0; i < payload_size; i++) | |
682 | u(8, payload[i] & 0xff, sei_payload); | |
683 | } | |
684 | ||
685 | vaapi_encode_h264_write_trailing_rbsp(pbc); | |
686 | } | |
687 | ||
2c62fcdf MT |
688 | static int vaapi_encode_h264_write_sequence_header(AVCodecContext *avctx, |
689 | char *data, size_t *data_len) | |
690 | { | |
691 | VAAPIEncodeContext *ctx = avctx->priv_data; | |
692 | PutBitContext pbc; | |
693 | char tmp[256]; | |
694 | int err; | |
695 | size_t nal_len, bit_len, bit_pos, next_len; | |
696 | ||
697 | bit_len = *data_len; | |
698 | bit_pos = 0; | |
699 | ||
700 | init_put_bits(&pbc, tmp, sizeof(tmp)); | |
701 | vaapi_encode_h264_write_sps(&pbc, ctx); | |
702 | nal_len = put_bits_count(&pbc); | |
703 | flush_put_bits(&pbc); | |
704 | ||
705 | next_len = bit_len - bit_pos; | |
706 | err = ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data + bit_pos / 8, | |
707 | &next_len, | |
708 | tmp, nal_len); | |
709 | if (err < 0) | |
710 | return err; | |
711 | bit_pos += next_len; | |
712 | ||
713 | init_put_bits(&pbc, tmp, sizeof(tmp)); | |
714 | vaapi_encode_h264_write_pps(&pbc, ctx); | |
715 | nal_len = put_bits_count(&pbc); | |
716 | flush_put_bits(&pbc); | |
717 | ||
718 | next_len = bit_len - bit_pos; | |
719 | err = ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data + bit_pos / 8, | |
720 | &next_len, | |
721 | tmp, nal_len); | |
722 | if (err < 0) | |
723 | return err; | |
724 | bit_pos += next_len; | |
725 | ||
726 | *data_len = bit_pos; | |
727 | return 0; | |
728 | } | |
729 | ||
730 | static int vaapi_encode_h264_write_slice_header(AVCodecContext *avctx, | |
731 | VAAPIEncodePicture *pic, | |
732 | VAAPIEncodeSlice *slice, | |
733 | char *data, size_t *data_len) | |
734 | { | |
735 | VAAPIEncodeContext *ctx = avctx->priv_data; | |
736 | PutBitContext pbc; | |
737 | char tmp[256]; | |
738 | size_t header_len; | |
739 | ||
740 | init_put_bits(&pbc, tmp, sizeof(tmp)); | |
741 | vaapi_encode_h264_write_slice_header2(&pbc, ctx, pic, slice); | |
742 | header_len = put_bits_count(&pbc); | |
743 | flush_put_bits(&pbc); | |
744 | ||
745 | return ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data, data_len, | |
746 | tmp, header_len); | |
747 | } | |
748 | ||
48e2967c MT |
749 | static int vaapi_encode_h264_write_extra_header(AVCodecContext *avctx, |
750 | VAAPIEncodePicture *pic, | |
751 | int index, int *type, | |
752 | char *data, size_t *data_len) | |
753 | { | |
754 | VAAPIEncodeContext *ctx = avctx->priv_data; | |
755 | PutBitContext pbc; | |
756 | char tmp[256]; | |
757 | size_t header_len; | |
758 | ||
759 | if (index == 0 && ctx->va_rc_mode == VA_RC_CBR) { | |
760 | *type = VAEncPackedHeaderH264_SEI; | |
761 | ||
762 | init_put_bits(&pbc, tmp, sizeof(tmp)); | |
763 | vaapi_encode_h264_write_sei(&pbc, ctx, pic); | |
764 | header_len = put_bits_count(&pbc); | |
765 | flush_put_bits(&pbc); | |
766 | ||
767 | return ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data, data_len, | |
768 | tmp, header_len); | |
769 | ||
770 | } else { | |
771 | return AVERROR_EOF; | |
772 | } | |
773 | } | |
774 | ||
2c62fcdf MT |
775 | static int vaapi_encode_h264_init_sequence_params(AVCodecContext *avctx) |
776 | { | |
777 | VAAPIEncodeContext *ctx = avctx->priv_data; | |
778 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; | |
779 | VAEncPictureParameterBufferH264 *vpic = ctx->codec_picture_params; | |
780 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
781 | VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params; | |
782 | int i; | |
783 | ||
784 | { | |
785 | vseq->seq_parameter_set_id = 0; | |
786 | ||
787 | vseq->level_idc = avctx->level; | |
788 | ||
789 | vseq->max_num_ref_frames = 2; | |
790 | ||
791 | vseq->picture_width_in_mbs = priv->mb_width; | |
792 | vseq->picture_height_in_mbs = priv->mb_height; | |
793 | ||
794 | vseq->seq_fields.bits.chroma_format_idc = 1; | |
795 | vseq->seq_fields.bits.frame_mbs_only_flag = 1; | |
796 | vseq->seq_fields.bits.direct_8x8_inference_flag = 1; | |
797 | vseq->seq_fields.bits.log2_max_frame_num_minus4 = 4; | |
798 | vseq->seq_fields.bits.pic_order_cnt_type = 0; | |
799 | ||
800 | if (ctx->input_width != ctx->aligned_width || | |
801 | ctx->input_height != ctx->aligned_height) { | |
802 | vseq->frame_cropping_flag = 1; | |
803 | ||
804 | vseq->frame_crop_left_offset = 0; | |
805 | vseq->frame_crop_right_offset = | |
806 | (ctx->aligned_width - ctx->input_width) / 2; | |
807 | vseq->frame_crop_top_offset = 0; | |
808 | vseq->frame_crop_bottom_offset = | |
809 | (ctx->aligned_height - ctx->input_height) / 2; | |
810 | } else { | |
811 | vseq->frame_cropping_flag = 0; | |
812 | } | |
6e8f66fc | 813 | |
081961f8 MT |
814 | vseq->vui_parameters_present_flag = 1; |
815 | if (avctx->sample_aspect_ratio.num != 0) { | |
816 | vseq->vui_fields.bits.aspect_ratio_info_present_flag = 1; | |
817 | // There is a large enum of these which we could support | |
818 | // individually rather than using the generic X/Y form? | |
819 | if (avctx->sample_aspect_ratio.num == | |
820 | avctx->sample_aspect_ratio.den) { | |
821 | vseq->aspect_ratio_idc = 1; | |
822 | } else { | |
823 | vseq->aspect_ratio_idc = 255; // Extended SAR. | |
824 | vseq->sar_width = avctx->sample_aspect_ratio.num; | |
825 | vseq->sar_height = avctx->sample_aspect_ratio.den; | |
826 | } | |
827 | } | |
828 | if (avctx->color_primaries != AVCOL_PRI_UNSPECIFIED || | |
829 | avctx->color_trc != AVCOL_TRC_UNSPECIFIED || | |
830 | avctx->colorspace != AVCOL_SPC_UNSPECIFIED) { | |
831 | mseq->video_signal_type_present_flag = 1; | |
832 | mseq->video_format = 5; // Unspecified. | |
833 | mseq->video_full_range_flag = 0; | |
834 | mseq->colour_description_present_flag = 1; | |
835 | // These enums are derived from the standard and hence | |
836 | // we can just use the values directly. | |
837 | mseq->colour_primaries = avctx->color_primaries; | |
838 | mseq->transfer_characteristics = avctx->color_trc; | |
839 | mseq->matrix_coefficients = avctx->colorspace; | |
840 | } | |
841 | ||
6e8f66fc | 842 | vseq->bits_per_second = avctx->bit_rate; |
081961f8 MT |
843 | |
844 | vseq->vui_fields.bits.timing_info_present_flag = 1; | |
6e8f66fc MT |
845 | if (avctx->framerate.num > 0 && avctx->framerate.den > 0) { |
846 | vseq->num_units_in_tick = avctx->framerate.num; | |
847 | vseq->time_scale = 2 * avctx->framerate.den; | |
081961f8 | 848 | mseq->fixed_frame_rate_flag = 1; |
6e8f66fc MT |
849 | } else { |
850 | vseq->num_units_in_tick = avctx->time_base.num; | |
851 | vseq->time_scale = 2 * avctx->time_base.den; | |
081961f8 | 852 | mseq->fixed_frame_rate_flag = 0; |
6e8f66fc MT |
853 | } |
854 | ||
48e2967c MT |
855 | if (ctx->va_rc_mode == VA_RC_CBR) { |
856 | priv->send_timing_sei = 1; | |
857 | mseq->nal_hrd_parameters_present_flag = 1; | |
858 | ||
859 | mseq->cpb_cnt_minus1 = 0; | |
860 | ||
861 | // Try to scale these to a sensible range so that the | |
862 | // golomb encode of the value is not overlong. | |
863 | mseq->bit_rate_scale = | |
3a9662af | 864 | av_clip_uintp2(av_log2(avctx->bit_rate) - 15 - 6, 4); |
48e2967c | 865 | mseq->bit_rate_value_minus1[0] = |
3a9662af | 866 | (avctx->bit_rate >> mseq->bit_rate_scale + 6) - 1; |
48e2967c MT |
867 | |
868 | mseq->cpb_size_scale = | |
3a9662af | 869 | av_clip_uintp2(av_log2(priv->hrd_params.hrd.buffer_size) - 15 - 4, 4); |
48e2967c | 870 | mseq->cpb_size_value_minus1[0] = |
3a9662af | 871 | (priv->hrd_params.hrd.buffer_size >> mseq->cpb_size_scale + 4) - 1; |
48e2967c MT |
872 | |
873 | // CBR mode isn't actually available here, despite naming. | |
874 | mseq->cbr_flag[0] = 0; | |
875 | ||
876 | mseq->initial_cpb_removal_delay_length_minus1 = 23; | |
877 | mseq->cpb_removal_delay_length_minus1 = 23; | |
878 | mseq->dpb_output_delay_length_minus1 = 7; | |
879 | mseq->time_offset_length = 0; | |
880 | ||
881 | // This calculation can easily overflow 32 bits. | |
882 | mseq->initial_cpb_removal_delay = 90000 * | |
883 | (uint64_t)priv->hrd_params.hrd.initial_buffer_fullness / | |
884 | priv->hrd_params.hrd.buffer_size; | |
885 | ||
886 | mseq->initial_cpb_removal_delay_offset = 0; | |
887 | } else { | |
888 | priv->send_timing_sei = 0; | |
889 | mseq->nal_hrd_parameters_present_flag = 0; | |
890 | } | |
891 | ||
6e8f66fc MT |
892 | vseq->intra_period = ctx->p_per_i * (ctx->b_per_p + 1); |
893 | vseq->intra_idr_period = vseq->intra_period; | |
894 | vseq->ip_period = ctx->b_per_p + 1; | |
2c62fcdf MT |
895 | } |
896 | ||
897 | { | |
898 | vpic->CurrPic.picture_id = VA_INVALID_ID; | |
899 | vpic->CurrPic.flags = VA_PICTURE_H264_INVALID; | |
900 | ||
901 | for (i = 0; i < FF_ARRAY_ELEMS(vpic->ReferenceFrames); i++) { | |
902 | vpic->ReferenceFrames[i].picture_id = VA_INVALID_ID; | |
903 | vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_INVALID; | |
904 | } | |
905 | ||
906 | vpic->coded_buf = VA_INVALID_ID; | |
907 | ||
908 | vpic->pic_parameter_set_id = 0; | |
909 | vpic->seq_parameter_set_id = 0; | |
910 | ||
911 | vpic->num_ref_idx_l0_active_minus1 = 0; | |
912 | vpic->num_ref_idx_l1_active_minus1 = 0; | |
913 | ||
914 | vpic->pic_fields.bits.entropy_coding_mode_flag = | |
915 | ((avctx->profile & 0xff) != 66); | |
916 | vpic->pic_fields.bits.weighted_pred_flag = 0; | |
917 | vpic->pic_fields.bits.weighted_bipred_idc = 0; | |
918 | vpic->pic_fields.bits.transform_8x8_mode_flag = | |
919 | ((avctx->profile & 0xff) >= 100); | |
920 | ||
921 | vpic->pic_init_qp = priv->fixed_qp_idr; | |
922 | } | |
923 | ||
924 | { | |
925 | mseq->profile_idc = avctx->profile & 0xff; | |
926 | ||
927 | if (avctx->profile & FF_PROFILE_H264_CONSTRAINED) | |
928 | mseq->constraint_set1_flag = 1; | |
929 | if (avctx->profile & FF_PROFILE_H264_INTRA) | |
930 | mseq->constraint_set3_flag = 1; | |
931 | } | |
932 | ||
933 | return 0; | |
934 | } | |
935 | ||
936 | static int vaapi_encode_h264_init_picture_params(AVCodecContext *avctx, | |
937 | VAAPIEncodePicture *pic) | |
938 | { | |
939 | VAAPIEncodeContext *ctx = avctx->priv_data; | |
940 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; | |
941 | VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; | |
942 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
943 | int i; | |
944 | ||
945 | if (pic->type == PICTURE_TYPE_IDR) { | |
946 | av_assert0(pic->display_order == pic->encode_order); | |
b51c7c6b MT |
947 | vpic->frame_num = 0; |
948 | priv->next_frame_num = 1; | |
48e2967c | 949 | priv->cpb_delay = 0; |
2c62fcdf | 950 | } else { |
b51c7c6b MT |
951 | vpic->frame_num = priv->next_frame_num; |
952 | if (pic->type != PICTURE_TYPE_B) { | |
953 | // nal_ref_idc != 0 | |
954 | ++priv->next_frame_num; | |
955 | } | |
48e2967c | 956 | ++priv->cpb_delay; |
2c62fcdf | 957 | } |
48e2967c | 958 | priv->dpb_delay = pic->display_order - pic->encode_order + 1; |
2c62fcdf | 959 | |
b51c7c6b | 960 | vpic->frame_num = vpic->frame_num & |
2c62fcdf MT |
961 | ((1 << (4 + vseq->seq_fields.bits.log2_max_frame_num_minus4)) - 1); |
962 | ||
963 | vpic->CurrPic.picture_id = pic->recon_surface; | |
964 | vpic->CurrPic.frame_idx = vpic->frame_num; | |
965 | vpic->CurrPic.flags = 0; | |
966 | vpic->CurrPic.TopFieldOrderCnt = pic->display_order; | |
967 | vpic->CurrPic.BottomFieldOrderCnt = pic->display_order; | |
968 | ||
969 | for (i = 0; i < pic->nb_refs; i++) { | |
970 | VAAPIEncodePicture *ref = pic->refs[i]; | |
b51c7c6b | 971 | av_assert0(ref && ref->encode_order < pic->encode_order); |
2c62fcdf | 972 | vpic->ReferenceFrames[i].picture_id = ref->recon_surface; |
b51c7c6b | 973 | vpic->ReferenceFrames[i].frame_idx = ref->encode_order; |
2c62fcdf MT |
974 | vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_SHORT_TERM_REFERENCE; |
975 | vpic->ReferenceFrames[i].TopFieldOrderCnt = ref->display_order; | |
976 | vpic->ReferenceFrames[i].BottomFieldOrderCnt = ref->display_order; | |
977 | } | |
978 | for (; i < FF_ARRAY_ELEMS(vpic->ReferenceFrames); i++) { | |
979 | vpic->ReferenceFrames[i].picture_id = VA_INVALID_ID; | |
980 | vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_INVALID; | |
981 | } | |
982 | ||
983 | vpic->coded_buf = pic->output_buffer; | |
984 | ||
985 | vpic->pic_fields.bits.idr_pic_flag = (pic->type == PICTURE_TYPE_IDR); | |
986 | vpic->pic_fields.bits.reference_pic_flag = (pic->type != PICTURE_TYPE_B); | |
987 | ||
988 | pic->nb_slices = 1; | |
989 | ||
990 | return 0; | |
991 | } | |
992 | ||
993 | static int vaapi_encode_h264_init_slice_params(AVCodecContext *avctx, | |
994 | VAAPIEncodePicture *pic, | |
995 | VAAPIEncodeSlice *slice) | |
996 | { | |
997 | VAAPIEncodeContext *ctx = avctx->priv_data; | |
998 | VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params; | |
999 | VAEncPictureParameterBufferH264 *vpic = pic->codec_picture_params; | |
1000 | VAEncSliceParameterBufferH264 *vslice = slice->codec_slice_params; | |
1001 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
1002 | VAAPIEncodeH264Slice *pslice; | |
1003 | VAAPIEncodeH264MiscSliceParams *mslice; | |
1004 | int i; | |
1005 | ||
1006 | slice->priv_data = av_mallocz(sizeof(*pslice)); | |
1007 | if (!slice->priv_data) | |
1008 | return AVERROR(ENOMEM); | |
1009 | pslice = slice->priv_data; | |
1010 | mslice = &pslice->misc_slice_params; | |
1011 | ||
1012 | if (pic->type == PICTURE_TYPE_IDR) | |
5c2fb561 | 1013 | mslice->nal_unit_type = H264_NAL_IDR_SLICE; |
2c62fcdf | 1014 | else |
5c2fb561 | 1015 | mslice->nal_unit_type = H264_NAL_SLICE; |
2c62fcdf MT |
1016 | |
1017 | switch (pic->type) { | |
1018 | case PICTURE_TYPE_IDR: | |
1019 | vslice->slice_type = SLICE_TYPE_I; | |
1020 | mslice->nal_ref_idc = 3; | |
1021 | break; | |
1022 | case PICTURE_TYPE_I: | |
1023 | vslice->slice_type = SLICE_TYPE_I; | |
1024 | mslice->nal_ref_idc = 2; | |
1025 | break; | |
1026 | case PICTURE_TYPE_P: | |
1027 | vslice->slice_type = SLICE_TYPE_P; | |
1028 | mslice->nal_ref_idc = 1; | |
1029 | break; | |
1030 | case PICTURE_TYPE_B: | |
1031 | vslice->slice_type = SLICE_TYPE_B; | |
1032 | mslice->nal_ref_idc = 0; | |
1033 | break; | |
1034 | default: | |
1035 | av_assert0(0 && "invalid picture type"); | |
1036 | } | |
1037 | ||
1038 | // Only one slice per frame. | |
1039 | vslice->macroblock_address = 0; | |
1040 | vslice->num_macroblocks = priv->mb_width * priv->mb_height; | |
1041 | ||
1042 | vslice->macroblock_info = VA_INVALID_ID; | |
1043 | ||
1044 | vslice->pic_parameter_set_id = vpic->pic_parameter_set_id; | |
1045 | vslice->idr_pic_id = priv->idr_pic_count++; | |
1046 | ||
1047 | vslice->pic_order_cnt_lsb = pic->display_order & | |
1048 | ((1 << (4 + vseq->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4)) - 1); | |
1049 | ||
1050 | for (i = 0; i < FF_ARRAY_ELEMS(vslice->RefPicList0); i++) { | |
1051 | vslice->RefPicList0[i].picture_id = VA_INVALID_ID; | |
1052 | vslice->RefPicList0[i].flags = VA_PICTURE_H264_INVALID; | |
1053 | vslice->RefPicList1[i].picture_id = VA_INVALID_ID; | |
1054 | vslice->RefPicList1[i].flags = VA_PICTURE_H264_INVALID; | |
1055 | } | |
1056 | ||
1057 | av_assert0(pic->nb_refs <= 2); | |
1058 | if (pic->nb_refs >= 1) { | |
41ed7ab4 | 1059 | // Backward reference for P- or B-frame. |
2c62fcdf MT |
1060 | av_assert0(pic->type == PICTURE_TYPE_P || |
1061 | pic->type == PICTURE_TYPE_B); | |
1062 | ||
1063 | vslice->num_ref_idx_l0_active_minus1 = 0; | |
1064 | vslice->RefPicList0[0] = vpic->ReferenceFrames[0]; | |
1065 | } | |
1066 | if (pic->nb_refs >= 2) { | |
41ed7ab4 | 1067 | // Forward reference for B-frame. |
2c62fcdf MT |
1068 | av_assert0(pic->type == PICTURE_TYPE_B); |
1069 | ||
1070 | vslice->num_ref_idx_l1_active_minus1 = 0; | |
1071 | vslice->RefPicList1[0] = vpic->ReferenceFrames[1]; | |
1072 | } | |
1073 | ||
1074 | if (pic->type == PICTURE_TYPE_B) | |
1075 | vslice->slice_qp_delta = priv->fixed_qp_b - vpic->pic_init_qp; | |
1076 | else if (pic->type == PICTURE_TYPE_P) | |
1077 | vslice->slice_qp_delta = priv->fixed_qp_p - vpic->pic_init_qp; | |
1078 | else | |
1079 | vslice->slice_qp_delta = priv->fixed_qp_idr - vpic->pic_init_qp; | |
1080 | ||
1081 | vslice->direct_spatial_mv_pred_flag = 1; | |
1082 | ||
1083 | return 0; | |
1084 | } | |
1085 | ||
6e8f66fc MT |
1086 | static av_cold int vaapi_encode_h264_init_constant_bitrate(AVCodecContext *avctx) |
1087 | { | |
1088 | VAAPIEncodeContext *ctx = avctx->priv_data; | |
1089 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
1090 | int hrd_buffer_size; | |
1091 | int hrd_initial_buffer_fullness; | |
1092 | ||
1093 | if (avctx->rc_buffer_size) | |
1094 | hrd_buffer_size = avctx->rc_buffer_size; | |
1095 | else | |
1096 | hrd_buffer_size = avctx->bit_rate; | |
1097 | if (avctx->rc_initial_buffer_occupancy) | |
1098 | hrd_initial_buffer_fullness = avctx->rc_initial_buffer_occupancy; | |
1099 | else | |
1100 | hrd_initial_buffer_fullness = hrd_buffer_size * 3 / 4; | |
1101 | ||
1102 | priv->rc_params.misc.type = VAEncMiscParameterTypeRateControl; | |
1103 | priv->rc_params.rc = (VAEncMiscParameterRateControl) { | |
1104 | .bits_per_second = avctx->bit_rate, | |
1105 | .target_percentage = 66, | |
1106 | .window_size = 1000, | |
1107 | .initial_qp = (avctx->qmax >= 0 ? avctx->qmax : 40), | |
1108 | .min_qp = (avctx->qmin >= 0 ? avctx->qmin : 18), | |
1109 | .basic_unit_size = 0, | |
1110 | }; | |
1111 | ctx->global_params[ctx->nb_global_params] = | |
1112 | &priv->rc_params.misc; | |
1113 | ctx->global_params_size[ctx->nb_global_params++] = | |
1114 | sizeof(priv->rc_params); | |
1115 | ||
1116 | priv->hrd_params.misc.type = VAEncMiscParameterTypeHRD; | |
1117 | priv->hrd_params.hrd = (VAEncMiscParameterHRD) { | |
1118 | .initial_buffer_fullness = hrd_initial_buffer_fullness, | |
1119 | .buffer_size = hrd_buffer_size, | |
1120 | }; | |
1121 | ctx->global_params[ctx->nb_global_params] = | |
1122 | &priv->hrd_params.misc; | |
1123 | ctx->global_params_size[ctx->nb_global_params++] = | |
1124 | sizeof(priv->hrd_params); | |
1125 | ||
1126 | // These still need to be set for pic_init_qp/slice_qp_delta. | |
1127 | priv->fixed_qp_idr = 26; | |
1128 | priv->fixed_qp_p = 26; | |
1129 | priv->fixed_qp_b = 26; | |
1130 | ||
1131 | av_log(avctx, AV_LOG_DEBUG, "Using constant-bitrate = %d bps.\n", | |
1132 | avctx->bit_rate); | |
1133 | return 0; | |
1134 | } | |
1135 | ||
1136 | static av_cold int vaapi_encode_h264_init_fixed_qp(AVCodecContext *avctx) | |
1137 | { | |
1138 | VAAPIEncodeContext *ctx = avctx->priv_data; | |
1139 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
9629701c | 1140 | VAAPIEncodeH264Options *opt = ctx->codec_options; |
6e8f66fc | 1141 | |
9629701c | 1142 | priv->fixed_qp_p = opt->qp; |
6e8f66fc MT |
1143 | if (avctx->i_quant_factor > 0.0) |
1144 | priv->fixed_qp_idr = (int)((priv->fixed_qp_p * avctx->i_quant_factor + | |
1145 | avctx->i_quant_offset) + 0.5); | |
1146 | else | |
1147 | priv->fixed_qp_idr = priv->fixed_qp_p; | |
1148 | if (avctx->b_quant_factor > 0.0) | |
1149 | priv->fixed_qp_b = (int)((priv->fixed_qp_p * avctx->b_quant_factor + | |
1150 | avctx->b_quant_offset) + 0.5); | |
1151 | else | |
1152 | priv->fixed_qp_b = priv->fixed_qp_p; | |
1153 | ||
1154 | av_log(avctx, AV_LOG_DEBUG, "Using fixed QP = " | |
41ed7ab4 | 1155 | "%d / %d / %d for IDR- / P- / B-frames.\n", |
6e8f66fc MT |
1156 | priv->fixed_qp_idr, priv->fixed_qp_p, priv->fixed_qp_b); |
1157 | return 0; | |
1158 | } | |
1159 | ||
2c62fcdf MT |
1160 | static av_cold int vaapi_encode_h264_init_internal(AVCodecContext *avctx) |
1161 | { | |
f6b85523 MT |
1162 | static const VAConfigAttrib default_config_attributes[] = { |
1163 | { .type = VAConfigAttribRTFormat, | |
1164 | .value = VA_RT_FORMAT_YUV420 }, | |
f6b85523 MT |
1165 | { .type = VAConfigAttribEncPackedHeaders, |
1166 | .value = (VA_ENC_PACKED_HEADER_SEQUENCE | | |
1167 | VA_ENC_PACKED_HEADER_SLICE) }, | |
1168 | }; | |
1169 | ||
2c62fcdf MT |
1170 | VAAPIEncodeContext *ctx = avctx->priv_data; |
1171 | VAAPIEncodeH264Context *priv = ctx->priv_data; | |
fcf536b1 | 1172 | VAAPIEncodeH264Options *opt = ctx->codec_options; |
6e8f66fc | 1173 | int i, err; |
2c62fcdf MT |
1174 | |
1175 | switch (avctx->profile) { | |
1176 | case FF_PROFILE_H264_CONSTRAINED_BASELINE: | |
1177 | ctx->va_profile = VAProfileH264ConstrainedBaseline; | |
1178 | break; | |
1179 | case FF_PROFILE_H264_BASELINE: | |
1180 | ctx->va_profile = VAProfileH264Baseline; | |
1181 | break; | |
1182 | case FF_PROFILE_H264_MAIN: | |
1183 | ctx->va_profile = VAProfileH264Main; | |
1184 | break; | |
1185 | case FF_PROFILE_H264_EXTENDED: | |
1186 | av_log(avctx, AV_LOG_ERROR, "H.264 extended profile " | |
1187 | "is not supported.\n"); | |
1188 | return AVERROR_PATCHWELCOME; | |
1189 | case FF_PROFILE_UNKNOWN: | |
1190 | case FF_PROFILE_H264_HIGH: | |
1191 | ctx->va_profile = VAProfileH264High; | |
1192 | break; | |
1193 | case FF_PROFILE_H264_HIGH_10: | |
1194 | case FF_PROFILE_H264_HIGH_10_INTRA: | |
1195 | av_log(avctx, AV_LOG_ERROR, "H.264 10-bit profiles " | |
1196 | "are not supported.\n"); | |
1197 | return AVERROR_PATCHWELCOME; | |
1198 | case FF_PROFILE_H264_HIGH_422: | |
1199 | case FF_PROFILE_H264_HIGH_422_INTRA: | |
1200 | case FF_PROFILE_H264_HIGH_444: | |
1201 | case FF_PROFILE_H264_HIGH_444_PREDICTIVE: | |
1202 | case FF_PROFILE_H264_HIGH_444_INTRA: | |
1203 | case FF_PROFILE_H264_CAVLC_444: | |
1204 | av_log(avctx, AV_LOG_ERROR, "H.264 non-4:2:0 profiles " | |
1205 | "are not supported.\n"); | |
1206 | return AVERROR_PATCHWELCOME; | |
1207 | default: | |
1208 | av_log(avctx, AV_LOG_ERROR, "Unknown H.264 profile %d.\n", | |
1209 | avctx->profile); | |
1210 | return AVERROR(EINVAL); | |
1211 | } | |
a86aa160 MT |
1212 | if (opt->low_power) { |
1213 | #if VA_CHECK_VERSION(0, 39, 1) | |
1214 | ctx->va_entrypoint = VAEntrypointEncSliceLP; | |
1215 | #else | |
1216 | av_log(avctx, AV_LOG_ERROR, "Low-power encoding is not " | |
1217 | "supported with this VAAPI version.\n"); | |
1218 | return AVERROR(EINVAL); | |
1219 | #endif | |
1220 | } else { | |
1221 | ctx->va_entrypoint = VAEntrypointEncSlice; | |
1222 | } | |
2c62fcdf | 1223 | |
2c62fcdf MT |
1224 | ctx->input_width = avctx->width; |
1225 | ctx->input_height = avctx->height; | |
1226 | ctx->aligned_width = FFALIGN(ctx->input_width, 16); | |
1227 | ctx->aligned_height = FFALIGN(ctx->input_height, 16); | |
1228 | priv->mb_width = ctx->aligned_width / 16; | |
1229 | priv->mb_height = ctx->aligned_height / 16; | |
1230 | ||
f6b85523 MT |
1231 | for (i = 0; i < FF_ARRAY_ELEMS(default_config_attributes); i++) { |
1232 | ctx->config_attributes[ctx->nb_config_attributes++] = | |
1233 | default_config_attributes[i]; | |
1234 | } | |
1235 | ||
6e8f66fc MT |
1236 | if (avctx->bit_rate > 0) { |
1237 | ctx->va_rc_mode = VA_RC_CBR; | |
1238 | err = vaapi_encode_h264_init_constant_bitrate(avctx); | |
1239 | } else { | |
1240 | ctx->va_rc_mode = VA_RC_CQP; | |
1241 | err = vaapi_encode_h264_init_fixed_qp(avctx); | |
1242 | } | |
1243 | if (err < 0) | |
1244 | return err; | |
1245 | ||
1246 | ctx->config_attributes[ctx->nb_config_attributes++] = (VAConfigAttrib) { | |
1247 | .type = VAConfigAttribRateControl, | |
1248 | .value = ctx->va_rc_mode, | |
1249 | }; | |
2c62fcdf | 1250 | |
fcf536b1 MT |
1251 | if (opt->quality > 0) { |
1252 | #if VA_CHECK_VERSION(0, 36, 0) | |
1253 | priv->quality_params.misc.type = | |
1254 | VAEncMiscParameterTypeQualityLevel; | |
1255 | priv->quality_params.quality.quality_level = opt->quality; | |
1256 | ||
1257 | ctx->global_params[ctx->nb_global_params] = | |
1258 | &priv->quality_params.misc; | |
1259 | ctx->global_params_size[ctx->nb_global_params++] = | |
1260 | sizeof(priv->quality_params); | |
1261 | #else | |
1262 | av_log(avctx, AV_LOG_WARNING, "The encode quality option is not " | |
1263 | "supported with this VAAPI version.\n"); | |
1264 | #endif | |
1265 | } | |
1266 | ||
2c62fcdf MT |
1267 | ctx->nb_recon_frames = 20; |
1268 | ||
1269 | return 0; | |
1270 | } | |
1271 | ||
1272 | static VAAPIEncodeType vaapi_encode_type_h264 = { | |
1273 | .priv_data_size = sizeof(VAAPIEncodeH264Context), | |
1274 | ||
1275 | .init = &vaapi_encode_h264_init_internal, | |
1276 | ||
1277 | .sequence_params_size = sizeof(VAEncSequenceParameterBufferH264), | |
1278 | .init_sequence_params = &vaapi_encode_h264_init_sequence_params, | |
1279 | ||
1280 | .picture_params_size = sizeof(VAEncPictureParameterBufferH264), | |
1281 | .init_picture_params = &vaapi_encode_h264_init_picture_params, | |
1282 | ||
1283 | .slice_params_size = sizeof(VAEncSliceParameterBufferH264), | |
1284 | .init_slice_params = &vaapi_encode_h264_init_slice_params, | |
1285 | ||
1286 | .sequence_header_type = VAEncPackedHeaderSequence, | |
1287 | .write_sequence_header = &vaapi_encode_h264_write_sequence_header, | |
1288 | ||
1289 | .slice_header_type = VAEncPackedHeaderH264_Slice, | |
1290 | .write_slice_header = &vaapi_encode_h264_write_slice_header, | |
48e2967c MT |
1291 | |
1292 | .write_extra_header = &vaapi_encode_h264_write_extra_header, | |
2c62fcdf MT |
1293 | }; |
1294 | ||
1295 | static av_cold int vaapi_encode_h264_init(AVCodecContext *avctx) | |
1296 | { | |
1297 | return ff_vaapi_encode_init(avctx, &vaapi_encode_type_h264); | |
1298 | } | |
1299 | ||
9629701c MT |
1300 | #define OFFSET(x) (offsetof(VAAPIEncodeContext, codec_options_data) + \ |
1301 | offsetof(VAAPIEncodeH264Options, x)) | |
1302 | #define FLAGS (AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM) | |
1303 | static const AVOption vaapi_encode_h264_options[] = { | |
41ed7ab4 | 1304 | { "qp", "Constant QP (for P-frames; scaled by qfactor/qoffset for I/B)", |
9629701c | 1305 | OFFSET(qp), AV_OPT_TYPE_INT, { .i64 = 20 }, 0, 52, FLAGS }, |
fcf536b1 | 1306 | { "quality", "Set encode quality (trades off against speed, higher is faster)", |
a86aa160 MT |
1307 | OFFSET(quality), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 8, FLAGS }, |
1308 | { "low_power", "Use low-power encoding mode (experimental: only supported " | |
1309 | "on some platforms, does not support all features)", | |
1310 | OFFSET(low_power), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS }, | |
9629701c MT |
1311 | { NULL }, |
1312 | }; | |
1313 | ||
2c62fcdf MT |
1314 | static const AVCodecDefault vaapi_encode_h264_defaults[] = { |
1315 | { "profile", "100" }, | |
1316 | { "level", "51" }, | |
1317 | { "b", "0" }, | |
1318 | { "bf", "2" }, | |
1319 | { "g", "120" }, | |
2c62fcdf MT |
1320 | { "i_qfactor", "1.0" }, |
1321 | { "i_qoffset", "0.0" }, | |
1322 | { "b_qfactor", "1.2" }, | |
1323 | { "b_qoffset", "0.0" }, | |
1324 | { NULL }, | |
1325 | }; | |
1326 | ||
1327 | static const AVClass vaapi_encode_h264_class = { | |
1328 | .class_name = "h264_vaapi", | |
1329 | .item_name = av_default_item_name, | |
9629701c | 1330 | .option = vaapi_encode_h264_options, |
2c62fcdf MT |
1331 | .version = LIBAVUTIL_VERSION_INT, |
1332 | }; | |
1333 | ||
1334 | AVCodec ff_h264_vaapi_encoder = { | |
1335 | .name = "h264_vaapi", | |
1336 | .long_name = NULL_IF_CONFIG_SMALL("H.264/AVC (VAAPI)"), | |
1337 | .type = AVMEDIA_TYPE_VIDEO, | |
1338 | .id = AV_CODEC_ID_H264, | |
9629701c MT |
1339 | .priv_data_size = (sizeof(VAAPIEncodeContext) + |
1340 | sizeof(VAAPIEncodeH264Options)), | |
2c62fcdf MT |
1341 | .init = &vaapi_encode_h264_init, |
1342 | .encode2 = &ff_vaapi_encode2, | |
1343 | .close = &ff_vaapi_encode_close, | |
1344 | .priv_class = &vaapi_encode_h264_class, | |
1345 | .capabilities = AV_CODEC_CAP_DELAY, | |
1346 | .defaults = vaapi_encode_h264_defaults, | |
1347 | .pix_fmts = (const enum AVPixelFormat[]) { | |
1348 | AV_PIX_FMT_VAAPI, | |
1349 | AV_PIX_FMT_NONE, | |
1350 | }, | |
1351 | }; |