8690a8565bb4fec9f525da7a6e503e5140758ab1
[ffmpeg.git] / libavcodec / vaapi_encode_h264.c
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"
28 #include "h264.h"
29 #include "internal.h"
30 #include "vaapi_encode.h"
31 #include "vaapi_encode_h26x.h"
32
33 enum {
34     SLICE_TYPE_P  = 0,
35     SLICE_TYPE_B  = 1,
36     SLICE_TYPE_I  = 2,
37     SLICE_TYPE_SP = 3,
38     SLICE_TYPE_SI = 4,
39 };
40
41 // This structure contains all possibly-useful per-sequence syntax elements
42 // which are not already contained in the various VAAPI structures.
43 typedef struct VAAPIEncodeH264MiscSequenceParams {
44     unsigned int profile_idc;
45     char constraint_set0_flag;
46     char constraint_set1_flag;
47     char constraint_set2_flag;
48     char constraint_set3_flag;
49     char constraint_set4_flag;
50     char constraint_set5_flag;
51
52     char separate_colour_plane_flag;
53     char qpprime_y_zero_transform_bypass_flag;
54
55     char gaps_in_frame_num_allowed_flag;
56     char delta_pic_order_always_zero_flag;
57     char bottom_field_pic_order_in_frame_present_flag;
58
59     unsigned int num_slice_groups_minus1;
60     unsigned int slice_group_map_type;
61
62     int pic_init_qs_minus26;
63
64     char vui_parameters_present_flag;
65 } VAAPIEncodeH264MiscSequenceParams;
66
67 // This structure contains all possibly-useful per-slice syntax elements
68 // which are not already contained in the various VAAPI structures.
69 typedef struct VAAPIEncodeH264MiscSliceParams {
70     unsigned int nal_unit_type;
71     unsigned int nal_ref_idc;
72
73     unsigned int colour_plane_id;
74     char field_pic_flag;
75     char bottom_field_flag;
76
77     unsigned int redundant_pic_cnt;
78
79     char sp_for_switch_flag;
80     int slice_qs_delta;
81
82     char ref_pic_list_modification_flag_l0;
83     char ref_pic_list_modification_flag_l1;
84
85     char no_output_of_prior_pics_flag;
86     char long_term_reference_flag;
87     char adaptive_ref_pic_marking_mode_flag;
88 } VAAPIEncodeH264MiscSliceParams;
89
90 typedef struct VAAPIEncodeH264Slice {
91     VAAPIEncodeH264MiscSliceParams misc_slice_params;
92 } VAAPIEncodeH264Slice;
93
94 typedef struct VAAPIEncodeH264Context {
95     VAAPIEncodeH264MiscSequenceParams misc_sequence_params;
96
97     int mb_width;
98     int mb_height;
99
100     int fixed_qp_idr;
101     int fixed_qp_p;
102     int fixed_qp_b;
103
104     int64_t idr_pic_count;
105     int64_t last_idr_frame;
106
107     // Rate control configuration.
108     struct {
109         VAEncMiscParameterBuffer misc;
110         VAEncMiscParameterRateControl rc;
111     } rc_params;
112     struct {
113         VAEncMiscParameterBuffer misc;
114         VAEncMiscParameterHRD hrd;
115     } hrd_params;
116
117 #if VA_CHECK_VERSION(0, 36, 0)
118     // Speed-quality tradeoff setting.
119     struct {
120         VAEncMiscParameterBuffer misc;
121         VAEncMiscParameterBufferQualityLevel quality;
122     } quality_params;
123 #endif
124 } VAAPIEncodeH264Context;
125
126 typedef struct VAAPIEncodeH264Options {
127     int qp;
128     int quality;
129 } VAAPIEncodeH264Options;
130
131
132 #define vseq_var(name)     vseq->name, name
133 #define vseq_field(name)   vseq->seq_fields.bits.name, name
134 #define vpic_var(name)     vpic->name, name
135 #define vpic_field(name)   vpic->pic_fields.bits.name, name
136 #define vslice_var(name)   vslice->name, name
137 #define vslice_field(name) vslice->slice_fields.bits.name, name
138 #define mseq_var(name)     mseq->name, name
139 #define mslice_var(name)   mslice->name, name
140
141 static void vaapi_encode_h264_write_nal_header(PutBitContext *pbc,
142                                                int nal_unit_type, int nal_ref_idc)
143 {
144     u(1, 0, forbidden_zero_bit);
145     u(2, nal_ref_idc, nal_ref_idc);
146     u(5, nal_unit_type, nal_unit_type);
147 }
148
149 static void vaapi_encode_h264_write_trailing_rbsp(PutBitContext *pbc)
150 {
151     u(1, 1, rbsp_stop_one_bit);
152     while (put_bits_count(pbc) & 7)
153         u(1, 0, rbsp_alignment_zero_bit);
154 }
155
156 static void vaapi_encode_h264_write_sps(PutBitContext *pbc,
157                                         VAAPIEncodeContext *ctx)
158 {
159     VAEncSequenceParameterBufferH264  *vseq = ctx->codec_sequence_params;
160     VAAPIEncodeH264Context            *priv = ctx->priv_data;
161     VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params;
162     int i;
163
164     vaapi_encode_h264_write_nal_header(pbc, NAL_SPS, 3);
165
166     u(8, mseq_var(profile_idc));
167     u(1, mseq_var(constraint_set0_flag));
168     u(1, mseq_var(constraint_set1_flag));
169     u(1, mseq_var(constraint_set2_flag));
170     u(1, mseq_var(constraint_set3_flag));
171     u(1, mseq_var(constraint_set4_flag));
172     u(1, mseq_var(constraint_set5_flag));
173     u(2, 0, reserved_zero_2bits);
174
175     u(8, vseq_var(level_idc));
176
177     ue(vseq_var(seq_parameter_set_id));
178
179     if (mseq->profile_idc == 100 || mseq->profile_idc == 110 ||
180         mseq->profile_idc == 122 || mseq->profile_idc == 244 ||
181         mseq->profile_idc ==  44 || mseq->profile_idc ==  83 ||
182         mseq->profile_idc ==  86 || mseq->profile_idc == 118 ||
183         mseq->profile_idc == 128 || mseq->profile_idc == 138) {
184         ue(vseq_field(chroma_format_idc));
185
186         if (vseq->seq_fields.bits.chroma_format_idc == 3)
187             u(1, mseq_var(separate_colour_plane_flag));
188
189         ue(vseq_var(bit_depth_luma_minus8));
190         ue(vseq_var(bit_depth_chroma_minus8));
191
192         u(1, mseq_var(qpprime_y_zero_transform_bypass_flag));
193
194         u(1, vseq_field(seq_scaling_matrix_present_flag));
195         if (vseq->seq_fields.bits.seq_scaling_matrix_present_flag) {
196             av_assert0(0 && "scaling matrices not supported");
197         }
198     }
199
200     ue(vseq_field(log2_max_frame_num_minus4));
201     ue(vseq_field(pic_order_cnt_type));
202
203     if (vseq->seq_fields.bits.pic_order_cnt_type == 0) {
204         ue(vseq_field(log2_max_pic_order_cnt_lsb_minus4));
205     } else if (vseq->seq_fields.bits.pic_order_cnt_type == 1) {
206         u(1, mseq_var(delta_pic_order_always_zero_flag));
207         se(vseq_var(offset_for_non_ref_pic));
208         se(vseq_var(offset_for_top_to_bottom_field));
209         ue(vseq_var(num_ref_frames_in_pic_order_cnt_cycle));
210
211         for (i = 0; i < vseq->num_ref_frames_in_pic_order_cnt_cycle; i++)
212             se(vseq_var(offset_for_ref_frame[i]));
213     }
214
215     ue(vseq_var(max_num_ref_frames));
216     u(1, mseq_var(gaps_in_frame_num_allowed_flag));
217
218     ue(vseq->picture_width_in_mbs  - 1, pic_width_in_mbs_minus1);
219     ue(vseq->picture_height_in_mbs - 1, pic_height_in_mbs_minus1);
220
221     u(1, vseq_field(frame_mbs_only_flag));
222     if (!vseq->seq_fields.bits.frame_mbs_only_flag)
223         u(1, vseq_field(mb_adaptive_frame_field_flag));
224
225     u(1, vseq_field(direct_8x8_inference_flag));
226
227     u(1, vseq_var(frame_cropping_flag));
228     if (vseq->frame_cropping_flag) {
229         ue(vseq_var(frame_crop_left_offset));
230         ue(vseq_var(frame_crop_right_offset));
231         ue(vseq_var(frame_crop_top_offset));
232         ue(vseq_var(frame_crop_bottom_offset));
233     }
234
235     u(1, mseq_var(vui_parameters_present_flag));
236
237     vaapi_encode_h264_write_trailing_rbsp(pbc);
238 }
239
240 static void vaapi_encode_h264_write_pps(PutBitContext *pbc,
241                                         VAAPIEncodeContext *ctx)
242 {
243     VAEncPictureParameterBufferH264   *vpic = ctx->codec_picture_params;
244     VAAPIEncodeH264Context            *priv = ctx->priv_data;
245     VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params;
246
247     vaapi_encode_h264_write_nal_header(pbc, NAL_PPS, 3);
248
249     ue(vpic_var(pic_parameter_set_id));
250     ue(vpic_var(seq_parameter_set_id));
251
252     u(1, vpic_field(entropy_coding_mode_flag));
253     u(1, mseq_var(bottom_field_pic_order_in_frame_present_flag));
254
255     ue(mseq_var(num_slice_groups_minus1));
256     if (mseq->num_slice_groups_minus1 > 0) {
257         ue(mseq_var(slice_group_map_type));
258         av_assert0(0 && "slice groups not supported");
259     }
260
261     ue(vpic_var(num_ref_idx_l0_active_minus1));
262     ue(vpic_var(num_ref_idx_l1_active_minus1));
263
264     u(1, vpic_field(weighted_pred_flag));
265     u(2, vpic_field(weighted_bipred_idc));
266
267     se(vpic->pic_init_qp - 26, pic_init_qp_minus26);
268     se(mseq_var(pic_init_qs_minus26));
269     se(vpic_var(chroma_qp_index_offset));
270
271     u(1, vpic_field(deblocking_filter_control_present_flag));
272     u(1, vpic_field(constrained_intra_pred_flag));
273     u(1, vpic_field(redundant_pic_cnt_present_flag));
274     u(1, vpic_field(transform_8x8_mode_flag));
275
276     u(1, vpic_field(pic_scaling_matrix_present_flag));
277     if (vpic->pic_fields.bits.pic_scaling_matrix_present_flag) {
278         av_assert0(0 && "scaling matrices not supported");
279     }
280
281     se(vpic_var(second_chroma_qp_index_offset));
282
283     vaapi_encode_h264_write_trailing_rbsp(pbc);
284 }
285
286 static void vaapi_encode_h264_write_slice_header2(PutBitContext *pbc,
287                                                   VAAPIEncodeContext *ctx,
288                                                   VAAPIEncodePicture *pic,
289                                                   VAAPIEncodeSlice *slice)
290 {
291     VAEncSequenceParameterBufferH264  *vseq = ctx->codec_sequence_params;
292     VAEncPictureParameterBufferH264   *vpic = pic->codec_picture_params;
293     VAEncSliceParameterBufferH264   *vslice = slice->codec_slice_params;
294     VAAPIEncodeH264Context            *priv = ctx->priv_data;
295     VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params;
296     VAAPIEncodeH264Slice            *pslice = slice->priv_data;
297     VAAPIEncodeH264MiscSliceParams  *mslice = &pslice->misc_slice_params;
298
299     vaapi_encode_h264_write_nal_header(pbc, mslice->nal_unit_type,
300                                        mslice->nal_ref_idc);
301
302     ue(vslice->macroblock_address, first_mb_in_slice);
303     ue(vslice_var(slice_type));
304     ue(vpic_var(pic_parameter_set_id));
305
306     if (mseq->separate_colour_plane_flag) {
307         u(2, mslice_var(colour_plane_id));
308     }
309
310     u(4 + vseq->seq_fields.bits.log2_max_frame_num_minus4,
311       (vpic->frame_num &
312        ((1 << (4 + vseq->seq_fields.bits.log2_max_frame_num_minus4)) - 1)),
313       frame_num);
314
315     if (!vseq->seq_fields.bits.frame_mbs_only_flag) {
316         u(1, mslice_var(field_pic_flag));
317         if (mslice->field_pic_flag)
318             u(1, mslice_var(bottom_field_flag));
319     }
320
321     if (vpic->pic_fields.bits.idr_pic_flag) {
322         ue(vslice_var(idr_pic_id));
323     }
324
325     if (vseq->seq_fields.bits.pic_order_cnt_type == 0) {
326         u(4 + vseq->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4,
327           vslice_var(pic_order_cnt_lsb));
328         if (mseq->bottom_field_pic_order_in_frame_present_flag &&
329             !mslice->field_pic_flag) {
330             se(vslice_var(delta_pic_order_cnt_bottom));
331         }
332     }
333
334     if (vseq->seq_fields.bits.pic_order_cnt_type == 1 &&
335         !vseq->seq_fields.bits.delta_pic_order_always_zero_flag) {
336         se(vslice_var(delta_pic_order_cnt[0]));
337         if (mseq->bottom_field_pic_order_in_frame_present_flag &&
338             !mslice->field_pic_flag) {
339             se(vslice_var(delta_pic_order_cnt[1]));
340         }
341     }
342
343     if (vpic->pic_fields.bits.redundant_pic_cnt_present_flag) {
344         ue(mslice_var(redundant_pic_cnt));
345     }
346
347     if (vslice->slice_type == SLICE_TYPE_B) {
348         u(1, vslice_var(direct_spatial_mv_pred_flag));
349     }
350
351     if (vslice->slice_type == SLICE_TYPE_P ||
352         vslice->slice_type == SLICE_TYPE_SP ||
353         vslice->slice_type == SLICE_TYPE_B) {
354         u(1, vslice_var(num_ref_idx_active_override_flag));
355         if (vslice->num_ref_idx_active_override_flag) {
356             ue(vslice_var(num_ref_idx_l0_active_minus1));
357             if (vslice->slice_type == SLICE_TYPE_B)
358                 ue(vslice_var(num_ref_idx_l1_active_minus1));
359         }
360     }
361
362     if (mslice->nal_unit_type == 20 || mslice->nal_unit_type == 21) {
363         av_assert0(0 && "no MVC support");
364     } else {
365         if (vslice->slice_type % 5 != 2 && vslice->slice_type % 5 != 4) {
366             u(1, mslice_var(ref_pic_list_modification_flag_l0));
367             if (mslice->ref_pic_list_modification_flag_l0) {
368                 av_assert0(0 && "ref pic list modification");
369             }
370         }
371         if (vslice->slice_type % 5 == 1) {
372             u(1, mslice_var(ref_pic_list_modification_flag_l1));
373             if (mslice->ref_pic_list_modification_flag_l1) {
374                 av_assert0(0 && "ref pic list modification");
375             }
376         }
377     }
378
379     if ((vpic->pic_fields.bits.weighted_pred_flag &&
380          (vslice->slice_type == SLICE_TYPE_P ||
381           vslice->slice_type == SLICE_TYPE_SP)) ||
382         (vpic->pic_fields.bits.weighted_bipred_idc == 1 &&
383          vslice->slice_type == SLICE_TYPE_B)) {
384         av_assert0(0 && "prediction weights not supported");
385     }
386
387     av_assert0(mslice->nal_ref_idc > 0 ==
388                vpic->pic_fields.bits.reference_pic_flag);
389     if (mslice->nal_ref_idc != 0) {
390         if (vpic->pic_fields.bits.idr_pic_flag) {
391             u(1, mslice_var(no_output_of_prior_pics_flag));
392             u(1, mslice_var(long_term_reference_flag));
393         } else {
394             u(1, mslice_var(adaptive_ref_pic_marking_mode_flag));
395             if (mslice->adaptive_ref_pic_marking_mode_flag) {
396                 av_assert0(0 && "MMCOs not supported");
397             }
398         }
399     }
400
401     if (vpic->pic_fields.bits.entropy_coding_mode_flag &&
402         vslice->slice_type != SLICE_TYPE_I &&
403         vslice->slice_type != SLICE_TYPE_SI) {
404         ue(vslice_var(cabac_init_idc));
405     }
406
407     se(vslice_var(slice_qp_delta));
408     if (vslice->slice_type == SLICE_TYPE_SP ||
409         vslice->slice_type == SLICE_TYPE_SI) {
410         if (vslice->slice_type == SLICE_TYPE_SP)
411             u(1, mslice_var(sp_for_switch_flag));
412         se(mslice_var(slice_qs_delta));
413     }
414
415     if (vpic->pic_fields.bits.deblocking_filter_control_present_flag) {
416         ue(vslice_var(disable_deblocking_filter_idc));
417         if (vslice->disable_deblocking_filter_idc != 1) {
418             se(vslice_var(slice_alpha_c0_offset_div2));
419             se(vslice_var(slice_beta_offset_div2));
420         }
421     }
422
423     if (mseq->num_slice_groups_minus1 > 0 &&
424         mseq->slice_group_map_type >= 3 && mseq->slice_group_map_type <= 5) {
425         av_assert0(0 && "slice groups not supported");
426     }
427
428     // No alignment - this need not be a byte boundary.
429 }
430
431 static int vaapi_encode_h264_write_sequence_header(AVCodecContext *avctx,
432                                                    char *data, size_t *data_len)
433 {
434     VAAPIEncodeContext *ctx = avctx->priv_data;
435     PutBitContext pbc;
436     char tmp[256];
437     int err;
438     size_t nal_len, bit_len, bit_pos, next_len;
439
440     bit_len = *data_len;
441     bit_pos = 0;
442
443     init_put_bits(&pbc, tmp, sizeof(tmp));
444     vaapi_encode_h264_write_sps(&pbc, ctx);
445     nal_len = put_bits_count(&pbc);
446     flush_put_bits(&pbc);
447
448     next_len = bit_len - bit_pos;
449     err = ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data + bit_pos / 8,
450                                                        &next_len,
451                                                        tmp, nal_len);
452     if (err < 0)
453         return err;
454     bit_pos += next_len;
455
456     init_put_bits(&pbc, tmp, sizeof(tmp));
457     vaapi_encode_h264_write_pps(&pbc, ctx);
458     nal_len = put_bits_count(&pbc);
459     flush_put_bits(&pbc);
460
461     next_len = bit_len - bit_pos;
462     err = ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data + bit_pos / 8,
463                                                        &next_len,
464                                                        tmp, nal_len);
465     if (err < 0)
466         return err;
467     bit_pos += next_len;
468
469     *data_len = bit_pos;
470     return 0;
471 }
472
473 static int vaapi_encode_h264_write_slice_header(AVCodecContext *avctx,
474                                                 VAAPIEncodePicture *pic,
475                                                 VAAPIEncodeSlice *slice,
476                                                 char *data, size_t *data_len)
477 {
478     VAAPIEncodeContext *ctx = avctx->priv_data;
479     PutBitContext pbc;
480     char tmp[256];
481     size_t header_len;
482
483     init_put_bits(&pbc, tmp, sizeof(tmp));
484     vaapi_encode_h264_write_slice_header2(&pbc, ctx, pic, slice);
485     header_len = put_bits_count(&pbc);
486     flush_put_bits(&pbc);
487
488     return ff_vaapi_encode_h26x_nal_unit_to_byte_stream(data, data_len,
489                                                         tmp, header_len);
490 }
491
492 static int vaapi_encode_h264_init_sequence_params(AVCodecContext *avctx)
493 {
494     VAAPIEncodeContext                 *ctx = avctx->priv_data;
495     VAEncSequenceParameterBufferH264  *vseq = ctx->codec_sequence_params;
496     VAEncPictureParameterBufferH264   *vpic = ctx->codec_picture_params;
497     VAAPIEncodeH264Context            *priv = ctx->priv_data;
498     VAAPIEncodeH264MiscSequenceParams *mseq = &priv->misc_sequence_params;
499     int i;
500
501     {
502         vseq->seq_parameter_set_id = 0;
503
504         vseq->level_idc = avctx->level;
505
506         vseq->max_num_ref_frames = 2;
507
508         vseq->picture_width_in_mbs  = priv->mb_width;
509         vseq->picture_height_in_mbs = priv->mb_height;
510
511         vseq->seq_fields.bits.chroma_format_idc = 1;
512         vseq->seq_fields.bits.frame_mbs_only_flag = 1;
513         vseq->seq_fields.bits.direct_8x8_inference_flag = 1;
514         vseq->seq_fields.bits.log2_max_frame_num_minus4 = 4;
515         vseq->seq_fields.bits.pic_order_cnt_type = 0;
516
517         if (ctx->input_width  != ctx->aligned_width ||
518             ctx->input_height != ctx->aligned_height) {
519             vseq->frame_cropping_flag = 1;
520
521             vseq->frame_crop_left_offset   = 0;
522             vseq->frame_crop_right_offset  =
523                 (ctx->aligned_width - ctx->input_width) / 2;
524             vseq->frame_crop_top_offset    = 0;
525             vseq->frame_crop_bottom_offset =
526                 (ctx->aligned_height - ctx->input_height) / 2;
527         } else {
528             vseq->frame_cropping_flag = 0;
529         }
530
531         vseq->bits_per_second = avctx->bit_rate;
532         if (avctx->framerate.num > 0 && avctx->framerate.den > 0) {
533             vseq->num_units_in_tick = avctx->framerate.num;
534             vseq->time_scale        = 2 * avctx->framerate.den;
535         } else {
536             vseq->num_units_in_tick = avctx->time_base.num;
537             vseq->time_scale        = 2 * avctx->time_base.den;
538         }
539
540         vseq->intra_period     = ctx->p_per_i * (ctx->b_per_p + 1);
541         vseq->intra_idr_period = vseq->intra_period;
542         vseq->ip_period        = ctx->b_per_p + 1;
543     }
544
545     {
546         vpic->CurrPic.picture_id = VA_INVALID_ID;
547         vpic->CurrPic.flags      = VA_PICTURE_H264_INVALID;
548
549         for (i = 0; i < FF_ARRAY_ELEMS(vpic->ReferenceFrames); i++) {
550             vpic->ReferenceFrames[i].picture_id = VA_INVALID_ID;
551             vpic->ReferenceFrames[i].flags      = VA_PICTURE_H264_INVALID;
552         }
553
554         vpic->coded_buf = VA_INVALID_ID;
555
556         vpic->pic_parameter_set_id = 0;
557         vpic->seq_parameter_set_id = 0;
558
559         vpic->num_ref_idx_l0_active_minus1 = 0;
560         vpic->num_ref_idx_l1_active_minus1 = 0;
561
562         vpic->pic_fields.bits.entropy_coding_mode_flag =
563             ((avctx->profile & 0xff) != 66);
564         vpic->pic_fields.bits.weighted_pred_flag = 0;
565         vpic->pic_fields.bits.weighted_bipred_idc = 0;
566         vpic->pic_fields.bits.transform_8x8_mode_flag =
567             ((avctx->profile & 0xff) >= 100);
568
569         vpic->pic_init_qp = priv->fixed_qp_idr;
570     }
571
572     {
573         mseq->profile_idc = avctx->profile & 0xff;
574
575         if (avctx->profile & FF_PROFILE_H264_CONSTRAINED)
576             mseq->constraint_set1_flag = 1;
577         if (avctx->profile & FF_PROFILE_H264_INTRA)
578             mseq->constraint_set3_flag = 1;
579     }
580
581     return 0;
582 }
583
584 static int vaapi_encode_h264_init_picture_params(AVCodecContext *avctx,
585                                                  VAAPIEncodePicture *pic)
586 {
587     VAAPIEncodeContext                *ctx = avctx->priv_data;
588     VAEncSequenceParameterBufferH264 *vseq = ctx->codec_sequence_params;
589     VAEncPictureParameterBufferH264  *vpic = pic->codec_picture_params;
590     VAAPIEncodeH264Context           *priv = ctx->priv_data;
591     int i;
592
593     if (pic->type == PICTURE_TYPE_IDR) {
594         av_assert0(pic->display_order == pic->encode_order);
595         priv->last_idr_frame = pic->display_order;
596     } else {
597         av_assert0(pic->display_order > priv->last_idr_frame);
598     }
599
600     vpic->frame_num = (pic->encode_order - priv->last_idr_frame) &
601         ((1 << (4 + vseq->seq_fields.bits.log2_max_frame_num_minus4)) - 1);
602
603     vpic->CurrPic.picture_id          = pic->recon_surface;
604     vpic->CurrPic.frame_idx           = vpic->frame_num;
605     vpic->CurrPic.flags               = 0;
606     vpic->CurrPic.TopFieldOrderCnt    = pic->display_order;
607     vpic->CurrPic.BottomFieldOrderCnt = pic->display_order;
608
609     for (i = 0; i < pic->nb_refs; i++) {
610         VAAPIEncodePicture *ref = pic->refs[i];
611         av_assert0(ref && ref->encode_order >= priv->last_idr_frame);
612         vpic->ReferenceFrames[i].picture_id = ref->recon_surface;
613         vpic->ReferenceFrames[i].frame_idx =
614             ref->encode_order - priv->last_idr_frame;
615         vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_SHORT_TERM_REFERENCE;
616         vpic->ReferenceFrames[i].TopFieldOrderCnt    = ref->display_order;
617         vpic->ReferenceFrames[i].BottomFieldOrderCnt = ref->display_order;
618     }
619     for (; i < FF_ARRAY_ELEMS(vpic->ReferenceFrames); i++) {
620         vpic->ReferenceFrames[i].picture_id = VA_INVALID_ID;
621         vpic->ReferenceFrames[i].flags = VA_PICTURE_H264_INVALID;
622     }
623
624     vpic->coded_buf = pic->output_buffer;
625
626     vpic->pic_fields.bits.idr_pic_flag = (pic->type == PICTURE_TYPE_IDR);
627     vpic->pic_fields.bits.reference_pic_flag = (pic->type != PICTURE_TYPE_B);
628
629     pic->nb_slices = 1;
630
631     return 0;
632 }
633
634 static int vaapi_encode_h264_init_slice_params(AVCodecContext *avctx,
635                                                VAAPIEncodePicture *pic,
636                                                VAAPIEncodeSlice *slice)
637 {
638     VAAPIEncodeContext                 *ctx = avctx->priv_data;
639     VAEncSequenceParameterBufferH264  *vseq = ctx->codec_sequence_params;
640     VAEncPictureParameterBufferH264   *vpic = pic->codec_picture_params;
641     VAEncSliceParameterBufferH264   *vslice = slice->codec_slice_params;
642     VAAPIEncodeH264Context            *priv = ctx->priv_data;
643     VAAPIEncodeH264Slice            *pslice;
644     VAAPIEncodeH264MiscSliceParams  *mslice;
645     int i;
646
647     slice->priv_data = av_mallocz(sizeof(*pslice));
648     if (!slice->priv_data)
649         return AVERROR(ENOMEM);
650     pslice = slice->priv_data;
651     mslice = &pslice->misc_slice_params;
652
653     if (pic->type == PICTURE_TYPE_IDR)
654         mslice->nal_unit_type = NAL_IDR_SLICE;
655     else
656         mslice->nal_unit_type = NAL_SLICE;
657
658     switch (pic->type) {
659     case PICTURE_TYPE_IDR:
660         vslice->slice_type  = SLICE_TYPE_I;
661         mslice->nal_ref_idc = 3;
662         break;
663     case PICTURE_TYPE_I:
664         vslice->slice_type  = SLICE_TYPE_I;
665         mslice->nal_ref_idc = 2;
666         break;
667     case PICTURE_TYPE_P:
668         vslice->slice_type  = SLICE_TYPE_P;
669         mslice->nal_ref_idc = 1;
670         break;
671     case PICTURE_TYPE_B:
672         vslice->slice_type  = SLICE_TYPE_B;
673         mslice->nal_ref_idc = 0;
674         break;
675     default:
676         av_assert0(0 && "invalid picture type");
677     }
678
679     // Only one slice per frame.
680     vslice->macroblock_address = 0;
681     vslice->num_macroblocks = priv->mb_width * priv->mb_height;
682
683     vslice->macroblock_info = VA_INVALID_ID;
684
685     vslice->pic_parameter_set_id = vpic->pic_parameter_set_id;
686     vslice->idr_pic_id = priv->idr_pic_count++;
687
688     vslice->pic_order_cnt_lsb = pic->display_order &
689         ((1 << (4 + vseq->seq_fields.bits.log2_max_pic_order_cnt_lsb_minus4)) - 1);
690
691     for (i = 0; i < FF_ARRAY_ELEMS(vslice->RefPicList0); i++) {
692         vslice->RefPicList0[i].picture_id = VA_INVALID_ID;
693         vslice->RefPicList0[i].flags      = VA_PICTURE_H264_INVALID;
694         vslice->RefPicList1[i].picture_id = VA_INVALID_ID;
695         vslice->RefPicList1[i].flags      = VA_PICTURE_H264_INVALID;
696     }
697
698     av_assert0(pic->nb_refs <= 2);
699     if (pic->nb_refs >= 1) {
700         // Backward reference for P- or B-frame.
701         av_assert0(pic->type == PICTURE_TYPE_P ||
702                    pic->type == PICTURE_TYPE_B);
703
704         vslice->num_ref_idx_l0_active_minus1 = 0;
705         vslice->RefPicList0[0] = vpic->ReferenceFrames[0];
706     }
707     if (pic->nb_refs >= 2) {
708         // Forward reference for B-frame.
709         av_assert0(pic->type == PICTURE_TYPE_B);
710
711         vslice->num_ref_idx_l1_active_minus1 = 0;
712         vslice->RefPicList1[0] = vpic->ReferenceFrames[1];
713     }
714
715     if (pic->type == PICTURE_TYPE_B)
716         vslice->slice_qp_delta = priv->fixed_qp_b - vpic->pic_init_qp;
717     else if (pic->type == PICTURE_TYPE_P)
718         vslice->slice_qp_delta = priv->fixed_qp_p - vpic->pic_init_qp;
719     else
720         vslice->slice_qp_delta = priv->fixed_qp_idr - vpic->pic_init_qp;
721
722     vslice->direct_spatial_mv_pred_flag = 1;
723
724     return 0;
725 }
726
727 static av_cold int vaapi_encode_h264_init_constant_bitrate(AVCodecContext *avctx)
728 {
729     VAAPIEncodeContext      *ctx = avctx->priv_data;
730     VAAPIEncodeH264Context *priv = ctx->priv_data;
731     int hrd_buffer_size;
732     int hrd_initial_buffer_fullness;
733
734     if (avctx->rc_buffer_size)
735         hrd_buffer_size = avctx->rc_buffer_size;
736     else
737         hrd_buffer_size = avctx->bit_rate;
738     if (avctx->rc_initial_buffer_occupancy)
739         hrd_initial_buffer_fullness = avctx->rc_initial_buffer_occupancy;
740     else
741         hrd_initial_buffer_fullness = hrd_buffer_size * 3 / 4;
742
743     priv->rc_params.misc.type = VAEncMiscParameterTypeRateControl;
744     priv->rc_params.rc = (VAEncMiscParameterRateControl) {
745         .bits_per_second   = avctx->bit_rate,
746         .target_percentage = 66,
747         .window_size       = 1000,
748         .initial_qp        = (avctx->qmax >= 0 ? avctx->qmax : 40),
749         .min_qp            = (avctx->qmin >= 0 ? avctx->qmin : 18),
750         .basic_unit_size   = 0,
751     };
752     ctx->global_params[ctx->nb_global_params] =
753         &priv->rc_params.misc;
754     ctx->global_params_size[ctx->nb_global_params++] =
755         sizeof(priv->rc_params);
756
757     priv->hrd_params.misc.type = VAEncMiscParameterTypeHRD;
758     priv->hrd_params.hrd = (VAEncMiscParameterHRD) {
759         .initial_buffer_fullness = hrd_initial_buffer_fullness,
760         .buffer_size             = hrd_buffer_size,
761     };
762     ctx->global_params[ctx->nb_global_params] =
763         &priv->hrd_params.misc;
764     ctx->global_params_size[ctx->nb_global_params++] =
765         sizeof(priv->hrd_params);
766
767     // These still need to be  set for pic_init_qp/slice_qp_delta.
768     priv->fixed_qp_idr = 26;
769     priv->fixed_qp_p   = 26;
770     priv->fixed_qp_b   = 26;
771
772     av_log(avctx, AV_LOG_DEBUG, "Using constant-bitrate = %d bps.\n",
773            avctx->bit_rate);
774     return 0;
775 }
776
777 static av_cold int vaapi_encode_h264_init_fixed_qp(AVCodecContext *avctx)
778 {
779     VAAPIEncodeContext      *ctx = avctx->priv_data;
780     VAAPIEncodeH264Context *priv = ctx->priv_data;
781     VAAPIEncodeH264Options  *opt = ctx->codec_options;
782
783     priv->fixed_qp_p = opt->qp;
784     if (avctx->i_quant_factor > 0.0)
785         priv->fixed_qp_idr = (int)((priv->fixed_qp_p * avctx->i_quant_factor +
786                                     avctx->i_quant_offset) + 0.5);
787     else
788         priv->fixed_qp_idr = priv->fixed_qp_p;
789     if (avctx->b_quant_factor > 0.0)
790         priv->fixed_qp_b = (int)((priv->fixed_qp_p * avctx->b_quant_factor +
791                                   avctx->b_quant_offset) + 0.5);
792     else
793         priv->fixed_qp_b = priv->fixed_qp_p;
794
795     av_log(avctx, AV_LOG_DEBUG, "Using fixed QP = "
796            "%d / %d / %d for IDR- / P- / B-frames.\n",
797            priv->fixed_qp_idr, priv->fixed_qp_p, priv->fixed_qp_b);
798     return 0;
799 }
800
801 static av_cold int vaapi_encode_h264_init_internal(AVCodecContext *avctx)
802 {
803     static const VAConfigAttrib default_config_attributes[] = {
804         { .type  = VAConfigAttribRTFormat,
805           .value = VA_RT_FORMAT_YUV420 },
806         { .type  = VAConfigAttribEncPackedHeaders,
807           .value = (VA_ENC_PACKED_HEADER_SEQUENCE |
808                     VA_ENC_PACKED_HEADER_SLICE) },
809     };
810
811     VAAPIEncodeContext      *ctx = avctx->priv_data;
812     VAAPIEncodeH264Context *priv = ctx->priv_data;
813     VAAPIEncodeH264Options  *opt = ctx->codec_options;
814     int i, err;
815
816     switch (avctx->profile) {
817     case FF_PROFILE_H264_CONSTRAINED_BASELINE:
818         ctx->va_profile = VAProfileH264ConstrainedBaseline;
819         break;
820     case FF_PROFILE_H264_BASELINE:
821         ctx->va_profile = VAProfileH264Baseline;
822         break;
823     case FF_PROFILE_H264_MAIN:
824         ctx->va_profile = VAProfileH264Main;
825         break;
826     case FF_PROFILE_H264_EXTENDED:
827         av_log(avctx, AV_LOG_ERROR, "H.264 extended profile "
828                "is not supported.\n");
829         return AVERROR_PATCHWELCOME;
830     case FF_PROFILE_UNKNOWN:
831     case FF_PROFILE_H264_HIGH:
832         ctx->va_profile = VAProfileH264High;
833         break;
834     case FF_PROFILE_H264_HIGH_10:
835     case FF_PROFILE_H264_HIGH_10_INTRA:
836         av_log(avctx, AV_LOG_ERROR, "H.264 10-bit profiles "
837                "are not supported.\n");
838         return AVERROR_PATCHWELCOME;
839     case FF_PROFILE_H264_HIGH_422:
840     case FF_PROFILE_H264_HIGH_422_INTRA:
841     case FF_PROFILE_H264_HIGH_444:
842     case FF_PROFILE_H264_HIGH_444_PREDICTIVE:
843     case FF_PROFILE_H264_HIGH_444_INTRA:
844     case FF_PROFILE_H264_CAVLC_444:
845         av_log(avctx, AV_LOG_ERROR, "H.264 non-4:2:0 profiles "
846                "are not supported.\n");
847         return AVERROR_PATCHWELCOME;
848     default:
849         av_log(avctx, AV_LOG_ERROR, "Unknown H.264 profile %d.\n",
850                avctx->profile);
851         return AVERROR(EINVAL);
852     }
853     ctx->va_entrypoint = VAEntrypointEncSlice;
854
855     ctx->input_width    = avctx->width;
856     ctx->input_height   = avctx->height;
857     ctx->aligned_width  = FFALIGN(ctx->input_width,  16);
858     ctx->aligned_height = FFALIGN(ctx->input_height, 16);
859     priv->mb_width      = ctx->aligned_width  / 16;
860     priv->mb_height     = ctx->aligned_height / 16;
861
862     for (i = 0; i < FF_ARRAY_ELEMS(default_config_attributes); i++) {
863         ctx->config_attributes[ctx->nb_config_attributes++] =
864             default_config_attributes[i];
865     }
866
867     if (avctx->bit_rate > 0) {
868         ctx->va_rc_mode = VA_RC_CBR;
869         err = vaapi_encode_h264_init_constant_bitrate(avctx);
870     } else {
871         ctx->va_rc_mode = VA_RC_CQP;
872         err = vaapi_encode_h264_init_fixed_qp(avctx);
873     }
874     if (err < 0)
875         return err;
876
877     ctx->config_attributes[ctx->nb_config_attributes++] = (VAConfigAttrib) {
878         .type  = VAConfigAttribRateControl,
879         .value = ctx->va_rc_mode,
880     };
881
882     if (opt->quality > 0) {
883 #if VA_CHECK_VERSION(0, 36, 0)
884         priv->quality_params.misc.type =
885             VAEncMiscParameterTypeQualityLevel;
886         priv->quality_params.quality.quality_level = opt->quality;
887
888         ctx->global_params[ctx->nb_global_params] =
889             &priv->quality_params.misc;
890         ctx->global_params_size[ctx->nb_global_params++] =
891             sizeof(priv->quality_params);
892 #else
893         av_log(avctx, AV_LOG_WARNING, "The encode quality option is not "
894                "supported with this VAAPI version.\n");
895 #endif
896     }
897
898     ctx->nb_recon_frames = 20;
899
900     return 0;
901 }
902
903 static VAAPIEncodeType vaapi_encode_type_h264 = {
904     .priv_data_size        = sizeof(VAAPIEncodeH264Context),
905
906     .init                  = &vaapi_encode_h264_init_internal,
907
908     .sequence_params_size  = sizeof(VAEncSequenceParameterBufferH264),
909     .init_sequence_params  = &vaapi_encode_h264_init_sequence_params,
910
911     .picture_params_size   = sizeof(VAEncPictureParameterBufferH264),
912     .init_picture_params   = &vaapi_encode_h264_init_picture_params,
913
914     .slice_params_size     = sizeof(VAEncSliceParameterBufferH264),
915     .init_slice_params     = &vaapi_encode_h264_init_slice_params,
916
917     .sequence_header_type  = VAEncPackedHeaderSequence,
918     .write_sequence_header = &vaapi_encode_h264_write_sequence_header,
919
920     .slice_header_type     = VAEncPackedHeaderH264_Slice,
921     .write_slice_header    = &vaapi_encode_h264_write_slice_header,
922 };
923
924 static av_cold int vaapi_encode_h264_init(AVCodecContext *avctx)
925 {
926     return ff_vaapi_encode_init(avctx, &vaapi_encode_type_h264);
927 }
928
929 #define OFFSET(x) (offsetof(VAAPIEncodeContext, codec_options_data) + \
930                    offsetof(VAAPIEncodeH264Options, x))
931 #define FLAGS (AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_ENCODING_PARAM)
932 static const AVOption vaapi_encode_h264_options[] = {
933     { "qp", "Constant QP (for P-frames; scaled by qfactor/qoffset for I/B)",
934       OFFSET(qp), AV_OPT_TYPE_INT, { .i64 = 20 }, 0, 52, FLAGS },
935     { "quality", "Set encode quality (trades off against speed, higher is faster)",
936       OFFSET(quality), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 2, FLAGS },
937     { NULL },
938 };
939
940 static const AVCodecDefault vaapi_encode_h264_defaults[] = {
941     { "profile",        "100" },
942     { "level",          "51"  },
943     { "b",              "0"   },
944     { "bf",             "2"   },
945     { "g",              "120" },
946     { "i_qfactor",      "1.0" },
947     { "i_qoffset",      "0.0" },
948     { "b_qfactor",      "1.2" },
949     { "b_qoffset",      "0.0" },
950     { NULL },
951 };
952
953 static const AVClass vaapi_encode_h264_class = {
954     .class_name = "h264_vaapi",
955     .item_name  = av_default_item_name,
956     .option     = vaapi_encode_h264_options,
957     .version    = LIBAVUTIL_VERSION_INT,
958 };
959
960 AVCodec ff_h264_vaapi_encoder = {
961     .name           = "h264_vaapi",
962     .long_name      = NULL_IF_CONFIG_SMALL("H.264/AVC (VAAPI)"),
963     .type           = AVMEDIA_TYPE_VIDEO,
964     .id             = AV_CODEC_ID_H264,
965     .priv_data_size = (sizeof(VAAPIEncodeContext) +
966                        sizeof(VAAPIEncodeH264Options)),
967     .init           = &vaapi_encode_h264_init,
968     .encode2        = &ff_vaapi_encode2,
969     .close          = &ff_vaapi_encode_close,
970     .priv_class     = &vaapi_encode_h264_class,
971     .capabilities   = AV_CODEC_CAP_DELAY,
972     .defaults       = vaapi_encode_h264_defaults,
973     .pix_fmts = (const enum AVPixelFormat[]) {
974         AV_PIX_FMT_VAAPI,
975         AV_PIX_FMT_NONE,
976     },
977 };