1 /*
2 * Copyright (c) 2019 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include <memory>
12 #include <vector>
13 #include "./ivfenc.h"
14 #include "vp9/common/vp9_entropymode.h"
15 #include "vp9/common/vp9_enums.h"
16 #include "vp9/common/vp9_onyxc_int.h"
17 #include "vp9/vp9_iface_common.h"
18 #include "vp9/encoder/vp9_encoder.h"
19 #include "vp9/encoder/vp9_firstpass.h"
20 #include "vp9/simple_encode.h"
21 #include "vp9/vp9_cx_iface.h"
22
23 namespace vp9 {
24
get_plane_height(vpx_img_fmt_t img_fmt,int frame_height,int plane)25 static int get_plane_height(vpx_img_fmt_t img_fmt, int frame_height,
26 int plane) {
27 assert(plane < 3);
28 if (plane == 0) {
29 return frame_height;
30 }
31 switch (img_fmt) {
32 case VPX_IMG_FMT_I420:
33 case VPX_IMG_FMT_I440:
34 case VPX_IMG_FMT_YV12:
35 case VPX_IMG_FMT_I42016:
36 case VPX_IMG_FMT_I44016: return (frame_height + 1) >> 1;
37 default: return frame_height;
38 }
39 }
40
get_plane_width(vpx_img_fmt_t img_fmt,int frame_width,int plane)41 static int get_plane_width(vpx_img_fmt_t img_fmt, int frame_width, int plane) {
42 assert(plane < 3);
43 if (plane == 0) {
44 return frame_width;
45 }
46 switch (img_fmt) {
47 case VPX_IMG_FMT_I420:
48 case VPX_IMG_FMT_YV12:
49 case VPX_IMG_FMT_I422:
50 case VPX_IMG_FMT_I42016:
51 case VPX_IMG_FMT_I42216: return (frame_width + 1) >> 1;
52 default: return frame_width;
53 }
54 }
55
56 // TODO(angiebird): Merge this function with vpx_img_plane_width()
img_plane_width(const vpx_image_t * img,int plane)57 static int img_plane_width(const vpx_image_t *img, int plane) {
58 if (plane > 0 && img->x_chroma_shift > 0)
59 return (img->d_w + 1) >> img->x_chroma_shift;
60 else
61 return img->d_w;
62 }
63
64 // TODO(angiebird): Merge this function with vpx_img_plane_height()
img_plane_height(const vpx_image_t * img,int plane)65 static int img_plane_height(const vpx_image_t *img, int plane) {
66 if (plane > 0 && img->y_chroma_shift > 0)
67 return (img->d_h + 1) >> img->y_chroma_shift;
68 else
69 return img->d_h;
70 }
71
72 // TODO(angiebird): Merge this function with vpx_img_read()
img_read(vpx_image_t * img,FILE * file)73 static int img_read(vpx_image_t *img, FILE *file) {
74 int plane;
75
76 for (plane = 0; plane < 3; ++plane) {
77 unsigned char *buf = img->planes[plane];
78 const int stride = img->stride[plane];
79 const int w = img_plane_width(img, plane) *
80 ((img->fmt & VPX_IMG_FMT_HIGHBITDEPTH) ? 2 : 1);
81 const int h = img_plane_height(img, plane);
82 int y;
83
84 for (y = 0; y < h; ++y) {
85 if (fread(buf, 1, w, file) != (size_t)w) return 0;
86 buf += stride;
87 }
88 }
89
90 return 1;
91 }
92
93 // Assume every config in VP9EncoderConfig is less than 100 characters.
94 #define ENCODE_CONFIG_BUF_SIZE 100
95 struct EncodeConfig {
96 char name[ENCODE_CONFIG_BUF_SIZE];
97 char value[ENCODE_CONFIG_BUF_SIZE];
98 };
99
100 class SimpleEncode::EncodeImpl {
101 public:
102 VP9_COMP *cpi;
103 vpx_img_fmt_t img_fmt;
104 vpx_image_t tmp_img;
105 std::vector<FIRSTPASS_STATS> first_pass_stats;
106 std::vector<EncodeConfig> encode_config_list;
107 };
108
init_encoder(const VP9EncoderConfig * oxcf,vpx_img_fmt_t img_fmt)109 static VP9_COMP *init_encoder(const VP9EncoderConfig *oxcf,
110 vpx_img_fmt_t img_fmt) {
111 VP9_COMP *cpi;
112 BufferPool *buffer_pool = (BufferPool *)vpx_calloc(1, sizeof(*buffer_pool));
113 vp9_initialize_enc();
114 cpi = vp9_create_compressor(oxcf, buffer_pool);
115 vp9_update_compressor_with_img_fmt(cpi, img_fmt);
116 return cpi;
117 }
118
free_encoder(VP9_COMP * cpi)119 static void free_encoder(VP9_COMP *cpi) {
120 BufferPool *buffer_pool = cpi->common.buffer_pool;
121 vp9_remove_compressor(cpi);
122 // buffer_pool needs to be free after cpi because buffer_pool contains
123 // allocated buffers that will be free in vp9_remove_compressor()
124 vpx_free(buffer_pool);
125 }
126
make_vpx_rational(int num,int den)127 static INLINE vpx_rational_t make_vpx_rational(int num, int den) {
128 vpx_rational_t v;
129 v.num = num;
130 v.den = den;
131 return v;
132 }
133
134 static INLINE FrameType
get_frame_type_from_update_type(FRAME_UPDATE_TYPE update_type)135 get_frame_type_from_update_type(FRAME_UPDATE_TYPE update_type) {
136 switch (update_type) {
137 case KF_UPDATE: return kFrameTypeKey;
138 case ARF_UPDATE: return kFrameTypeAltRef;
139 case GF_UPDATE: return kFrameTypeGolden;
140 case OVERLAY_UPDATE: return kFrameTypeOverlay;
141 case LF_UPDATE: return kFrameTypeInter;
142 default:
143 fprintf(stderr, "Unsupported update_type %d\n", update_type);
144 abort();
145 return kFrameTypeInter;
146 }
147 }
148
update_partition_info(const PARTITION_INFO * input_partition_info,const int num_rows_4x4,const int num_cols_4x4,PartitionInfo * output_partition_info)149 static void update_partition_info(const PARTITION_INFO *input_partition_info,
150 const int num_rows_4x4,
151 const int num_cols_4x4,
152 PartitionInfo *output_partition_info) {
153 const int num_units_4x4 = num_rows_4x4 * num_cols_4x4;
154 for (int i = 0; i < num_units_4x4; ++i) {
155 output_partition_info[i].row = input_partition_info[i].row;
156 output_partition_info[i].column = input_partition_info[i].column;
157 output_partition_info[i].row_start = input_partition_info[i].row_start;
158 output_partition_info[i].column_start =
159 input_partition_info[i].column_start;
160 output_partition_info[i].width = input_partition_info[i].width;
161 output_partition_info[i].height = input_partition_info[i].height;
162 }
163 }
164
165 // translate MV_REFERENCE_FRAME to RefFrameType
mv_ref_frame_to_ref_frame_type(MV_REFERENCE_FRAME mv_ref_frame)166 static RefFrameType mv_ref_frame_to_ref_frame_type(
167 MV_REFERENCE_FRAME mv_ref_frame) {
168 switch (mv_ref_frame) {
169 case LAST_FRAME: return kRefFrameTypeLast;
170 case GOLDEN_FRAME: return kRefFrameTypePast;
171 case ALTREF_FRAME: return kRefFrameTypeFuture;
172 default: return kRefFrameTypeNone;
173 }
174 }
175
update_motion_vector_info(const MOTION_VECTOR_INFO * input_motion_vector_info,const int num_rows_4x4,const int num_cols_4x4,MotionVectorInfo * output_motion_vector_info,int motion_vector_scale)176 static void update_motion_vector_info(
177 const MOTION_VECTOR_INFO *input_motion_vector_info, const int num_rows_4x4,
178 const int num_cols_4x4, MotionVectorInfo *output_motion_vector_info,
179 int motion_vector_scale) {
180 const int num_units_4x4 = num_rows_4x4 * num_cols_4x4;
181 for (int i = 0; i < num_units_4x4; ++i) {
182 const MV_REFERENCE_FRAME *in_ref_frame =
183 input_motion_vector_info[i].ref_frame;
184 output_motion_vector_info[i].mv_count =
185 (in_ref_frame[0] == INTRA_FRAME) ? 0
186 : ((in_ref_frame[1] == NONE) ? 1 : 2);
187 if (in_ref_frame[0] == NONE) {
188 fprintf(stderr, "in_ref_frame[0] shouldn't be NONE\n");
189 abort();
190 }
191 output_motion_vector_info[i].ref_frame[0] =
192 mv_ref_frame_to_ref_frame_type(in_ref_frame[0]);
193 output_motion_vector_info[i].ref_frame[1] =
194 mv_ref_frame_to_ref_frame_type(in_ref_frame[1]);
195 output_motion_vector_info[i].mv_row[0] =
196 (double)input_motion_vector_info[i].mv[0].as_mv.row /
197 motion_vector_scale;
198 output_motion_vector_info[i].mv_column[0] =
199 (double)input_motion_vector_info[i].mv[0].as_mv.col /
200 motion_vector_scale;
201 output_motion_vector_info[i].mv_row[1] =
202 (double)input_motion_vector_info[i].mv[1].as_mv.row /
203 motion_vector_scale;
204 output_motion_vector_info[i].mv_column[1] =
205 (double)input_motion_vector_info[i].mv[1].as_mv.col /
206 motion_vector_scale;
207 }
208 }
209
update_frame_counts(const FRAME_COUNTS * input_counts,FrameCounts * output_counts)210 static void update_frame_counts(const FRAME_COUNTS *input_counts,
211 FrameCounts *output_counts) {
212 // Init array sizes.
213 output_counts->y_mode.resize(BLOCK_SIZE_GROUPS);
214 for (int i = 0; i < BLOCK_SIZE_GROUPS; ++i) {
215 output_counts->y_mode[i].resize(INTRA_MODES);
216 }
217
218 output_counts->uv_mode.resize(INTRA_MODES);
219 for (int i = 0; i < INTRA_MODES; ++i) {
220 output_counts->uv_mode[i].resize(INTRA_MODES);
221 }
222
223 output_counts->partition.resize(PARTITION_CONTEXTS);
224 for (int i = 0; i < PARTITION_CONTEXTS; ++i) {
225 output_counts->partition[i].resize(PARTITION_TYPES);
226 }
227
228 output_counts->coef.resize(TX_SIZES);
229 output_counts->eob_branch.resize(TX_SIZES);
230 for (int i = 0; i < TX_SIZES; ++i) {
231 output_counts->coef[i].resize(PLANE_TYPES);
232 output_counts->eob_branch[i].resize(PLANE_TYPES);
233 for (int j = 0; j < PLANE_TYPES; ++j) {
234 output_counts->coef[i][j].resize(REF_TYPES);
235 output_counts->eob_branch[i][j].resize(REF_TYPES);
236 for (int k = 0; k < REF_TYPES; ++k) {
237 output_counts->coef[i][j][k].resize(COEF_BANDS);
238 output_counts->eob_branch[i][j][k].resize(COEF_BANDS);
239 for (int l = 0; l < COEF_BANDS; ++l) {
240 output_counts->coef[i][j][k][l].resize(COEFF_CONTEXTS);
241 output_counts->eob_branch[i][j][k][l].resize(COEFF_CONTEXTS);
242 for (int m = 0; m < COEFF_CONTEXTS; ++m) {
243 output_counts->coef[i][j][k][l][m].resize(UNCONSTRAINED_NODES + 1);
244 }
245 }
246 }
247 }
248 }
249
250 output_counts->switchable_interp.resize(SWITCHABLE_FILTER_CONTEXTS);
251 for (int i = 0; i < SWITCHABLE_FILTER_CONTEXTS; ++i) {
252 output_counts->switchable_interp[i].resize(SWITCHABLE_FILTERS);
253 }
254
255 output_counts->inter_mode.resize(INTER_MODE_CONTEXTS);
256 for (int i = 0; i < INTER_MODE_CONTEXTS; ++i) {
257 output_counts->inter_mode[i].resize(INTER_MODES);
258 }
259
260 output_counts->intra_inter.resize(INTRA_INTER_CONTEXTS);
261 for (int i = 0; i < INTRA_INTER_CONTEXTS; ++i) {
262 output_counts->intra_inter[i].resize(2);
263 }
264
265 output_counts->comp_inter.resize(COMP_INTER_CONTEXTS);
266 for (int i = 0; i < COMP_INTER_CONTEXTS; ++i) {
267 output_counts->comp_inter[i].resize(2);
268 }
269
270 output_counts->single_ref.resize(REF_CONTEXTS);
271 for (int i = 0; i < REF_CONTEXTS; ++i) {
272 output_counts->single_ref[i].resize(2);
273 for (int j = 0; j < 2; ++j) {
274 output_counts->single_ref[i][j].resize(2);
275 }
276 }
277
278 output_counts->comp_ref.resize(REF_CONTEXTS);
279 for (int i = 0; i < REF_CONTEXTS; ++i) {
280 output_counts->comp_ref[i].resize(2);
281 }
282
283 output_counts->skip.resize(SKIP_CONTEXTS);
284 for (int i = 0; i < SKIP_CONTEXTS; ++i) {
285 output_counts->skip[i].resize(2);
286 }
287
288 output_counts->tx.p32x32.resize(TX_SIZE_CONTEXTS);
289 output_counts->tx.p16x16.resize(TX_SIZE_CONTEXTS);
290 output_counts->tx.p8x8.resize(TX_SIZE_CONTEXTS);
291 for (int i = 0; i < TX_SIZE_CONTEXTS; i++) {
292 output_counts->tx.p32x32[i].resize(TX_SIZES);
293 output_counts->tx.p16x16[i].resize(TX_SIZES - 1);
294 output_counts->tx.p8x8[i].resize(TX_SIZES - 2);
295 }
296 output_counts->tx.tx_totals.resize(TX_SIZES);
297
298 output_counts->mv.joints.resize(MV_JOINTS);
299 output_counts->mv.comps.resize(2);
300 for (int i = 0; i < 2; ++i) {
301 output_counts->mv.comps[i].sign.resize(2);
302 output_counts->mv.comps[i].classes.resize(MV_CLASSES);
303 output_counts->mv.comps[i].class0.resize(CLASS0_SIZE);
304 output_counts->mv.comps[i].bits.resize(MV_OFFSET_BITS);
305 for (int j = 0; j < MV_OFFSET_BITS; ++j) {
306 output_counts->mv.comps[i].bits[j].resize(2);
307 }
308 output_counts->mv.comps[i].class0_fp.resize(CLASS0_SIZE);
309 for (int j = 0; j < CLASS0_SIZE; ++j) {
310 output_counts->mv.comps[i].class0_fp[j].resize(MV_FP_SIZE);
311 }
312 output_counts->mv.comps[i].fp.resize(MV_FP_SIZE);
313 output_counts->mv.comps[i].class0_hp.resize(2);
314 output_counts->mv.comps[i].hp.resize(2);
315 }
316
317 // Populate counts.
318 for (int i = 0; i < BLOCK_SIZE_GROUPS; ++i) {
319 for (int j = 0; j < INTRA_MODES; ++j) {
320 output_counts->y_mode[i][j] = input_counts->y_mode[i][j];
321 }
322 }
323 for (int i = 0; i < INTRA_MODES; ++i) {
324 for (int j = 0; j < INTRA_MODES; ++j) {
325 output_counts->uv_mode[i][j] = input_counts->uv_mode[i][j];
326 }
327 }
328 for (int i = 0; i < PARTITION_CONTEXTS; ++i) {
329 for (int j = 0; j < PARTITION_TYPES; ++j) {
330 output_counts->partition[i][j] = input_counts->partition[i][j];
331 }
332 }
333 for (int i = 0; i < TX_SIZES; ++i) {
334 for (int j = 0; j < PLANE_TYPES; ++j) {
335 for (int k = 0; k < REF_TYPES; ++k) {
336 for (int l = 0; l < COEF_BANDS; ++l) {
337 for (int m = 0; m < COEFF_CONTEXTS; ++m) {
338 output_counts->eob_branch[i][j][k][l][m] =
339 input_counts->eob_branch[i][j][k][l][m];
340 for (int n = 0; n < UNCONSTRAINED_NODES + 1; n++) {
341 output_counts->coef[i][j][k][l][m][n] =
342 input_counts->coef[i][j][k][l][m][n];
343 }
344 }
345 }
346 }
347 }
348 }
349 for (int i = 0; i < SWITCHABLE_FILTER_CONTEXTS; ++i) {
350 for (int j = 0; j < SWITCHABLE_FILTERS; ++j) {
351 output_counts->switchable_interp[i][j] =
352 input_counts->switchable_interp[i][j];
353 }
354 }
355 for (int i = 0; i < INTER_MODE_CONTEXTS; ++i) {
356 for (int j = 0; j < INTER_MODES; ++j) {
357 output_counts->inter_mode[i][j] = input_counts->inter_mode[i][j];
358 }
359 }
360 for (int i = 0; i < INTRA_INTER_CONTEXTS; ++i) {
361 for (int j = 0; j < 2; ++j) {
362 output_counts->intra_inter[i][j] = input_counts->intra_inter[i][j];
363 }
364 }
365 for (int i = 0; i < COMP_INTER_CONTEXTS; ++i) {
366 for (int j = 0; j < 2; ++j) {
367 output_counts->comp_inter[i][j] = input_counts->comp_inter[i][j];
368 }
369 }
370 for (int i = 0; i < REF_CONTEXTS; ++i) {
371 for (int j = 0; j < 2; ++j) {
372 for (int k = 0; k < 2; ++k) {
373 output_counts->single_ref[i][j][k] = input_counts->single_ref[i][j][k];
374 }
375 }
376 }
377 for (int i = 0; i < REF_CONTEXTS; ++i) {
378 for (int j = 0; j < 2; ++j) {
379 output_counts->comp_ref[i][j] = input_counts->comp_ref[i][j];
380 }
381 }
382 for (int i = 0; i < SKIP_CONTEXTS; ++i) {
383 for (int j = 0; j < 2; ++j) {
384 output_counts->skip[i][j] = input_counts->skip[i][j];
385 }
386 }
387 for (int i = 0; i < TX_SIZE_CONTEXTS; i++) {
388 for (int j = 0; j < TX_SIZES; j++) {
389 output_counts->tx.p32x32[i][j] = input_counts->tx.p32x32[i][j];
390 }
391 for (int j = 0; j < TX_SIZES - 1; j++) {
392 output_counts->tx.p16x16[i][j] = input_counts->tx.p16x16[i][j];
393 }
394 for (int j = 0; j < TX_SIZES - 2; j++) {
395 output_counts->tx.p8x8[i][j] = input_counts->tx.p8x8[i][j];
396 }
397 }
398 for (int i = 0; i < TX_SIZES; i++) {
399 output_counts->tx.tx_totals[i] = input_counts->tx.tx_totals[i];
400 }
401 for (int i = 0; i < MV_JOINTS; i++) {
402 output_counts->mv.joints[i] = input_counts->mv.joints[i];
403 }
404 for (int k = 0; k < 2; k++) {
405 const nmv_component_counts *const comps_t = &input_counts->mv.comps[k];
406 for (int i = 0; i < 2; i++) {
407 output_counts->mv.comps[k].sign[i] = comps_t->sign[i];
408 output_counts->mv.comps[k].class0_hp[i] = comps_t->class0_hp[i];
409 output_counts->mv.comps[k].hp[i] = comps_t->hp[i];
410 }
411 for (int i = 0; i < MV_CLASSES; i++) {
412 output_counts->mv.comps[k].classes[i] = comps_t->classes[i];
413 }
414 for (int i = 0; i < CLASS0_SIZE; i++) {
415 output_counts->mv.comps[k].class0[i] = comps_t->class0[i];
416 for (int j = 0; j < MV_FP_SIZE; j++) {
417 output_counts->mv.comps[k].class0_fp[i][j] = comps_t->class0_fp[i][j];
418 }
419 }
420 for (int i = 0; i < MV_OFFSET_BITS; i++) {
421 for (int j = 0; j < 2; j++) {
422 output_counts->mv.comps[k].bits[i][j] = comps_t->bits[i][j];
423 }
424 }
425 for (int i = 0; i < MV_FP_SIZE; i++) {
426 output_counts->mv.comps[k].fp[i] = comps_t->fp[i];
427 }
428 }
429 }
430
output_image_buffer(const ImageBuffer & image_buffer,std::FILE * out_file)431 void output_image_buffer(const ImageBuffer &image_buffer, std::FILE *out_file) {
432 for (int plane = 0; plane < 3; ++plane) {
433 const int w = image_buffer.plane_width[plane];
434 const int h = image_buffer.plane_height[plane];
435 const uint8_t *buf = image_buffer.plane_buffer[plane].get();
436 fprintf(out_file, "%d %d\n", h, w);
437 for (int i = 0; i < w * h; ++i) {
438 fprintf(out_file, "%d ", (int)buf[i]);
439 }
440 fprintf(out_file, "\n");
441 }
442 }
443
init_image_buffer(ImageBuffer * image_buffer,int frame_width,int frame_height,vpx_img_fmt_t img_fmt)444 static bool init_image_buffer(ImageBuffer *image_buffer, int frame_width,
445 int frame_height, vpx_img_fmt_t img_fmt) {
446 for (int plane = 0; plane < 3; ++plane) {
447 const int w = get_plane_width(img_fmt, frame_width, plane);
448 const int h = get_plane_height(img_fmt, frame_height, plane);
449 image_buffer->plane_width[plane] = w;
450 image_buffer->plane_height[plane] = h;
451 image_buffer->plane_buffer[plane].reset(new (std::nothrow) uint8_t[w * h]);
452 if (image_buffer->plane_buffer[plane].get() == nullptr) {
453 return false;
454 }
455 }
456 return true;
457 }
458
ImageBuffer_to_IMAGE_BUFFER(const ImageBuffer & image_buffer,IMAGE_BUFFER * image_buffer_c)459 static void ImageBuffer_to_IMAGE_BUFFER(const ImageBuffer &image_buffer,
460 IMAGE_BUFFER *image_buffer_c) {
461 image_buffer_c->allocated = 1;
462 for (int plane = 0; plane < 3; ++plane) {
463 image_buffer_c->plane_width[plane] = image_buffer.plane_width[plane];
464 image_buffer_c->plane_height[plane] = image_buffer.plane_height[plane];
465 image_buffer_c->plane_buffer[plane] =
466 image_buffer.plane_buffer[plane].get();
467 }
468 }
469
get_max_coding_data_byte_size(int frame_width,int frame_height)470 static size_t get_max_coding_data_byte_size(int frame_width, int frame_height) {
471 return frame_width * frame_height * 3;
472 }
473
init_encode_frame_result(EncodeFrameResult * encode_frame_result,int frame_width,int frame_height,vpx_img_fmt_t img_fmt)474 static bool init_encode_frame_result(EncodeFrameResult *encode_frame_result,
475 int frame_width, int frame_height,
476 vpx_img_fmt_t img_fmt) {
477 const size_t max_coding_data_byte_size =
478 get_max_coding_data_byte_size(frame_width, frame_height);
479
480 encode_frame_result->coding_data.reset(
481 new (std::nothrow) uint8_t[max_coding_data_byte_size]);
482
483 encode_frame_result->num_rows_4x4 = get_num_unit_4x4(frame_height);
484 encode_frame_result->num_cols_4x4 = get_num_unit_4x4(frame_width);
485 encode_frame_result->partition_info.resize(encode_frame_result->num_rows_4x4 *
486 encode_frame_result->num_cols_4x4);
487 encode_frame_result->motion_vector_info.resize(
488 encode_frame_result->num_rows_4x4 * encode_frame_result->num_cols_4x4);
489
490 if (encode_frame_result->coding_data.get() == nullptr) {
491 return false;
492 }
493 return init_image_buffer(&encode_frame_result->coded_frame, frame_width,
494 frame_height, img_fmt);
495 }
496
encode_frame_result_update_rq_history(const RATE_QINDEX_HISTORY * rq_history,EncodeFrameResult * encode_frame_result)497 static void encode_frame_result_update_rq_history(
498 const RATE_QINDEX_HISTORY *rq_history,
499 EncodeFrameResult *encode_frame_result) {
500 encode_frame_result->recode_count = rq_history->recode_count;
501 for (int i = 0; i < encode_frame_result->recode_count; ++i) {
502 const int q_index = rq_history->q_index_history[i];
503 const int rate = rq_history->rate_history[i];
504 encode_frame_result->q_index_history.push_back(q_index);
505 encode_frame_result->rate_history.push_back(rate);
506 }
507 }
508
update_encode_frame_result(EncodeFrameResult * encode_frame_result,const ENCODE_FRAME_RESULT * encode_frame_info)509 static void update_encode_frame_result(
510 EncodeFrameResult *encode_frame_result,
511 const ENCODE_FRAME_RESULT *encode_frame_info) {
512 encode_frame_result->coding_data_bit_size =
513 encode_frame_result->coding_data_byte_size * 8;
514 encode_frame_result->show_idx = encode_frame_info->show_idx;
515 encode_frame_result->coding_idx = encode_frame_info->frame_coding_index;
516 assert(kRefFrameTypeMax == MAX_INTER_REF_FRAMES);
517 for (int i = 0; i < kRefFrameTypeMax; ++i) {
518 encode_frame_result->ref_frame_info.coding_indexes[i] =
519 encode_frame_info->ref_frame_coding_indexes[i];
520 encode_frame_result->ref_frame_info.valid_list[i] =
521 encode_frame_info->ref_frame_valid_list[i];
522 }
523 encode_frame_result->frame_type =
524 get_frame_type_from_update_type(encode_frame_info->update_type);
525 encode_frame_result->psnr = encode_frame_info->psnr;
526 encode_frame_result->sse = encode_frame_info->sse;
527 encode_frame_result->quantize_index = encode_frame_info->quantize_index;
528 update_partition_info(encode_frame_info->partition_info,
529 encode_frame_result->num_rows_4x4,
530 encode_frame_result->num_cols_4x4,
531 &encode_frame_result->partition_info[0]);
532 update_motion_vector_info(encode_frame_info->motion_vector_info,
533 encode_frame_result->num_rows_4x4,
534 encode_frame_result->num_cols_4x4,
535 &encode_frame_result->motion_vector_info[0],
536 kMotionVectorSubPixelPrecision);
537 update_frame_counts(&encode_frame_info->frame_counts,
538 &encode_frame_result->frame_counts);
539 encode_frame_result_update_rq_history(&encode_frame_info->rq_history,
540 encode_frame_result);
541 }
542
IncreaseGroupOfPictureIndex(GroupOfPicture * group_of_picture)543 static void IncreaseGroupOfPictureIndex(GroupOfPicture *group_of_picture) {
544 ++group_of_picture->next_encode_frame_index;
545 }
546
IsGroupOfPictureFinished(const GroupOfPicture & group_of_picture)547 static int IsGroupOfPictureFinished(const GroupOfPicture &group_of_picture) {
548 return static_cast<size_t>(group_of_picture.next_encode_frame_index) ==
549 group_of_picture.encode_frame_list.size();
550 }
551
operator ==(const RefFrameInfo & a,const RefFrameInfo & b)552 bool operator==(const RefFrameInfo &a, const RefFrameInfo &b) {
553 bool match = true;
554 for (int i = 0; i < kRefFrameTypeMax; ++i) {
555 match &= a.coding_indexes[i] == b.coding_indexes[i];
556 match &= a.valid_list[i] == b.valid_list[i];
557 }
558 return match;
559 }
560
InitRefFrameInfo(RefFrameInfo * ref_frame_info)561 static void InitRefFrameInfo(RefFrameInfo *ref_frame_info) {
562 for (int i = 0; i < kRefFrameTypeMax; ++i) {
563 ref_frame_info->coding_indexes[i] = -1;
564 ref_frame_info->valid_list[i] = 0;
565 }
566 }
567
568 // After finishing coding a frame, this function will update the coded frame
569 // into the ref_frame_info based on the frame_type and the coding_index.
PostUpdateRefFrameInfo(FrameType frame_type,int frame_coding_index,RefFrameInfo * ref_frame_info)570 static void PostUpdateRefFrameInfo(FrameType frame_type, int frame_coding_index,
571 RefFrameInfo *ref_frame_info) {
572 // This part is written based on the logics in vp9_configure_buffer_updates()
573 // and update_ref_frames()
574 int *ref_frame_coding_indexes = ref_frame_info->coding_indexes;
575 switch (frame_type) {
576 case kFrameTypeKey:
577 ref_frame_coding_indexes[kRefFrameTypeLast] = frame_coding_index;
578 ref_frame_coding_indexes[kRefFrameTypePast] = frame_coding_index;
579 ref_frame_coding_indexes[kRefFrameTypeFuture] = frame_coding_index;
580 break;
581 case kFrameTypeInter:
582 ref_frame_coding_indexes[kRefFrameTypeLast] = frame_coding_index;
583 break;
584 case kFrameTypeAltRef:
585 ref_frame_coding_indexes[kRefFrameTypeFuture] = frame_coding_index;
586 break;
587 case kFrameTypeOverlay:
588 // Reserve the past coding_index in the future slot. This logic is from
589 // update_ref_frames() with condition vp9_preserve_existing_gf() == 1
590 // TODO(angiebird): Invetegate why we need this.
591 ref_frame_coding_indexes[kRefFrameTypeFuture] =
592 ref_frame_coding_indexes[kRefFrameTypePast];
593 ref_frame_coding_indexes[kRefFrameTypePast] = frame_coding_index;
594 break;
595 case kFrameTypeGolden:
596 ref_frame_coding_indexes[kRefFrameTypePast] = frame_coding_index;
597 ref_frame_coding_indexes[kRefFrameTypeLast] = frame_coding_index;
598 break;
599 }
600
601 // This part is written based on the logics in get_ref_frame_flags() but we
602 // rename the flags alt, golden to future, past respectively. Mark
603 // non-duplicated reference frames as valid. The priorities are
604 // kRefFrameTypeLast > kRefFrameTypePast > kRefFrameTypeFuture.
605 const int last_index = ref_frame_coding_indexes[kRefFrameTypeLast];
606 const int past_index = ref_frame_coding_indexes[kRefFrameTypePast];
607 const int future_index = ref_frame_coding_indexes[kRefFrameTypeFuture];
608
609 int *ref_frame_valid_list = ref_frame_info->valid_list;
610 for (int ref_frame_idx = 0; ref_frame_idx < kRefFrameTypeMax;
611 ++ref_frame_idx) {
612 ref_frame_valid_list[ref_frame_idx] = 1;
613 }
614
615 if (past_index == last_index) {
616 ref_frame_valid_list[kRefFrameTypePast] = 0;
617 }
618
619 if (future_index == last_index) {
620 ref_frame_valid_list[kRefFrameTypeFuture] = 0;
621 }
622
623 if (future_index == past_index) {
624 ref_frame_valid_list[kRefFrameTypeFuture] = 0;
625 }
626 }
627
SetGroupOfPicture(int first_is_key_frame,int use_alt_ref,int coding_frame_count,int first_show_idx,int last_gop_use_alt_ref,int start_coding_index,const RefFrameInfo & start_ref_frame_info,GroupOfPicture * group_of_picture)628 static void SetGroupOfPicture(int first_is_key_frame, int use_alt_ref,
629 int coding_frame_count, int first_show_idx,
630 int last_gop_use_alt_ref, int start_coding_index,
631 const RefFrameInfo &start_ref_frame_info,
632 GroupOfPicture *group_of_picture) {
633 // Clean up the state of previous group of picture.
634 group_of_picture->encode_frame_list.clear();
635 group_of_picture->next_encode_frame_index = 0;
636 group_of_picture->show_frame_count = coding_frame_count - use_alt_ref;
637 group_of_picture->start_show_index = first_show_idx;
638 group_of_picture->start_coding_index = start_coding_index;
639 group_of_picture->first_is_key_frame = first_is_key_frame;
640 group_of_picture->use_alt_ref = use_alt_ref;
641 group_of_picture->last_gop_use_alt_ref = last_gop_use_alt_ref;
642
643 // We need to make a copy of start reference frame info because we
644 // use it to simulate the ref frame update.
645 RefFrameInfo ref_frame_info = start_ref_frame_info;
646
647 {
648 // First frame in the group of pictures. It's either key frame or show inter
649 // frame.
650 EncodeFrameInfo encode_frame_info;
651 // Set frame_type
652 if (first_is_key_frame) {
653 encode_frame_info.frame_type = kFrameTypeKey;
654 } else {
655 if (last_gop_use_alt_ref) {
656 encode_frame_info.frame_type = kFrameTypeOverlay;
657 } else {
658 encode_frame_info.frame_type = kFrameTypeGolden;
659 }
660 }
661
662 encode_frame_info.show_idx = first_show_idx;
663 encode_frame_info.coding_index = start_coding_index;
664
665 encode_frame_info.ref_frame_info = ref_frame_info;
666 PostUpdateRefFrameInfo(encode_frame_info.frame_type,
667 encode_frame_info.coding_index, &ref_frame_info);
668
669 group_of_picture->encode_frame_list.push_back(encode_frame_info);
670 }
671
672 const int show_frame_count = coding_frame_count - use_alt_ref;
673 if (use_alt_ref) {
674 // If there is alternate reference, it is always coded at the second place.
675 // Its show index (or timestamp) is at the last of this group
676 EncodeFrameInfo encode_frame_info;
677 encode_frame_info.frame_type = kFrameTypeAltRef;
678 encode_frame_info.show_idx = first_show_idx + show_frame_count;
679 encode_frame_info.coding_index = start_coding_index + 1;
680
681 encode_frame_info.ref_frame_info = ref_frame_info;
682 PostUpdateRefFrameInfo(encode_frame_info.frame_type,
683 encode_frame_info.coding_index, &ref_frame_info);
684
685 group_of_picture->encode_frame_list.push_back(encode_frame_info);
686 }
687
688 // Encode the rest show inter frames.
689 for (int i = 1; i < show_frame_count; ++i) {
690 EncodeFrameInfo encode_frame_info;
691 encode_frame_info.frame_type = kFrameTypeInter;
692 encode_frame_info.show_idx = first_show_idx + i;
693 encode_frame_info.coding_index = start_coding_index + use_alt_ref + i;
694
695 encode_frame_info.ref_frame_info = ref_frame_info;
696 PostUpdateRefFrameInfo(encode_frame_info.frame_type,
697 encode_frame_info.coding_index, &ref_frame_info);
698
699 group_of_picture->encode_frame_list.push_back(encode_frame_info);
700 }
701 }
702
703 // Gets group of picture information from VP9's decision, and update
704 // |group_of_picture| accordingly.
705 // This is called at the starting of encoding of each group of picture.
UpdateGroupOfPicture(const VP9_COMP * cpi,int start_coding_index,const RefFrameInfo & start_ref_frame_info,GroupOfPicture * group_of_picture)706 static void UpdateGroupOfPicture(const VP9_COMP *cpi, int start_coding_index,
707 const RefFrameInfo &start_ref_frame_info,
708 GroupOfPicture *group_of_picture) {
709 int first_is_key_frame;
710 int use_alt_ref;
711 int coding_frame_count;
712 int first_show_idx;
713 int last_gop_use_alt_ref;
714 vp9_get_next_group_of_picture(cpi, &first_is_key_frame, &use_alt_ref,
715 &coding_frame_count, &first_show_idx,
716 &last_gop_use_alt_ref);
717 SetGroupOfPicture(first_is_key_frame, use_alt_ref, coding_frame_count,
718 first_show_idx, last_gop_use_alt_ref, start_coding_index,
719 start_ref_frame_info, group_of_picture);
720 }
721
722 #define SET_STRUCT_VALUE(config, structure, ret, field) \
723 if (strcmp(config.name, #field) == 0) { \
724 structure->field = atoi(config.value); \
725 ret = 1; \
726 }
727
UpdateEncodeConfig(const EncodeConfig & config,VP9EncoderConfig * oxcf)728 static void UpdateEncodeConfig(const EncodeConfig &config,
729 VP9EncoderConfig *oxcf) {
730 int ret = 0;
731 SET_STRUCT_VALUE(config, oxcf, ret, key_freq);
732 SET_STRUCT_VALUE(config, oxcf, ret, two_pass_vbrmin_section);
733 SET_STRUCT_VALUE(config, oxcf, ret, two_pass_vbrmax_section);
734 SET_STRUCT_VALUE(config, oxcf, ret, under_shoot_pct);
735 SET_STRUCT_VALUE(config, oxcf, ret, over_shoot_pct);
736 SET_STRUCT_VALUE(config, oxcf, ret, max_threads);
737 SET_STRUCT_VALUE(config, oxcf, ret, frame_parallel_decoding_mode);
738 SET_STRUCT_VALUE(config, oxcf, ret, tile_columns);
739 SET_STRUCT_VALUE(config, oxcf, ret, arnr_max_frames);
740 SET_STRUCT_VALUE(config, oxcf, ret, arnr_strength);
741 SET_STRUCT_VALUE(config, oxcf, ret, lag_in_frames);
742 SET_STRUCT_VALUE(config, oxcf, ret, encode_breakout);
743 SET_STRUCT_VALUE(config, oxcf, ret, enable_tpl_model);
744 SET_STRUCT_VALUE(config, oxcf, ret, enable_auto_arf);
745 if (ret == 0) {
746 fprintf(stderr, "Ignored unsupported encode_config %s\n", config.name);
747 }
748 }
749
GetEncodeConfig(int frame_width,int frame_height,vpx_rational_t frame_rate,int target_bitrate,int encode_speed,vpx_enc_pass enc_pass,const std::vector<EncodeConfig> & encode_config_list)750 static VP9EncoderConfig GetEncodeConfig(
751 int frame_width, int frame_height, vpx_rational_t frame_rate,
752 int target_bitrate, int encode_speed, vpx_enc_pass enc_pass,
753 const std::vector<EncodeConfig> &encode_config_list) {
754 VP9EncoderConfig oxcf =
755 vp9_get_encoder_config(frame_width, frame_height, frame_rate,
756 target_bitrate, encode_speed, enc_pass);
757 for (const auto &config : encode_config_list) {
758 UpdateEncodeConfig(config, &oxcf);
759 }
760 if (enc_pass == VPX_RC_FIRST_PASS) {
761 oxcf.lag_in_frames = 0;
762 }
763 return oxcf;
764 }
765
SimpleEncode(int frame_width,int frame_height,int frame_rate_num,int frame_rate_den,int target_bitrate,int num_frames,const char * infile_path,const char * outfile_path)766 SimpleEncode::SimpleEncode(int frame_width, int frame_height,
767 int frame_rate_num, int frame_rate_den,
768 int target_bitrate, int num_frames,
769 const char *infile_path, const char *outfile_path) {
770 impl_ptr_ = std::unique_ptr<EncodeImpl>(new EncodeImpl());
771 frame_width_ = frame_width;
772 frame_height_ = frame_height;
773 frame_rate_num_ = frame_rate_num;
774 frame_rate_den_ = frame_rate_den;
775 target_bitrate_ = target_bitrate;
776 num_frames_ = num_frames;
777 encode_speed_ = 0;
778
779 frame_coding_index_ = 0;
780 show_frame_count_ = 0;
781
782 key_frame_group_index_ = 0;
783 key_frame_group_size_ = 0;
784
785 // TODO(angirbid): Should we keep a file pointer here or keep the file_path?
786 assert(infile_path != nullptr);
787 in_file_ = fopen(infile_path, "r");
788 if (outfile_path != nullptr) {
789 out_file_ = fopen(outfile_path, "w");
790 } else {
791 out_file_ = nullptr;
792 }
793 impl_ptr_->cpi = nullptr;
794 impl_ptr_->img_fmt = VPX_IMG_FMT_I420;
795
796 InitRefFrameInfo(&ref_frame_info_);
797 }
798
SetEncodeSpeed(int encode_speed)799 void SimpleEncode::SetEncodeSpeed(int encode_speed) {
800 encode_speed_ = encode_speed;
801 }
802
SetEncodeConfig(const char * name,const char * value)803 StatusCode SimpleEncode::SetEncodeConfig(const char *name, const char *value) {
804 if (name == nullptr || value == nullptr) {
805 fprintf(stderr, "SetEncodeConfig: null pointer, name %p value %p\n", name,
806 value);
807 return StatusError;
808 }
809 EncodeConfig config;
810 snprintf(config.name, ENCODE_CONFIG_BUF_SIZE, "%s", name);
811 snprintf(config.value, ENCODE_CONFIG_BUF_SIZE, "%s", value);
812 impl_ptr_->encode_config_list.push_back(config);
813 return StatusOk;
814 }
815
DumpEncodeConfigs(int pass,FILE * fp)816 StatusCode SimpleEncode::DumpEncodeConfigs(int pass, FILE *fp) {
817 if (fp == nullptr) {
818 fprintf(stderr, "DumpEncodeConfigs: null pointer, fp %p\n", fp);
819 return StatusError;
820 }
821 vpx_enc_pass enc_pass;
822 if (pass == 1) {
823 enc_pass = VPX_RC_FIRST_PASS;
824 } else {
825 enc_pass = VPX_RC_LAST_PASS;
826 }
827 const vpx_rational_t frame_rate =
828 make_vpx_rational(frame_rate_num_, frame_rate_den_);
829 const VP9EncoderConfig oxcf =
830 GetEncodeConfig(frame_width_, frame_height_, frame_rate, target_bitrate_,
831 encode_speed_, enc_pass, impl_ptr_->encode_config_list);
832 vp9_dump_encoder_config(&oxcf, fp);
833 return StatusOk;
834 }
835
ComputeFirstPassStats()836 void SimpleEncode::ComputeFirstPassStats() {
837 vpx_rational_t frame_rate =
838 make_vpx_rational(frame_rate_num_, frame_rate_den_);
839 const VP9EncoderConfig oxcf = GetEncodeConfig(
840 frame_width_, frame_height_, frame_rate, target_bitrate_, encode_speed_,
841 VPX_RC_FIRST_PASS, impl_ptr_->encode_config_list);
842 VP9_COMP *cpi = init_encoder(&oxcf, impl_ptr_->img_fmt);
843 struct lookahead_ctx *lookahead = cpi->lookahead;
844 int i;
845 int use_highbitdepth = 0;
846 const int num_rows_16x16 = get_num_unit_16x16(frame_height_);
847 const int num_cols_16x16 = get_num_unit_16x16(frame_width_);
848 #if CONFIG_VP9_HIGHBITDEPTH
849 use_highbitdepth = cpi->common.use_highbitdepth;
850 #endif
851 vpx_image_t img;
852 vpx_img_alloc(&img, impl_ptr_->img_fmt, frame_width_, frame_height_, 1);
853 rewind(in_file_);
854 impl_ptr_->first_pass_stats.clear();
855 for (i = 0; i < num_frames_; ++i) {
856 assert(!vp9_lookahead_full(lookahead));
857 if (img_read(&img, in_file_)) {
858 int next_show_idx = vp9_lookahead_next_show_idx(lookahead);
859 int64_t ts_start =
860 timebase_units_to_ticks(&oxcf.g_timebase_in_ts, next_show_idx);
861 int64_t ts_end =
862 timebase_units_to_ticks(&oxcf.g_timebase_in_ts, next_show_idx + 1);
863 YV12_BUFFER_CONFIG sd;
864 image2yuvconfig(&img, &sd);
865 vp9_lookahead_push(lookahead, &sd, ts_start, ts_end, use_highbitdepth, 0);
866 {
867 int64_t time_stamp;
868 int64_t time_end;
869 int flush = 1; // Makes vp9_get_compressed_data process a frame
870 size_t size;
871 unsigned int frame_flags = 0;
872 ENCODE_FRAME_RESULT encode_frame_info;
873 vp9_init_encode_frame_result(&encode_frame_info);
874 // TODO(angiebird): Call vp9_first_pass directly
875 vp9_get_compressed_data(cpi, &frame_flags, &size, nullptr, &time_stamp,
876 &time_end, flush, &encode_frame_info);
877 // vp9_get_compressed_data only generates first pass stats not
878 // compresses data
879 assert(size == 0);
880 // Get vp9 first pass motion vector info.
881 std::vector<MotionVectorInfo> mv_info(num_rows_16x16 * num_cols_16x16);
882 update_motion_vector_info(cpi->fp_motion_vector_info, num_rows_16x16,
883 num_cols_16x16, mv_info.data(),
884 kMotionVectorFullPixelPrecision);
885 fp_motion_vector_info_.push_back(mv_info);
886 }
887 impl_ptr_->first_pass_stats.push_back(vp9_get_frame_stats(&cpi->twopass));
888 }
889 }
890 vp9_end_first_pass(cpi);
891 // TODO(angiebird): Store the total_stats apart form first_pass_stats
892 impl_ptr_->first_pass_stats.push_back(vp9_get_total_stats(&cpi->twopass));
893 free_encoder(cpi);
894 rewind(in_file_);
895 vpx_img_free(&img);
896
897 // Generate key_frame_map based on impl_ptr_->first_pass_stats.
898 key_frame_map_ = ComputeKeyFrameMap();
899 }
900
ObserveFirstPassStats()901 std::vector<std::vector<double>> SimpleEncode::ObserveFirstPassStats() {
902 std::vector<std::vector<double>> output_stats;
903 // TODO(angiebird): This function make several assumptions of
904 // FIRSTPASS_STATS. 1) All elements in FIRSTPASS_STATS are double except the
905 // last one. 2) The last entry of first_pass_stats is the total_stats.
906 // Change the code structure, so that we don't have to make these assumptions
907
908 // Note the last entry of first_pass_stats is the total_stats, we don't need
909 // it.
910 for (size_t i = 0; i < impl_ptr_->first_pass_stats.size() - 1; ++i) {
911 double *buf_start =
912 reinterpret_cast<double *>(&impl_ptr_->first_pass_stats[i]);
913 // We use - 1 here because the last member in FIRSTPASS_STATS is not double
914 double *buf_end =
915 buf_start + sizeof(impl_ptr_->first_pass_stats[i]) / sizeof(*buf_end) -
916 1;
917 std::vector<double> this_stats(buf_start, buf_end);
918 output_stats.push_back(this_stats);
919 }
920 return output_stats;
921 }
922
923 std::vector<std::vector<MotionVectorInfo>>
ObserveFirstPassMotionVectors()924 SimpleEncode::ObserveFirstPassMotionVectors() {
925 return fp_motion_vector_info_;
926 }
927
SetExternalGroupOfPicturesMap(int * gop_map,int gop_map_size)928 void SimpleEncode::SetExternalGroupOfPicturesMap(int *gop_map,
929 int gop_map_size) {
930 for (int i = 0; i < gop_map_size; ++i) {
931 gop_map_.push_back(gop_map[i]);
932 }
933 // The following will check and modify gop_map_ to make sure the
934 // gop_map_ satisfies the constraints.
935 // 1) Each key frame position should be at the start of a gop.
936 // 2) The last gop should not use an alt ref.
937 assert(gop_map_.size() == key_frame_map_.size());
938 int last_gop_start = 0;
939 for (int i = 0; static_cast<size_t>(i) < gop_map_.size(); ++i) {
940 if (key_frame_map_[i] == 1 && gop_map_[i] == 0) {
941 fprintf(stderr, "Add an extra gop start at show_idx %d\n", i);
942 // Insert a gop start at key frame location.
943 gop_map_[i] |= kGopMapFlagStart;
944 gop_map_[i] |= kGopMapFlagUseAltRef;
945 }
946 if (gop_map_[i] & kGopMapFlagStart) {
947 last_gop_start = i;
948 }
949 }
950 if (gop_map_[last_gop_start] & kGopMapFlagUseAltRef) {
951 fprintf(stderr,
952 "Last group of pictures starting at show_idx %d shouldn't use alt "
953 "ref\n",
954 last_gop_start);
955 gop_map_[last_gop_start] &= ~kGopMapFlagUseAltRef;
956 }
957 }
958
ObserveExternalGroupOfPicturesMap()959 std::vector<int> SimpleEncode::ObserveExternalGroupOfPicturesMap() {
960 return gop_map_;
961 }
962
963 template <typename T>
GetVectorData(const std::vector<T> & v)964 T *GetVectorData(const std::vector<T> &v) {
965 if (v.empty()) {
966 return nullptr;
967 }
968 return const_cast<T *>(v.data());
969 }
970
GetGopCommand(const std::vector<int> & gop_map,int start_show_index)971 static GOP_COMMAND GetGopCommand(const std::vector<int> &gop_map,
972 int start_show_index) {
973 GOP_COMMAND gop_command;
974 if (gop_map.size() > 0) {
975 assert(static_cast<size_t>(start_show_index) < gop_map.size());
976 assert((gop_map[start_show_index] & kGopMapFlagStart) != 0);
977 int end_show_index = start_show_index + 1;
978 // gop_map[end_show_index] & kGopMapFlagStart == 0 means this is
979 // the start of a gop.
980 while (static_cast<size_t>(end_show_index) < gop_map.size() &&
981 (gop_map[end_show_index] & kGopMapFlagStart) == 0) {
982 ++end_show_index;
983 }
984 const int show_frame_count = end_show_index - start_show_index;
985 int use_alt_ref = (gop_map[start_show_index] & kGopMapFlagUseAltRef) != 0;
986 if (static_cast<size_t>(end_show_index) == gop_map.size()) {
987 // This is the last gop group, there must be no altref.
988 use_alt_ref = 0;
989 }
990 gop_command_on(&gop_command, show_frame_count, use_alt_ref);
991 } else {
992 gop_command_off(&gop_command);
993 }
994 return gop_command;
995 }
996
StartEncode()997 void SimpleEncode::StartEncode() {
998 assert(impl_ptr_->first_pass_stats.size() > 0);
999 vpx_rational_t frame_rate =
1000 make_vpx_rational(frame_rate_num_, frame_rate_den_);
1001 VP9EncoderConfig oxcf = GetEncodeConfig(
1002 frame_width_, frame_height_, frame_rate, target_bitrate_, encode_speed_,
1003 VPX_RC_LAST_PASS, impl_ptr_->encode_config_list);
1004
1005 vpx_fixed_buf_t stats;
1006 stats.buf = GetVectorData(impl_ptr_->first_pass_stats);
1007 stats.sz = sizeof(impl_ptr_->first_pass_stats[0]) *
1008 impl_ptr_->first_pass_stats.size();
1009
1010 vp9_set_first_pass_stats(&oxcf, &stats);
1011 assert(impl_ptr_->cpi == nullptr);
1012 impl_ptr_->cpi = init_encoder(&oxcf, impl_ptr_->img_fmt);
1013 vpx_img_alloc(&impl_ptr_->tmp_img, impl_ptr_->img_fmt, frame_width_,
1014 frame_height_, 1);
1015
1016 frame_coding_index_ = 0;
1017 show_frame_count_ = 0;
1018
1019 UpdateKeyFrameGroup(show_frame_count_);
1020
1021 const GOP_COMMAND gop_command = GetGopCommand(gop_map_, show_frame_count_);
1022 encode_command_set_gop_command(&impl_ptr_->cpi->encode_command, gop_command);
1023 UpdateGroupOfPicture(impl_ptr_->cpi, frame_coding_index_, ref_frame_info_,
1024 &group_of_picture_);
1025 rewind(in_file_);
1026
1027 if (out_file_ != nullptr) {
1028 const char *fourcc = "VP90";
1029 // In SimpleEncode, we use time_base = 1 / TICKS_PER_SEC.
1030 // Based on that, the ivf_timestamp for each image is set to
1031 // show_idx * TICKS_PER_SEC / frame_rate
1032 // such that each image's actual timestamp in seconds can be computed as
1033 // ivf_timestamp * time_base == show_idx / frame_rate
1034 // TODO(angiebird): 1) Add unit test for ivf timestamp.
1035 // 2) Simplify the frame_rate setting process.
1036 vpx_rational_t time_base = make_vpx_rational(1, TICKS_PER_SEC);
1037 ivf_write_file_header_with_video_info(out_file_, *(const uint32_t *)fourcc,
1038 num_frames_, frame_width_,
1039 frame_height_, time_base);
1040 }
1041 }
1042
EndEncode()1043 void SimpleEncode::EndEncode() {
1044 free_encoder(impl_ptr_->cpi);
1045 impl_ptr_->cpi = nullptr;
1046 vpx_img_free(&impl_ptr_->tmp_img);
1047 rewind(in_file_);
1048 }
1049
UpdateKeyFrameGroup(int key_frame_show_index)1050 void SimpleEncode::UpdateKeyFrameGroup(int key_frame_show_index) {
1051 const VP9_COMP *cpi = impl_ptr_->cpi;
1052 key_frame_group_index_ = 0;
1053 key_frame_group_size_ = vp9_get_frames_to_next_key(
1054 &cpi->oxcf, &cpi->frame_info, &cpi->twopass.first_pass_info,
1055 key_frame_show_index, cpi->rc.min_gf_interval);
1056 assert(key_frame_group_size_ > 0);
1057 // Init the reference frame info when a new key frame group appears.
1058 InitRefFrameInfo(&ref_frame_info_);
1059 }
1060
PostUpdateKeyFrameGroupIndex(FrameType frame_type)1061 void SimpleEncode::PostUpdateKeyFrameGroupIndex(FrameType frame_type) {
1062 if (frame_type != kFrameTypeAltRef) {
1063 // key_frame_group_index_ only counts show frames
1064 ++key_frame_group_index_;
1065 }
1066 }
1067
GetKeyFrameGroupSize() const1068 int SimpleEncode::GetKeyFrameGroupSize() const { return key_frame_group_size_; }
1069
ObserveGroupOfPicture() const1070 GroupOfPicture SimpleEncode::ObserveGroupOfPicture() const {
1071 return group_of_picture_;
1072 }
1073
GetNextEncodeFrameInfo() const1074 EncodeFrameInfo SimpleEncode::GetNextEncodeFrameInfo() const {
1075 return group_of_picture_
1076 .encode_frame_list[group_of_picture_.next_encode_frame_index];
1077 }
1078
PostUpdateState(const EncodeFrameResult & encode_frame_result)1079 void SimpleEncode::PostUpdateState(
1080 const EncodeFrameResult &encode_frame_result) {
1081 // This function needs to be called before the increament of
1082 // frame_coding_index_
1083 PostUpdateRefFrameInfo(encode_frame_result.frame_type, frame_coding_index_,
1084 &ref_frame_info_);
1085 ++frame_coding_index_;
1086 if (encode_frame_result.frame_type != kFrameTypeAltRef) {
1087 // Only kFrameTypeAltRef is not a show frame
1088 ++show_frame_count_;
1089 }
1090
1091 PostUpdateKeyFrameGroupIndex(encode_frame_result.frame_type);
1092 if (key_frame_group_index_ == key_frame_group_size_) {
1093 UpdateKeyFrameGroup(show_frame_count_);
1094 }
1095
1096 IncreaseGroupOfPictureIndex(&group_of_picture_);
1097 if (IsGroupOfPictureFinished(group_of_picture_)) {
1098 const GOP_COMMAND gop_command = GetGopCommand(gop_map_, show_frame_count_);
1099 encode_command_set_gop_command(&impl_ptr_->cpi->encode_command,
1100 gop_command);
1101 // This function needs to be called after ref_frame_info_ is updated
1102 // properly in PostUpdateRefFrameInfo() and UpdateKeyFrameGroup().
1103 UpdateGroupOfPicture(impl_ptr_->cpi, frame_coding_index_, ref_frame_info_,
1104 &group_of_picture_);
1105 }
1106 }
1107
EncodeFrame(EncodeFrameResult * encode_frame_result)1108 void SimpleEncode::EncodeFrame(EncodeFrameResult *encode_frame_result) {
1109 VP9_COMP *cpi = impl_ptr_->cpi;
1110 struct lookahead_ctx *lookahead = cpi->lookahead;
1111 int use_highbitdepth = 0;
1112 #if CONFIG_VP9_HIGHBITDEPTH
1113 use_highbitdepth = cpi->common.use_highbitdepth;
1114 #endif
1115 // The lookahead's size is set to oxcf->lag_in_frames.
1116 // We want to fill lookahead to it's max capacity if possible so that the
1117 // encoder can construct alt ref frame in time.
1118 // In the other words, we hope vp9_get_compressed_data to encode a frame
1119 // every time in the function
1120 while (!vp9_lookahead_full(lookahead)) {
1121 // TODO(angiebird): Check whether we can move this file read logics to
1122 // lookahead
1123 if (img_read(&impl_ptr_->tmp_img, in_file_)) {
1124 int next_show_idx = vp9_lookahead_next_show_idx(lookahead);
1125 int64_t ts_start =
1126 timebase_units_to_ticks(&cpi->oxcf.g_timebase_in_ts, next_show_idx);
1127 int64_t ts_end = timebase_units_to_ticks(&cpi->oxcf.g_timebase_in_ts,
1128 next_show_idx + 1);
1129 YV12_BUFFER_CONFIG sd;
1130 image2yuvconfig(&impl_ptr_->tmp_img, &sd);
1131 vp9_lookahead_push(lookahead, &sd, ts_start, ts_end, use_highbitdepth, 0);
1132 } else {
1133 break;
1134 }
1135 }
1136
1137 if (init_encode_frame_result(encode_frame_result, frame_width_, frame_height_,
1138 impl_ptr_->img_fmt)) {
1139 int64_t time_stamp;
1140 int64_t time_end;
1141 int flush = 1; // Make vp9_get_compressed_data encode a frame
1142 unsigned int frame_flags = 0;
1143 ENCODE_FRAME_RESULT encode_frame_info;
1144 vp9_init_encode_frame_result(&encode_frame_info);
1145 ImageBuffer_to_IMAGE_BUFFER(encode_frame_result->coded_frame,
1146 &encode_frame_info.coded_frame);
1147 vp9_get_compressed_data(cpi, &frame_flags,
1148 &encode_frame_result->coding_data_byte_size,
1149 encode_frame_result->coding_data.get(), &time_stamp,
1150 &time_end, flush, &encode_frame_info);
1151 if (out_file_ != nullptr) {
1152 ivf_write_frame_header(out_file_, time_stamp,
1153 encode_frame_result->coding_data_byte_size);
1154 fwrite(encode_frame_result->coding_data.get(), 1,
1155 encode_frame_result->coding_data_byte_size, out_file_);
1156 }
1157
1158 // vp9_get_compressed_data is expected to encode a frame every time, so the
1159 // data size should be greater than zero.
1160 if (encode_frame_result->coding_data_byte_size <= 0) {
1161 fprintf(stderr, "Coding data size <= 0\n");
1162 abort();
1163 }
1164 const size_t max_coding_data_byte_size =
1165 get_max_coding_data_byte_size(frame_width_, frame_height_);
1166 if (encode_frame_result->coding_data_byte_size >
1167 max_coding_data_byte_size) {
1168 fprintf(stderr, "Coding data size exceeds the maximum.\n");
1169 abort();
1170 }
1171
1172 update_encode_frame_result(encode_frame_result, &encode_frame_info);
1173 PostUpdateState(*encode_frame_result);
1174 } else {
1175 // TODO(angiebird): Clean up encode_frame_result.
1176 fprintf(stderr, "init_encode_frame_result() failed.\n");
1177 this->EndEncode();
1178 }
1179 }
1180
EncodeFrameWithQuantizeIndex(EncodeFrameResult * encode_frame_result,int quantize_index)1181 void SimpleEncode::EncodeFrameWithQuantizeIndex(
1182 EncodeFrameResult *encode_frame_result, int quantize_index) {
1183 encode_command_set_external_quantize_index(&impl_ptr_->cpi->encode_command,
1184 quantize_index);
1185 EncodeFrame(encode_frame_result);
1186 encode_command_reset_external_quantize_index(&impl_ptr_->cpi->encode_command);
1187 }
1188
EncodeFrameWithTargetFrameBits(EncodeFrameResult * encode_frame_result,int target_frame_bits,double percent_diff)1189 void SimpleEncode::EncodeFrameWithTargetFrameBits(
1190 EncodeFrameResult *encode_frame_result, int target_frame_bits,
1191 double percent_diff) {
1192 encode_command_set_target_frame_bits(&impl_ptr_->cpi->encode_command,
1193 target_frame_bits, percent_diff);
1194 EncodeFrame(encode_frame_result);
1195 encode_command_reset_target_frame_bits(&impl_ptr_->cpi->encode_command);
1196 }
1197
GetCodingFrameNumFromGopMap(const std::vector<int> & gop_map)1198 static int GetCodingFrameNumFromGopMap(const std::vector<int> &gop_map) {
1199 int start_show_index = 0;
1200 int coding_frame_count = 0;
1201 while (static_cast<size_t>(start_show_index) < gop_map.size()) {
1202 const GOP_COMMAND gop_command = GetGopCommand(gop_map, start_show_index);
1203 start_show_index += gop_command.show_frame_count;
1204 coding_frame_count += gop_command_coding_frame_count(&gop_command);
1205 }
1206 assert(start_show_index == gop_map.size());
1207 return coding_frame_count;
1208 }
1209
GetCodingFrameNum() const1210 int SimpleEncode::GetCodingFrameNum() const {
1211 assert(impl_ptr_->first_pass_stats.size() > 0);
1212 if (gop_map_.size() > 0) {
1213 return GetCodingFrameNumFromGopMap(gop_map_);
1214 }
1215
1216 // These are the default settings for now.
1217 const int multi_layer_arf = 0;
1218 const int allow_alt_ref = 1;
1219 vpx_rational_t frame_rate =
1220 make_vpx_rational(frame_rate_num_, frame_rate_den_);
1221 const VP9EncoderConfig oxcf = GetEncodeConfig(
1222 frame_width_, frame_height_, frame_rate, target_bitrate_, encode_speed_,
1223 VPX_RC_LAST_PASS, impl_ptr_->encode_config_list);
1224 FRAME_INFO frame_info = vp9_get_frame_info(&oxcf);
1225 FIRST_PASS_INFO first_pass_info;
1226 fps_init_first_pass_info(&first_pass_info,
1227 GetVectorData(impl_ptr_->first_pass_stats),
1228 num_frames_);
1229 return vp9_get_coding_frame_num(&oxcf, &frame_info, &first_pass_info,
1230 multi_layer_arf, allow_alt_ref);
1231 }
1232
ComputeKeyFrameMap() const1233 std::vector<int> SimpleEncode::ComputeKeyFrameMap() const {
1234 // The last entry of first_pass_stats is the overall stats.
1235 assert(impl_ptr_->first_pass_stats.size() == num_frames_ + 1);
1236 vpx_rational_t frame_rate =
1237 make_vpx_rational(frame_rate_num_, frame_rate_den_);
1238 const VP9EncoderConfig oxcf = GetEncodeConfig(
1239 frame_width_, frame_height_, frame_rate, target_bitrate_, encode_speed_,
1240 VPX_RC_LAST_PASS, impl_ptr_->encode_config_list);
1241 FRAME_INFO frame_info = vp9_get_frame_info(&oxcf);
1242 FIRST_PASS_INFO first_pass_info;
1243 fps_init_first_pass_info(&first_pass_info,
1244 GetVectorData(impl_ptr_->first_pass_stats),
1245 num_frames_);
1246 std::vector<int> key_frame_map(num_frames_, 0);
1247 vp9_get_key_frame_map(&oxcf, &frame_info, &first_pass_info,
1248 GetVectorData(key_frame_map));
1249 return key_frame_map;
1250 }
1251
ObserveKeyFrameMap() const1252 std::vector<int> SimpleEncode::ObserveKeyFrameMap() const {
1253 return key_frame_map_;
1254 }
1255
GetFramePixelCount() const1256 uint64_t SimpleEncode::GetFramePixelCount() const {
1257 assert(frame_width_ % 2 == 0);
1258 assert(frame_height_ % 2 == 0);
1259 switch (impl_ptr_->img_fmt) {
1260 case VPX_IMG_FMT_I420: return frame_width_ * frame_height_ * 3 / 2;
1261 case VPX_IMG_FMT_I422: return frame_width_ * frame_height_ * 2;
1262 case VPX_IMG_FMT_I444: return frame_width_ * frame_height_ * 3;
1263 case VPX_IMG_FMT_I440: return frame_width_ * frame_height_ * 2;
1264 case VPX_IMG_FMT_I42016: return frame_width_ * frame_height_ * 3 / 2;
1265 case VPX_IMG_FMT_I42216: return frame_width_ * frame_height_ * 2;
1266 case VPX_IMG_FMT_I44416: return frame_width_ * frame_height_ * 3;
1267 case VPX_IMG_FMT_I44016: return frame_width_ * frame_height_ * 2;
1268 default: return 0;
1269 }
1270 }
1271
~SimpleEncode()1272 SimpleEncode::~SimpleEncode() {
1273 if (in_file_ != nullptr) {
1274 fclose(in_file_);
1275 }
1276 if (out_file_ != nullptr) {
1277 fclose(out_file_);
1278 }
1279 }
1280
1281 } // namespace vp9
1282