FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libavcodec/opus/enc.c
Date: 2026-06-06 18:10:07
Exec Total Coverage
Lines: 160 405 39.5%
Functions: 9 18 50.0%
Branches: 60 216 27.8%
Line Branch Exec Source
1 /*
2 * Opus encoder
3 * Copyright (c) 2017 Rostislav Pehlivanov <atomnuker@gmail.com>
4 *
5 * This file is part of FFmpeg.
6 *
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22 #include <float.h>
23
24 #include "encode.h"
25 #include "enc.h"
26 #include "pvq.h"
27 #include "enc_psy.h"
28 #include "tab.h"
29
30 #include "libavutil/channel_layout.h"
31 #include "libavutil/float_dsp.h"
32 #include "libavutil/mem.h"
33 #include "libavutil/mem_internal.h"
34 #include "libavutil/opt.h"
35 #include "bytestream.h"
36 #include "audio_frame_queue.h"
37 #include "codec_internal.h"
38
39 typedef struct OpusEncContext {
40 AVClass *av_class;
41 OpusEncOptions options;
42 OpusPsyContext psyctx;
43 AVCodecContext *avctx;
44 AudioFrameQueue afq;
45 AVFloatDSPContext *dsp;
46 AVTXContext *tx[CELT_BLOCK_NB];
47 av_tx_fn tx_fn[CELT_BLOCK_NB];
48 CeltPVQ *pvq;
49 struct FFBufQueue bufqueue;
50
51 uint8_t enc_id[64];
52 int enc_id_bits;
53
54 OpusPacketInfo packet;
55
56 int channels;
57
58 CeltFrame *frame;
59 OpusRangeCoder *rc;
60
61 /* Actual energy the decoder will have */
62 float last_quantized_energy[OPUS_MAX_CHANNELS][CELT_MAX_BANDS];
63
64 DECLARE_ALIGNED(32, float, scratch)[2048];
65 } OpusEncContext;
66
67 1 static void opus_write_extradata(AVCodecContext *avctx)
68 {
69 1 uint8_t *bs = avctx->extradata;
70
71 1 bytestream_put_buffer(&bs, "OpusHead", 8);
72 1 bytestream_put_byte (&bs, 0x1);
73 1 bytestream_put_byte (&bs, avctx->ch_layout.nb_channels);
74 1 bytestream_put_le16 (&bs, avctx->initial_padding);
75 1 bytestream_put_le32 (&bs, avctx->sample_rate);
76 1 bytestream_put_le16 (&bs, 0x0);
77 1 bytestream_put_byte (&bs, 0x0); /* Default layout */
78 1 }
79
80 2 static int opus_gen_toc(OpusEncContext *s, uint8_t *toc, int *size, int *fsize_needed)
81 {
82 2 int tmp = 0x0, extended_toc = 0;
83 static const int toc_cfg[][OPUS_MODE_NB][OPUS_BANDWITH_NB] = {
84 /* Silk Hybrid Celt Layer */
85 /* NB MB WB SWB FB NB MB WB SWB FB NB MB WB SWB FB Bandwidth */
86 { { 0, 0, 0, 0, 0 }, { 0, 0, 0, 0, 0 }, { 17, 0, 21, 25, 29 } }, /* 2.5 ms */
87 { { 0, 0, 0, 0, 0 }, { 0, 0, 0, 0, 0 }, { 18, 0, 22, 26, 30 } }, /* 5 ms */
88 { { 1, 5, 9, 0, 0 }, { 0, 0, 0, 13, 15 }, { 19, 0, 23, 27, 31 } }, /* 10 ms */
89 { { 2, 6, 10, 0, 0 }, { 0, 0, 0, 14, 16 }, { 20, 0, 24, 28, 32 } }, /* 20 ms */
90 { { 3, 7, 11, 0, 0 }, { 0, 0, 0, 0, 0 }, { 0, 0, 0, 0, 0 } }, /* 40 ms */
91 { { 4, 8, 12, 0, 0 }, { 0, 0, 0, 0, 0 }, { 0, 0, 0, 0, 0 } }, /* 60 ms */
92 };
93 2 int cfg = toc_cfg[s->packet.framesize][s->packet.mode][s->packet.bandwidth];
94 2 *fsize_needed = 0;
95
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (!cfg)
96 ✗ return 1;
97
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (s->packet.frames == 2) { /* 2 packets */
98 ✗ if (s->frame[0].framebits == s->frame[1].framebits) { /* same size */
99 ✗ tmp = 0x1;
100 } else { /* different size */
101 ✗ tmp = 0x2;
102 ✗ *fsize_needed = 1; /* put frame sizes in the packet */
103 }
104
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 } else if (s->packet.frames > 2) {
105 ✗ tmp = 0x3;
106 ✗ extended_toc = 1;
107 }
108
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 tmp |= (s->channels > 1) << 2; /* Stereo or mono */
109 2 tmp |= (cfg - 1) << 3; /* codec configuration */
110 2 *toc++ = tmp;
111
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (extended_toc) {
112 ✗ for (int i = 0; i < (s->packet.frames - 1); i++)
113 ✗ *fsize_needed |= (s->frame[i].framebits != s->frame[i + 1].framebits);
114 ✗ tmp = (*fsize_needed) << 7; /* vbr flag */
115 ✗ tmp |= (0) << 6; /* padding flag */
116 ✗ tmp |= s->packet.frames;
117 ✗ *toc++ = tmp;
118 }
119 2 *size = 1 + extended_toc;
120 2 return 0;
121 }
122
123 2 static void celt_frame_setup_input(OpusEncContext *s, CeltFrame *f)
124 {
125 2 AVFrame *cur = NULL;
126 2 const int subframesize = s->avctx->frame_size;
127 2 int subframes = OPUS_BLOCK_SIZE(s->packet.framesize) / subframesize;
128
129 2 cur = ff_bufqueue_get(&s->bufqueue);
130
131
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
 6 for (int ch = 0; ch < f->channels; ch++) {
132 4 CeltBlock *b = &f->block[ch];
133 4 const char *input = cur->extended_data[ch];
134 4 size_t bps = av_get_bytes_per_sample(cur->format);
135 /* The MDCT overlap is the trailing CELT_OVERLAP samples of the
136 * previous packet's last frame. Because the encoder advertises
137 * AV_CODEC_CAP_SMALL_LAST_FRAME, that frame can be shorter than
138 * CELT_OVERLAP; in that case, zero-pad the leading part of the
139 * overlap buffer and copy only what's available. */
140 4 int n = FFMIN(cur->nb_samples, CELT_OVERLAP);
141
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
 4 if (n < CELT_OVERLAP) {
142 ✗ memset(b->overlap, 0, (CELT_OVERLAP - n) * bps);
143 }
144 4 memcpy((char *)b->overlap + (CELT_OVERLAP - n) * bps,
145 4 input + (cur->nb_samples - n) * bps,
146 n * bps);
147 }
148
149 2 av_frame_free(&cur);
150
151
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
 4 for (int sf = 0; sf < subframes; sf++) {
152
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (sf != (subframes - 1))
153 ✗ cur = ff_bufqueue_get(&s->bufqueue);
154 else
155 2 cur = ff_bufqueue_peek(&s->bufqueue, 0);
156
157
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
 6 for (int ch = 0; ch < f->channels; ch++) {
158 4 CeltBlock *b = &f->block[ch];
159 4 const void *input = cur->extended_data[ch];
160 4 const size_t bps = av_get_bytes_per_sample(cur->format);
161 4 const size_t left = (subframesize - cur->nb_samples)*bps;
162 4 const size_t len = FFMIN(subframesize, cur->nb_samples)*bps;
163 4 memcpy(&b->samples[sf*subframesize], input, len);
164 4 memset(&b->samples[cur->nb_samples], 0, left);
165 }
166
167 /* Last frame isn't popped off and freed yet - we need it for overlap */
168
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (sf != (subframes - 1))
169 ✗ av_frame_free(&cur);
170 }
171 2 }
172
173 /* Apply the pre emphasis filter */
174 ✗ static void celt_apply_preemph_filter(OpusEncContext *s, CeltFrame *f)
175 {
176 ✗ const int subframesize = s->avctx->frame_size;
177 ✗ const int subframes = OPUS_BLOCK_SIZE(s->packet.framesize) / subframesize;
178 ✗ const float c = ff_opus_deemph_weights[0];
179
180 /* Filter overlap */
181 ✗ for (int ch = 0; ch < f->channels; ch++) {
182 ✗ CeltBlock *b = &f->block[ch];
183 ✗ float m = b->emph_coeff;
184 ✗ for (int i = 0; i < CELT_OVERLAP; i++) {
185 ✗ float sample = b->overlap[i];
186 ✗ b->overlap[i] = sample - m;
187 ✗ m = sample * c;
188 }
189 ✗ b->emph_coeff = m;
190 }
191
192 /* Filter the samples but do not update the last subframe's coeff - overlap ^^^ */
193 ✗ for (int sf = 0; sf < subframes; sf++) {
194 ✗ for (int ch = 0; ch < f->channels; ch++) {
195 ✗ CeltBlock *b = &f->block[ch];
196 ✗ float m = b->emph_coeff;
197 ✗ for (int i = 0; i < subframesize; i++) {
198 ✗ float sample = b->samples[sf*subframesize + i];
199 ✗ b->samples[sf*subframesize + i] = sample - m;
200 ✗ m = sample * c;
201 }
202 ✗ if (sf != (subframes - 1))
203 ✗ b->emph_coeff = m;
204 }
205 }
206 ✗ }
207
208 /* Create the window and do the mdct */
209 ✗ static void celt_frame_mdct(OpusEncContext *s, CeltFrame *f)
210 {
211 ✗ float *win = s->scratch, *temp = s->scratch + 1920;
212
213 ✗ if (f->transient) {
214 ✗ for (int ch = 0; ch < f->channels; ch++) {
215 ✗ CeltBlock *b = &f->block[ch];
216 ✗ float *src1 = b->overlap;
217 ✗ for (int t = 0; t < f->blocks; t++) {
218 ✗ float *src2 = &b->samples[CELT_OVERLAP*t];
219 ✗ s->dsp->vector_fmul(win, src1, ff_celt_window, 128);
220 ✗ s->dsp->vector_fmul_reverse(&win[CELT_OVERLAP], src2,
221 ff_celt_window_padded, 128);
222 ✗ src1 = src2;
223 ✗ s->tx_fn[0](s->tx[0], b->coeffs + t, win, sizeof(float)*f->blocks);
224 }
225 }
226 } else {
227 ✗ int blk_len = OPUS_BLOCK_SIZE(f->size), wlen = OPUS_BLOCK_SIZE(f->size + 1);
228 ✗ int rwin = blk_len - CELT_OVERLAP, lap_dst = (wlen - blk_len - CELT_OVERLAP) >> 1;
229 ✗ memset(win, 0, wlen*sizeof(float));
230 ✗ for (int ch = 0; ch < f->channels; ch++) {
231 ✗ CeltBlock *b = &f->block[ch];
232
233 /* Overlap */
234 ✗ s->dsp->vector_fmul(temp, b->overlap, ff_celt_window, 128);
235 ✗ memcpy(win + lap_dst, temp, CELT_OVERLAP*sizeof(float));
236
237 /* Samples, flat top window */
238 ✗ memcpy(&win[lap_dst + CELT_OVERLAP], b->samples, rwin*sizeof(float));
239
240 /* Samples, windowed */
241 ✗ s->dsp->vector_fmul_reverse(temp, b->samples + rwin,
242 ff_celt_window_padded, 128);
243 ✗ memcpy(win + lap_dst + blk_len, temp, CELT_OVERLAP*sizeof(float));
244
245 ✗ s->tx_fn[f->size](s->tx[f->size], b->coeffs, win, sizeof(float));
246 }
247 }
248
249 ✗ for (int ch = 0; ch < f->channels; ch++) {
250 ✗ CeltBlock *block = &f->block[ch];
251 ✗ for (int i = 0; i < CELT_MAX_BANDS; i++) {
252 ✗ float ener = 0.0f;
253 ✗ int band_offset = ff_celt_freq_bands[i] << f->size;
254 ✗ int band_size = ff_celt_freq_range[i] << f->size;
255 ✗ float *coeffs = &block->coeffs[band_offset];
256
257 ✗ for (int j = 0; j < band_size; j++)
258 ✗ ener += coeffs[j]*coeffs[j];
259
260 ✗ block->lin_energy[i] = sqrtf(ener) + FLT_EPSILON;
261 ✗ ener = 1.0f/block->lin_energy[i];
262
263 ✗ for (int j = 0; j < band_size; j++)
264 ✗ coeffs[j] *= ener;
265
266 ✗ block->energy[i] = log2f(block->lin_energy[i]) - ff_celt_mean_energy[i];
267
268 /* CELT_ENERGY_SILENCE is what the decoder uses and its not -infinity */
269 ✗ block->energy[i] = FFMAX(block->energy[i], CELT_ENERGY_SILENCE);
270 }
271 }
272 ✗ }
273
274 ✗ static void celt_enc_tf(CeltFrame *f, OpusRangeCoder *rc)
275 {
276 ✗ int tf_select = 0, diff = 0, tf_changed = 0, tf_select_needed;
277 ✗ int bits = f->transient ? 2 : 4;
278
279 ✗ tf_select_needed = ((f->size && (opus_rc_tell(rc) + bits + 1) <= f->framebits));
280
281 ✗ for (int i = f->start_band; i < f->end_band; i++) {
282 ✗ if ((opus_rc_tell(rc) + bits + tf_select_needed) <= f->framebits) {
283 ✗ const int tbit = (diff ^ 1) == f->tf_change[i];
284 ✗ ff_opus_rc_enc_log(rc, tbit, bits);
285 ✗ diff ^= tbit;
286 ✗ tf_changed |= diff;
287 }
288 ✗ bits = f->transient ? 4 : 5;
289 }
290
291 ✗ if (tf_select_needed && ff_celt_tf_select[f->size][f->transient][0][tf_changed] !=
292 ✗ ff_celt_tf_select[f->size][f->transient][1][tf_changed]) {
293 ✗ ff_opus_rc_enc_log(rc, f->tf_select, 1);
294 ✗ tf_select = f->tf_select;
295 }
296
297 ✗ for (int i = f->start_band; i < f->end_band; i++)
298 ✗ f->tf_change[i] = ff_celt_tf_select[f->size][f->transient][tf_select][f->tf_change[i]];
299 ✗ }
300
301 ✗ static void celt_enc_quant_pfilter(OpusRangeCoder *rc, CeltFrame *f)
302 {
303 ✗ float gain = f->pf_gain;
304 ✗ int txval, octave = f->pf_octave, period = f->pf_period, tapset = f->pf_tapset;
305
306 ✗ ff_opus_rc_enc_log(rc, f->pfilter, 1);
307 ✗ if (!f->pfilter)
308 ✗ return;
309
310 /* Octave */
311 ✗ txval = FFMIN(octave, 6);
312 ✗ ff_opus_rc_enc_uint(rc, txval, 6);
313 ✗ octave = txval;
314 /* Period */
315 ✗ txval = av_clip(period - (16 << octave) + 1, 0, (1 << (4 + octave)) - 1);
316 ✗ ff_opus_rc_put_raw(rc, period, 4 + octave);
317 ✗ period = txval + (16 << octave) - 1;
318 /* Gain */
319 ✗ txval = FFMIN(((int)(gain / 0.09375f)) - 1, 7);
320 ✗ ff_opus_rc_put_raw(rc, txval, 3);
321 ✗ gain = 0.09375f * (txval + 1);
322 /* Tapset */
323 ✗ if ((opus_rc_tell(rc) + 2) <= f->framebits)
324 ✗ ff_opus_rc_enc_cdf(rc, tapset, ff_celt_model_tapset);
325 else
326 ✗ tapset = 0;
327 /* Finally create the coeffs */
328 ✗ for (int i = 0; i < 2; i++) {
329 ✗ CeltBlock *block = &f->block[i];
330
331 ✗ block->pf_period_new = FFMAX(period, CELT_POSTFILTER_MINPERIOD);
332 ✗ block->pf_gains_new[0] = gain * ff_celt_postfilter_taps[tapset][0];
333 ✗ block->pf_gains_new[1] = gain * ff_celt_postfilter_taps[tapset][1];
334 ✗ block->pf_gains_new[2] = gain * ff_celt_postfilter_taps[tapset][2];
335 }
336 }
337
338 ✗ static void exp_quant_coarse(OpusRangeCoder *rc, CeltFrame *f,
339 float last_energy[][CELT_MAX_BANDS], int intra)
340 {
341 ✗ float alpha, beta, prev[2] = { 0, 0 };
342 ✗ const uint8_t *pmod = ff_celt_coarse_energy_dist[f->size][intra];
343
344 /* Inter is really just differential coding */
345 ✗ if (opus_rc_tell(rc) + 3 <= f->framebits)
346 ✗ ff_opus_rc_enc_log(rc, intra, 3);
347 else
348 ✗ intra = 0;
349
350 ✗ if (intra) {
351 ✗ alpha = 0.0f;
352 ✗ beta = 1.0f - (4915.0f/32768.0f);
353 } else {
354 ✗ alpha = ff_celt_alpha_coef[f->size];
355 ✗ beta = ff_celt_beta_coef[f->size];
356 }
357
358 ✗ for (int i = f->start_band; i < f->end_band; i++) {
359 ✗ for (int ch = 0; ch < f->channels; ch++) {
360 ✗ CeltBlock *block = &f->block[ch];
361 ✗ const int left = f->framebits - opus_rc_tell(rc);
362 ✗ const float last = FFMAX(-9.0f, last_energy[ch][i]);
363 ✗ float diff = block->energy[i] - prev[ch] - last*alpha;
364 ✗ int q_en = lrintf(diff);
365 ✗ if (left >= 15) {
366 ✗ ff_opus_rc_enc_laplace(rc, &q_en, pmod[i << 1] << 7, pmod[(i << 1) + 1] << 6);
367 ✗ } else if (left >= 2) {
368 ✗ q_en = av_clip(q_en, -1, 1);
369 ✗ ff_opus_rc_enc_cdf(rc, 2*q_en + 3*(q_en < 0), ff_celt_model_energy_small);
370 ✗ } else if (left >= 1) {
371 ✗ q_en = av_clip(q_en, -1, 0);
372 ✗ ff_opus_rc_enc_log(rc, (q_en & 1), 1);
373 ✗ } else q_en = -1;
374
375 ✗ block->error_energy[i] = q_en - diff;
376 ✗ prev[ch] += beta * q_en;
377 }
378 }
379 ✗ }
380
381 ✗ static void celt_quant_coarse(CeltFrame *f, OpusRangeCoder *rc,
382 float last_energy[][CELT_MAX_BANDS])
383 {
384 uint32_t inter, intra;
385 ✗ OPUS_RC_CHECKPOINT_SPAWN(rc);
386
387 ✗ exp_quant_coarse(rc, f, last_energy, 1);
388 ✗ intra = OPUS_RC_CHECKPOINT_BITS(rc);
389
390 ✗ OPUS_RC_CHECKPOINT_ROLLBACK(rc);
391
392 ✗ exp_quant_coarse(rc, f, last_energy, 0);
393 ✗ inter = OPUS_RC_CHECKPOINT_BITS(rc);
394
395 ✗ if (inter > intra) { /* Unlikely */
396 ✗ OPUS_RC_CHECKPOINT_ROLLBACK(rc);
397 ✗ exp_quant_coarse(rc, f, last_energy, 1);
398 }
399 ✗ }
400
401 ✗ static void celt_quant_fine(CeltFrame *f, OpusRangeCoder *rc)
402 {
403 ✗ for (int i = f->start_band; i < f->end_band; i++) {
404 ✗ if (!f->fine_bits[i])
405 ✗ continue;
406 ✗ for (int ch = 0; ch < f->channels; ch++) {
407 ✗ CeltBlock *block = &f->block[ch];
408 ✗ int quant, lim = (1 << f->fine_bits[i]);
409 ✗ float offset, diff = 0.5f - block->error_energy[i];
410 ✗ quant = av_clip(floor(diff*lim), 0, lim - 1);
411 ✗ ff_opus_rc_put_raw(rc, quant, f->fine_bits[i]);
412 ✗ offset = 0.5f - ((quant + 0.5f) * (1 << (14 - f->fine_bits[i])) / 16384.0f);
413 ✗ block->error_energy[i] -= offset;
414 }
415 }
416 ✗ }
417
418 ✗ static void celt_quant_final(OpusEncContext *s, OpusRangeCoder *rc, CeltFrame *f)
419 {
420 ✗ for (int priority = 0; priority < 2; priority++) {
421 ✗ for (int i = f->start_band; i < f->end_band && (f->framebits - opus_rc_tell(rc)) >= f->channels; i++) {
422 ✗ if (f->fine_priority[i] != priority || f->fine_bits[i] >= CELT_MAX_FINE_BITS)
423 ✗ continue;
424 ✗ for (int ch = 0; ch < f->channels; ch++) {
425 ✗ CeltBlock *block = &f->block[ch];
426 ✗ const float err = block->error_energy[i];
427 ✗ const float offset = 0.5f * (1 << (14 - f->fine_bits[i] - 1)) / 16384.0f;
428 ✗ const int sign = FFABS(err + offset) < FFABS(err - offset);
429 ✗ ff_opus_rc_put_raw(rc, sign, 1);
430 ✗ block->error_energy[i] -= offset*(1 - 2*sign);
431 }
432 }
433 }
434 ✗ }
435
436 2 static void celt_encode_frame(OpusEncContext *s, OpusRangeCoder *rc,
437 CeltFrame *f, int index)
438 {
439 2 ff_opus_rc_enc_init(rc);
440
441 2 ff_opus_psy_celt_frame_init(&s->psyctx, f, index);
442
443 2 celt_frame_setup_input(s, f);
444
445
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (f->silence) {
446
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (f->framebits >= 16)
447 ✗ ff_opus_rc_enc_log(rc, 1, 15); /* Silence (if using explicit signalling) */
448
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
 6 for (int ch = 0; ch < s->channels; ch++)
449 4 memset(s->last_quantized_energy[ch], 0.0f, sizeof(float)*CELT_MAX_BANDS);
450 2 return;
451 }
452
453 /* Filters */
454 ✗ celt_apply_preemph_filter(s, f);
455 ✗ if (f->pfilter) {
456 ✗ ff_opus_rc_enc_log(rc, 0, 15);
457 ✗ celt_enc_quant_pfilter(rc, f);
458 }
459
460 /* Transform */
461 ✗ celt_frame_mdct(s, f);
462
463 /* Need to handle transient/non-transient switches at any point during analysis */
464 ✗ while (ff_opus_psy_celt_frame_process(&s->psyctx, f, index))
465 ✗ celt_frame_mdct(s, f);
466
467 ✗ ff_opus_rc_enc_init(rc);
468
469 /* Silence */
470 ✗ ff_opus_rc_enc_log(rc, 0, 15);
471
472 /* Pitch filter */
473 ✗ if (!f->start_band && opus_rc_tell(rc) + 16 <= f->framebits)
474 ✗ celt_enc_quant_pfilter(rc, f);
475
476 /* Transient flag */
477 ✗ if (f->size && opus_rc_tell(rc) + 3 <= f->framebits)
478 ✗ ff_opus_rc_enc_log(rc, f->transient, 3);
479
480 /* Main encoding */
481 ✗ celt_quant_coarse (f, rc, s->last_quantized_energy);
482 ✗ celt_enc_tf (f, rc);
483 ✗ ff_celt_bitalloc (f, rc, 1);
484 ✗ celt_quant_fine (f, rc);
485 ✗ ff_celt_quant_bands(f, rc);
486
487 /* Anticollapse bit */
488 ✗ if (f->anticollapse_needed)
489 ✗ ff_opus_rc_put_raw(rc, f->anticollapse, 1);
490
491 /* Final per-band energy adjustments from leftover bits */
492 ✗ celt_quant_final(s, rc, f);
493
494 ✗ for (int ch = 0; ch < f->channels; ch++) {
495 ✗ CeltBlock *block = &f->block[ch];
496 ✗ for (int i = 0; i < CELT_MAX_BANDS; i++)
497 ✗ s->last_quantized_energy[ch][i] = block->energy[i] + block->error_energy[i];
498 }
499 }
500
501 ✗ static inline int write_opuslacing(uint8_t *dst, int v)
502 {
503 ✗ dst[0] = FFMIN(v - FFALIGN(v - 255, 4), v);
504 ✗ dst[1] = v - dst[0] >> 2;
505 ✗ return 1 + (v >= 252);
506 }
507
508 2 static void opus_packet_assembler(OpusEncContext *s, AVPacket *avpkt)
509 {
510 int offset, fsize_needed;
511
512 /* Write toc */
513 2 opus_gen_toc(s, avpkt->data, &offset, &fsize_needed);
514
515 /* Frame sizes if needed */
516
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (fsize_needed) {
517 ✗ for (int i = 0; i < s->packet.frames - 1; i++) {
518 ✗ offset += write_opuslacing(avpkt->data + offset,
519 ✗ s->frame[i].framebits >> 3);
520 }
521 }
522
523 /* Packets */
524
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
 4 for (int i = 0; i < s->packet.frames; i++) {
525 2 ff_opus_rc_enc_end(&s->rc[i], avpkt->data + offset,
526 2 s->frame[i].framebits >> 3);
527 2 offset += s->frame[i].framebits >> 3;
528 }
529
530 2 avpkt->size = offset;
531 2 }
532
533 /* Used as overlap for the first frame and padding for the last encoded packet */
534 1 static AVFrame *spawn_empty_frame(OpusEncContext *s)
535 {
536 1 AVFrame *f = av_frame_alloc();
537 int ret;
538
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!f)
539 ✗ return NULL;
540 1 f->format = s->avctx->sample_fmt;
541 1 f->nb_samples = s->avctx->frame_size;
542 1 ret = av_channel_layout_copy(&f->ch_layout, &s->avctx->ch_layout);
543
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (ret < 0) {
544 ✗ av_frame_free(&f);
545 ✗ return NULL;
546 }
547
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if (av_frame_get_buffer(f, 4)) {
548 ✗ av_frame_free(&f);
549 ✗ return NULL;
550 }
551
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 1 times.
 3 for (int i = 0; i < s->channels; i++) {
552 2 size_t bps = av_get_bytes_per_sample(f->format);
553 2 memset(f->extended_data[i], 0, bps*f->nb_samples);
554 }
555 1 return f;
556 }
557
558 5 static int opus_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
559 const AVFrame *frame, int *got_packet_ptr)
560 {
561 5 OpusEncContext *s = avctx->priv_data;
562 5 int ret, frame_size, discard_padding, alloc_size = 0;
563
564
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 3 times.
 5 if (frame) { /* Add new frame to queue */
565
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
 2 if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
566 ✗ return ret;
567 2 ff_bufqueue_add(avctx, &s->bufqueue, av_frame_clone(frame));
568 } else {
569 3 ff_opus_psy_signal_eof(&s->psyctx);
570
3/4
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 1 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
 3 if (!s->afq.remaining_samples || !avctx->frame_num)
571 1 return 0; /* We've been flushed and there's nothing left to encode */
572 }
573
574 /* Run the psychoacoustic system */
575
2/2
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 2 times.
 4 if (ff_opus_psy_process(&s->psyctx, &s->packet))
576 2 return 0;
577
578 2 frame_size = OPUS_BLOCK_SIZE(s->packet.framesize);
579
580
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (!frame) {
581 /* This can go negative, that's not a problem, we only pad if positive */
582 2 int pad_empty = s->packet.frames*(frame_size/s->avctx->frame_size) - s->bufqueue.available + 1;
583 /* Pad with empty 2.5 ms frames to whatever framesize was decided,
584 * this should only happen at the very last flush frame. The frames
585 * allocated here will be freed (because they have no other references)
586 * after they get used by celt_frame_setup_input() */
587
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 for (int i = 0; i < pad_empty; i++) {
588 ✗ AVFrame *empty = spawn_empty_frame(s);
589 ✗ if (!empty)
590 ✗ return AVERROR(ENOMEM);
591 ✗ ff_bufqueue_add(avctx, &s->bufqueue, empty);
592 }
593 }
594
595
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
 4 for (int i = 0; i < s->packet.frames; i++) {
596 2 celt_encode_frame(s, &s->rc[i], &s->frame[i], i);
597 2 alloc_size += s->frame[i].framebits >> 3;
598 }
599
600 /* Worst case toc + the frame lengths if needed */
601 2 alloc_size += 2 + s->packet.frames*2;
602
603
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
 2 if ((ret = ff_alloc_packet(avctx, avpkt, alloc_size)) < 0)
604 ✗ return ret;
605
606 /* Assemble packet */
607 2 opus_packet_assembler(s, avpkt);
608
609 /* Update the psychoacoustic system */
610 2 ff_opus_psy_postencode_update(&s->psyctx, s->frame);
611
612 /* Remove samples from queue and skip if needed */
613 2 ff_af_queue_remove(&s->afq, s->packet.frames*frame_size, &avpkt->pts, &avpkt->duration);
614
615 2 discard_padding = s->packet.frames*frame_size - ff_samples_from_time_base(avctx, avpkt->duration);
616
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
 2 if (discard_padding > 0) {
617 1 uint8_t *side = av_packet_new_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, 10);
618
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!side)
619 ✗ return AVERROR(ENOMEM);
620 1 AV_WL32(&side[4], discard_padding);
621 }
622
623 2 *got_packet_ptr = 1;
624
625 2 return 0;
626 }
627
628 1 static av_cold int opus_encode_end(AVCodecContext *avctx)
629 {
630 1 OpusEncContext *s = avctx->priv_data;
631
632
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1 times.
 5 for (int i = 0; i < CELT_BLOCK_NB; i++)
633 4 av_tx_uninit(&s->tx[i]);
634
635 1 ff_celt_pvq_uninit(&s->pvq);
636 1 av_freep(&s->dsp);
637 1 av_freep(&s->frame);
638 1 av_freep(&s->rc);
639 1 ff_af_queue_close(&s->afq);
640 1 ff_opus_psy_end(&s->psyctx);
641 1 ff_bufqueue_discard_all(&s->bufqueue);
642
643 1 return 0;
644 }
645
646 1 static av_cold int opus_encode_init(AVCodecContext *avctx)
647 {
648 int ret, max_frames;
649 1 OpusEncContext *s = avctx->priv_data;
650
651 1 s->avctx = avctx;
652 1 s->channels = avctx->ch_layout.nb_channels;
653
654 1 int max_delay_samples = (s->options.max_delay_ms * s->avctx->sample_rate) / 1000;
655 1 avctx->frame_size = OPUS_BLOCK_SIZE(FFMIN(OPUS_SAMPLES_TO_BLOCK_SIZE(max_delay_samples), CELT_BLOCK_960));
656 /* Initial padding will change if SILK is ever supported */
657 1 avctx->initial_padding = 120;
658
659
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
 1 if (!avctx->bit_rate) {
660 1 int coupled = ff_opus_default_coupled_streams[s->channels - 1];
661 1 avctx->bit_rate = coupled*(96000) + (s->channels - coupled*2)*(48000);
662 ✗ } else if (avctx->bit_rate < 6000 || avctx->bit_rate > 255000 * s->channels) {
663 ✗ int64_t clipped_rate = av_clip(avctx->bit_rate, 6000, 255000 * s->channels);
664 ✗ av_log(avctx, AV_LOG_ERROR, "Unsupported bitrate %"PRId64" kbps, clipping to %"PRId64" kbps\n",
665 ✗ avctx->bit_rate/1000, clipped_rate/1000);
666 ✗ avctx->bit_rate = clipped_rate;
667 }
668
669 /* Extradata */
670 1 avctx->extradata_size = 19;
671 1 avctx->extradata = av_malloc(avctx->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
672
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!avctx->extradata)
673 ✗ return AVERROR(ENOMEM);
674 1 opus_write_extradata(avctx);
675
676 1 ff_af_queue_init(avctx, &s->afq);
677
678
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if ((ret = ff_celt_pvq_init(&s->pvq, 1)) < 0)
679 ✗ return ret;
680
681
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if (!(s->dsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT)))
682 ✗ return AVERROR(ENOMEM);
683
684 /* I have no idea why a base scaling factor of 68 works, could be the twiddles */
685
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1 times.
 5 for (int i = 0; i < CELT_BLOCK_NB; i++) {
686 4 const float scale = 68 << (CELT_BLOCK_NB - 1 - i);
687
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 4 times.
 4 if ((ret = av_tx_init(&s->tx[i], &s->tx_fn[i], AV_TX_FLOAT_MDCT, 0, 15 << (i + 3), &scale, 0)))
688 ✗ return AVERROR(ENOMEM);
689 }
690
691 /* Zero out previous energy (matters for inter first frame) */
692
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 1 times.
 3 for (int ch = 0; ch < s->channels; ch++)
693 2 memset(s->last_quantized_energy[ch], 0.0f, sizeof(float)*CELT_MAX_BANDS);
694
695 /* Allocate an empty frame to use as overlap for the first frame of audio */
696 1 ff_bufqueue_add(avctx, &s->bufqueue, spawn_empty_frame(s));
697
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if (!ff_bufqueue_peek(&s->bufqueue, 0))
698 ✗ return AVERROR(ENOMEM);
699
700
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if ((ret = ff_opus_psy_init(&s->psyctx, s->avctx, &s->bufqueue, &s->options)))
701 ✗ return ret;
702
703 /* Frame structs and range coder buffers */
704
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
 1 max_frames = ceilf(FFMIN(s->options.max_delay_ms, 120.0f)/2.5f);
705 1 s->frame = av_malloc(max_frames*sizeof(CeltFrame));
706
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!s->frame)
707 ✗ return AVERROR(ENOMEM);
708 1 s->rc = av_malloc(max_frames*sizeof(OpusRangeCoder));
709
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!s->rc)
710 ✗ return AVERROR(ENOMEM);
711
712
2/2
✓ Branch 0 taken 48 times.
✓ Branch 1 taken 1 times.
 49 for (int i = 0; i < max_frames; i++) {
713 48 s->frame[i].dsp = s->dsp;
714 48 s->frame[i].avctx = s->avctx;
715 48 s->frame[i].seed = 0;
716 48 s->frame[i].pvq = s->pvq;
717 48 s->frame[i].apply_phase_inv = s->options.apply_phase_inv;
718 48 s->frame[i].block[0].emph_coeff = s->frame[i].block[1].emph_coeff = 0.0f;
719 }
720
721 1 return 0;
722 }
723
724 #define OPUSENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
725 static const AVOption opusenc_options[] = {
726 { "opus_delay", "Maximum delay in milliseconds", offsetof(OpusEncContext, options.max_delay_ms), AV_OPT_TYPE_FLOAT, { .dbl = OPUS_MAX_LOOKAHEAD }, 2.5f, OPUS_MAX_LOOKAHEAD, OPUSENC_FLAGS, .unit = "max_delay_ms" },
727 { "apply_phase_inv", "Apply intensity stereo phase inversion", offsetof(OpusEncContext, options.apply_phase_inv), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, OPUSENC_FLAGS, .unit = "apply_phase_inv" },
728 { NULL },
729 };
730
731 static const AVClass opusenc_class = {
732 .class_name = "Opus encoder",
733 .item_name = av_default_item_name,
734 .option = opusenc_options,
735 .version = LIBAVUTIL_VERSION_INT,
736 };
737
738 static const FFCodecDefault opusenc_defaults[] = {
739 { "b", "0" },
740 { "compression_level", "10" },
741 { NULL },
742 };
743
744 const FFCodec ff_opus_encoder = {
745 .p.name = "opus",
746 CODEC_LONG_NAME("Opus"),
747 .p.type = AVMEDIA_TYPE_AUDIO,
748 .p.id = AV_CODEC_ID_OPUS,
749 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
750 AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_EXPERIMENTAL,
751 .defaults = opusenc_defaults,
752 .p.priv_class = &opusenc_class,
753 .priv_data_size = sizeof(OpusEncContext),
754 .init = opus_encode_init,
755 FF_CODEC_ENCODE_CB(opus_encode_frame),
756 .close = opus_encode_end,
757 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
758 CODEC_SAMPLERATES(48000),
759 CODEC_CH_LAYOUTS(AV_CHANNEL_LAYOUT_MONO, AV_CHANNEL_LAYOUT_STEREO),
760 CODEC_SAMPLEFMTS(AV_SAMPLE_FMT_FLTP),
761 };
762