FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libavcodec/opus/enc.c
Date: 2026-08-31 23:16:59
Exec Total Coverage
Lines: 161 407 39.6%
Functions: 9 18 50.0%
Branches: 61 218 28.0%
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 "enc.h"
25 #include "pvq.h"
26 #include "enc_psy.h"
27 #include "tab.h"
28
29 #include "libavutil/channel_layout.h"
30 #include "libavutil/float_dsp.h"
31 #include "libavutil/mem.h"
32 #include "libavutil/mem_internal.h"
33 #include "libavutil/opt.h"
34
35 #include "libavcodec/audio_frame_queue.h"
36 #include "libavcodec/bytestream.h"
37 #include "libavcodec/codec_internal.h"
38 #include "libavcodec/encode.h"
39
40 typedef struct OpusEncContext {
41 AVClass *av_class;
42 OpusEncOptions options;
43 OpusPsyContext psyctx;
44 AVCodecContext *avctx;
45 AudioFrameQueue afq;
46 AVFloatDSPContext *dsp;
47 AVTXContext *tx[CELT_BLOCK_NB];
48 av_tx_fn tx_fn[CELT_BLOCK_NB];
49 CeltPVQ *pvq;
50 struct FFBufQueue bufqueue;
51
52 uint8_t enc_id[64];
53 int enc_id_bits;
54
55 OpusPacketInfo packet;
56
57 int channels;
58
59 CeltFrame *frame;
60 OpusRangeCoder *rc;
61
62 /* Actual energy the decoder will have */
63 float last_quantized_energy[OPUS_MAX_CHANNELS][CELT_MAX_BANDS];
64
65 DECLARE_ALIGNED(32, float, scratch)[2048];
66 } OpusEncContext;
67
68 1 static void opus_write_extradata(AVCodecContext *avctx)
69 {
70 1 uint8_t *bs = avctx->extradata;
71
72 1 bytestream_put_buffer(&bs, "OpusHead", 8);
73 1 bytestream_put_byte (&bs, 0x1);
74 1 bytestream_put_byte (&bs, avctx->ch_layout.nb_channels);
75 1 bytestream_put_le16 (&bs, avctx->initial_padding);
76 1 bytestream_put_le32 (&bs, avctx->sample_rate);
77 1 bytestream_put_le16 (&bs, 0x0);
78 1 bytestream_put_byte (&bs, 0x0); /* Default layout */
79 1 }
80
81 2 static int opus_gen_toc(OpusEncContext *s, uint8_t *toc, int *size, int *fsize_needed)
82 {
83 2 int tmp = 0x0, extended_toc = 0;
84 static const int toc_cfg[][OPUS_MODE_NB][OPUS_BANDWITH_NB] = {
85 /* Silk Hybrid Celt Layer */
86 /* NB MB WB SWB FB NB MB WB SWB FB NB MB WB SWB FB Bandwidth */
87 { { 0, 0, 0, 0, 0 }, { 0, 0, 0, 0, 0 }, { 17, 0, 21, 25, 29 } }, /* 2.5 ms */
88 { { 0, 0, 0, 0, 0 }, { 0, 0, 0, 0, 0 }, { 18, 0, 22, 26, 30 } }, /* 5 ms */
89 { { 1, 5, 9, 0, 0 }, { 0, 0, 0, 13, 15 }, { 19, 0, 23, 27, 31 } }, /* 10 ms */
90 { { 2, 6, 10, 0, 0 }, { 0, 0, 0, 14, 16 }, { 20, 0, 24, 28, 32 } }, /* 20 ms */
91 { { 3, 7, 11, 0, 0 }, { 0, 0, 0, 0, 0 }, { 0, 0, 0, 0, 0 } }, /* 40 ms */
92 { { 4, 8, 12, 0, 0 }, { 0, 0, 0, 0, 0 }, { 0, 0, 0, 0, 0 } }, /* 60 ms */
93 };
94 2 int cfg = toc_cfg[s->packet.framesize][s->packet.mode][s->packet.bandwidth];
95 2 *fsize_needed = 0;
96
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (!cfg)
97 ✗ return 1;
98
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (s->packet.frames == 2) { /* 2 packets */
99 ✗ if (s->frame[0].framebits == s->frame[1].framebits) { /* same size */
100 ✗ tmp = 0x1;
101 } else { /* different size */
102 ✗ tmp = 0x2;
103 ✗ *fsize_needed = 1; /* put frame sizes in the packet */
104 }
105
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 } else if (s->packet.frames > 2) {
106 ✗ tmp = 0x3;
107 ✗ extended_toc = 1;
108 }
109
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 tmp |= (s->channels > 1) << 2; /* Stereo or mono */
110 2 tmp |= (cfg - 1) << 3; /* codec configuration */
111 2 *toc++ = tmp;
112
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (extended_toc) {
113 ✗ for (int i = 0; i < (s->packet.frames - 1); i++)
114 ✗ *fsize_needed |= (s->frame[i].framebits != s->frame[i + 1].framebits);
115 ✗ tmp = (*fsize_needed) << 7; /* vbr flag */
116 ✗ tmp |= (0) << 6; /* padding flag */
117 ✗ tmp |= s->packet.frames;
118 ✗ *toc++ = tmp;
119 }
120 2 *size = 1 + extended_toc;
121 2 return 0;
122 }
123
124 2 static void celt_frame_setup_input(OpusEncContext *s, CeltFrame *f)
125 {
126 2 AVFrame *cur = NULL;
127 2 const int subframesize = s->avctx->frame_size;
128 2 int subframes = OPUS_BLOCK_SIZE(s->packet.framesize) / subframesize;
129
130 2 cur = ff_bufqueue_get(&s->bufqueue);
131
132
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
 6 for (int ch = 0; ch < f->channels; ch++) {
133 4 CeltBlock *b = &f->block[ch];
134 4 const char *input = cur->extended_data[ch];
135 4 size_t bps = av_get_bytes_per_sample(cur->format);
136 /* The MDCT overlap is the trailing CELT_OVERLAP samples of the
137 * previous packet's last frame. Because the encoder advertises
138 * AV_CODEC_CAP_SMALL_LAST_FRAME, that frame can be shorter than
139 * CELT_OVERLAP; in that case, zero-pad the leading part of the
140 * overlap buffer and copy only what's available. */
141 4 int n = FFMIN(cur->nb_samples, CELT_OVERLAP);
142
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
 4 if (n < CELT_OVERLAP) {
143 ✗ memset(b->overlap, 0, (CELT_OVERLAP - n) * bps);
144 }
145 4 memcpy((char *)b->overlap + (CELT_OVERLAP - n) * bps,
146 4 input + (cur->nb_samples - n) * bps,
147 n * bps);
148 }
149
150 2 av_frame_free(&cur);
151
152
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
 4 for (int sf = 0; sf < subframes; sf++) {
153
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (sf != (subframes - 1))
154 ✗ cur = ff_bufqueue_get(&s->bufqueue);
155 else
156 2 cur = ff_bufqueue_peek(&s->bufqueue, 0);
157
158
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
 6 for (int ch = 0; ch < f->channels; ch++) {
159 4 CeltBlock *b = &f->block[ch];
160 4 const void *input = cur->extended_data[ch];
161 4 const size_t bps = av_get_bytes_per_sample(cur->format);
162 4 const size_t left = (subframesize - cur->nb_samples)*bps;
163 4 const size_t len = FFMIN(subframesize, cur->nb_samples)*bps;
164 4 memcpy(&b->samples[sf*subframesize], input, len);
165 4 memset(&b->samples[cur->nb_samples], 0, left);
166 }
167
168 /* Last frame isn't popped off and freed yet - we need it for overlap */
169
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (sf != (subframes - 1))
170 ✗ av_frame_free(&cur);
171 }
172 2 }
173
174 /* Apply the pre emphasis filter */
175 ✗ static void celt_apply_preemph_filter(OpusEncContext *s, CeltFrame *f)
176 {
177 ✗ const int subframesize = s->avctx->frame_size;
178 ✗ const int subframes = OPUS_BLOCK_SIZE(s->packet.framesize) / subframesize;
179 ✗ const float c = ff_opus_deemph_weights[0];
180
181 /* Filter overlap */
182 ✗ for (int ch = 0; ch < f->channels; ch++) {
183 ✗ CeltBlock *b = &f->block[ch];
184 ✗ float m = b->emph_coeff;
185 ✗ for (int i = 0; i < CELT_OVERLAP; i++) {
186 ✗ float sample = b->overlap[i];
187 ✗ b->overlap[i] = sample - m;
188 ✗ m = sample * c;
189 }
190 ✗ b->emph_coeff = m;
191 }
192
193 /* Filter the samples but do not update the last subframe's coeff - overlap ^^^ */
194 ✗ for (int sf = 0; sf < subframes; sf++) {
195 ✗ for (int ch = 0; ch < f->channels; ch++) {
196 ✗ CeltBlock *b = &f->block[ch];
197 ✗ float m = b->emph_coeff;
198 ✗ for (int i = 0; i < subframesize; i++) {
199 ✗ float sample = b->samples[sf*subframesize + i];
200 ✗ b->samples[sf*subframesize + i] = sample - m;
201 ✗ m = sample * c;
202 }
203 ✗ if (sf != (subframes - 1))
204 ✗ b->emph_coeff = m;
205 }
206 }
207 ✗ }
208
209 /* Create the window and do the mdct */
210 ✗ static void celt_frame_mdct(OpusEncContext *s, CeltFrame *f)
211 {
212 ✗ float *win = s->scratch, *temp = s->scratch + 1920;
213
214 ✗ if (f->transient) {
215 ✗ for (int ch = 0; ch < f->channels; ch++) {
216 ✗ CeltBlock *b = &f->block[ch];
217 ✗ float *src1 = b->overlap;
218 ✗ for (int t = 0; t < f->blocks; t++) {
219 ✗ float *src2 = &b->samples[CELT_OVERLAP*t];
220 ✗ s->dsp->vector_fmul(win, src1, ff_celt_window, 128);
221 ✗ s->dsp->vector_fmul_reverse(&win[CELT_OVERLAP], src2,
222 ff_celt_window_padded, 128);
223 ✗ src1 = src2;
224 ✗ s->tx_fn[0](s->tx[0], b->coeffs + t, win, sizeof(float)*f->blocks);
225 }
226 }
227 } else {
228 ✗ int blk_len = OPUS_BLOCK_SIZE(f->size), wlen = OPUS_BLOCK_SIZE(f->size + 1);
229 ✗ int rwin = blk_len - CELT_OVERLAP, lap_dst = (wlen - blk_len - CELT_OVERLAP) >> 1;
230 ✗ memset(win, 0, wlen*sizeof(float));
231 ✗ for (int ch = 0; ch < f->channels; ch++) {
232 ✗ CeltBlock *b = &f->block[ch];
233
234 /* Overlap */
235 ✗ s->dsp->vector_fmul(temp, b->overlap, ff_celt_window, 128);
236 ✗ memcpy(win + lap_dst, temp, CELT_OVERLAP*sizeof(float));
237
238 /* Samples, flat top window */
239 ✗ memcpy(&win[lap_dst + CELT_OVERLAP], b->samples, rwin*sizeof(float));
240
241 /* Samples, windowed */
242 ✗ s->dsp->vector_fmul_reverse(temp, b->samples + rwin,
243 ff_celt_window_padded, 128);
244 ✗ memcpy(win + lap_dst + blk_len, temp, CELT_OVERLAP*sizeof(float));
245
246 ✗ s->tx_fn[f->size](s->tx[f->size], b->coeffs, win, sizeof(float));
247 }
248 }
249
250 ✗ for (int ch = 0; ch < f->channels; ch++) {
251 ✗ CeltBlock *block = &f->block[ch];
252 ✗ for (int i = 0; i < CELT_MAX_BANDS; i++) {
253 ✗ float ener = 0.0f;
254 ✗ int band_offset = ff_celt_freq_bands[i] << f->size;
255 ✗ int band_size = ff_celt_freq_range[i] << f->size;
256 ✗ float *coeffs = &block->coeffs[band_offset];
257
258 ✗ for (int j = 0; j < band_size; j++)
259 ✗ ener += coeffs[j]*coeffs[j];
260
261 ✗ block->lin_energy[i] = sqrtf(ener) + FLT_EPSILON;
262 ✗ ener = 1.0f/block->lin_energy[i];
263
264 ✗ for (int j = 0; j < band_size; j++)
265 ✗ coeffs[j] *= ener;
266
267 ✗ block->energy[i] = log2f(block->lin_energy[i]) - ff_celt_mean_energy[i];
268
269 /* CELT_ENERGY_SILENCE is what the decoder uses and its not -infinity */
270 ✗ block->energy[i] = FFMAX(block->energy[i], CELT_ENERGY_SILENCE);
271 }
272 }
273 ✗ }
274
275 ✗ static void celt_enc_tf(CeltFrame *f, OpusRangeCoder *rc)
276 {
277 ✗ int tf_select = 0, diff = 0, tf_changed = 0, tf_select_needed;
278 ✗ int bits = f->transient ? 2 : 4;
279
280 ✗ tf_select_needed = ((f->size && (opus_rc_tell(rc) + bits + 1) <= f->framebits));
281
282 ✗ for (int i = f->start_band; i < f->end_band; i++) {
283 ✗ if ((opus_rc_tell(rc) + bits + tf_select_needed) <= f->framebits) {
284 ✗ const int tbit = (diff ^ 1) == f->tf_change[i];
285 ✗ ff_opus_rc_enc_log(rc, tbit, bits);
286 ✗ diff ^= tbit;
287 ✗ tf_changed |= diff;
288 }
289 ✗ bits = f->transient ? 4 : 5;
290 }
291
292 ✗ if (tf_select_needed && ff_celt_tf_select[f->size][f->transient][0][tf_changed] !=
293 ✗ ff_celt_tf_select[f->size][f->transient][1][tf_changed]) {
294 ✗ ff_opus_rc_enc_log(rc, f->tf_select, 1);
295 ✗ tf_select = f->tf_select;
296 }
297
298 ✗ for (int i = f->start_band; i < f->end_band; i++)
299 ✗ f->tf_change[i] = ff_celt_tf_select[f->size][f->transient][tf_select][f->tf_change[i]];
300 ✗ }
301
302 ✗ static void celt_enc_quant_pfilter(OpusRangeCoder *rc, CeltFrame *f)
303 {
304 ✗ float gain = f->pf_gain;
305 ✗ int txval, octave = f->pf_octave, period = f->pf_period, tapset = f->pf_tapset;
306
307 ✗ ff_opus_rc_enc_log(rc, f->pfilter, 1);
308 ✗ if (!f->pfilter)
309 ✗ return;
310
311 /* Octave */
312 ✗ txval = FFMIN(octave, 6);
313 ✗ ff_opus_rc_enc_uint(rc, txval, 6);
314 ✗ octave = txval;
315 /* Period */
316 ✗ txval = av_clip(period - (16 << octave) + 1, 0, (1 << (4 + octave)) - 1);
317 ✗ ff_opus_rc_put_raw(rc, period, 4 + octave);
318 ✗ period = txval + (16 << octave) - 1;
319 /* Gain */
320 ✗ txval = FFMIN(((int)(gain / 0.09375f)) - 1, 7);
321 ✗ ff_opus_rc_put_raw(rc, txval, 3);
322 ✗ gain = 0.09375f * (txval + 1);
323 /* Tapset */
324 ✗ if ((opus_rc_tell(rc) + 2) <= f->framebits)
325 ✗ ff_opus_rc_enc_cdf(rc, tapset, ff_celt_model_tapset);
326 else
327 ✗ tapset = 0;
328 /* Finally create the coeffs */
329 ✗ for (int i = 0; i < 2; i++) {
330 ✗ CeltBlock *block = &f->block[i];
331
332 ✗ block->pf_period_new = FFMAX(period, CELT_POSTFILTER_MINPERIOD);
333 ✗ block->pf_gains_new[0] = gain * ff_celt_postfilter_taps[tapset][0];
334 ✗ block->pf_gains_new[1] = gain * ff_celt_postfilter_taps[tapset][1];
335 ✗ block->pf_gains_new[2] = gain * ff_celt_postfilter_taps[tapset][2];
336 }
337 }
338
339 ✗ static void exp_quant_coarse(OpusRangeCoder *rc, CeltFrame *f,
340 float last_energy[][CELT_MAX_BANDS], int intra)
341 {
342 ✗ float alpha, beta, prev[2] = { 0, 0 };
343 ✗ const uint8_t *pmod = ff_celt_coarse_energy_dist[f->size][intra];
344
345 /* Inter is really just differential coding */
346 ✗ if (opus_rc_tell(rc) + 3 <= f->framebits)
347 ✗ ff_opus_rc_enc_log(rc, intra, 3);
348 else
349 ✗ intra = 0;
350
351 ✗ if (intra) {
352 ✗ alpha = 0.0f;
353 ✗ beta = 1.0f - (4915.0f/32768.0f);
354 } else {
355 ✗ alpha = ff_celt_alpha_coef[f->size];
356 ✗ beta = ff_celt_beta_coef[f->size];
357 }
358
359 ✗ for (int i = f->start_band; i < f->end_band; i++) {
360 ✗ for (int ch = 0; ch < f->channels; ch++) {
361 ✗ CeltBlock *block = &f->block[ch];
362 ✗ const int left = f->framebits - opus_rc_tell(rc);
363 ✗ const float last = FFMAX(-9.0f, last_energy[ch][i]);
364 ✗ float diff = block->energy[i] - prev[ch] - last*alpha;
365 ✗ int q_en = lrintf(diff);
366 ✗ if (left >= 15) {
367 ✗ ff_opus_rc_enc_laplace(rc, &q_en, pmod[i << 1] << 7, pmod[(i << 1) + 1] << 6);
368 ✗ } else if (left >= 2) {
369 ✗ q_en = av_clip(q_en, -1, 1);
370 ✗ ff_opus_rc_enc_cdf(rc, 2*q_en + 3*(q_en < 0), ff_celt_model_energy_small);
371 ✗ } else if (left >= 1) {
372 ✗ q_en = av_clip(q_en, -1, 0);
373 ✗ ff_opus_rc_enc_log(rc, (q_en & 1), 1);
374 ✗ } else q_en = -1;
375
376 ✗ block->error_energy[i] = q_en - diff;
377 ✗ prev[ch] += beta * q_en;
378 }
379 }
380 ✗ }
381
382 ✗ static void celt_quant_coarse(CeltFrame *f, OpusRangeCoder *rc,
383 float last_energy[][CELT_MAX_BANDS])
384 {
385 uint32_t inter, intra;
386 ✗ OPUS_RC_CHECKPOINT_SPAWN(rc);
387
388 ✗ exp_quant_coarse(rc, f, last_energy, 1);
389 ✗ intra = OPUS_RC_CHECKPOINT_BITS(rc);
390
391 ✗ OPUS_RC_CHECKPOINT_ROLLBACK(rc);
392
393 ✗ exp_quant_coarse(rc, f, last_energy, 0);
394 ✗ inter = OPUS_RC_CHECKPOINT_BITS(rc);
395
396 ✗ if (inter > intra) { /* Unlikely */
397 ✗ OPUS_RC_CHECKPOINT_ROLLBACK(rc);
398 ✗ exp_quant_coarse(rc, f, last_energy, 1);
399 }
400 ✗ }
401
402 ✗ static void celt_quant_fine(CeltFrame *f, OpusRangeCoder *rc)
403 {
404 ✗ for (int i = f->start_band; i < f->end_band; i++) {
405 ✗ if (!f->fine_bits[i])
406 ✗ continue;
407 ✗ for (int ch = 0; ch < f->channels; ch++) {
408 ✗ CeltBlock *block = &f->block[ch];
409 ✗ int quant, lim = (1 << f->fine_bits[i]);
410 ✗ float offset, diff = 0.5f - block->error_energy[i];
411 ✗ quant = av_clip(floor(diff*lim), 0, lim - 1);
412 ✗ ff_opus_rc_put_raw(rc, quant, f->fine_bits[i]);
413 ✗ offset = 0.5f - ((quant + 0.5f) * (1 << (14 - f->fine_bits[i])) / 16384.0f);
414 ✗ block->error_energy[i] -= offset;
415 }
416 }
417 ✗ }
418
419 ✗ static void celt_quant_final(OpusEncContext *s, OpusRangeCoder *rc, CeltFrame *f)
420 {
421 ✗ for (int priority = 0; priority < 2; priority++) {
422 ✗ for (int i = f->start_band; i < f->end_band && (f->framebits - opus_rc_tell(rc)) >= f->channels; i++) {
423 ✗ if (f->fine_priority[i] != priority || f->fine_bits[i] >= CELT_MAX_FINE_BITS)
424 ✗ continue;
425 ✗ for (int ch = 0; ch < f->channels; ch++) {
426 ✗ CeltBlock *block = &f->block[ch];
427 ✗ const float err = block->error_energy[i];
428 ✗ const float offset = 0.5f * (1 << (14 - f->fine_bits[i] - 1)) / 16384.0f;
429 ✗ const int sign = FFABS(err + offset) < FFABS(err - offset);
430 ✗ ff_opus_rc_put_raw(rc, sign, 1);
431 ✗ block->error_energy[i] -= offset*(1 - 2*sign);
432 }
433 }
434 }
435 ✗ }
436
437 2 static void celt_encode_frame(OpusEncContext *s, OpusRangeCoder *rc,
438 CeltFrame *f, int index)
439 {
440 2 ff_opus_rc_enc_init(rc);
441
442 2 ff_opus_psy_celt_frame_init(&s->psyctx, f, index);
443
444 2 celt_frame_setup_input(s, f);
445
446
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (f->silence) {
447
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (f->framebits >= 16)
448 ✗ ff_opus_rc_enc_log(rc, 1, 15); /* Silence (if using explicit signalling) */
449
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
 6 for (int ch = 0; ch < s->channels; ch++)
450 4 memset(s->last_quantized_energy[ch], 0.0f, sizeof(float)*CELT_MAX_BANDS);
451 2 return;
452 }
453
454 /* Filters */
455 ✗ celt_apply_preemph_filter(s, f);
456 ✗ if (f->pfilter) {
457 ✗ ff_opus_rc_enc_log(rc, 0, 15);
458 ✗ celt_enc_quant_pfilter(rc, f);
459 }
460
461 /* Transform */
462 ✗ celt_frame_mdct(s, f);
463
464 /* Need to handle transient/non-transient switches at any point during analysis */
465 ✗ while (ff_opus_psy_celt_frame_process(&s->psyctx, f, index))
466 ✗ celt_frame_mdct(s, f);
467
468 ✗ ff_opus_rc_enc_init(rc);
469
470 /* Silence */
471 ✗ ff_opus_rc_enc_log(rc, 0, 15);
472
473 /* Pitch filter */
474 ✗ if (!f->start_band && opus_rc_tell(rc) + 16 <= f->framebits)
475 ✗ celt_enc_quant_pfilter(rc, f);
476
477 /* Transient flag */
478 ✗ if (f->size && opus_rc_tell(rc) + 3 <= f->framebits)
479 ✗ ff_opus_rc_enc_log(rc, f->transient, 3);
480
481 /* Main encoding */
482 ✗ celt_quant_coarse (f, rc, s->last_quantized_energy);
483 ✗ celt_enc_tf (f, rc);
484 ✗ ff_celt_bitalloc (f, rc, 1);
485 ✗ celt_quant_fine (f, rc);
486 ✗ ff_celt_quant_bands(f, rc);
487
488 /* Anticollapse bit */
489 ✗ if (f->anticollapse_needed)
490 ✗ ff_opus_rc_put_raw(rc, f->anticollapse, 1);
491
492 /* Final per-band energy adjustments from leftover bits */
493 ✗ celt_quant_final(s, rc, f);
494
495 ✗ for (int ch = 0; ch < f->channels; ch++) {
496 ✗ CeltBlock *block = &f->block[ch];
497 ✗ for (int i = 0; i < CELT_MAX_BANDS; i++)
498 ✗ s->last_quantized_energy[ch][i] = block->energy[i] + block->error_energy[i];
499 }
500 }
501
502 ✗ static inline int write_opuslacing(uint8_t *dst, int v)
503 {
504 ✗ dst[0] = FFMIN(v - FFALIGN(v - 255, 4), v);
505 ✗ dst[1] = v - dst[0] >> 2;
506 ✗ return 1 + (v >= 252);
507 }
508
509 2 static void opus_packet_assembler(OpusEncContext *s, AVPacket *avpkt)
510 {
511 int offset, fsize_needed;
512
513 /* Write toc */
514 2 opus_gen_toc(s, avpkt->data, &offset, &fsize_needed);
515
516 /* Frame sizes if needed */
517
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (fsize_needed) {
518 ✗ for (int i = 0; i < s->packet.frames - 1; i++) {
519 ✗ offset += write_opuslacing(avpkt->data + offset,
520 ✗ s->frame[i].framebits >> 3);
521 }
522 }
523
524 /* Packets */
525
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
 4 for (int i = 0; i < s->packet.frames; i++) {
526 2 ff_opus_rc_enc_end(&s->rc[i], avpkt->data + offset,
527 2 s->frame[i].framebits >> 3);
528 2 offset += s->frame[i].framebits >> 3;
529 }
530
531 2 avpkt->size = offset;
532 2 }
533
534 /* Used as overlap for the first frame and padding for the last encoded packet */
535 1 static AVFrame *spawn_empty_frame(OpusEncContext *s)
536 {
537 1 AVFrame *f = av_frame_alloc();
538 int ret;
539
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!f)
540 ✗ return NULL;
541 1 f->format = s->avctx->sample_fmt;
542 1 f->nb_samples = s->avctx->frame_size;
543 1 ret = av_channel_layout_copy(&f->ch_layout, &s->avctx->ch_layout);
544
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (ret < 0) {
545 ✗ av_frame_free(&f);
546 ✗ return NULL;
547 }
548
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if (av_frame_get_buffer(f, 4)) {
549 ✗ av_frame_free(&f);
550 ✗ return NULL;
551 }
552
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 1 times.
 3 for (int i = 0; i < s->channels; i++) {
553 2 size_t bps = av_get_bytes_per_sample(f->format);
554 2 memset(f->extended_data[i], 0, bps*f->nb_samples);
555 }
556 1 return f;
557 }
558
559 5 static int opus_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
560 const AVFrame *frame, int *got_packet_ptr)
561 {
562 5 OpusEncContext *s = avctx->priv_data;
563 5 int ret, frame_size, discard_padding, alloc_size = 0;
564
565
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 3 times.
 5 if (frame) { /* Add new frame to queue */
566
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
 2 if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
567 ✗ return ret;
568 2 ff_bufqueue_add(avctx, &s->bufqueue, av_frame_clone(frame));
569 } else {
570 3 ff_opus_psy_signal_eof(&s->psyctx);
571
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)
572 1 return 0; /* We've been flushed and there's nothing left to encode */
573 }
574
575 /* Run the psychoacoustic system */
576
2/2
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 2 times.
 4 if (ff_opus_psy_process(&s->psyctx, &s->packet))
577 2 return 0;
578
579 2 frame_size = OPUS_BLOCK_SIZE(s->packet.framesize);
580
581
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (!frame) {
582 /* This can go negative, that's not a problem, we only pad if positive */
583 2 int pad_empty = s->packet.frames*(frame_size/s->avctx->frame_size) - s->bufqueue.available + 1;
584 /* Pad with empty 2.5 ms frames to whatever framesize was decided,
585 * this should only happen at the very last flush frame. The frames
586 * allocated here will be freed (because they have no other references)
587 * after they get used by celt_frame_setup_input() */
588
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 for (int i = 0; i < pad_empty; i++) {
589 ✗ AVFrame *empty = spawn_empty_frame(s);
590 ✗ if (!empty)
591 ✗ return AVERROR(ENOMEM);
592 ✗ ff_bufqueue_add(avctx, &s->bufqueue, empty);
593 }
594 }
595
596
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
 4 for (int i = 0; i < s->packet.frames; i++) {
597 2 celt_encode_frame(s, &s->rc[i], &s->frame[i], i);
598 2 alloc_size += s->frame[i].framebits >> 3;
599 }
600
601 /* Worst case toc + the frame lengths if needed */
602 2 alloc_size += 2 + s->packet.frames*2;
603
604
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
 2 if ((ret = ff_alloc_packet(avctx, avpkt, alloc_size)) < 0)
605 ✗ return ret;
606
607 /* Assemble packet */
608 2 opus_packet_assembler(s, avpkt);
609
610 /* Update the psychoacoustic system */
611 2 ff_opus_psy_postencode_update(&s->psyctx, s->frame);
612
613 /* Remove samples from queue and skip if needed */
614 2 ret = ff_af_queue_remove(&s->afq, s->packet.frames*frame_size, avpkt);
615
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (ret < 0)
616 ✗ return ret;
617
618 2 discard_padding = s->packet.frames*frame_size - ff_samples_from_time_base(avctx, avpkt->duration);
619
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
 2 if (discard_padding > 0) {
620 1 uint8_t *side = av_packet_new_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, 10);
621
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!side)
622 ✗ return AVERROR(ENOMEM);
623 1 AV_WL32(&side[4], discard_padding);
624 }
625
626 2 *got_packet_ptr = 1;
627
628 2 return 0;
629 }
630
631 1 static av_cold int opus_encode_end(AVCodecContext *avctx)
632 {
633 1 OpusEncContext *s = avctx->priv_data;
634
635
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1 times.
 5 for (int i = 0; i < CELT_BLOCK_NB; i++)
636 4 av_tx_uninit(&s->tx[i]);
637
638 1 ff_celt_pvq_uninit(&s->pvq);
639 1 av_freep(&s->dsp);
640 1 av_freep(&s->frame);
641 1 av_freep(&s->rc);
642 1 ff_af_queue_close(&s->afq);
643 1 ff_opus_psy_end(&s->psyctx);
644 1 ff_bufqueue_discard_all(&s->bufqueue);
645
646 1 return 0;
647 }
648
649 1 static av_cold int opus_encode_init(AVCodecContext *avctx)
650 {
651 int ret, max_frames;
652 1 OpusEncContext *s = avctx->priv_data;
653
654 1 s->avctx = avctx;
655 1 s->channels = avctx->ch_layout.nb_channels;
656
657 1 int max_delay_samples = (s->options.max_delay_ms * s->avctx->sample_rate) / 1000;
658 1 avctx->frame_size = OPUS_BLOCK_SIZE(FFMIN(OPUS_SAMPLES_TO_BLOCK_SIZE(max_delay_samples), CELT_BLOCK_960));
659 /* Initial padding will change if SILK is ever supported */
660 1 avctx->initial_padding = 120;
661
662
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
 1 if (!avctx->bit_rate) {
663 1 int coupled = ff_opus_default_coupled_streams[s->channels - 1];
664 1 avctx->bit_rate = coupled*(96000) + (s->channels - coupled*2)*(48000);
665 ✗ } else if (avctx->bit_rate < 6000 || avctx->bit_rate > 255000 * s->channels) {
666 ✗ int64_t clipped_rate = av_clip(avctx->bit_rate, 6000, 255000 * s->channels);
667 ✗ av_log(avctx, AV_LOG_ERROR, "Unsupported bitrate %"PRId64" kbps, clipping to %"PRId64" kbps\n",
668 ✗ avctx->bit_rate/1000, clipped_rate/1000);
669 ✗ avctx->bit_rate = clipped_rate;
670 }
671
672 /* Extradata */
673 1 avctx->extradata_size = 19;
674 1 avctx->extradata = av_malloc(avctx->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
675
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!avctx->extradata)
676 ✗ return AVERROR(ENOMEM);
677 1 opus_write_extradata(avctx);
678
679 1 ff_af_queue_init(avctx, &s->afq);
680
681
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if ((ret = ff_celt_pvq_init(&s->pvq, 1)) < 0)
682 ✗ return ret;
683
684
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)))
685 ✗ return AVERROR(ENOMEM);
686
687 /* I have no idea why a base scaling factor of 68 works, could be the twiddles */
688
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1 times.
 5 for (int i = 0; i < CELT_BLOCK_NB; i++) {
689 4 const float scale = 68 << (CELT_BLOCK_NB - 1 - i);
690
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)))
691 ✗ return AVERROR(ENOMEM);
692 }
693
694 /* Zero out previous energy (matters for inter first frame) */
695
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 1 times.
 3 for (int ch = 0; ch < s->channels; ch++)
696 2 memset(s->last_quantized_energy[ch], 0.0f, sizeof(float)*CELT_MAX_BANDS);
697
698 /* Allocate an empty frame to use as overlap for the first frame of audio */
699 1 ff_bufqueue_add(avctx, &s->bufqueue, spawn_empty_frame(s));
700
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if (!ff_bufqueue_peek(&s->bufqueue, 0))
701 ✗ return AVERROR(ENOMEM);
702
703
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)))
704 ✗ return ret;
705
706 /* Frame structs and range coder buffers */
707
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);
708 1 s->frame = av_malloc(max_frames*sizeof(CeltFrame));
709
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!s->frame)
710 ✗ return AVERROR(ENOMEM);
711 1 s->rc = av_malloc(max_frames*sizeof(OpusRangeCoder));
712
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!s->rc)
713 ✗ return AVERROR(ENOMEM);
714
715
2/2
✓ Branch 0 taken 48 times.
✓ Branch 1 taken 1 times.
 49 for (int i = 0; i < max_frames; i++) {
716 48 s->frame[i].dsp = s->dsp;
717 48 s->frame[i].avctx = s->avctx;
718 48 s->frame[i].seed = 0;
719 48 s->frame[i].pvq = s->pvq;
720 48 s->frame[i].apply_phase_inv = s->options.apply_phase_inv;
721 48 s->frame[i].block[0].emph_coeff = s->frame[i].block[1].emph_coeff = 0.0f;
722 }
723
724 1 return 0;
725 }
726
727 #define OPUSENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
728 static const AVOption opusenc_options[] = {
729 { "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" },
730 { "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" },
731 { NULL },
732 };
733
734 static const AVClass opusenc_class = {
735 .class_name = "Opus encoder",
736 .item_name = av_default_item_name,
737 .option = opusenc_options,
738 .version = LIBAVUTIL_VERSION_INT,
739 };
740
741 static const FFCodecDefault opusenc_defaults[] = {
742 { "b", "0" },
743 { "compression_level", "10" },
744 { NULL },
745 };
746
747 const FFCodec ff_opus_encoder = {
748 .p.name = "opus",
749 CODEC_LONG_NAME("Opus"),
750 .p.type = AVMEDIA_TYPE_AUDIO,
751 .p.id = AV_CODEC_ID_OPUS,
752 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
753 AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_EXPERIMENTAL,
754 .defaults = opusenc_defaults,
755 .p.priv_class = &opusenc_class,
756 .priv_data_size = sizeof(OpusEncContext),
757 .init = opus_encode_init,
758 FF_CODEC_ENCODE_CB(opus_encode_frame),
759 .close = opus_encode_end,
760 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
761 CODEC_SAMPLERATES(48000),
762 CODEC_CH_LAYOUTS(AV_CHANNEL_LAYOUT_MONO, AV_CHANNEL_LAYOUT_STEREO),
763 CODEC_SAMPLEFMTS(AV_SAMPLE_FMT_FLTP),
764 };
765