FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libavcodec/opus/enc.c
Date: 2026-09-27 12:50:02
Exec Total Coverage
Lines: 163 401 40.6%
Functions: 9 18 50.0%
Branches: 63 216 29.2%
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 float *input = (const float *)cur->extended_data[ch];
135 /* The MDCT overlap is the trailing CELT_OVERLAP samples of the
136 * previous packet's last frame, as they were when that frame was
137 * encoded. Because the encoder advertises AV_CODEC_CAP_SMALL_LAST_FRAME,
138 * that frame may have been shorter than frame_size, in which case it
139 * was zero padded, so the overlap has to be zero padded the same way. */
140 4 const int start = subframesize - CELT_OVERLAP;
141 4 const int n = av_clip(cur->nb_samples - start, 0, CELT_OVERLAP);
142
1/2
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
 4 if (n > 0)
143 4 memcpy(b->overlap, input + start, n * sizeof(float));
144 4 memset(b->overlap + n, 0, (CELT_OVERLAP - n) * sizeof(float));
145 }
146
147 2 av_frame_free(&cur);
148
149
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
 4 for (int sf = 0; sf < subframes; sf++) {
150
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (sf != (subframes - 1))
151 ✗ cur = ff_bufqueue_get(&s->bufqueue);
152 else
153 2 cur = ff_bufqueue_peek(&s->bufqueue, 0);
154
155
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
 6 for (int ch = 0; ch < f->channels; ch++) {
156 4 CeltBlock *b = &f->block[ch];
157 4 const float *input = (const float *)cur->extended_data[ch];
158 4 float *dst = &b->samples[sf * subframesize];
159 4 const int n = FFMIN(cur->nb_samples, subframesize);
160 4 memcpy(dst, input, n * sizeof(float));
161 4 memset(dst + n, 0, (subframesize - n) * sizeof(float));
162 }
163
164 /* Last frame isn't popped off and freed yet - we need it for overlap */
165
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (sf != (subframes - 1))
166 ✗ av_frame_free(&cur);
167 }
168 2 }
169
170 /* Apply the pre emphasis filter */
171 ✗ static void celt_apply_preemph_filter(OpusEncContext *s, CeltFrame *f)
172 {
173 ✗ const int frame_len = OPUS_BLOCK_SIZE(s->packet.framesize);
174 ✗ const float c = ff_opus_deemph_weights[0];
175
176 ✗ for (int ch = 0; ch < f->channels; ch++) {
177 ✗ CeltBlock *b = &f->block[ch];
178 ✗ float m = b->emph_coeff;
179
180 /* Filter the overlap (the trailing CELT_OVERLAP samples of the previous frame) */
181 ✗ for (int i = 0; i < CELT_OVERLAP; i++) {
182 ✗ float sample = b->overlap[i];
183 ✗ b->overlap[i] = sample - m;
184 ✗ m = sample * c;
185 }
186
187 /* Filter the samples. The trailing CELT_OVERLAP samples are filtered
188 * again as the next frame's overlap, so the filter state saved for
189 * the next frame is the one from right before them. */
190 ✗ for (int i = 0; i < frame_len; i++) {
191 ✗ float sample = b->samples[i];
192 ✗ if (i == frame_len - CELT_OVERLAP)
193 ✗ b->emph_coeff = m;
194 ✗ b->samples[i] = sample - m;
195 ✗ m = sample * c;
196 }
197 }
198 ✗ }
199
200 /* Create the window and do the mdct */
201 ✗ static void celt_frame_mdct(OpusEncContext *s, CeltFrame *f)
202 {
203 ✗ float *win = s->scratch, *temp = s->scratch + 1920;
204
205 ✗ if (f->transient) {
206 ✗ for (int ch = 0; ch < f->channels; ch++) {
207 ✗ CeltBlock *b = &f->block[ch];
208 ✗ float *src1 = b->overlap;
209 ✗ for (int t = 0; t < f->blocks; t++) {
210 ✗ float *src2 = &b->samples[CELT_OVERLAP*t];
211 ✗ s->dsp->vector_fmul(win, src1, ff_celt_window, 128);
212 ✗ s->dsp->vector_fmul_reverse(&win[CELT_OVERLAP], src2,
213 ff_celt_window_padded, 128);
214 ✗ src1 = src2;
215 ✗ s->tx_fn[0](s->tx[0], b->coeffs + t, win, sizeof(float)*f->blocks);
216 }
217 }
218 } else {
219 ✗ int blk_len = OPUS_BLOCK_SIZE(f->size), wlen = OPUS_BLOCK_SIZE(f->size + 1);
220 ✗ int rwin = blk_len - CELT_OVERLAP, lap_dst = (wlen - blk_len - CELT_OVERLAP) >> 1;
221 ✗ memset(win, 0, wlen*sizeof(float));
222 ✗ for (int ch = 0; ch < f->channels; ch++) {
223 ✗ CeltBlock *b = &f->block[ch];
224
225 /* Overlap */
226 ✗ s->dsp->vector_fmul(temp, b->overlap, ff_celt_window, 128);
227 ✗ memcpy(win + lap_dst, temp, CELT_OVERLAP*sizeof(float));
228
229 /* Samples, flat top window */
230 ✗ memcpy(&win[lap_dst + CELT_OVERLAP], b->samples, rwin*sizeof(float));
231
232 /* Samples, windowed */
233 ✗ s->dsp->vector_fmul_reverse(temp, b->samples + rwin,
234 ff_celt_window_padded, 128);
235 ✗ memcpy(win + lap_dst + blk_len, temp, CELT_OVERLAP*sizeof(float));
236
237 ✗ s->tx_fn[f->size](s->tx[f->size], b->coeffs, win, sizeof(float));
238 }
239 }
240
241 ✗ for (int ch = 0; ch < f->channels; ch++) {
242 ✗ CeltBlock *block = &f->block[ch];
243 ✗ for (int i = 0; i < CELT_MAX_BANDS; i++) {
244 ✗ float ener = 0.0f;
245 ✗ int band_offset = ff_celt_freq_bands[i] << f->size;
246 ✗ int band_size = ff_celt_freq_range[i] << f->size;
247 ✗ float *coeffs = &block->coeffs[band_offset];
248
249 ✗ for (int j = 0; j < band_size; j++)
250 ✗ ener += coeffs[j]*coeffs[j];
251
252 ✗ block->lin_energy[i] = sqrtf(ener) + FLT_EPSILON;
253 ✗ ener = 1.0f/block->lin_energy[i];
254
255 ✗ for (int j = 0; j < band_size; j++)
256 ✗ coeffs[j] *= ener;
257
258 ✗ block->energy[i] = log2f(block->lin_energy[i]) - ff_celt_mean_energy[i];
259
260 /* CELT_ENERGY_SILENCE is what the decoder uses and its not -infinity */
261 ✗ block->energy[i] = FFMAX(block->energy[i], CELT_ENERGY_SILENCE);
262 }
263 }
264 ✗ }
265
266 ✗ static void celt_enc_tf(CeltFrame *f, OpusRangeCoder *rc)
267 {
268 ✗ int tf_select = 0, diff = 0, tf_changed = 0, tf_select_needed;
269 ✗ int bits = f->transient ? 2 : 4;
270
271 ✗ tf_select_needed = ((f->size && (opus_rc_tell(rc) + bits + 1) <= f->framebits));
272
273 ✗ for (int i = f->start_band; i < f->end_band; i++) {
274 ✗ if ((opus_rc_tell(rc) + bits + tf_select_needed) <= f->framebits) {
275 ✗ const int tbit = (diff ^ 1) == f->tf_change[i];
276 ✗ ff_opus_rc_enc_log(rc, tbit, bits);
277 ✗ diff ^= tbit;
278 ✗ tf_changed |= diff;
279 }
280 ✗ bits = f->transient ? 4 : 5;
281 }
282
283 ✗ if (tf_select_needed && ff_celt_tf_select[f->size][f->transient][0][tf_changed] !=
284 ✗ ff_celt_tf_select[f->size][f->transient][1][tf_changed]) {
285 ✗ ff_opus_rc_enc_log(rc, f->tf_select, 1);
286 ✗ tf_select = f->tf_select;
287 }
288
289 ✗ for (int i = f->start_band; i < f->end_band; i++)
290 ✗ f->tf_change[i] = ff_celt_tf_select[f->size][f->transient][tf_select][f->tf_change[i]];
291 ✗ }
292
293 ✗ static void celt_enc_quant_pfilter(OpusRangeCoder *rc, CeltFrame *f)
294 {
295 ✗ float gain = f->pf_gain;
296 ✗ int txval, octave = f->pf_octave, period = f->pf_period, tapset = f->pf_tapset;
297
298 ✗ ff_opus_rc_enc_log(rc, f->pfilter, 1);
299 ✗ if (!f->pfilter)
300 ✗ return;
301
302 /* Octave */
303 ✗ txval = FFMIN(octave, 6);
304 ✗ ff_opus_rc_enc_uint(rc, txval, 6);
305 ✗ octave = txval;
306 /* Period */
307 ✗ txval = av_clip(period - (16 << octave) + 1, 0, (1 << (4 + octave)) - 1);
308 ✗ ff_opus_rc_put_raw(rc, period, 4 + octave);
309 ✗ period = txval + (16 << octave) - 1;
310 /* Gain */
311 ✗ txval = FFMIN(((int)(gain / 0.09375f)) - 1, 7);
312 ✗ ff_opus_rc_put_raw(rc, txval, 3);
313 ✗ gain = 0.09375f * (txval + 1);
314 /* Tapset */
315 ✗ if ((opus_rc_tell(rc) + 2) <= f->framebits)
316 ✗ ff_opus_rc_enc_cdf(rc, tapset, ff_celt_model_tapset);
317 else
318 ✗ tapset = 0;
319 /* Finally create the coeffs */
320 ✗ for (int i = 0; i < 2; i++) {
321 ✗ CeltBlock *block = &f->block[i];
322
323 ✗ block->pf_period_new = FFMAX(period, CELT_POSTFILTER_MINPERIOD);
324 ✗ block->pf_gains_new[0] = gain * ff_celt_postfilter_taps[tapset][0];
325 ✗ block->pf_gains_new[1] = gain * ff_celt_postfilter_taps[tapset][1];
326 ✗ block->pf_gains_new[2] = gain * ff_celt_postfilter_taps[tapset][2];
327 }
328 }
329
330 ✗ static void exp_quant_coarse(OpusRangeCoder *rc, CeltFrame *f,
331 float last_energy[][CELT_MAX_BANDS], int intra)
332 {
333 ✗ float alpha, beta, prev[2] = { 0, 0 };
334 ✗ const uint8_t *pmod = ff_celt_coarse_energy_dist[f->size][intra];
335
336 /* Inter is really just differential coding */
337 ✗ if (opus_rc_tell(rc) + 3 <= f->framebits)
338 ✗ ff_opus_rc_enc_log(rc, intra, 3);
339 else
340 ✗ intra = 0;
341
342 ✗ if (intra) {
343 ✗ alpha = 0.0f;
344 ✗ beta = 1.0f - (4915.0f/32768.0f);
345 } else {
346 ✗ alpha = ff_celt_alpha_coef[f->size];
347 ✗ beta = ff_celt_beta_coef[f->size];
348 }
349
350 ✗ for (int i = f->start_band; i < f->end_band; i++) {
351 ✗ for (int ch = 0; ch < f->channels; ch++) {
352 ✗ CeltBlock *block = &f->block[ch];
353 ✗ const int left = f->framebits - opus_rc_tell(rc);
354 ✗ const float last = FFMAX(-9.0f, last_energy[ch][i]);
355 ✗ float diff = block->energy[i] - prev[ch] - last*alpha;
356 ✗ int q_en = lrintf(diff);
357 ✗ if (left >= 15) {
358 ✗ ff_opus_rc_enc_laplace(rc, &q_en, pmod[i << 1] << 7, pmod[(i << 1) + 1] << 6);
359 ✗ } else if (left >= 2) {
360 ✗ q_en = av_clip(q_en, -1, 1);
361 ✗ ff_opus_rc_enc_cdf(rc, 2*q_en + 3*(q_en < 0), ff_celt_model_energy_small);
362 ✗ } else if (left >= 1) {
363 ✗ q_en = av_clip(q_en, -1, 0);
364 ✗ ff_opus_rc_enc_log(rc, (q_en & 1), 1);
365 ✗ } else q_en = -1;
366
367 ✗ block->error_energy[i] = q_en - diff;
368 ✗ prev[ch] += beta * q_en;
369 }
370 }
371 ✗ }
372
373 ✗ static void celt_quant_coarse(CeltFrame *f, OpusRangeCoder *rc,
374 float last_energy[][CELT_MAX_BANDS])
375 {
376 uint32_t inter, intra;
377 ✗ OPUS_RC_CHECKPOINT_SPAWN(rc);
378
379 ✗ exp_quant_coarse(rc, f, last_energy, 1);
380 ✗ intra = OPUS_RC_CHECKPOINT_BITS(rc);
381
382 ✗ OPUS_RC_CHECKPOINT_ROLLBACK(rc);
383
384 ✗ exp_quant_coarse(rc, f, last_energy, 0);
385 ✗ inter = OPUS_RC_CHECKPOINT_BITS(rc);
386
387 ✗ if (inter > intra) { /* Unlikely */
388 ✗ OPUS_RC_CHECKPOINT_ROLLBACK(rc);
389 ✗ exp_quant_coarse(rc, f, last_energy, 1);
390 }
391 ✗ }
392
393 ✗ static void celt_quant_fine(CeltFrame *f, OpusRangeCoder *rc)
394 {
395 ✗ for (int i = f->start_band; i < f->end_band; i++) {
396 ✗ if (!f->fine_bits[i])
397 ✗ continue;
398 ✗ for (int ch = 0; ch < f->channels; ch++) {
399 ✗ CeltBlock *block = &f->block[ch];
400 ✗ int quant, lim = (1 << f->fine_bits[i]);
401 ✗ float offset, diff = 0.5f - block->error_energy[i];
402 ✗ quant = av_clip(floor(diff*lim), 0, lim - 1);
403 ✗ ff_opus_rc_put_raw(rc, quant, f->fine_bits[i]);
404 ✗ offset = 0.5f - ((quant + 0.5f) * (1 << (14 - f->fine_bits[i])) / 16384.0f);
405 ✗ block->error_energy[i] -= offset;
406 }
407 }
408 ✗ }
409
410 ✗ static void celt_quant_final(OpusEncContext *s, OpusRangeCoder *rc, CeltFrame *f)
411 {
412 ✗ for (int priority = 0; priority < 2; priority++) {
413 ✗ for (int i = f->start_band; i < f->end_band && (f->framebits - opus_rc_tell(rc)) >= f->channels; i++) {
414 ✗ if (f->fine_priority[i] != priority || f->fine_bits[i] >= CELT_MAX_FINE_BITS)
415 ✗ continue;
416 ✗ for (int ch = 0; ch < f->channels; ch++) {
417 ✗ CeltBlock *block = &f->block[ch];
418 ✗ const float err = block->error_energy[i];
419 ✗ const float offset = 0.5f * (1 << (14 - f->fine_bits[i] - 1)) / 16384.0f;
420 ✗ const int sign = FFABS(err + offset) < FFABS(err - offset);
421 ✗ ff_opus_rc_put_raw(rc, sign, 1);
422 ✗ block->error_energy[i] -= offset*(1 - 2*sign);
423 }
424 }
425 }
426 ✗ }
427
428 2 static void celt_encode_frame(OpusEncContext *s, OpusRangeCoder *rc,
429 CeltFrame *f, int index)
430 {
431 2 ff_opus_rc_enc_init(rc);
432
433 2 ff_opus_psy_celt_frame_init(&s->psyctx, f, index);
434
435 2 celt_frame_setup_input(s, f);
436
437
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (f->silence) {
438
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (f->framebits >= 16)
439 2 ff_opus_rc_enc_log(rc, 1, 15); /* Silence (if using explicit signalling) */
440
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
 6 for (int ch = 0; ch < s->channels; ch++) {
441 /* The decoder sets all band energies to CELT_ENERGY_SILENCE on
442 * a silence frame, and predicts the next frame's from them. */
443
2/2
✓ Branch 0 taken 84 times.
✓ Branch 1 taken 4 times.
 88 for (int i = 0; i < CELT_MAX_BANDS; i++)
444 84 s->last_quantized_energy[ch][i] = CELT_ENERGY_SILENCE;
445 /* The frame is all zeros, so this is the pre emphasis filter state
446 * at the point the next frame's overlap starts */
447 4 f->block[ch].emph_coeff = 0.0f;
448 }
449 2 return;
450 }
451
452 /* Filters */
453 ✗ celt_apply_preemph_filter(s, f);
454 ✗ if (f->pfilter) {
455 ✗ ff_opus_rc_enc_log(rc, 0, 15);
456 ✗ celt_enc_quant_pfilter(rc, f);
457 }
458
459 /* Transform */
460 ✗ celt_frame_mdct(s, f);
461
462 /* Need to handle transient/non-transient switches at any point during analysis */
463 ✗ while (ff_opus_psy_celt_frame_process(&s->psyctx, f, index))
464 ✗ celt_frame_mdct(s, f);
465
466 ✗ ff_opus_rc_enc_init(rc);
467
468 /* Silence */
469 ✗ ff_opus_rc_enc_log(rc, 0, 15);
470
471 /* Pitch filter */
472 ✗ if (!f->start_band && opus_rc_tell(rc) + 16 <= f->framebits)
473 ✗ celt_enc_quant_pfilter(rc, f);
474
475 /* Transient flag */
476 ✗ if (f->size && opus_rc_tell(rc) + 3 <= f->framebits)
477 ✗ ff_opus_rc_enc_log(rc, f->transient, 3);
478
479 /* Main encoding */
480 ✗ celt_quant_coarse (f, rc, s->last_quantized_energy);
481 ✗ celt_enc_tf (f, rc);
482 ✗ ff_celt_bitalloc (f, rc, 1);
483 ✗ celt_quant_fine (f, rc);
484 ✗ ff_celt_quant_bands(f, rc);
485
486 /* Anticollapse bit */
487 ✗ if (f->anticollapse_needed)
488 ✗ ff_opus_rc_put_raw(rc, f->anticollapse, 1);
489
490 /* Final per-band energy adjustments from leftover bits */
491 ✗ celt_quant_final(s, rc, f);
492
493 ✗ for (int ch = 0; ch < f->channels; ch++) {
494 ✗ CeltBlock *block = &f->block[ch];
495 ✗ for (int i = 0; i < CELT_MAX_BANDS; i++)
496 ✗ s->last_quantized_energy[ch][i] = block->energy[i] + block->error_energy[i];
497 }
498 }
499
500 ✗ static inline int write_opuslacing(uint8_t *dst, int v)
501 {
502 ✗ dst[0] = FFMIN(v - FFALIGN(v - 255, 4), v);
503 ✗ dst[1] = v - dst[0] >> 2;
504 ✗ return 1 + (v >= 252);
505 }
506
507 2 static void opus_packet_assembler(OpusEncContext *s, AVPacket *avpkt)
508 {
509 int offset, fsize_needed;
510
511 /* Write toc */
512 2 opus_gen_toc(s, avpkt->data, &offset, &fsize_needed);
513
514 /* Frame sizes if needed */
515
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (fsize_needed) {
516 ✗ for (int i = 0; i < s->packet.frames - 1; i++) {
517 ✗ offset += write_opuslacing(avpkt->data + offset,
518 ✗ s->frame[i].framebits >> 3);
519 }
520 }
521
522 /* Packets */
523
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
 4 for (int i = 0; i < s->packet.frames; i++) {
524 2 ff_opus_rc_enc_end(&s->rc[i], avpkt->data + offset,
525 2 s->frame[i].framebits >> 3);
526 2 offset += s->frame[i].framebits >> 3;
527 }
528
529 2 avpkt->size = offset;
530 2 }
531
532 /* Used as overlap for the first frame and padding for the last encoded packet */
533 1 static AVFrame *spawn_empty_frame(OpusEncContext *s)
534 {
535 1 AVFrame *f = av_frame_alloc();
536 int ret;
537
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!f)
538 ✗ return NULL;
539 1 f->format = s->avctx->sample_fmt;
540 1 f->nb_samples = s->avctx->frame_size;
541 1 ret = av_channel_layout_copy(&f->ch_layout, &s->avctx->ch_layout);
542
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (ret < 0) {
543 ✗ av_frame_free(&f);
544 ✗ return NULL;
545 }
546
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if (av_frame_get_buffer(f, 4)) {
547 ✗ av_frame_free(&f);
548 ✗ return NULL;
549 }
550
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 1 times.
 3 for (int i = 0; i < s->channels; i++) {
551 2 size_t bps = av_get_bytes_per_sample(f->format);
552 2 memset(f->extended_data[i], 0, bps*f->nb_samples);
553 }
554 1 return f;
555 }
556
557 5 static int opus_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
558 const AVFrame *frame, int *got_packet_ptr)
559 {
560 5 OpusEncContext *s = avctx->priv_data;
561 5 int ret, frame_size, discard_padding, alloc_size = 0;
562
563
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 3 times.
 5 if (frame) { /* Add new frame to queue */
564
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
 2 if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
565 ✗ return ret;
566 2 ff_bufqueue_add(avctx, &s->bufqueue, av_frame_clone(frame));
567 } else {
568 3 ff_opus_psy_signal_eof(&s->psyctx);
569
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)
570 1 return 0; /* We've been flushed and there's nothing left to encode */
571 }
572
573 /* Run the psychoacoustic system */
574
2/2
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 2 times.
 4 if (ff_opus_psy_process(&s->psyctx, &s->packet))
575 2 return 0;
576
577 2 frame_size = OPUS_BLOCK_SIZE(s->packet.framesize);
578
579
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (!frame) {
580 /* This can go negative, that's not a problem, we only pad if positive */
581 2 int pad_empty = s->packet.frames*(frame_size/s->avctx->frame_size) - s->bufqueue.available + 1;
582 /* Pad with empty 2.5 ms frames to whatever framesize was decided,
583 * this should only happen at the very last flush frame. The frames
584 * allocated here will be freed (because they have no other references)
585 * after they get used by celt_frame_setup_input() */
586
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 for (int i = 0; i < pad_empty; i++) {
587 ✗ AVFrame *empty = spawn_empty_frame(s);
588 ✗ if (!empty)
589 ✗ return AVERROR(ENOMEM);
590 ✗ ff_bufqueue_add(avctx, &s->bufqueue, empty);
591 }
592 }
593
594
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
 4 for (int i = 0; i < s->packet.frames; i++) {
595 2 celt_encode_frame(s, &s->rc[i], &s->frame[i], i);
596 2 alloc_size += s->frame[i].framebits >> 3;
597 }
598
599 /* Worst case toc + the frame lengths if needed */
600 2 alloc_size += 2 + s->packet.frames*2;
601
602
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
 2 if ((ret = ff_alloc_packet(avctx, avpkt, alloc_size)) < 0)
603 ✗ return ret;
604
605 /* Assemble packet */
606 2 opus_packet_assembler(s, avpkt);
607
608 /* Update the psychoacoustic system */
609 2 ff_opus_psy_postencode_update(&s->psyctx, s->frame);
610
611 /* Remove samples from queue and skip if needed */
612 2 ret = ff_af_queue_remove(&s->afq, s->packet.frames*frame_size, avpkt);
613
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (ret < 0)
614 ✗ return ret;
615
616 2 discard_padding = s->packet.frames*frame_size - ff_samples_from_time_base(avctx, avpkt->duration);
617
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
 2 if (discard_padding > 0) {
618 1 uint8_t *side = av_packet_new_side_data(avpkt, AV_PKT_DATA_SKIP_SAMPLES, 10);
619
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!side)
620 ✗ return AVERROR(ENOMEM);
621 1 AV_WL32(&side[4], discard_padding);
622 }
623
624 2 *got_packet_ptr = 1;
625
626 2 return 0;
627 }
628
629 1 static av_cold int opus_encode_end(AVCodecContext *avctx)
630 {
631 1 OpusEncContext *s = avctx->priv_data;
632
633
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1 times.
 5 for (int i = 0; i < CELT_BLOCK_NB; i++)
634 4 av_tx_uninit(&s->tx[i]);
635
636 1 ff_celt_pvq_uninit(&s->pvq);
637 1 av_freep(&s->dsp);
638 1 av_freep(&s->frame);
639 1 av_freep(&s->rc);
640 1 ff_af_queue_close(&s->afq);
641 1 ff_opus_psy_end(&s->psyctx);
642 1 ff_bufqueue_discard_all(&s->bufqueue);
643
644 1 return 0;
645 }
646
647 1 static av_cold int opus_encode_init(AVCodecContext *avctx)
648 {
649 int ret, max_frames;
650 1 OpusEncContext *s = avctx->priv_data;
651
652 1 s->avctx = avctx;
653 1 s->channels = avctx->ch_layout.nb_channels;
654
655 1 int max_delay_samples = (s->options.max_delay_ms * s->avctx->sample_rate) / 1000;
656 1 avctx->frame_size = OPUS_BLOCK_SIZE(FFMIN(OPUS_SAMPLES_TO_BLOCK_SIZE(max_delay_samples), CELT_BLOCK_960));
657 /* Initial padding will change if SILK is ever supported */
658 1 avctx->initial_padding = 120;
659
660
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
 1 if (!avctx->bit_rate) {
661 1 int coupled = ff_opus_default_coupled_streams[s->channels - 1];
662 1 avctx->bit_rate = coupled*(96000) + (s->channels - coupled*2)*(48000);
663 ✗ } else if (avctx->bit_rate < 6000 || avctx->bit_rate > 255000 * s->channels) {
664 ✗ int64_t clipped_rate = av_clip(avctx->bit_rate, 6000, 255000 * s->channels);
665 ✗ av_log(avctx, AV_LOG_ERROR, "Unsupported bitrate %"PRId64" kbps, clipping to %"PRId64" kbps\n",
666 ✗ avctx->bit_rate/1000, clipped_rate/1000);
667 ✗ avctx->bit_rate = clipped_rate;
668 }
669
670 /* Extradata */
671 1 avctx->extradata_size = 19;
672 1 avctx->extradata = av_malloc(avctx->extradata_size + AV_INPUT_BUFFER_PADDING_SIZE);
673
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!avctx->extradata)
674 ✗ return AVERROR(ENOMEM);
675 1 opus_write_extradata(avctx);
676
677 1 ff_af_queue_init(avctx, &s->afq);
678
679
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if ((ret = ff_celt_pvq_init(&s->pvq, 1)) < 0)
680 ✗ return ret;
681
682
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)))
683 ✗ return AVERROR(ENOMEM);
684
685 /* I have no idea why a base scaling factor of 68 works, could be the twiddles */
686
2/2
✓ Branch 0 taken 4 times.
✓ Branch 1 taken 1 times.
 5 for (int i = 0; i < CELT_BLOCK_NB; i++) {
687 4 const float scale = 68 << (CELT_BLOCK_NB - 1 - i);
688
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)))
689 ✗ return AVERROR(ENOMEM);
690 }
691
692 /* Zero out previous energy (matters for inter first frame) */
693
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 1 times.
 3 for (int ch = 0; ch < s->channels; ch++)
694 2 memset(s->last_quantized_energy[ch], 0.0f, sizeof(float)*CELT_MAX_BANDS);
695
696 /* Allocate an empty frame to use as overlap for the first frame of audio */
697 1 ff_bufqueue_add(avctx, &s->bufqueue, spawn_empty_frame(s));
698
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
 1 if (!ff_bufqueue_peek(&s->bufqueue, 0))
699 ✗ return AVERROR(ENOMEM);
700
701
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)))
702 ✗ return ret;
703
704 /* Frame structs and range coder buffers */
705
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);
706 1 s->frame = av_malloc(max_frames*sizeof(CeltFrame));
707
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!s->frame)
708 ✗ return AVERROR(ENOMEM);
709 1 s->rc = av_malloc(max_frames*sizeof(OpusRangeCoder));
710
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 1 if (!s->rc)
711 ✗ return AVERROR(ENOMEM);
712
713
2/2
✓ Branch 0 taken 48 times.
✓ Branch 1 taken 1 times.
 49 for (int i = 0; i < max_frames; i++) {
714 48 s->frame[i].dsp = s->dsp;
715 48 s->frame[i].avctx = s->avctx;
716 48 s->frame[i].seed = 0;
717 48 s->frame[i].pvq = s->pvq;
718 48 s->frame[i].apply_phase_inv = s->options.apply_phase_inv;
719 48 s->frame[i].block[0].emph_coeff = s->frame[i].block[1].emph_coeff = 0.0f;
720 }
721
722 1 return 0;
723 }
724
725 #define OPUSENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
726 static const AVOption opusenc_options[] = {
727 { "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" },
728 { "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" },
729 { NULL },
730 };
731
732 static const AVClass opusenc_class = {
733 .class_name = "Opus encoder",
734 .item_name = av_default_item_name,
735 .option = opusenc_options,
736 .version = LIBAVUTIL_VERSION_INT,
737 };
738
739 static const FFCodecDefault opusenc_defaults[] = {
740 { "b", "0" },
741 { "compression_level", "10" },
742 { NULL },
743 };
744
745 const FFCodec ff_opus_encoder = {
746 .p.name = "opus",
747 CODEC_LONG_NAME("Opus"),
748 .p.type = AVMEDIA_TYPE_AUDIO,
749 .p.id = AV_CODEC_ID_OPUS,
750 .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY |
751 AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_EXPERIMENTAL,
752 .defaults = opusenc_defaults,
753 .p.priv_class = &opusenc_class,
754 .priv_data_size = sizeof(OpusEncContext),
755 .init = opus_encode_init,
756 FF_CODEC_ENCODE_CB(opus_encode_frame),
757 .close = opus_encode_end,
758 .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
759 CODEC_SAMPLERATES(48000),
760 CODEC_CH_LAYOUTS(AV_CHANNEL_LAYOUT_MONO, AV_CHANNEL_LAYOUT_STEREO),
761 CODEC_SAMPLEFMTS(AV_SAMPLE_FMT_FLTP),
762 };
763