blob: dc14cc2d2a688e0afcbf079d6f941ea54527f1b2
1 | /* |
2 | * Copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at> |
3 | * Copyright (c) 2012 Justin Ruggles <justin.ruggles@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 "libavutil/common.h" |
23 | #include "libavutil/libm.h" |
24 | #include "libavutil/log.h" |
25 | #include "internal.h" |
26 | #include "resample.h" |
27 | #include "audio_data.h" |
28 | |
29 | |
30 | /* double template */ |
31 | #define CONFIG_RESAMPLE_DBL |
32 | #include "resample_template.c" |
33 | #undef CONFIG_RESAMPLE_DBL |
34 | |
35 | /* float template */ |
36 | #define CONFIG_RESAMPLE_FLT |
37 | #include "resample_template.c" |
38 | #undef CONFIG_RESAMPLE_FLT |
39 | |
40 | /* s32 template */ |
41 | #define CONFIG_RESAMPLE_S32 |
42 | #include "resample_template.c" |
43 | #undef CONFIG_RESAMPLE_S32 |
44 | |
45 | /* s16 template */ |
46 | #include "resample_template.c" |
47 | |
48 | |
49 | /* 0th order modified Bessel function of the first kind. */ |
50 | static double bessel(double x) |
51 | { |
52 | double v = 1; |
53 | double lastv = 0; |
54 | double t = 1; |
55 | int i; |
56 | |
57 | x = x * x / 4; |
58 | for (i = 1; v != lastv; i++) { |
59 | lastv = v; |
60 | t *= x / (i * i); |
61 | v += t; |
62 | } |
63 | return v; |
64 | } |
65 | |
66 | /* Build a polyphase filterbank. */ |
67 | static int build_filter(ResampleContext *c, double factor) |
68 | { |
69 | int ph, i; |
70 | double x, y, w; |
71 | double *tab; |
72 | int tap_count = c->filter_length; |
73 | int phase_count = 1 << c->phase_shift; |
74 | const int center = (tap_count - 1) / 2; |
75 | |
76 | tab = av_malloc(tap_count * sizeof(*tab)); |
77 | if (!tab) |
78 | return AVERROR(ENOMEM); |
79 | |
80 | for (ph = 0; ph < phase_count; ph++) { |
81 | double norm = 0; |
82 | for (i = 0; i < tap_count; i++) { |
83 | x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor; |
84 | if (x == 0) y = 1.0; |
85 | else y = sin(x) / x; |
86 | switch (c->filter_type) { |
87 | case AV_RESAMPLE_FILTER_TYPE_CUBIC: { |
88 | const float d = -0.5; //first order derivative = -0.5 |
89 | x = fabs(((double)(i - center) - (double)ph / phase_count) * factor); |
90 | if (x < 1.0) y = 1 - 3 * x*x + 2 * x*x*x + d * ( -x*x + x*x*x); |
91 | else y = d * (-4 + 8 * x - 5 * x*x + x*x*x); |
92 | break; |
93 | } |
94 | case AV_RESAMPLE_FILTER_TYPE_BLACKMAN_NUTTALL: |
95 | w = 2.0 * x / (factor * tap_count) + M_PI; |
96 | y *= 0.3635819 - 0.4891775 * cos( w) + |
97 | 0.1365995 * cos(2 * w) - |
98 | 0.0106411 * cos(3 * w); |
99 | break; |
100 | case AV_RESAMPLE_FILTER_TYPE_KAISER: |
101 | w = 2.0 * x / (factor * tap_count * M_PI); |
102 | y *= bessel(c->kaiser_beta * sqrt(FFMAX(1 - w * w, 0))); |
103 | break; |
104 | } |
105 | |
106 | tab[i] = y; |
107 | norm += y; |
108 | } |
109 | /* normalize so that an uniform color remains the same */ |
110 | for (i = 0; i < tap_count; i++) |
111 | tab[i] = tab[i] / norm; |
112 | |
113 | c->set_filter(c->filter_bank, tab, ph, tap_count); |
114 | } |
115 | |
116 | av_free(tab); |
117 | return 0; |
118 | } |
119 | |
120 | ResampleContext *ff_audio_resample_init(AVAudioResampleContext *avr) |
121 | { |
122 | ResampleContext *c; |
123 | int out_rate = avr->out_sample_rate; |
124 | int in_rate = avr->in_sample_rate; |
125 | double factor = FFMIN(out_rate * avr->cutoff / in_rate, 1.0); |
126 | int phase_count = 1 << avr->phase_shift; |
127 | int felem_size; |
128 | |
129 | if (avr->internal_sample_fmt != AV_SAMPLE_FMT_S16P && |
130 | avr->internal_sample_fmt != AV_SAMPLE_FMT_S32P && |
131 | avr->internal_sample_fmt != AV_SAMPLE_FMT_FLTP && |
132 | avr->internal_sample_fmt != AV_SAMPLE_FMT_DBLP) { |
133 | av_log(avr, AV_LOG_ERROR, "Unsupported internal format for " |
134 | "resampling: %s\n", |
135 | av_get_sample_fmt_name(avr->internal_sample_fmt)); |
136 | return NULL; |
137 | } |
138 | c = av_mallocz(sizeof(*c)); |
139 | if (!c) |
140 | return NULL; |
141 | |
142 | c->avr = avr; |
143 | c->phase_shift = avr->phase_shift; |
144 | c->phase_mask = phase_count - 1; |
145 | c->linear = avr->linear_interp; |
146 | c->filter_length = FFMAX((int)ceil(avr->filter_size / factor), 1); |
147 | c->filter_type = avr->filter_type; |
148 | c->kaiser_beta = avr->kaiser_beta; |
149 | |
150 | switch (avr->internal_sample_fmt) { |
151 | case AV_SAMPLE_FMT_DBLP: |
152 | c->resample_one = c->linear ? resample_linear_dbl : resample_one_dbl; |
153 | c->resample_nearest = resample_nearest_dbl; |
154 | c->set_filter = set_filter_dbl; |
155 | break; |
156 | case AV_SAMPLE_FMT_FLTP: |
157 | c->resample_one = c->linear ? resample_linear_flt : resample_one_flt; |
158 | c->resample_nearest = resample_nearest_flt; |
159 | c->set_filter = set_filter_flt; |
160 | break; |
161 | case AV_SAMPLE_FMT_S32P: |
162 | c->resample_one = c->linear ? resample_linear_s32 : resample_one_s32; |
163 | c->resample_nearest = resample_nearest_s32; |
164 | c->set_filter = set_filter_s32; |
165 | break; |
166 | case AV_SAMPLE_FMT_S16P: |
167 | c->resample_one = c->linear ? resample_linear_s16 : resample_one_s16; |
168 | c->resample_nearest = resample_nearest_s16; |
169 | c->set_filter = set_filter_s16; |
170 | break; |
171 | } |
172 | |
173 | if (ARCH_AARCH64) |
174 | ff_audio_resample_init_aarch64(c, avr->internal_sample_fmt); |
175 | if (ARCH_ARM) |
176 | ff_audio_resample_init_arm(c, avr->internal_sample_fmt); |
177 | |
178 | felem_size = av_get_bytes_per_sample(avr->internal_sample_fmt); |
179 | c->filter_bank = av_mallocz(c->filter_length * (phase_count + 1) * felem_size); |
180 | if (!c->filter_bank) |
181 | goto error; |
182 | |
183 | if (build_filter(c, factor) < 0) |
184 | goto error; |
185 | |
186 | memcpy(&c->filter_bank[(c->filter_length * phase_count + 1) * felem_size], |
187 | c->filter_bank, (c->filter_length - 1) * felem_size); |
188 | memcpy(&c->filter_bank[c->filter_length * phase_count * felem_size], |
189 | &c->filter_bank[(c->filter_length - 1) * felem_size], felem_size); |
190 | |
191 | c->compensation_distance = 0; |
192 | if (!av_reduce(&c->src_incr, &c->dst_incr, out_rate, |
193 | in_rate * (int64_t)phase_count, INT32_MAX / 2)) |
194 | goto error; |
195 | c->ideal_dst_incr = c->dst_incr; |
196 | |
197 | c->padding_size = (c->filter_length - 1) / 2; |
198 | c->initial_padding_filled = 0; |
199 | c->index = 0; |
200 | c->frac = 0; |
201 | |
202 | /* allocate internal buffer */ |
203 | c->buffer = ff_audio_data_alloc(avr->resample_channels, c->padding_size, |
204 | avr->internal_sample_fmt, |
205 | "resample buffer"); |
206 | if (!c->buffer) |
207 | goto error; |
208 | c->buffer->nb_samples = c->padding_size; |
209 | c->initial_padding_samples = c->padding_size; |
210 | |
211 | av_log(avr, AV_LOG_DEBUG, "resample: %s from %d Hz to %d Hz\n", |
212 | av_get_sample_fmt_name(avr->internal_sample_fmt), |
213 | avr->in_sample_rate, avr->out_sample_rate); |
214 | |
215 | return c; |
216 | |
217 | error: |
218 | ff_audio_data_free(&c->buffer); |
219 | av_free(c->filter_bank); |
220 | av_free(c); |
221 | return NULL; |
222 | } |
223 | |
224 | void ff_audio_resample_free(ResampleContext **c) |
225 | { |
226 | if (!*c) |
227 | return; |
228 | ff_audio_data_free(&(*c)->buffer); |
229 | av_free((*c)->filter_bank); |
230 | av_freep(c); |
231 | } |
232 | |
233 | int avresample_set_compensation(AVAudioResampleContext *avr, int sample_delta, |
234 | int compensation_distance) |
235 | { |
236 | ResampleContext *c; |
237 | |
238 | if (compensation_distance < 0) |
239 | return AVERROR(EINVAL); |
240 | if (!compensation_distance && sample_delta) |
241 | return AVERROR(EINVAL); |
242 | |
243 | if (!avr->resample_needed) { |
244 | av_log(avr, AV_LOG_ERROR, "Unable to set resampling compensation\n"); |
245 | return AVERROR(EINVAL); |
246 | } |
247 | c = avr->resample; |
248 | c->compensation_distance = compensation_distance; |
249 | if (compensation_distance) { |
250 | c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr * |
251 | (int64_t)sample_delta / compensation_distance; |
252 | } else { |
253 | c->dst_incr = c->ideal_dst_incr; |
254 | } |
255 | |
256 | return 0; |
257 | } |
258 | |
259 | static int resample(ResampleContext *c, void *dst, const void *src, |
260 | int *consumed, int src_size, int dst_size, int update_ctx, |
261 | int nearest_neighbour) |
262 | { |
263 | int dst_index; |
264 | unsigned int index = c->index; |
265 | int frac = c->frac; |
266 | int dst_incr_frac = c->dst_incr % c->src_incr; |
267 | int dst_incr = c->dst_incr / c->src_incr; |
268 | int compensation_distance = c->compensation_distance; |
269 | |
270 | if (!dst != !src) |
271 | return AVERROR(EINVAL); |
272 | |
273 | if (nearest_neighbour) { |
274 | uint64_t index2 = ((uint64_t)index) << 32; |
275 | int64_t incr = (1LL << 32) * c->dst_incr / c->src_incr; |
276 | dst_size = FFMIN(dst_size, |
277 | (src_size-1-index) * (int64_t)c->src_incr / |
278 | c->dst_incr); |
279 | |
280 | if (dst) { |
281 | for(dst_index = 0; dst_index < dst_size; dst_index++) { |
282 | c->resample_nearest(dst, dst_index, src, index2 >> 32); |
283 | index2 += incr; |
284 | } |
285 | } else { |
286 | dst_index = dst_size; |
287 | } |
288 | index += dst_index * dst_incr; |
289 | index += (frac + dst_index * (int64_t)dst_incr_frac) / c->src_incr; |
290 | frac = (frac + dst_index * (int64_t)dst_incr_frac) % c->src_incr; |
291 | } else { |
292 | for (dst_index = 0; dst_index < dst_size; dst_index++) { |
293 | int sample_index = index >> c->phase_shift; |
294 | |
295 | if (sample_index + c->filter_length > src_size) |
296 | break; |
297 | |
298 | if (dst) |
299 | c->resample_one(c, dst, dst_index, src, index, frac); |
300 | |
301 | frac += dst_incr_frac; |
302 | index += dst_incr; |
303 | if (frac >= c->src_incr) { |
304 | frac -= c->src_incr; |
305 | index++; |
306 | } |
307 | if (dst_index + 1 == compensation_distance) { |
308 | compensation_distance = 0; |
309 | dst_incr_frac = c->ideal_dst_incr % c->src_incr; |
310 | dst_incr = c->ideal_dst_incr / c->src_incr; |
311 | } |
312 | } |
313 | } |
314 | if (consumed) |
315 | *consumed = index >> c->phase_shift; |
316 | |
317 | if (update_ctx) { |
318 | index &= c->phase_mask; |
319 | |
320 | if (compensation_distance) { |
321 | compensation_distance -= dst_index; |
322 | if (compensation_distance <= 0) |
323 | return AVERROR_BUG; |
324 | } |
325 | c->frac = frac; |
326 | c->index = index; |
327 | c->dst_incr = dst_incr_frac + c->src_incr*dst_incr; |
328 | c->compensation_distance = compensation_distance; |
329 | } |
330 | |
331 | return dst_index; |
332 | } |
333 | |
334 | int ff_audio_resample(ResampleContext *c, AudioData *dst, AudioData *src) |
335 | { |
336 | int ch, in_samples, in_leftover, consumed = 0, out_samples = 0; |
337 | int ret = AVERROR(EINVAL); |
338 | int nearest_neighbour = (c->compensation_distance == 0 && |
339 | c->filter_length == 1 && |
340 | c->phase_shift == 0); |
341 | |
342 | in_samples = src ? src->nb_samples : 0; |
343 | in_leftover = c->buffer->nb_samples; |
344 | |
345 | /* add input samples to the internal buffer */ |
346 | if (src) { |
347 | ret = ff_audio_data_combine(c->buffer, in_leftover, src, 0, in_samples); |
348 | if (ret < 0) |
349 | return ret; |
350 | } else if (in_leftover <= c->final_padding_samples) { |
351 | /* no remaining samples to flush */ |
352 | return 0; |
353 | } |
354 | |
355 | if (!c->initial_padding_filled) { |
356 | int bps = av_get_bytes_per_sample(c->avr->internal_sample_fmt); |
357 | int i; |
358 | |
359 | if (src && c->buffer->nb_samples < 2 * c->padding_size) |
360 | return 0; |
361 | |
362 | for (i = 0; i < c->padding_size; i++) |
363 | for (ch = 0; ch < c->buffer->channels; ch++) { |
364 | if (c->buffer->nb_samples > 2 * c->padding_size - i) { |
365 | memcpy(c->buffer->data[ch] + bps * i, |
366 | c->buffer->data[ch] + bps * (2 * c->padding_size - i), bps); |
367 | } else { |
368 | memset(c->buffer->data[ch] + bps * i, 0, bps); |
369 | } |
370 | } |
371 | c->initial_padding_filled = 1; |
372 | } |
373 | |
374 | if (!src && !c->final_padding_filled) { |
375 | int bps = av_get_bytes_per_sample(c->avr->internal_sample_fmt); |
376 | int i; |
377 | |
378 | ret = ff_audio_data_realloc(c->buffer, |
379 | FFMAX(in_samples, in_leftover) + |
380 | c->padding_size); |
381 | if (ret < 0) { |
382 | av_log(c->avr, AV_LOG_ERROR, "Error reallocating resampling buffer\n"); |
383 | return AVERROR(ENOMEM); |
384 | } |
385 | |
386 | for (i = 0; i < c->padding_size; i++) |
387 | for (ch = 0; ch < c->buffer->channels; ch++) { |
388 | if (in_leftover > i) { |
389 | memcpy(c->buffer->data[ch] + bps * (in_leftover + i), |
390 | c->buffer->data[ch] + bps * (in_leftover - i - 1), |
391 | bps); |
392 | } else { |
393 | memset(c->buffer->data[ch] + bps * (in_leftover + i), |
394 | 0, bps); |
395 | } |
396 | } |
397 | c->buffer->nb_samples += c->padding_size; |
398 | c->final_padding_samples = c->padding_size; |
399 | c->final_padding_filled = 1; |
400 | } |
401 | |
402 | |
403 | /* calculate output size and reallocate output buffer if needed */ |
404 | /* TODO: try to calculate this without the dummy resample() run */ |
405 | if (!dst->read_only && dst->allow_realloc) { |
406 | out_samples = resample(c, NULL, NULL, NULL, c->buffer->nb_samples, |
407 | INT_MAX, 0, nearest_neighbour); |
408 | ret = ff_audio_data_realloc(dst, out_samples); |
409 | if (ret < 0) { |
410 | av_log(c->avr, AV_LOG_ERROR, "error reallocating output\n"); |
411 | return ret; |
412 | } |
413 | } |
414 | |
415 | /* resample each channel plane */ |
416 | for (ch = 0; ch < c->buffer->channels; ch++) { |
417 | out_samples = resample(c, (void *)dst->data[ch], |
418 | (const void *)c->buffer->data[ch], &consumed, |
419 | c->buffer->nb_samples, dst->allocated_samples, |
420 | ch + 1 == c->buffer->channels, nearest_neighbour); |
421 | } |
422 | if (out_samples < 0) { |
423 | av_log(c->avr, AV_LOG_ERROR, "error during resampling\n"); |
424 | return out_samples; |
425 | } |
426 | |
427 | /* drain consumed samples from the internal buffer */ |
428 | ff_audio_data_drain(c->buffer, consumed); |
429 | c->initial_padding_samples = FFMAX(c->initial_padding_samples - consumed, 0); |
430 | |
431 | av_log(c->avr, AV_LOG_TRACE, "resampled %d in + %d leftover to %d out + %d leftover\n", |
432 | in_samples, in_leftover, out_samples, c->buffer->nb_samples); |
433 | |
434 | dst->nb_samples = out_samples; |
435 | return 0; |
436 | } |
437 | |
438 | int avresample_get_delay(AVAudioResampleContext *avr) |
439 | { |
440 | ResampleContext *c = avr->resample; |
441 | |
442 | if (!avr->resample_needed || !avr->resample) |
443 | return 0; |
444 | |
445 | return FFMAX(c->buffer->nb_samples - c->padding_size, 0); |
446 | } |
447 |