| 1 | /* |
| 2 | * Wavesynth pseudo-codec |
| 3 | * Copyright (c) 2011 Nicolas George |
| 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/intreadwrite.h" |
| 23 | #include "libavutil/log.h" |
| 24 | #include "avcodec.h" |
| 25 | #include "internal.h" |
| 26 | |
| 27 | |
| 28 | #define SIN_BITS 14 |
| 29 | #define WS_MAX_CHANNELS 32 |
| 30 | #define INF_TS 0x7FFFFFFFFFFFFFFF |
| 31 | |
| 32 | #define PINK_UNIT 128 |
| 33 | |
| 34 | /* |
| 35 | Format of the extradata and packets |
| 36 | |
| 37 | THIS INFORMATION IS NOT PART OF THE PUBLIC API OR ABI. |
| 38 | IT CAN CHANGE WITHOUT NOTIFICATION. |
| 39 | |
| 40 | All numbers are in little endian. |
| 41 | |
| 42 | The codec extradata define a set of intervals with uniform content. |
| 43 | Overlapping intervals are added together. |
| 44 | |
| 45 | extradata: |
| 46 | uint32 number of intervals |
| 47 | ... intervals |
| 48 | |
| 49 | interval: |
| 50 | int64 start timestamp; time_base must be 1/sample_rate; |
| 51 | start timestamps must be in ascending order |
| 52 | int64 end timestamp |
| 53 | uint32 type |
| 54 | uint32 channels mask |
| 55 | ... additional information, depends on type |
| 56 | |
| 57 | sine interval (type fourcc "SINE"): |
| 58 | int32 start frequency, in 1/(1<<16) Hz |
| 59 | int32 end frequency |
| 60 | int32 start amplitude, 1<<16 is the full amplitude |
| 61 | int32 end amplitude |
| 62 | uint32 start phase, 0 is sin(0), 0x20000000 is sin(pi/2), etc.; |
| 63 | n | (1<<31) means to match the phase of previous channel #n |
| 64 | |
| 65 | pink noise interval (type fourcc "NOIS"): |
| 66 | int32 start amplitude |
| 67 | int32 end amplitude |
| 68 | |
| 69 | The input packets encode the time and duration of the requested segment. |
| 70 | |
| 71 | packet: |
| 72 | int64 start timestamp |
| 73 | int32 duration |
| 74 | |
| 75 | */ |
| 76 | |
| 77 | enum ws_interval_type { |
| 78 | WS_SINE = MKTAG('S','I','N','E'), |
| 79 | WS_NOISE = MKTAG('N','O','I','S'), |
| 80 | }; |
| 81 | |
| 82 | struct ws_interval { |
| 83 | int64_t ts_start, ts_end; |
| 84 | uint64_t phi0, dphi0, ddphi; |
| 85 | uint64_t amp0, damp; |
| 86 | uint64_t phi, dphi, amp; |
| 87 | uint32_t channels; |
| 88 | enum ws_interval_type type; |
| 89 | int next; |
| 90 | }; |
| 91 | |
| 92 | struct wavesynth_context { |
| 93 | int64_t cur_ts; |
| 94 | int64_t next_ts; |
| 95 | int32_t *sin; |
| 96 | struct ws_interval *inter; |
| 97 | uint32_t dither_state; |
| 98 | uint32_t pink_state; |
| 99 | int32_t pink_pool[PINK_UNIT]; |
| 100 | unsigned pink_need, pink_pos; |
| 101 | int nb_inter; |
| 102 | int cur_inter; |
| 103 | int next_inter; |
| 104 | }; |
| 105 | |
| 106 | #define LCG_A 1284865837 |
| 107 | #define LCG_C 4150755663 |
| 108 | #define LCG_AI 849225893 /* A*AI = 1 [mod 1<<32] */ |
| 109 | |
| 110 | static uint32_t lcg_next(uint32_t *s) |
| 111 | { |
| 112 | *s = *s * LCG_A + LCG_C; |
| 113 | return *s; |
| 114 | } |
| 115 | |
| 116 | static void lcg_seek(uint32_t *s, int64_t dt) |
| 117 | { |
| 118 | uint32_t a, c, t = *s; |
| 119 | |
| 120 | if (dt >= 0) { |
| 121 | a = LCG_A; |
| 122 | c = LCG_C; |
| 123 | } else { /* coefficients for a step backward */ |
| 124 | a = LCG_AI; |
| 125 | c = (uint32_t)(LCG_AI * LCG_C); |
| 126 | dt = -dt; |
| 127 | } |
| 128 | while (dt) { |
| 129 | if (dt & 1) |
| 130 | t = a * t + c; |
| 131 | c *= a + 1; /* coefficients for a double step */ |
| 132 | a *= a; |
| 133 | dt >>= 1; |
| 134 | } |
| 135 | *s = t; |
| 136 | } |
| 137 | |
| 138 | /* Emulate pink noise by summing white noise at the sampling frequency, |
| 139 | * white noise at half the sampling frequency (each value taken twice), |
| 140 | * etc., with a total of 8 octaves. |
| 141 | * This is known as the Voss-McCartney algorithm. */ |
| 142 | |
| 143 | static void pink_fill(struct wavesynth_context *ws) |
| 144 | { |
| 145 | int32_t vt[7] = { 0 }, v = 0; |
| 146 | int i, j; |
| 147 | |
| 148 | ws->pink_pos = 0; |
| 149 | if (!ws->pink_need) |
| 150 | return; |
| 151 | for (i = 0; i < PINK_UNIT; i++) { |
| 152 | for (j = 0; j < 7; j++) { |
| 153 | if ((i >> j) & 1) |
| 154 | break; |
| 155 | v -= vt[j]; |
| 156 | vt[j] = (int32_t)lcg_next(&ws->pink_state) >> 3; |
| 157 | v += vt[j]; |
| 158 | } |
| 159 | ws->pink_pool[i] = v + ((int32_t)lcg_next(&ws->pink_state) >> 3); |
| 160 | } |
| 161 | lcg_next(&ws->pink_state); /* so we use exactly 256 steps */ |
| 162 | } |
| 163 | |
| 164 | /** |
| 165 | * @return (1<<64) * a / b, without overflow, if a < b |
| 166 | */ |
| 167 | static uint64_t frac64(uint64_t a, uint64_t b) |
| 168 | { |
| 169 | uint64_t r = 0; |
| 170 | int i; |
| 171 | |
| 172 | if (b < (uint64_t)1 << 32) { /* b small, use two 32-bits steps */ |
| 173 | a <<= 32; |
| 174 | return ((a / b) << 32) | ((a % b) << 32) / b; |
| 175 | } |
| 176 | if (b < (uint64_t)1 << 48) { /* b medium, use four 16-bits steps */ |
| 177 | for (i = 0; i < 4; i++) { |
| 178 | a <<= 16; |
| 179 | r = (r << 16) | (a / b); |
| 180 | a %= b; |
| 181 | } |
| 182 | return r; |
| 183 | } |
| 184 | for (i = 63; i >= 0; i--) { |
| 185 | if (a >= (uint64_t)1 << 63 || a << 1 >= b) { |
| 186 | r |= (uint64_t)1 << i; |
| 187 | a = (a << 1) - b; |
| 188 | } else { |
| 189 | a <<= 1; |
| 190 | } |
| 191 | } |
| 192 | return r; |
| 193 | } |
| 194 | |
| 195 | static uint64_t phi_at(struct ws_interval *in, int64_t ts) |
| 196 | { |
| 197 | uint64_t dt = ts - in->ts_start; |
| 198 | uint64_t dt2 = dt & 1 ? /* dt * (dt - 1) / 2 without overflow */ |
| 199 | dt * ((dt - 1) >> 1) : (dt >> 1) * (dt - 1); |
| 200 | return in->phi0 + dt * in->dphi0 + dt2 * in->ddphi; |
| 201 | } |
| 202 | |
| 203 | static void wavesynth_seek(struct wavesynth_context *ws, int64_t ts) |
| 204 | { |
| 205 | int *last, i; |
| 206 | struct ws_interval *in; |
| 207 | |
| 208 | last = &ws->cur_inter; |
| 209 | for (i = 0; i < ws->nb_inter; i++) { |
| 210 | in = &ws->inter[i]; |
| 211 | if (ts < in->ts_start) |
| 212 | break; |
| 213 | if (ts >= in->ts_end) |
| 214 | continue; |
| 215 | *last = i; |
| 216 | last = &in->next; |
| 217 | in->phi = phi_at(in, ts); |
| 218 | in->dphi = in->dphi0 + (ts - in->ts_start) * in->ddphi; |
| 219 | in->amp = in->amp0 + (ts - in->ts_start) * in->damp; |
| 220 | } |
| 221 | ws->next_inter = i; |
| 222 | ws->next_ts = i < ws->nb_inter ? ws->inter[i].ts_start : INF_TS; |
| 223 | *last = -1; |
| 224 | lcg_seek(&ws->dither_state, ts - ws->cur_ts); |
| 225 | if (ws->pink_need) { |
| 226 | int64_t pink_ts_cur = (ws->cur_ts + PINK_UNIT - 1) & ~(PINK_UNIT - 1); |
| 227 | int64_t pink_ts_next = ts & ~(PINK_UNIT - 1); |
| 228 | int pos = ts & (PINK_UNIT - 1); |
| 229 | lcg_seek(&ws->pink_state, (pink_ts_next - pink_ts_cur) << 1); |
| 230 | if (pos) { |
| 231 | pink_fill(ws); |
| 232 | ws->pink_pos = pos; |
| 233 | } else { |
| 234 | ws->pink_pos = PINK_UNIT; |
| 235 | } |
| 236 | } |
| 237 | ws->cur_ts = ts; |
| 238 | } |
| 239 | |
| 240 | static int wavesynth_parse_extradata(AVCodecContext *avc) |
| 241 | { |
| 242 | struct wavesynth_context *ws = avc->priv_data; |
| 243 | struct ws_interval *in; |
| 244 | uint8_t *edata, *edata_end; |
| 245 | int32_t f1, f2, a1, a2; |
| 246 | uint32_t phi; |
| 247 | int64_t dphi1, dphi2, dt, cur_ts = -0x8000000000000000; |
| 248 | int i; |
| 249 | |
| 250 | if (avc->extradata_size < 4) |
| 251 | return AVERROR(EINVAL); |
| 252 | edata = avc->extradata; |
| 253 | edata_end = edata + avc->extradata_size; |
| 254 | ws->nb_inter = AV_RL32(edata); |
| 255 | edata += 4; |
| 256 | if (ws->nb_inter < 0) |
| 257 | return AVERROR(EINVAL); |
| 258 | ws->inter = av_calloc(ws->nb_inter, sizeof(*ws->inter)); |
| 259 | if (!ws->inter) |
| 260 | return AVERROR(ENOMEM); |
| 261 | for (i = 0; i < ws->nb_inter; i++) { |
| 262 | in = &ws->inter[i]; |
| 263 | if (edata_end - edata < 24) |
| 264 | return AVERROR(EINVAL); |
| 265 | in->ts_start = AV_RL64(edata + 0); |
| 266 | in->ts_end = AV_RL64(edata + 8); |
| 267 | in->type = AV_RL32(edata + 16); |
| 268 | in->channels = AV_RL32(edata + 20); |
| 269 | edata += 24; |
| 270 | if (in->ts_start < cur_ts || in->ts_end <= in->ts_start) |
| 271 | return AVERROR(EINVAL); |
| 272 | cur_ts = in->ts_start; |
| 273 | dt = in->ts_end - in->ts_start; |
| 274 | switch (in->type) { |
| 275 | case WS_SINE: |
| 276 | if (edata_end - edata < 20) |
| 277 | return AVERROR(EINVAL); |
| 278 | f1 = AV_RL32(edata + 0); |
| 279 | f2 = AV_RL32(edata + 4); |
| 280 | a1 = AV_RL32(edata + 8); |
| 281 | a2 = AV_RL32(edata + 12); |
| 282 | phi = AV_RL32(edata + 16); |
| 283 | edata += 20; |
| 284 | dphi1 = frac64(f1, (int64_t)avc->sample_rate << 16); |
| 285 | dphi2 = frac64(f2, (int64_t)avc->sample_rate << 16); |
| 286 | in->dphi0 = dphi1; |
| 287 | in->ddphi = (dphi2 - dphi1) / dt; |
| 288 | if (phi & 0x80000000) { |
| 289 | phi &= ~0x80000000; |
| 290 | if (phi >= i) |
| 291 | return AVERROR(EINVAL); |
| 292 | in->phi0 = phi_at(&ws->inter[phi], in->ts_start); |
| 293 | } else { |
| 294 | in->phi0 = (uint64_t)phi << 33; |
| 295 | } |
| 296 | break; |
| 297 | case WS_NOISE: |
| 298 | if (edata_end - edata < 8) |
| 299 | return AVERROR(EINVAL); |
| 300 | a1 = AV_RL32(edata + 0); |
| 301 | a2 = AV_RL32(edata + 4); |
| 302 | edata += 8; |
| 303 | break; |
| 304 | default: |
| 305 | return AVERROR(EINVAL); |
| 306 | } |
| 307 | in->amp0 = (int64_t)a1 << 32; |
| 308 | in->damp = (((int64_t)a2 << 32) - ((int64_t)a1 << 32)) / dt; |
| 309 | } |
| 310 | if (edata != edata_end) |
| 311 | return AVERROR(EINVAL); |
| 312 | return 0; |
| 313 | } |
| 314 | |
| 315 | static av_cold int wavesynth_init(AVCodecContext *avc) |
| 316 | { |
| 317 | struct wavesynth_context *ws = avc->priv_data; |
| 318 | int i, r; |
| 319 | |
| 320 | if (avc->channels > WS_MAX_CHANNELS) { |
| 321 | av_log(avc, AV_LOG_ERROR, |
| 322 | "This implementation is limited to %d channels.\n", |
| 323 | WS_MAX_CHANNELS); |
| 324 | return AVERROR(EINVAL); |
| 325 | } |
| 326 | r = wavesynth_parse_extradata(avc); |
| 327 | if (r < 0) { |
| 328 | av_log(avc, AV_LOG_ERROR, "Invalid intervals definitions.\n"); |
| 329 | goto fail; |
| 330 | } |
| 331 | ws->sin = av_malloc(sizeof(*ws->sin) << SIN_BITS); |
| 332 | if (!ws->sin) { |
| 333 | r = AVERROR(ENOMEM); |
| 334 | goto fail; |
| 335 | } |
| 336 | for (i = 0; i < 1 << SIN_BITS; i++) |
| 337 | ws->sin[i] = floor(32767 * sin(2 * M_PI * i / (1 << SIN_BITS))); |
| 338 | ws->dither_state = MKTAG('D','I','T','H'); |
| 339 | for (i = 0; i < ws->nb_inter; i++) |
| 340 | ws->pink_need += ws->inter[i].type == WS_NOISE; |
| 341 | ws->pink_state = MKTAG('P','I','N','K'); |
| 342 | ws->pink_pos = PINK_UNIT; |
| 343 | wavesynth_seek(ws, 0); |
| 344 | avc->sample_fmt = AV_SAMPLE_FMT_S16; |
| 345 | return 0; |
| 346 | |
| 347 | fail: |
| 348 | av_free(ws->inter); |
| 349 | av_free(ws->sin); |
| 350 | return r; |
| 351 | } |
| 352 | |
| 353 | static void wavesynth_synth_sample(struct wavesynth_context *ws, int64_t ts, |
| 354 | int32_t *channels) |
| 355 | { |
| 356 | int32_t amp, val, *cv; |
| 357 | struct ws_interval *in; |
| 358 | int i, *last, pink; |
| 359 | uint32_t c, all_ch = 0; |
| 360 | |
| 361 | i = ws->cur_inter; |
| 362 | last = &ws->cur_inter; |
| 363 | if (ws->pink_pos == PINK_UNIT) |
| 364 | pink_fill(ws); |
| 365 | pink = ws->pink_pool[ws->pink_pos++] >> 16; |
| 366 | while (i >= 0) { |
| 367 | in = &ws->inter[i]; |
| 368 | i = in->next; |
| 369 | if (ts >= in->ts_end) { |
| 370 | *last = i; |
| 371 | continue; |
| 372 | } |
| 373 | last = &in->next; |
| 374 | amp = in->amp >> 32; |
| 375 | in->amp += in->damp; |
| 376 | switch (in->type) { |
| 377 | case WS_SINE: |
| 378 | val = amp * ws->sin[in->phi >> (64 - SIN_BITS)]; |
| 379 | in->phi += in->dphi; |
| 380 | in->dphi += in->ddphi; |
| 381 | break; |
| 382 | case WS_NOISE: |
| 383 | val = amp * pink; |
| 384 | break; |
| 385 | default: |
| 386 | val = 0; |
| 387 | } |
| 388 | all_ch |= in->channels; |
| 389 | for (c = in->channels, cv = channels; c; c >>= 1, cv++) |
| 390 | if (c & 1) |
| 391 | *cv += val; |
| 392 | } |
| 393 | val = (int32_t)lcg_next(&ws->dither_state) >> 16; |
| 394 | for (c = all_ch, cv = channels; c; c >>= 1, cv++) |
| 395 | if (c & 1) |
| 396 | *cv += val; |
| 397 | } |
| 398 | |
| 399 | static void wavesynth_enter_intervals(struct wavesynth_context *ws, int64_t ts) |
| 400 | { |
| 401 | int *last, i; |
| 402 | struct ws_interval *in; |
| 403 | |
| 404 | last = &ws->cur_inter; |
| 405 | for (i = ws->cur_inter; i >= 0; i = ws->inter[i].next) |
| 406 | last = &ws->inter[i].next; |
| 407 | for (i = ws->next_inter; i < ws->nb_inter; i++) { |
| 408 | in = &ws->inter[i]; |
| 409 | if (ts < in->ts_start) |
| 410 | break; |
| 411 | if (ts >= in->ts_end) |
| 412 | continue; |
| 413 | *last = i; |
| 414 | last = &in->next; |
| 415 | in->phi = in->phi0; |
| 416 | in->dphi = in->dphi0; |
| 417 | in->amp = in->amp0; |
| 418 | } |
| 419 | ws->next_inter = i; |
| 420 | ws->next_ts = i < ws->nb_inter ? ws->inter[i].ts_start : INF_TS; |
| 421 | *last = -1; |
| 422 | } |
| 423 | |
| 424 | static int wavesynth_decode(AVCodecContext *avc, void *rframe, int *rgot_frame, |
| 425 | AVPacket *packet) |
| 426 | { |
| 427 | struct wavesynth_context *ws = avc->priv_data; |
| 428 | AVFrame *frame = rframe; |
| 429 | int64_t ts; |
| 430 | int duration; |
| 431 | int s, c, r; |
| 432 | int16_t *pcm; |
| 433 | int32_t channels[WS_MAX_CHANNELS]; |
| 434 | |
| 435 | *rgot_frame = 0; |
| 436 | if (packet->size != 12) |
| 437 | return AVERROR_INVALIDDATA; |
| 438 | ts = AV_RL64(packet->data); |
| 439 | if (ts != ws->cur_ts) |
| 440 | wavesynth_seek(ws, ts); |
| 441 | duration = AV_RL32(packet->data + 8); |
| 442 | if (duration <= 0) |
| 443 | return AVERROR(EINVAL); |
| 444 | frame->nb_samples = duration; |
| 445 | r = ff_get_buffer(avc, frame, 0); |
| 446 | if (r < 0) |
| 447 | return r; |
| 448 | pcm = (int16_t *)frame->data[0]; |
| 449 | for (s = 0; s < duration; s++, ts++) { |
| 450 | memset(channels, 0, avc->channels * sizeof(*channels)); |
| 451 | if (ts >= ws->next_ts) |
| 452 | wavesynth_enter_intervals(ws, ts); |
| 453 | wavesynth_synth_sample(ws, ts, channels); |
| 454 | for (c = 0; c < avc->channels; c++) |
| 455 | *(pcm++) = channels[c] >> 16; |
| 456 | } |
| 457 | ws->cur_ts += duration; |
| 458 | *rgot_frame = 1; |
| 459 | return packet->size; |
| 460 | } |
| 461 | |
| 462 | static av_cold int wavesynth_close(AVCodecContext *avc) |
| 463 | { |
| 464 | struct wavesynth_context *ws = avc->priv_data; |
| 465 | |
| 466 | av_free(ws->sin); |
| 467 | av_free(ws->inter); |
| 468 | return 0; |
| 469 | } |
| 470 | |
| 471 | AVCodec ff_ffwavesynth_decoder = { |
| 472 | .name = "wavesynth", |
| 473 | .long_name = NULL_IF_CONFIG_SMALL("Wave synthesis pseudo-codec"), |
| 474 | .type = AVMEDIA_TYPE_AUDIO, |
| 475 | .id = AV_CODEC_ID_FFWAVESYNTH, |
| 476 | .priv_data_size = sizeof(struct wavesynth_context), |
| 477 | .init = wavesynth_init, |
| 478 | .close = wavesynth_close, |
| 479 | .decode = wavesynth_decode, |
| 480 | .capabilities = CODEC_CAP_DR1, |
| 481 | }; |