Showing the changes to libavfilter (1 of 1 file). Show all
Unified · Split
| @@ -26,6 +26,7 @@ · 25 unmodified lines | ||
| 26 | 26 | #include "libavcodec/avfft.h" |
| 27 | 27 | #include "libavutil/eval.h" |
| 28 | 28 | #include "audio.h" |
| 29 | +#include "filters.h" | |
| 29 | 30 | #include "window_func.h" |
| 30 | 31 | |
| 31 | 32 | typedef struct AFFTFiltContext { |
| @@ -46,7 +47,7 @@ typedef struct AFFTFiltContext { · 14 unmodified lines | ||
| 46 | 47 | int hop_size; |
| 47 | 48 | float overlap; |
| 48 | 49 | AVFrame *buffer; |
| 49 | − int start, end; | |
| 50 | + int eof; | |
| 50 | 51 | int win_func; |
| 51 | 52 | float win_scale; |
| 52 | 53 | float *window_func_lut; |
| @@ -240,7 +241,7 @@ static int config_input(AVFilterLink *inlink) · 187 unmodified lines | ||
| 240 | 241 | return ret; |
| 241 | 242 | } |
| 242 | 243 | |
| 243 | −static int filter_frame(AVFilterLink *inlink, AVFrame *frame) | |
| 244 | +static int filter_frame(AVFilterLink *inlink) | |
| 244 | 245 | { |
| 245 | 246 | AVFilterContext *ctx = inlink->dst; |
| 246 | 247 | AVFilterLink *outlink = ctx->outputs[0]; |
| @@ -249,140 +250,163 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *frame) · 2 unmodified lines | ||
| 249 | 250 | const float f = 1. / s->win_scale; |
| 250 | 251 | double values[VAR_VARS_NB]; |
| 251 | 252 | AVFrame *out, *in = NULL; |
| 252 | − int ch, n, ret, i, j, k; | |
| 253 | − int start = s->start, end = s->end; | |
| 253 | + int ch, n, ret, i; | |
| 254 | 254 | |
| 255 | − if (s->pts == AV_NOPTS_VALUE) | |
| 256 | − s->pts = frame->pts; | |
| 255 | + if (!in) { | |
| 256 | + in = ff_get_audio_buffer(outlink, window_size); | |
| 257 | + if (!in) | |
| 258 | + return AVERROR(ENOMEM); | |
| 259 | + } | |
| 257 | 260 | |
| 258 | − ret = av_audio_fifo_write(s->fifo, (void **)frame->extended_data, frame->nb_samples); | |
| 259 | − av_frame_free(&frame); | |
| 261 | + ret = av_audio_fifo_peek(s->fifo, (void **)in->extended_data, window_size); | |
| 260 | 262 | if (ret < 0) |
| 261 | − return ret; | |
| 263 | + goto fail; | |
| 262 | 264 | |
| 263 | − while (av_audio_fifo_size(s->fifo) >= window_size) { | |
| 264 | − if (!in) { | |
| 265 | − in = ff_get_audio_buffer(outlink, window_size); | |
| 266 | − if (!in) | |
| 267 | − return AVERROR(ENOMEM); | |
| 265 | + for (ch = 0; ch < inlink->channels; ch++) { | |
| 266 | + const float *src = (float *)in->extended_data[ch]; | |
| 267 | + FFTComplex *fft_data = s->fft_data[ch]; | |
| 268 | + | |
| 269 | + for (n = 0; n < in->nb_samples; n++) { | |
| 270 | + fft_data[n].re = src[n] * s->window_func_lut[n]; | |
| 271 | + fft_data[n].im = 0; | |
| 268 | 272 | } |
| 269 | 273 | |
| 270 | − ret = av_audio_fifo_peek(s->fifo, (void **)in->extended_data, window_size); | |
| 271 | − if (ret < 0) | |
| 272 | − break; | |
| 274 | + for (; n < window_size; n++) { | |
| 275 | + fft_data[n].re = 0; | |
| 276 | + fft_data[n].im = 0; | |
| 277 | + } | |
| 278 | + } | |
| 273 | 279 | |
| 274 | − for (ch = 0; ch < inlink->channels; ch++) { | |
| 275 | − const float *src = (float *)in->extended_data[ch]; | |
| 276 | − FFTComplex *fft_data = s->fft_data[ch]; | |
| 280 | + values[VAR_PTS] = s->pts; | |
| 281 | + values[VAR_SAMPLE_RATE] = inlink->sample_rate; | |
| 282 | + values[VAR_NBBINS] = window_size / 2; | |
| 283 | + values[VAR_CHANNELS] = inlink->channels; | |
| 277 | 284 | |
| 278 | − for (n = 0; n < in->nb_samples; n++) { | |
| 279 | − fft_data[n].re = src[n] * s->window_func_lut[n]; | |
| 280 | − fft_data[n].im = 0; | |
| 281 | − } | |
| 285 | + for (ch = 0; ch < inlink->channels; ch++) { | |
| 286 | + FFTComplex *fft_data = s->fft_data[ch]; | |
| 282 | 287 | |
| 283 | − for (; n < window_size; n++) { | |
| 284 | − fft_data[n].re = 0; | |
| 285 | − fft_data[n].im = 0; | |
| 286 | − } | |
| 287 | − } | |
| 288 | + av_fft_permute(s->fft, fft_data); | |
| 289 | + av_fft_calc(s->fft, fft_data); | |
| 290 | + } | |
| 288 | 291 | |
| 289 | − values[VAR_PTS] = s->pts; | |
| 290 | − values[VAR_SAMPLE_RATE] = inlink->sample_rate; | |
| 291 | − values[VAR_NBBINS] = window_size / 2; | |
| 292 | − values[VAR_CHANNELS] = inlink->channels; | |
| 292 | + for (ch = 0; ch < inlink->channels; ch++) { | |
| 293 | + FFTComplex *fft_data = s->fft_data[ch]; | |
| 294 | + FFTComplex *fft_temp = s->fft_temp[ch]; | |
| 295 | + float *buf = (float *)s->buffer->extended_data[ch]; | |
| 296 | + int x; | |
| 297 | + values[VAR_CHANNEL] = ch; | |
| 293 | 298 | |
| 294 | − for (ch = 0; ch < inlink->channels; ch++) { | |
| 295 | − FFTComplex *fft_data = s->fft_data[ch]; | |
| 299 | + for (n = 0; n <= window_size / 2; n++) { | |
| 300 | + float fr, fi; | |
| 296 | 301 | |
| 297 | − av_fft_permute(s->fft, fft_data); | |
| 298 | − av_fft_calc(s->fft, fft_data); | |
| 299 | − } | |
| 302 | + values[VAR_BIN] = n; | |
| 303 | + values[VAR_REAL] = fft_data[n].re; | |
| 304 | + values[VAR_IMAG] = fft_data[n].im; | |
| 300 | 305 | |
| 301 | − for (ch = 0; ch < inlink->channels; ch++) { | |
| 302 | − FFTComplex *fft_data = s->fft_data[ch]; | |
| 303 | − FFTComplex *fft_temp = s->fft_temp[ch]; | |
| 304 | − float *buf = (float *)s->buffer->extended_data[ch]; | |
| 305 | − int x; | |
| 306 | − values[VAR_CHANNEL] = ch; | |
| 306 | + fr = av_expr_eval(s->real[ch], values, s); | |
| 307 | + fi = av_expr_eval(s->imag[ch], values, s); | |
| 307 | 308 | |
| 308 | − for (n = 0; n <= window_size / 2; n++) { | |
| 309 | − float fr, fi; | |
| 309 | + fft_temp[n].re = fr; | |
| 310 | + fft_temp[n].im = fi; | |
| 311 | + } | |
| 310 | 312 | |
| 311 | − values[VAR_BIN] = n; | |
| 312 | − values[VAR_REAL] = fft_data[n].re; | |
| 313 | − values[VAR_IMAG] = fft_data[n].im; | |
| 313 | + for (n = window_size / 2 + 1, x = window_size / 2 - 1; n < window_size; n++, x--) { | |
| 314 | + fft_temp[n].re = fft_temp[x].re; | |
| 315 | + fft_temp[n].im = -fft_temp[x].im; | |
| 316 | + } | |
| 314 | 317 | |
| 315 | − fr = av_expr_eval(s->real[ch], values, s); | |
| 316 | − fi = av_expr_eval(s->imag[ch], values, s); | |
| 318 | + av_fft_permute(s->ifft, fft_temp); | |
| 319 | + av_fft_calc(s->ifft, fft_temp); | |
| 317 | 320 | |
| 318 | − fft_temp[n].re = fr; | |
| 319 | − fft_temp[n].im = fi; | |
| 320 | − } | |
| 321 | + for (i = 0; i < window_size; i++) { | |
| 322 | + buf[i] += s->fft_temp[ch][i].re * f; | |
| 323 | + } | |
| 324 | + } | |
| 321 | 325 | |
| 322 | − for (n = window_size / 2 + 1, x = window_size / 2 - 1; n < window_size; n++, x--) { | |
| 323 | − fft_temp[n].re = fft_temp[x].re; | |
| 324 | − fft_temp[n].im = -fft_temp[x].im; | |
| 325 | − } | |
| 326 | + out = ff_get_audio_buffer(outlink, s->hop_size); | |
| 327 | + if (!out) { | |
| 328 | + ret = AVERROR(ENOMEM); | |
| 329 | + goto fail; | |
| 330 | + } | |
| 326 | 331 | |
| 327 | − av_fft_permute(s->ifft, fft_temp); | |
| 328 | − av_fft_calc(s->ifft, fft_temp); | |
| 332 | + out->pts = s->pts; | |
| 333 | + s->pts += s->hop_size; | |
| 329 | 334 | |
| 330 | − start = s->start; | |
| 331 | − end = s->end; | |
| 332 | − k = end; | |
| 333 | − for (i = 0, j = start; j < k && i < window_size; i++, j++) { | |
| 334 | − buf[j] += s->fft_temp[ch][i].re * f; | |
| 335 | − } | |
| 335 | + for (ch = 0; ch < inlink->channels; ch++) { | |
| 336 | + float *dst = (float *)out->extended_data[ch]; | |
| 337 | + float *buf = (float *)s->buffer->extended_data[ch]; | |
| 336 | 338 | |
| 337 | − for (; i < window_size; i++, j++) { | |
| 338 | − buf[j] = s->fft_temp[ch][i].re * f; | |
| 339 | − } | |
| 339 | + for (n = 0; n < s->hop_size; n++) | |
| 340 | + dst[n] = buf[n] * (1.f - s->overlap); | |
| 341 | + memmove(buf, buf + s->hop_size, window_size * 4); | |
| 342 | + } | |
| 340 | 343 | |
| 341 | − start += s->hop_size; | |
| 342 | − end = j; | |
| 343 | − } | |
| 344 | + ret = ff_filter_frame(outlink, out); | |
| 345 | + if (ret < 0) | |
| 346 | + goto fail; | |
| 344 | 347 | |
| 345 | − s->start = start; | |
| 346 | − s->end = end; | |
| 348 | + av_audio_fifo_drain(s->fifo, s->hop_size); | |
| 347 | 349 | |
| 348 | − if (start >= window_size) { | |
| 349 | − float *dst, *buf; | |
| 350 | +fail: | |
| 351 | + av_frame_free(&in); | |
| 352 | + return ret < 0 ? ret : 0; | |
| 353 | +} | |
| 350 | 354 | |
| 351 | − start -= window_size; | |
| 352 | − end -= window_size; | |
| 355 | +static int activate(AVFilterContext *ctx) | |
| 356 | +{ | |
| 357 | + AVFilterLink *inlink = ctx->inputs[0]; | |
| 358 | + AVFilterLink *outlink = ctx->outputs[0]; | |
| 359 | + AFFTFiltContext *s = ctx->priv; | |
| 360 | + AVFrame *in = NULL; | |
| 361 | + int ret = 0, status; | |
| 362 | + int64_t pts; | |
| 353 | 363 | |
| 354 | − s->start = start; | |
| 355 | − s->end = end; | |
| 364 | + FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink); | |
| 356 | 365 | |
| 357 | − out = ff_get_audio_buffer(outlink, window_size); | |
| 358 | − if (!out) { | |
| 359 | − ret = AVERROR(ENOMEM); | |
| 360 | − break; | |
| 361 | − } | |
| 366 | + if (!s->eof && av_audio_fifo_size(s->fifo) < s->window_size) { | |
| 367 | + ret = ff_inlink_consume_frame(inlink, &in); | |
| 368 | + if (ret < 0) | |
| 369 | + return ret; | |
| 362 | 370 | |
| 363 | − out->pts = s->pts; | |
| 364 | − s->pts += window_size; | |
| 371 | + if (ret > 0) { | |
| 372 | + ret = av_audio_fifo_write(s->fifo, (void **)in->extended_data, | |
| 373 | + in->nb_samples); | |
| 374 | + if (ret >= 0 && s->pts == AV_NOPTS_VALUE) | |
| 375 | + s->pts = in->pts; | |
| 365 | 376 | |
| 366 | − for (ch = 0; ch < inlink->channels; ch++) { | |
| 367 | − dst = (float *)out->extended_data[ch]; | |
| 368 | − buf = (float *)s->buffer->extended_data[ch]; | |
| 377 | + av_frame_free(&in); | |
| 378 | + if (ret < 0) | |
| 379 | + return ret; | |
| 380 | + } | |
| 381 | + } | |
| 369 | 382 | |
| 370 | − for (n = 0; n < window_size; n++) { | |
| 371 | − dst[n] = buf[n] * (1 - s->overlap); | |
| 372 | − } | |
| 373 | − memmove(buf, buf + window_size, window_size * 4); | |
| 374 | − } | |
| 383 | + if ((av_audio_fifo_size(s->fifo) >= s->window_size) || | |
| 384 | + (av_audio_fifo_size(s->fifo) > 0 && s->eof)) { | |
| 385 | + ret = filter_frame(inlink); | |
| 386 | + if (av_audio_fifo_size(s->fifo) >= s->window_size) | |
| 387 | + ff_filter_set_ready(ctx, 100); | |
| 388 | + return ret; | |
| 389 | + } | |
| 375 | 390 | |
| 376 | − ret = ff_filter_frame(outlink, out); | |
| 377 | − if (ret < 0) | |
| 378 | − break; | |
| 391 | + if (!s->eof && ff_inlink_acknowledge_status(inlink, &status, &pts)) { | |
| 392 | + if (status == AVERROR_EOF) { | |
| 393 | + s->eof = 1; | |
| 394 | + if (av_audio_fifo_size(s->fifo) >= 0) { | |
| 395 | + ff_filter_set_ready(ctx, 100); | |
| 396 | + return 0; | |
| 397 | + } | |
| 379 | 398 | } |
| 399 | + } | |
| 380 | 400 | |
| 381 | − av_audio_fifo_drain(s->fifo, s->hop_size); | |
| 401 | + if (s->eof && av_audio_fifo_size(s->fifo) <= 0) { | |
| 402 | + ff_outlink_set_status(outlink, AVERROR_EOF, s->pts); | |
| 403 | + return 0; | |
| 382 | 404 | } |
| 383 | 405 | |
| 384 | − av_frame_free(&in); | |
| 385 | − return ret < 0 ? ret : 0; | |
| 406 | + if (!s->eof) | |
| 407 | + FF_FILTER_FORWARD_WANTED(outlink, inlink); | |
| 408 | + | |
| 409 | + return FFERROR_NOT_READY; | |
| 386 | 410 | } |
| 387 | 411 | |
| 388 | 412 | static int query_formats(AVFilterContext *ctx) |
| @@ -450,7 +474,6 @@ static const AVFilterPad inputs[] = { · 61 unmodified lines | ||
| 450 | 474 | .name = "default", |
| 451 | 475 | .type = AVMEDIA_TYPE_AUDIO, |
| 452 | 476 | .config_props = config_input, |
| 453 | − .filter_frame = filter_frame, | |
| 454 | 477 | }, |
| 455 | 478 | { NULL } |
| 456 | 479 | }; |
| @@ -470,6 +493,7 @@ AVFilter ff_af_afftfilt = { · 13 unmodified lines | ||
| 470 | 493 | .priv_class = &afftfilt_class, |
| 471 | 494 | .inputs = inputs, |
| 472 | 495 | .outputs = outputs, |
| 496 | + .activate = activate, | |
| 473 | 497 | .query_formats = query_formats, |
| 474 | 498 | .uninit = uninit, |
| 475 | 499 | }; |