3 * Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
5 * This file is part of FFmpeg.
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.
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.
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
26 * Mixes audio from multiple sources into a single output. The channel layout,
27 * sample rate, and sample format will be the same for all inputs and the
31 #include "libavutil/attributes.h"
32 #include "libavutil/audio_fifo.h"
33 #include "libavutil/avassert.h"
34 #include "libavutil/avstring.h"
35 #include "libavutil/channel_layout.h"
36 #include "libavutil/common.h"
37 #include "libavutil/float_dsp.h"
38 #include "libavutil/mathematics.h"
39 #include "libavutil/opt.h"
40 #include "libavutil/samplefmt.h"
48 #define INPUT_ON 1 /**< input is active */
49 #define INPUT_EOF 2 /**< input has reached EOF (may still be active) */
51 #define DURATION_LONGEST 0
52 #define DURATION_SHORTEST 1
53 #define DURATION_FIRST 2
56 typedef struct FrameInfo {
59 struct FrameInfo *next;
63 * Linked list used to store timestamps and frame sizes of all frames in the
64 * FIFO for the first input.
66 * This is needed to keep timestamps synchronized for the case where multiple
67 * input frames are pushed to the filter for processing before a frame is
68 * requested by the output link.
70 typedef struct FrameList {
77 static void frame_list_clear(FrameList *frame_list)
80 while (frame_list->list) {
81 FrameInfo *info = frame_list->list;
82 frame_list->list = info->next;
85 frame_list->nb_frames = 0;
86 frame_list->nb_samples = 0;
87 frame_list->end = NULL;
91 static int frame_list_next_frame_size(FrameList *frame_list)
93 if (!frame_list->list)
95 return frame_list->list->nb_samples;
98 static int64_t frame_list_next_pts(FrameList *frame_list)
100 if (!frame_list->list)
101 return AV_NOPTS_VALUE;
102 return frame_list->list->pts;
105 static void frame_list_remove_samples(FrameList *frame_list, int nb_samples)
107 if (nb_samples >= frame_list->nb_samples) {
108 frame_list_clear(frame_list);
110 int samples = nb_samples;
111 while (samples > 0) {
112 FrameInfo *info = frame_list->list;
114 if (info->nb_samples <= samples) {
115 samples -= info->nb_samples;
116 frame_list->list = info->next;
117 if (!frame_list->list)
118 frame_list->end = NULL;
119 frame_list->nb_frames--;
120 frame_list->nb_samples -= info->nb_samples;
123 info->nb_samples -= samples;
124 info->pts += samples;
125 frame_list->nb_samples -= samples;
132 static int frame_list_add_frame(FrameList *frame_list, int nb_samples, int64_t pts)
134 FrameInfo *info = av_malloc(sizeof(*info));
136 return AVERROR(ENOMEM);
137 info->nb_samples = nb_samples;
141 if (!frame_list->list) {
142 frame_list->list = info;
143 frame_list->end = info;
145 av_assert0(frame_list->end);
146 frame_list->end->next = info;
147 frame_list->end = info;
149 frame_list->nb_frames++;
150 frame_list->nb_samples += nb_samples;
155 /* FIXME: use directly links fifo */
157 typedef struct MixContext {
158 const AVClass *class; /**< class for AVOptions */
159 AVFloatDSPContext *fdsp;
161 int nb_inputs; /**< number of inputs */
162 int active_inputs; /**< number of input currently active */
163 int duration_mode; /**< mode for determining duration */
164 float dropout_transition; /**< transition time when an input drops out */
166 int nb_channels; /**< number of channels */
167 int sample_rate; /**< sample rate */
169 AVAudioFifo **fifos; /**< audio fifo for each input */
170 uint8_t *input_state; /**< current state of each input */
171 float *input_scale; /**< mixing scale factor for each input */
172 float scale_norm; /**< normalization factor for all inputs */
173 int64_t next_pts; /**< calculated pts for next output frame */
174 FrameList *frame_list; /**< list of frame info for the first input */
177 #define OFFSET(x) offsetof(MixContext, x)
178 #define A AV_OPT_FLAG_AUDIO_PARAM
179 #define F AV_OPT_FLAG_FILTERING_PARAM
180 static const AVOption amix_options[] = {
181 { "inputs", "Number of inputs.",
182 OFFSET(nb_inputs), AV_OPT_TYPE_INT, { .i64 = 2 }, 1, 1024, A|F },
183 { "duration", "How to determine the end-of-stream.",
184 OFFSET(duration_mode), AV_OPT_TYPE_INT, { .i64 = DURATION_LONGEST }, 0, 2, A|F, "duration" },
185 { "longest", "Duration of longest input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_LONGEST }, 0, 0, A|F, "duration" },
186 { "shortest", "Duration of shortest input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_SHORTEST }, 0, 0, A|F, "duration" },
187 { "first", "Duration of first input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_FIRST }, 0, 0, A|F, "duration" },
188 { "dropout_transition", "Transition time, in seconds, for volume "
189 "renormalization when an input stream ends.",
190 OFFSET(dropout_transition), AV_OPT_TYPE_FLOAT, { .dbl = 2.0 }, 0, INT_MAX, A|F },
194 AVFILTER_DEFINE_CLASS(amix);
197 * Update the scaling factors to apply to each input during mixing.
199 * This balances the full volume range between active inputs and handles
200 * volume transitions when EOF is encountered on an input but mixing continues
201 * with the remaining inputs.
203 static void calculate_scales(MixContext *s, int nb_samples)
207 if (s->scale_norm > s->active_inputs) {
208 s->scale_norm -= nb_samples / (s->dropout_transition * s->sample_rate);
209 s->scale_norm = FFMAX(s->scale_norm, s->active_inputs);
212 for (i = 0; i < s->nb_inputs; i++) {
213 if (s->input_state[i] & INPUT_ON)
214 s->input_scale[i] = 1.0f / s->scale_norm;
216 s->input_scale[i] = 0.0f;
220 static int config_output(AVFilterLink *outlink)
222 AVFilterContext *ctx = outlink->src;
223 MixContext *s = ctx->priv;
227 s->planar = av_sample_fmt_is_planar(outlink->format);
228 s->sample_rate = outlink->sample_rate;
229 outlink->time_base = (AVRational){ 1, outlink->sample_rate };
230 s->next_pts = AV_NOPTS_VALUE;
232 s->frame_list = av_mallocz(sizeof(*s->frame_list));
234 return AVERROR(ENOMEM);
236 s->fifos = av_mallocz_array(s->nb_inputs, sizeof(*s->fifos));
238 return AVERROR(ENOMEM);
240 s->nb_channels = outlink->channels;
241 for (i = 0; i < s->nb_inputs; i++) {
242 s->fifos[i] = av_audio_fifo_alloc(outlink->format, s->nb_channels, 1024);
244 return AVERROR(ENOMEM);
247 s->input_state = av_malloc(s->nb_inputs);
249 return AVERROR(ENOMEM);
250 memset(s->input_state, INPUT_ON, s->nb_inputs);
251 s->active_inputs = s->nb_inputs;
253 s->input_scale = av_mallocz_array(s->nb_inputs, sizeof(*s->input_scale));
255 return AVERROR(ENOMEM);
256 s->scale_norm = s->active_inputs;
257 calculate_scales(s, 0);
259 av_get_channel_layout_string(buf, sizeof(buf), -1, outlink->channel_layout);
261 av_log(ctx, AV_LOG_VERBOSE,
262 "inputs:%d fmt:%s srate:%d cl:%s\n", s->nb_inputs,
263 av_get_sample_fmt_name(outlink->format), outlink->sample_rate, buf);
269 * Read samples from the input FIFOs, mix, and write to the output link.
271 static int output_frame(AVFilterLink *outlink)
273 AVFilterContext *ctx = outlink->src;
274 MixContext *s = ctx->priv;
275 AVFrame *out_buf, *in_buf;
276 int nb_samples, ns, i;
278 if (s->input_state[0] & INPUT_ON) {
279 /* first input live: use the corresponding frame size */
280 nb_samples = frame_list_next_frame_size(s->frame_list);
281 for (i = 1; i < s->nb_inputs; i++) {
282 if (s->input_state[i] & INPUT_ON) {
283 ns = av_audio_fifo_size(s->fifos[i]);
284 if (ns < nb_samples) {
285 if (!(s->input_state[i] & INPUT_EOF))
286 /* unclosed input with not enough samples */
288 /* closed input to drain */
294 /* first input closed: use the available samples */
295 nb_samples = INT_MAX;
296 for (i = 1; i < s->nb_inputs; i++) {
297 if (s->input_state[i] & INPUT_ON) {
298 ns = av_audio_fifo_size(s->fifos[i]);
299 nb_samples = FFMIN(nb_samples, ns);
302 if (nb_samples == INT_MAX) {
303 ff_outlink_set_status(outlink, AVERROR_EOF, s->next_pts);
308 s->next_pts = frame_list_next_pts(s->frame_list);
309 frame_list_remove_samples(s->frame_list, nb_samples);
311 calculate_scales(s, nb_samples);
316 out_buf = ff_get_audio_buffer(outlink, nb_samples);
318 return AVERROR(ENOMEM);
320 in_buf = ff_get_audio_buffer(outlink, nb_samples);
322 av_frame_free(&out_buf);
323 return AVERROR(ENOMEM);
326 for (i = 0; i < s->nb_inputs; i++) {
327 if (s->input_state[i] & INPUT_ON) {
328 int planes, plane_size, p;
330 av_audio_fifo_read(s->fifos[i], (void **)in_buf->extended_data,
333 planes = s->planar ? s->nb_channels : 1;
334 plane_size = nb_samples * (s->planar ? 1 : s->nb_channels);
335 plane_size = FFALIGN(plane_size, 16);
337 if (out_buf->format == AV_SAMPLE_FMT_FLT ||
338 out_buf->format == AV_SAMPLE_FMT_FLTP) {
339 for (p = 0; p < planes; p++) {
340 s->fdsp->vector_fmac_scalar((float *)out_buf->extended_data[p],
341 (float *) in_buf->extended_data[p],
342 s->input_scale[i], plane_size);
345 for (p = 0; p < planes; p++) {
346 s->fdsp->vector_dmac_scalar((double *)out_buf->extended_data[p],
347 (double *) in_buf->extended_data[p],
348 s->input_scale[i], plane_size);
353 av_frame_free(&in_buf);
355 out_buf->pts = s->next_pts;
356 if (s->next_pts != AV_NOPTS_VALUE)
357 s->next_pts += nb_samples;
359 return ff_filter_frame(outlink, out_buf);
363 * Requests a frame, if needed, from each input link other than the first.
365 static int request_samples(AVFilterContext *ctx, int min_samples)
367 MixContext *s = ctx->priv;
370 av_assert0(s->nb_inputs > 1);
372 for (i = 1; i < s->nb_inputs; i++) {
373 if (!(s->input_state[i] & INPUT_ON) ||
374 (s->input_state[i] & INPUT_EOF))
376 if (av_audio_fifo_size(s->fifos[i]) >= min_samples)
378 ff_inlink_request_frame(ctx->inputs[i]);
380 return output_frame(ctx->outputs[0]);
384 * Calculates the number of active inputs and determines EOF based on the
387 * @return 0 if mixing should continue, or AVERROR_EOF if mixing should stop.
389 static int calc_active_inputs(MixContext *s)
392 int active_inputs = 0;
393 for (i = 0; i < s->nb_inputs; i++)
394 active_inputs += !!(s->input_state[i] & INPUT_ON);
395 s->active_inputs = active_inputs;
397 if (!active_inputs ||
398 (s->duration_mode == DURATION_FIRST && !(s->input_state[0] & INPUT_ON)) ||
399 (s->duration_mode == DURATION_SHORTEST && active_inputs != s->nb_inputs))
404 static int activate(AVFilterContext *ctx)
406 AVFilterLink *outlink = ctx->outputs[0];
407 MixContext *s = ctx->priv;
411 for (i = 0; i < s->nb_inputs; i++) {
412 AVFilterLink *inlink = ctx->inputs[i];
414 if ((ret = ff_inlink_consume_frame(ctx->inputs[i], &buf)) > 0) {
416 int64_t pts = av_rescale_q(buf->pts, inlink->time_base,
418 ret = frame_list_add_frame(s->frame_list, buf->nb_samples, pts);
425 ret = av_audio_fifo_write(s->fifos[i], (void **)buf->extended_data,
434 ret = output_frame(outlink);
440 for (i = 0; i < s->nb_inputs; i++) {
444 if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) {
445 if (status == AVERROR_EOF) {
447 s->input_state[i] = 0;
448 if (s->nb_inputs == 1) {
449 ff_outlink_set_status(outlink, status, pts);
453 s->input_state[i] |= INPUT_EOF;
454 if (av_audio_fifo_size(s->fifos[i]) == 0) {
455 s->input_state[i] = 0;
462 if (calc_active_inputs(s)) {
463 ff_outlink_set_status(outlink, AVERROR_EOF, s->next_pts);
467 if (ff_outlink_frame_wanted(outlink)) {
470 if (!(s->input_state[0] & INPUT_ON))
471 return request_samples(ctx, 1);
473 if (s->frame_list->nb_frames == 0) {
474 ff_inlink_request_frame(ctx->inputs[0]);
477 av_assert0(s->frame_list->nb_frames > 0);
479 wanted_samples = frame_list_next_frame_size(s->frame_list);
481 return request_samples(ctx, wanted_samples);
487 static av_cold int init(AVFilterContext *ctx)
489 MixContext *s = ctx->priv;
492 for (i = 0; i < s->nb_inputs; i++) {
494 AVFilterPad pad = { 0 };
496 snprintf(name, sizeof(name), "input%d", i);
497 pad.type = AVMEDIA_TYPE_AUDIO;
498 pad.name = av_strdup(name);
500 return AVERROR(ENOMEM);
502 if ((ret = ff_insert_inpad(ctx, i, &pad)) < 0) {
508 s->fdsp = avpriv_float_dsp_alloc(0);
510 return AVERROR(ENOMEM);
515 static av_cold void uninit(AVFilterContext *ctx)
518 MixContext *s = ctx->priv;
521 for (i = 0; i < s->nb_inputs; i++)
522 av_audio_fifo_free(s->fifos[i]);
525 frame_list_clear(s->frame_list);
526 av_freep(&s->frame_list);
527 av_freep(&s->input_state);
528 av_freep(&s->input_scale);
531 for (i = 0; i < ctx->nb_inputs; i++)
532 av_freep(&ctx->input_pads[i].name);
535 static int query_formats(AVFilterContext *ctx)
537 AVFilterFormats *formats = NULL;
538 AVFilterChannelLayouts *layouts;
541 layouts = ff_all_channel_counts();
543 ret = AVERROR(ENOMEM);
547 if ((ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT )) < 0 ||
548 (ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLTP)) < 0 ||
549 (ret = ff_add_format(&formats, AV_SAMPLE_FMT_DBL )) < 0 ||
550 (ret = ff_add_format(&formats, AV_SAMPLE_FMT_DBLP)) < 0 ||
551 (ret = ff_set_common_formats (ctx, formats)) < 0 ||
552 (ret = ff_set_common_channel_layouts(ctx, layouts)) < 0 ||
553 (ret = ff_set_common_samplerates(ctx, ff_all_samplerates())) < 0)
558 av_freep(&layouts->channel_layouts);
563 static const AVFilterPad avfilter_af_amix_outputs[] = {
566 .type = AVMEDIA_TYPE_AUDIO,
567 .config_props = config_output,
572 AVFilter ff_af_amix = {
574 .description = NULL_IF_CONFIG_SMALL("Audio mixing."),
575 .priv_size = sizeof(MixContext),
576 .priv_class = &amix_class,
579 .activate = activate,
580 .query_formats = query_formats,
582 .outputs = avfilter_af_amix_outputs,
583 .flags = AVFILTER_FLAG_DYNAMIC_INPUTS,