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"
47 #define INPUT_ON 1 /**< input is active */
48 #define INPUT_EOF 2 /**< input has reached EOF (may still be active) */
50 #define DURATION_LONGEST 0
51 #define DURATION_SHORTEST 1
52 #define DURATION_FIRST 2
55 typedef struct FrameInfo {
58 struct FrameInfo *next;
62 * Linked list used to store timestamps and frame sizes of all frames in the
63 * FIFO for the first input.
65 * This is needed to keep timestamps synchronized for the case where multiple
66 * input frames are pushed to the filter for processing before a frame is
67 * requested by the output link.
69 typedef struct FrameList {
76 static void frame_list_clear(FrameList *frame_list)
79 while (frame_list->list) {
80 FrameInfo *info = frame_list->list;
81 frame_list->list = info->next;
84 frame_list->nb_frames = 0;
85 frame_list->nb_samples = 0;
86 frame_list->end = NULL;
90 static int frame_list_next_frame_size(FrameList *frame_list)
92 if (!frame_list->list)
94 return frame_list->list->nb_samples;
97 static int64_t frame_list_next_pts(FrameList *frame_list)
99 if (!frame_list->list)
100 return AV_NOPTS_VALUE;
101 return frame_list->list->pts;
104 static void frame_list_remove_samples(FrameList *frame_list, int nb_samples)
106 if (nb_samples >= frame_list->nb_samples) {
107 frame_list_clear(frame_list);
109 int samples = nb_samples;
110 while (samples > 0) {
111 FrameInfo *info = frame_list->list;
113 if (info->nb_samples <= samples) {
114 samples -= info->nb_samples;
115 frame_list->list = info->next;
116 if (!frame_list->list)
117 frame_list->end = NULL;
118 frame_list->nb_frames--;
119 frame_list->nb_samples -= info->nb_samples;
122 info->nb_samples -= samples;
123 info->pts += samples;
124 frame_list->nb_samples -= samples;
131 static int frame_list_add_frame(FrameList *frame_list, int nb_samples, int64_t pts)
133 FrameInfo *info = av_malloc(sizeof(*info));
135 return AVERROR(ENOMEM);
136 info->nb_samples = nb_samples;
140 if (!frame_list->list) {
141 frame_list->list = info;
142 frame_list->end = info;
144 av_assert0(frame_list->end);
145 frame_list->end->next = info;
146 frame_list->end = info;
148 frame_list->nb_frames++;
149 frame_list->nb_samples += nb_samples;
155 typedef struct MixContext {
156 const AVClass *class; /**< class for AVOptions */
157 AVFloatDSPContext *fdsp;
159 int nb_inputs; /**< number of inputs */
160 int active_inputs; /**< number of input currently active */
161 int duration_mode; /**< mode for determining duration */
162 float dropout_transition; /**< transition time when an input drops out */
164 int nb_channels; /**< number of channels */
165 int sample_rate; /**< sample rate */
167 AVAudioFifo **fifos; /**< audio fifo for each input */
168 uint8_t *input_state; /**< current state of each input */
169 float *input_scale; /**< mixing scale factor for each input */
170 float scale_norm; /**< normalization factor for all inputs */
171 int64_t next_pts; /**< calculated pts for next output frame */
172 FrameList *frame_list; /**< list of frame info for the first input */
175 #define OFFSET(x) offsetof(MixContext, x)
176 #define A AV_OPT_FLAG_AUDIO_PARAM
177 #define F AV_OPT_FLAG_FILTERING_PARAM
178 static const AVOption amix_options[] = {
179 { "inputs", "Number of inputs.",
180 OFFSET(nb_inputs), AV_OPT_TYPE_INT, { .i64 = 2 }, 1, 1024, A|F },
181 { "duration", "How to determine the end-of-stream.",
182 OFFSET(duration_mode), AV_OPT_TYPE_INT, { .i64 = DURATION_LONGEST }, 0, 2, A|F, "duration" },
183 { "longest", "Duration of longest input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_LONGEST }, INT_MIN, INT_MAX, A|F, "duration" },
184 { "shortest", "Duration of shortest input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_SHORTEST }, INT_MIN, INT_MAX, A|F, "duration" },
185 { "first", "Duration of first input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_FIRST }, INT_MIN, INT_MAX, A|F, "duration" },
186 { "dropout_transition", "Transition time, in seconds, for volume "
187 "renormalization when an input stream ends.",
188 OFFSET(dropout_transition), AV_OPT_TYPE_FLOAT, { .dbl = 2.0 }, 0, INT_MAX, A|F },
192 AVFILTER_DEFINE_CLASS(amix);
195 * Update the scaling factors to apply to each input during mixing.
197 * This balances the full volume range between active inputs and handles
198 * volume transitions when EOF is encountered on an input but mixing continues
199 * with the remaining inputs.
201 static void calculate_scales(MixContext *s, int nb_samples)
205 if (s->scale_norm > s->active_inputs) {
206 s->scale_norm -= nb_samples / (s->dropout_transition * s->sample_rate);
207 s->scale_norm = FFMAX(s->scale_norm, s->active_inputs);
210 for (i = 0; i < s->nb_inputs; i++) {
211 if (s->input_state[i] & INPUT_ON)
212 s->input_scale[i] = 1.0f / s->scale_norm;
214 s->input_scale[i] = 0.0f;
218 static int config_output(AVFilterLink *outlink)
220 AVFilterContext *ctx = outlink->src;
221 MixContext *s = ctx->priv;
225 s->planar = av_sample_fmt_is_planar(outlink->format);
226 s->sample_rate = outlink->sample_rate;
227 outlink->time_base = (AVRational){ 1, outlink->sample_rate };
228 s->next_pts = AV_NOPTS_VALUE;
230 s->frame_list = av_mallocz(sizeof(*s->frame_list));
232 return AVERROR(ENOMEM);
234 s->fifos = av_mallocz_array(s->nb_inputs, sizeof(*s->fifos));
236 return AVERROR(ENOMEM);
238 s->nb_channels = outlink->channels;
239 for (i = 0; i < s->nb_inputs; i++) {
240 s->fifos[i] = av_audio_fifo_alloc(outlink->format, s->nb_channels, 1024);
242 return AVERROR(ENOMEM);
245 s->input_state = av_malloc(s->nb_inputs);
247 return AVERROR(ENOMEM);
248 memset(s->input_state, INPUT_ON, s->nb_inputs);
249 s->active_inputs = s->nb_inputs;
251 s->input_scale = av_mallocz_array(s->nb_inputs, sizeof(*s->input_scale));
253 return AVERROR(ENOMEM);
254 s->scale_norm = s->active_inputs;
255 calculate_scales(s, 0);
257 av_get_channel_layout_string(buf, sizeof(buf), -1, outlink->channel_layout);
259 av_log(ctx, AV_LOG_VERBOSE,
260 "inputs:%d fmt:%s srate:%d cl:%s\n", s->nb_inputs,
261 av_get_sample_fmt_name(outlink->format), outlink->sample_rate, buf);
266 static int calc_active_inputs(MixContext *s);
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, ret, i;
278 ret = calc_active_inputs(s);
282 if (s->input_state[0] & INPUT_ON) {
283 /* first input live: use the corresponding frame size */
284 nb_samples = frame_list_next_frame_size(s->frame_list);
285 for (i = 1; i < s->nb_inputs; i++) {
286 if (s->input_state[i] & INPUT_ON) {
287 ns = av_audio_fifo_size(s->fifos[i]);
288 if (ns < nb_samples) {
289 if (!(s->input_state[i] & INPUT_EOF))
290 /* unclosed input with not enough samples */
292 /* closed input to drain */
298 /* first input closed: use the available samples */
299 nb_samples = INT_MAX;
300 for (i = 1; i < s->nb_inputs; i++) {
301 if (s->input_state[i] & INPUT_ON) {
302 ns = av_audio_fifo_size(s->fifos[i]);
303 nb_samples = FFMIN(nb_samples, ns);
306 if (nb_samples == INT_MAX)
310 s->next_pts = frame_list_next_pts(s->frame_list);
311 frame_list_remove_samples(s->frame_list, nb_samples);
313 calculate_scales(s, nb_samples);
318 out_buf = ff_get_audio_buffer(outlink, nb_samples);
320 return AVERROR(ENOMEM);
322 in_buf = ff_get_audio_buffer(outlink, nb_samples);
324 av_frame_free(&out_buf);
325 return AVERROR(ENOMEM);
328 for (i = 0; i < s->nb_inputs; i++) {
329 if (s->input_state[i] & INPUT_ON) {
330 int planes, plane_size, p;
332 av_audio_fifo_read(s->fifos[i], (void **)in_buf->extended_data,
335 planes = s->planar ? s->nb_channels : 1;
336 plane_size = nb_samples * (s->planar ? 1 : s->nb_channels);
337 plane_size = FFALIGN(plane_size, 16);
339 if (out_buf->format == AV_SAMPLE_FMT_FLT ||
340 out_buf->format == AV_SAMPLE_FMT_FLTP) {
341 for (p = 0; p < planes; p++) {
342 s->fdsp->vector_fmac_scalar((float *)out_buf->extended_data[p],
343 (float *) in_buf->extended_data[p],
344 s->input_scale[i], plane_size);
347 for (p = 0; p < planes; p++) {
348 s->fdsp->vector_dmac_scalar((double *)out_buf->extended_data[p],
349 (double *) in_buf->extended_data[p],
350 s->input_scale[i], plane_size);
355 av_frame_free(&in_buf);
357 out_buf->pts = s->next_pts;
358 if (s->next_pts != AV_NOPTS_VALUE)
359 s->next_pts += nb_samples;
361 return ff_filter_frame(outlink, out_buf);
365 * Requests a frame, if needed, from each input link other than the first.
367 static int request_samples(AVFilterContext *ctx, int min_samples)
369 MixContext *s = ctx->priv;
372 av_assert0(s->nb_inputs > 1);
374 for (i = 1; i < s->nb_inputs; i++) {
376 if (!(s->input_state[i] & INPUT_ON))
378 if (av_audio_fifo_size(s->fifos[i]) >= min_samples)
380 ret = ff_request_frame(ctx->inputs[i]);
381 if (ret == AVERROR_EOF) {
382 s->input_state[i] |= INPUT_EOF;
383 if (av_audio_fifo_size(s->fifos[i]) == 0) {
384 s->input_state[i] = 0;
390 return output_frame(ctx->outputs[0]);
394 * Calculates the number of active inputs and determines EOF based on the
397 * @return 0 if mixing should continue, or AVERROR_EOF if mixing should stop.
399 static int calc_active_inputs(MixContext *s)
402 int active_inputs = 0;
403 for (i = 0; i < s->nb_inputs; i++)
404 active_inputs += !!(s->input_state[i] & INPUT_ON);
405 s->active_inputs = active_inputs;
407 if (!active_inputs ||
408 (s->duration_mode == DURATION_FIRST && !(s->input_state[0] & INPUT_ON)) ||
409 (s->duration_mode == DURATION_SHORTEST && active_inputs != s->nb_inputs))
414 static int request_frame(AVFilterLink *outlink)
416 AVFilterContext *ctx = outlink->src;
417 MixContext *s = ctx->priv;
421 ret = calc_active_inputs(s);
425 if (!(s->input_state[0] & INPUT_ON))
426 return request_samples(ctx, 1);
428 if (s->frame_list->nb_frames == 0) {
429 ret = ff_request_frame(ctx->inputs[0]);
430 if (ret == AVERROR_EOF) {
431 s->input_state[0] = 0;
432 if (s->nb_inputs == 1)
434 return output_frame(ctx->outputs[0]);
438 av_assert0(s->frame_list->nb_frames > 0);
440 wanted_samples = frame_list_next_frame_size(s->frame_list);
442 return request_samples(ctx, wanted_samples);
445 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
447 AVFilterContext *ctx = inlink->dst;
448 MixContext *s = ctx->priv;
449 AVFilterLink *outlink = ctx->outputs[0];
452 for (i = 0; i < ctx->nb_inputs; i++)
453 if (ctx->inputs[i] == inlink)
455 if (i >= ctx->nb_inputs) {
456 av_log(ctx, AV_LOG_ERROR, "unknown input link\n");
457 ret = AVERROR(EINVAL);
462 int64_t pts = av_rescale_q(buf->pts, inlink->time_base,
464 ret = frame_list_add_frame(s->frame_list, buf->nb_samples, pts);
469 ret = av_audio_fifo_write(s->fifos[i], (void **)buf->extended_data,
473 return output_frame(outlink);
481 static av_cold int init(AVFilterContext *ctx)
483 MixContext *s = ctx->priv;
486 for (i = 0; i < s->nb_inputs; i++) {
488 AVFilterPad pad = { 0 };
490 snprintf(name, sizeof(name), "input%d", i);
491 pad.type = AVMEDIA_TYPE_AUDIO;
492 pad.name = av_strdup(name);
494 return AVERROR(ENOMEM);
495 pad.filter_frame = filter_frame;
497 ff_insert_inpad(ctx, i, &pad);
500 s->fdsp = avpriv_float_dsp_alloc(0);
502 return AVERROR(ENOMEM);
507 static av_cold void uninit(AVFilterContext *ctx)
510 MixContext *s = ctx->priv;
513 for (i = 0; i < s->nb_inputs; i++)
514 av_audio_fifo_free(s->fifos[i]);
517 frame_list_clear(s->frame_list);
518 av_freep(&s->frame_list);
519 av_freep(&s->input_state);
520 av_freep(&s->input_scale);
523 for (i = 0; i < ctx->nb_inputs; i++)
524 av_freep(&ctx->input_pads[i].name);
527 static int query_formats(AVFilterContext *ctx)
529 AVFilterFormats *formats = NULL;
530 AVFilterChannelLayouts *layouts;
533 layouts = ff_all_channel_counts();
535 ret = AVERROR(ENOMEM);
539 if ((ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT )) < 0 ||
540 (ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLTP)) < 0 ||
541 (ret = ff_add_format(&formats, AV_SAMPLE_FMT_DBL )) < 0 ||
542 (ret = ff_add_format(&formats, AV_SAMPLE_FMT_DBLP)) < 0 ||
543 (ret = ff_set_common_formats (ctx, formats)) < 0 ||
544 (ret = ff_set_common_channel_layouts(ctx, layouts)) < 0 ||
545 (ret = ff_set_common_samplerates(ctx, ff_all_samplerates())) < 0)
550 av_freep(&layouts->channel_layouts);
555 static const AVFilterPad avfilter_af_amix_outputs[] = {
558 .type = AVMEDIA_TYPE_AUDIO,
559 .config_props = config_output,
560 .request_frame = request_frame
565 AVFilter ff_af_amix = {
567 .description = NULL_IF_CONFIG_SMALL("Audio mixing."),
568 .priv_size = sizeof(MixContext),
569 .priv_class = &amix_class,
572 .query_formats = query_formats,
574 .outputs = avfilter_af_amix_outputs,
575 .flags = AVFILTER_FLAG_DYNAMIC_INPUTS,