GCC Code Coverage Report
Directory: ../../../ffmpeg/ Exec Total Coverage
File: src/libavfilter/af_axcorrelate.c Lines: 0 180 0.0 %
Date: 2020-10-23 17:01:47 Branches: 0 101 0.0 %

Line Branch Exec Source
1
/*
2
 * Copyright (c) 2019 Paul B Mahol
3
 *
4
 * This file is part of FFmpeg.
5
 *
6
 * FFmpeg is free software; you can redistribute it and/or
7
 * modify it under the terms of the GNU Lesser General Public
8
 * License as published by the Free Software Foundation; either
9
 * version 2.1 of the License, or (at your option) any later version.
10
 *
11
 * FFmpeg is distributed in the hope that it will be useful,
12
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14
 * Lesser General Public License for more details.
15
 *
16
 * You should have received a copy of the GNU Lesser General Public
17
 * License along with FFmpeg; if not, write to the Free Software
18
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
 */
20
21
#include "libavutil/avassert.h"
22
#include "libavutil/audio_fifo.h"
23
#include "libavutil/channel_layout.h"
24
#include "libavutil/common.h"
25
#include "libavutil/opt.h"
26
27
#include "audio.h"
28
#include "avfilter.h"
29
#include "formats.h"
30
#include "filters.h"
31
#include "internal.h"
32
33
typedef struct AudioXCorrelateContext {
34
    const AVClass *class;
35
36
    int size;
37
    int algo;
38
    int64_t pts;
39
40
    AVAudioFifo *fifo[2];
41
    AVFrame *cache[2];
42
    AVFrame *mean_sum[2];
43
    AVFrame *num_sum;
44
    AVFrame *den_sum[2];
45
    int used;
46
47
    int (*xcorrelate)(AVFilterContext *ctx, AVFrame *out);
48
} AudioXCorrelateContext;
49
50
static int query_formats(AVFilterContext *ctx)
51
{
52
    AVFilterFormats *formats;
53
    AVFilterChannelLayouts *layouts;
54
    static const enum AVSampleFormat sample_fmts[] = {
55
        AV_SAMPLE_FMT_FLTP,
56
        AV_SAMPLE_FMT_NONE
57
    };
58
    int ret;
59
60
    layouts = ff_all_channel_counts();
61
    if (!layouts)
62
        return AVERROR(ENOMEM);
63
    ret = ff_set_common_channel_layouts(ctx, layouts);
64
    if (ret < 0)
65
        return ret;
66
67
    formats = ff_make_format_list(sample_fmts);
68
    if (!formats)
69
        return AVERROR(ENOMEM);
70
    ret = ff_set_common_formats(ctx, formats);
71
    if (ret < 0)
72
        return ret;
73
74
    formats = ff_all_samplerates();
75
    if (!formats)
76
        return AVERROR(ENOMEM);
77
    return ff_set_common_samplerates(ctx, formats);
78
}
79
80
static float mean_sum(const float *in, int size)
81
{
82
    float mean_sum = 0.f;
83
84
    for (int i = 0; i < size; i++)
85
        mean_sum += in[i];
86
87
    return mean_sum;
88
}
89
90
static float square_sum(const float *x, const float *y, int size)
91
{
92
    float square_sum = 0.f;
93
94
    for (int i = 0; i < size; i++)
95
        square_sum += x[i] * y[i];
96
97
    return square_sum;
98
}
99
100
static float xcorrelate(const float *x, const float *y, float sumx, float sumy, int size)
101
{
102
    const float xm = sumx / size, ym = sumy / size;
103
    float num = 0.f, den, den0 = 0.f, den1 = 0.f;
104
105
    for (int i = 0; i < size; i++) {
106
        float xd = x[i] - xm;
107
        float yd = y[i] - ym;
108
109
        num += xd * yd;
110
        den0 += xd * xd;
111
        den1 += yd * yd;
112
    }
113
114
    num /= size;
115
    den  = sqrtf((den0 * den1) / (size * size));
116
117
    return den <= 1e-6f ? 0.f : num / den;
118
}
119
120
static int xcorrelate_slow(AVFilterContext *ctx, AVFrame *out)
121
{
122
    AudioXCorrelateContext *s = ctx->priv;
123
    const int size = s->size;
124
    int used;
125
126
    for (int ch = 0; ch < out->channels; ch++) {
127
        const float *x = (const float *)s->cache[0]->extended_data[ch];
128
        const float *y = (const float *)s->cache[1]->extended_data[ch];
129
        float *sumx = (float *)s->mean_sum[0]->extended_data[ch];
130
        float *sumy = (float *)s->mean_sum[1]->extended_data[ch];
131
        float *dst = (float *)out->extended_data[ch];
132
133
        used = s->used;
134
        if (!used) {
135
            sumx[0] = mean_sum(x, size);
136
            sumy[0] = mean_sum(y, size);
137
            used = 1;
138
        }
139
140
        for (int n = 0; n < out->nb_samples; n++) {
141
            dst[n] = xcorrelate(x + n, y + n, sumx[0], sumy[0], size);
142
143
            sumx[0] -= x[n];
144
            sumx[0] += x[n + size];
145
            sumy[0] -= y[n];
146
            sumy[0] += y[n + size];
147
        }
148
    }
149
150
    return used;
151
}
152
153
static int xcorrelate_fast(AVFilterContext *ctx, AVFrame *out)
154
{
155
    AudioXCorrelateContext *s = ctx->priv;
156
    const int size = s->size;
157
    int used;
158
159
    for (int ch = 0; ch < out->channels; ch++) {
160
        const float *x = (const float *)s->cache[0]->extended_data[ch];
161
        const float *y = (const float *)s->cache[1]->extended_data[ch];
162
        float *num_sum = (float *)s->num_sum->extended_data[ch];
163
        float *den_sumx = (float *)s->den_sum[0]->extended_data[ch];
164
        float *den_sumy = (float *)s->den_sum[1]->extended_data[ch];
165
        float *dst = (float *)out->extended_data[ch];
166
167
        used = s->used;
168
        if (!used) {
169
            num_sum[0]  = square_sum(x, y, size);
170
            den_sumx[0] = square_sum(x, x, size);
171
            den_sumy[0] = square_sum(y, y, size);
172
            used = 1;
173
        }
174
175
        for (int n = 0; n < out->nb_samples; n++) {
176
            float num, den;
177
178
            num = num_sum[0] / size;
179
            den = sqrtf((den_sumx[0] * den_sumy[0]) / (size * size));
180
181
            dst[n] = den <= 1e-6f ? 0.f : num / den;
182
183
            num_sum[0]  -= x[n] * y[n];
184
            num_sum[0]  += x[n + size] * y[n + size];
185
            den_sumx[0] -= x[n] * x[n];
186
            den_sumx[0]  = FFMAX(den_sumx[0], 0.f);
187
            den_sumx[0] += x[n + size] * x[n + size];
188
            den_sumy[0] -= y[n] * y[n];
189
            den_sumy[0]  = FFMAX(den_sumy[0], 0.f);
190
            den_sumy[0] += y[n + size] * y[n + size];
191
        }
192
    }
193
194
    return used;
195
}
196
197
static int activate(AVFilterContext *ctx)
198
{
199
    AudioXCorrelateContext *s = ctx->priv;
200
    AVFrame *frame = NULL;
201
    int ret, status;
202
    int available;
203
    int64_t pts;
204
205
    FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
206
207
    for (int i = 0; i < 2; i++) {
208
        ret = ff_inlink_consume_frame(ctx->inputs[i], &frame);
209
        if (ret > 0) {
210
            if (s->pts == AV_NOPTS_VALUE)
211
                s->pts = frame->pts;
212
            ret = av_audio_fifo_write(s->fifo[i], (void **)frame->extended_data,
213
                                      frame->nb_samples);
214
            av_frame_free(&frame);
215
            if (ret < 0)
216
                return ret;
217
        }
218
    }
219
220
    available = FFMIN(av_audio_fifo_size(s->fifo[0]), av_audio_fifo_size(s->fifo[1]));
221
    if (available > s->size) {
222
        const int out_samples = available - s->size;
223
        AVFrame *out;
224
225
        if (!s->cache[0] || s->cache[0]->nb_samples < available) {
226
            av_frame_free(&s->cache[0]);
227
            s->cache[0] = ff_get_audio_buffer(ctx->outputs[0], available);
228
            if (!s->cache[0])
229
                return AVERROR(ENOMEM);
230
        }
231
232
        if (!s->cache[1] || s->cache[1]->nb_samples < available) {
233
            av_frame_free(&s->cache[1]);
234
            s->cache[1] = ff_get_audio_buffer(ctx->outputs[0], available);
235
            if (!s->cache[1])
236
                return AVERROR(ENOMEM);
237
        }
238
239
        ret = av_audio_fifo_peek(s->fifo[0], (void **)s->cache[0]->extended_data, available);
240
        if (ret < 0)
241
            return ret;
242
243
        ret = av_audio_fifo_peek(s->fifo[1], (void **)s->cache[1]->extended_data, available);
244
        if (ret < 0)
245
            return ret;
246
247
        out = ff_get_audio_buffer(ctx->outputs[0], out_samples);
248
        if (!out)
249
            return AVERROR(ENOMEM);
250
251
        s->used = s->xcorrelate(ctx, out);
252
253
        out->pts = s->pts;
254
        s->pts += out_samples;
255
256
        av_audio_fifo_drain(s->fifo[0], out_samples);
257
        av_audio_fifo_drain(s->fifo[1], out_samples);
258
259
        return ff_filter_frame(ctx->outputs[0], out);
260
    }
261
262
    if (av_audio_fifo_size(s->fifo[0]) > s->size &&
263
        av_audio_fifo_size(s->fifo[1]) > s->size) {
264
        ff_filter_set_ready(ctx, 10);
265
        return 0;
266
    }
267
268
    for (int i = 0; i < 2; i++) {
269
        if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) {
270
            ff_outlink_set_status(ctx->outputs[0], status, pts);
271
            return 0;
272
        }
273
    }
274
275
    if (ff_outlink_frame_wanted(ctx->outputs[0])) {
276
        for (int i = 0; i < 2; i++) {
277
            if (av_audio_fifo_size(s->fifo[i]) > s->size)
278
                continue;
279
            ff_inlink_request_frame(ctx->inputs[i]);
280
            return 0;
281
        }
282
    }
283
284
    return FFERROR_NOT_READY;
285
}
286
287
static int config_output(AVFilterLink *outlink)
288
{
289
    AVFilterContext *ctx = outlink->src;
290
    AVFilterLink *inlink = ctx->inputs[0];
291
    AudioXCorrelateContext *s = ctx->priv;
292
293
    s->pts = AV_NOPTS_VALUE;
294
295
    outlink->format = inlink->format;
296
    outlink->channels = inlink->channels;
297
    s->fifo[0] = av_audio_fifo_alloc(outlink->format, outlink->channels, s->size);
298
    s->fifo[1] = av_audio_fifo_alloc(outlink->format, outlink->channels, s->size);
299
    if (!s->fifo[0] || !s->fifo[1])
300
        return AVERROR(ENOMEM);
301
302
    s->mean_sum[0] = ff_get_audio_buffer(outlink, 1);
303
    s->mean_sum[1] = ff_get_audio_buffer(outlink, 1);
304
    s->num_sum = ff_get_audio_buffer(outlink, 1);
305
    s->den_sum[0] = ff_get_audio_buffer(outlink, 1);
306
    s->den_sum[1] = ff_get_audio_buffer(outlink, 1);
307
    if (!s->mean_sum[0] || !s->mean_sum[1] || !s->num_sum ||
308
        !s->den_sum[0] || !s->den_sum[1])
309
        return AVERROR(ENOMEM);
310
311
    switch (s->algo) {
312
    case 0: s->xcorrelate = xcorrelate_slow; break;
313
    case 1: s->xcorrelate = xcorrelate_fast; break;
314
    }
315
316
    return 0;
317
}
318
319
static av_cold void uninit(AVFilterContext *ctx)
320
{
321
    AudioXCorrelateContext *s = ctx->priv;
322
323
    av_audio_fifo_free(s->fifo[0]);
324
    av_audio_fifo_free(s->fifo[1]);
325
    av_frame_free(&s->cache[0]);
326
    av_frame_free(&s->cache[1]);
327
    av_frame_free(&s->mean_sum[0]);
328
    av_frame_free(&s->mean_sum[1]);
329
    av_frame_free(&s->num_sum);
330
    av_frame_free(&s->den_sum[0]);
331
    av_frame_free(&s->den_sum[1]);
332
}
333
334
static const AVFilterPad inputs[] = {
335
    {
336
        .name = "axcorrelate0",
337
        .type = AVMEDIA_TYPE_AUDIO,
338
    },
339
    {
340
        .name = "axcorrelate1",
341
        .type = AVMEDIA_TYPE_AUDIO,
342
    },
343
    { NULL }
344
};
345
346
static const AVFilterPad outputs[] = {
347
    {
348
        .name         = "default",
349
        .type         = AVMEDIA_TYPE_AUDIO,
350
        .config_props = config_output,
351
    },
352
    { NULL }
353
};
354
355
#define AF AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
356
#define OFFSET(x) offsetof(AudioXCorrelateContext, x)
357
358
static const AVOption axcorrelate_options[] = {
359
    { "size", "set segment size", OFFSET(size), AV_OPT_TYPE_INT,   {.i64=256}, 2, 131072, AF },
360
    { "algo", "set alghorithm",   OFFSET(algo), AV_OPT_TYPE_INT,   {.i64=0},   0,      1, AF, "algo" },
361
    { "slow", "slow algorithm",   0,            AV_OPT_TYPE_CONST, {.i64=0},   0,      0, AF, "algo" },
362
    { "fast", "fast algorithm",   0,            AV_OPT_TYPE_CONST, {.i64=1},   0,      0, AF, "algo" },
363
    { NULL }
364
};
365
366
AVFILTER_DEFINE_CLASS(axcorrelate);
367
368
AVFilter ff_af_axcorrelate = {
369
    .name           = "axcorrelate",
370
    .description    = NULL_IF_CONFIG_SMALL("Cross-correlate two audio streams."),
371
    .priv_size      = sizeof(AudioXCorrelateContext),
372
    .priv_class     = &axcorrelate_class,
373
    .query_formats  = query_formats,
374
    .activate       = activate,
375
    .uninit         = uninit,
376
    .inputs         = inputs,
377
    .outputs        = outputs,
378
};