avfilter/af_anlms: add double sample format support

Paul B Mahol committed 2 years ago

Commit 42e45ea1 parent ea6817d3 files · +194 −80Browse files

Unified · Split

3 files

doc/filters.texi +14 −0
libavfilter/af_anlms.c +39 −80
libavfilter/anlms_template.c +141 −0

@@ -2687,6 +2687,20 @@ Pass error signal estimated samples. · 2686 unmodified lines
2687		2687
2688	Default value is @var{o}.	2688	Default value is @var{o}.
2689	@end table	2689	@end table
		2690
		2691	@item precision
		2692	Set which precision to use when processing samples.
		2693
		2694	@table @option
		2695	@item auto
		2696	Auto pick internal sample format depending on other filters.
		2697
		2698	@item float
		2699	Always use single-floating point precision sample format.
		2700
		2701	@item double
		2702	Always use double-floating point precision sample format.
		2703	@end table
2690	@end table	2704	@end table
2691		2705
2692	@subsection Examples	2706	@subsection Examples

libavfilter/af_anlms.c+39−80

@@ -26,6 +26,7 @@ · 25 unmodified lines
26	#include "audio.h"	26	#include "audio.h"
27	#include "avfilter.h"	27	#include "avfilter.h"
28	#include "filters.h"	28	#include "filters.h"
		29	#include "formats.h"
29	#include "internal.h"	30	#include "internal.h"
30		31
31	enum OutModes {	32	enum OutModes {
@@ -45,6 +46,7 @@ typedef struct AudioNLMSContext { · 13 unmodified lines
45	float eps;	46	float eps;
46	float leakage;	47	float leakage;
47	int output_mode;	48	int output_mode;
		49	int precision;
48		50
49	int kernel_size;	51	int kernel_size;
50	AVFrame *offset;	52	AVFrame *offset;
@@ -56,6 +58,8 @@ typedef struct AudioNLMSContext { · 5 unmodified lines
56		58
57	int anlmf;	59	int anlmf;
58		60
		61	int (filter_channels)(AVFilterContext ctx, void *arg, int jobnr, int nb_jobs);
		62
59	AVFloatDSPContext *fdsp;	63	AVFloatDSPContext *fdsp;
60	} AudioNLMSContext;	64	} AudioNLMSContext;
61		65
@@ -74,93 +78,32 @@ static const AVOption anlms_options[] = { · 12 unmodified lines
74	{ "o", "output", 0, AV_OPT_TYPE_CONST, {.i64=OUT_MODE}, 0, 0, AT, "mode" },	78	{ "o", "output", 0, AV_OPT_TYPE_CONST, {.i64=OUT_MODE}, 0, 0, AT, "mode" },
75	{ "n", "noise", 0, AV_OPT_TYPE_CONST, {.i64=NOISE_MODE}, 0, 0, AT, "mode" },	79	{ "n", "noise", 0, AV_OPT_TYPE_CONST, {.i64=NOISE_MODE}, 0, 0, AT, "mode" },
76	{ "e", "error", 0, AV_OPT_TYPE_CONST, {.i64=ERROR_MODE}, 0, 0, AT, "mode" },	80	{ "e", "error", 0, AV_OPT_TYPE_CONST, {.i64=ERROR_MODE}, 0, 0, AT, "mode" },
		81	{ "precision", "set processing precision", OFFSET(precision), AV_OPT_TYPE_INT, {.i64=0}, 0, 2, A, "precision" },
		82	{ "auto", "set auto processing precision", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, A, "precision" },
		83	{ "float", "set single-floating point processing precision", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, A, "precision" },
		84	{ "double","set double-floating point processing precision", 0, AV_OPT_TYPE_CONST, {.i64=2}, 0, 0, A, "precision" },
77	{ NULL }	85	{ NULL }
78	};	86	};
79		87
80	AVFILTER_DEFINE_CLASS_EXT(anlms, "anlm(f\|s)", anlms_options);	88	AVFILTER_DEFINE_CLASS_EXT(anlms, "anlm(f\|s)", anlms_options);
81		89
82	static float fir_sample(AudioNLMSContext s, float sample, float delay,	90	static int query_formats(AVFilterContext *ctx)
83	float coeffs, float tmp, int *offset)
84	{	91	{
85	const int order = s->order;	92	AudioNLMSContext *s = ctx->priv;
86	float output;	93	static const enum AVSampleFormat sample_fmts[3][3] = {
87		94	{ AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_DBLP, AV_SAMPLE_FMT_NONE },
88	delay[*offset] = sample;	95	{ AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE },
89		96	{ AV_SAMPLE_FMT_DBLP, AV_SAMPLE_FMT_NONE },
90	memcpy(tmp, coeffs + order - offset, order sizeof(float));	97	};
91		98	int ret;
92	output = s->fdsp->scalarproduct_float(delay, tmp, s->kernel_size);
93
94	if (--(*offset) < 0)
95	*offset = order - 1;
96
97	return output;
98	}
99
100	static float process_sample(AudioNLMSContext *s, float input, float desired,
101	float delay, float coeffs, float tmp, int offsetp)
102	{
103	const int order = s->order;
104	const float leakage = s->leakage;
105	const float mu = s->mu;
106	const float a = 1.f - leakage;
107	float sum, output, e, norm, b;
108	int offset = *offsetp;
109
110	delay[offset + order] = input;
111
112	output = fir_sample(s, input, delay, coeffs, tmp, offsetp);
113	e = desired - output;
114
115	sum = s->fdsp->scalarproduct_float(delay, delay, s->kernel_size);
116
117	norm = s->eps + sum;
118	b = mu * e / norm;
119	if (s->anlmf)
120	b = e e;
121
122	memcpy(tmp, delay + offset, order * sizeof(float));
123
124	s->fdsp->vector_fmul_scalar(coeffs, coeffs, a, s->kernel_size);
125		99
126	s->fdsp->vector_fmac_scalar(coeffs, tmp, b, s->kernel_size);	100	if ((ret = ff_set_common_all_channel_counts(ctx)) < 0)
		101	return ret;
127		102
128	memcpy(coeffs + order, coeffs, order * sizeof(float));	103	if ((ret = ff_set_common_formats_from_list(ctx, sample_fmts[s->precision])) < 0)
		104	return ret;
129		105
130	switch (s->output_mode) {	106	return ff_set_common_all_samplerates(ctx);
131	case IN_MODE: output = input; break;
132	case DESIRED_MODE: output = desired; break;
133	case OUT_MODE: output = desired - output; break;
134	case NOISE_MODE: output = input - output; break;
135	case ERROR_MODE: break;
136	}
137	return output;
138	}
139
140	static int process_channels(AVFilterContext ctx, void arg, int jobnr, int nb_jobs)
141	{
142	AudioNLMSContext *s = ctx->priv;
143	AVFrame *out = arg;
144	const int start = (out->ch_layout.nb_channels * jobnr) / nb_jobs;
145	const int end = (out->ch_layout.nb_channels * (jobnr+1)) / nb_jobs;
146
147	for (int c = start; c < end; c++) {
148	const float input = (const float )s->frame[0]->extended_data[c];
149	const float desired = (const float )s->frame[1]->extended_data[c];
150	float delay = (float )s->delay->extended_data[c];
151	float coeffs = (float )s->coeffs->extended_data[c];
152	float tmp = (float )s->tmp->extended_data[c];
153	int offset = (int )s->offset->extended_data[c];
154	float output = (float )out->extended_data[c];
155
156	for (int n = 0; n < out->nb_samples; n++) {
157	output[n] = process_sample(s, input[n], desired[n], delay, coeffs, tmp, offset);
158	if (ctx->is_disabled)
159	output[n] = input[n];
160	}
161	}
162
163	return 0;
164	}	107	}
165		108
166	static int activate(AVFilterContext *ctx)	109	static int activate(AVFilterContext *ctx)
@@ -195,7 +138,7 @@ static int activate(AVFilterContext *ctx) · 28 unmodified lines
195	return AVERROR(ENOMEM);	138	return AVERROR(ENOMEM);
196	}	139	}
197		140
198	ff_filter_execute(ctx, process_channels, out, NULL,	141	ff_filter_execute(ctx, s->filter_channels, out, NULL,
199	FFMIN(ctx->outputs[0]->ch_layout.nb_channels, ff_filter_get_nb_threads(ctx)));	142	FFMIN(ctx->outputs[0]->ch_layout.nb_channels, ff_filter_get_nb_threads(ctx)));
200		143
201	out->pts = s->frame[0]->pts;	144	out->pts = s->frame[0]->pts;
@@ -228,6 +171,13 @@ static int activate(AVFilterContext *ctx) · 26 unmodified lines
228	return 0;	171	return 0;
229	}	172	}
230		173
		174	#define DEPTH 32
		175	#include "anlms_template.c"
		176
		177	#undef DEPTH
		178	#define DEPTH 64
		179	#include "anlms_template.c"
		180
231	static int config_output(AVFilterLink *outlink)	181	static int config_output(AVFilterLink *outlink)
232	{	182	{
233	AVFilterContext *ctx = outlink->src;	183	AVFilterContext *ctx = outlink->src;
@@ -247,6 +197,15 @@ static int config_output(AVFilterLink *outlink) · 13 unmodified lines
247	if (!s->delay \|\| !s->coeffs \|\| !s->offset \|\| !s->tmp)	197	if (!s->delay \|\| !s->coeffs \|\| !s->offset \|\| !s->tmp)
248	return AVERROR(ENOMEM);	198	return AVERROR(ENOMEM);
249		199
		200	switch (outlink->format) {
		201	case AV_SAMPLE_FMT_DBLP:
		202	s->filter_channels = filter_channels_double;
		203	break;
		204	case AV_SAMPLE_FMT_FLTP:
		205	s->filter_channels = filter_channels_float;
		206	break;
		207	}
		208
250	return 0;	209	return 0;
251	}	210	}
252		211
@@ -317,7 +276,7 @@ const AVFilter ff_af_anlmf = { · 64 unmodified lines
317	.activate = activate,	276	.activate = activate,
318	FILTER_INPUTS(inputs),	277	FILTER_INPUTS(inputs),
319	FILTER_OUTPUTS(outputs),	278	FILTER_OUTPUTS(outputs),
320	FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_FLTP),	279	FILTER_QUERY_FUNC(query_formats),
321	.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL \|	280	.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL \|
322	AVFILTER_FLAG_SLICE_THREADS,	281	AVFILTER_FLAG_SLICE_THREADS,
323	.process_command = ff_filter_process_command,	282	.process_command = ff_filter_process_command,

libavfilter/anlms_template.c+141−0

@@ -0,0 +1,141 @@
	1	/*
	2	* This file is part of FFmpeg.
	3	*
	4	* FFmpeg is free software; you can redistribute it and/or
	5	* modify it under the terms of the GNU Lesser General Public
	6	* License as published by the Free Software Foundation; either
	7	* version 2.1 of the License, or (at your option) any later version.
	8	*
	9	* FFmpeg is distributed in the hope that it will be useful,
	10	* but WITHOUT ANY WARRANTY; without even the implied warranty of
	11	* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	12	* Lesser General Public License for more details.
	13	*
	14	* You should have received a copy of the GNU Lesser General Public
	15	* License along with FFmpeg; if not, write to the Free Software
	16	* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	17	*/
	18
	19	#undef ONE
	20	#undef ftype
	21	#undef SAMPLE_FORMAT
	22	#if DEPTH == 32
	23	#define SAMPLE_FORMAT float
	24	#define ftype float
	25	#define ONE 1.f
	26	#else
	27	#define SAMPLE_FORMAT double
	28	#define ftype double
	29	#define ONE 1.0
	30	#endif
	31
	32	#define fn3(a,b) a##_##b
	33	#define fn2(a,b) fn3(a,b)
	34	#define fn(a) fn2(a, SAMPLE_FORMAT)
	35
	36	#if DEPTH == 64
	37	static double scalarproduct_double(const double v1, const double v2, int len)
	38	{
	39	double p = 0.0;
	40
	41	for (int i = 0; i < len; i++)
	42	p += v1[i] * v2[i];
	43
	44	return p;
	45	}
	46	#endif
	47
	48	static ftype fn(fir_sample)(AudioNLMSContext s, ftype sample, ftype delay,
	49	ftype coeffs, ftype tmp, int *offset)
	50	{
	51	const int order = s->order;
	52	ftype output;
	53
	54	delay[*offset] = sample;
	55
	56	memcpy(tmp, coeffs + order - offset, order sizeof(ftype));
	57
	58	#if DEPTH == 32
	59	output = s->fdsp->scalarproduct_float(delay, tmp, s->kernel_size);
	60	#else
	61	output = scalarproduct_double(delay, tmp, s->kernel_size);
	62	#endif
	63
	64	if (--(*offset) < 0)
	65	*offset = order - 1;
	66
	67	return output;
	68	}
	69
	70	static ftype fn(process_sample)(AudioNLMSContext *s, ftype input, ftype desired,
	71	ftype delay, ftype coeffs, ftype tmp, int offsetp)
	72	{
	73	const int order = s->order;
	74	const ftype leakage = s->leakage;
	75	const ftype mu = s->mu;
	76	const ftype a = ONE - leakage;
	77	ftype sum, output, e, norm, b;
	78	int offset = *offsetp;
	79
	80	delay[offset + order] = input;
	81
	82	output = fn(fir_sample)(s, input, delay, coeffs, tmp, offsetp);
	83	e = desired - output;
	84
	85	#if DEPTH == 32
	86	sum = s->fdsp->scalarproduct_float(delay, delay, s->kernel_size);
	87	#else
	88	sum = scalarproduct_double(delay, delay, s->kernel_size);
	89	#endif
	90	norm = s->eps + sum;
	91	b = mu * e / norm;
	92	if (s->anlmf)
	93	b = e e;
	94
	95	memcpy(tmp, delay + offset, order * sizeof(ftype));
	96
	97	#if DEPTH == 32
	98	s->fdsp->vector_fmul_scalar(coeffs, coeffs, a, s->kernel_size);
	99	s->fdsp->vector_fmac_scalar(coeffs, tmp, b, s->kernel_size);
	100	#else
	101	s->fdsp->vector_dmul_scalar(coeffs, coeffs, a, s->kernel_size);
	102	s->fdsp->vector_dmac_scalar(coeffs, tmp, b, s->kernel_size);
	103	#endif
	104
	105	memcpy(coeffs + order, coeffs, order * sizeof(ftype));
	106
	107	switch (s->output_mode) {
	108	case IN_MODE: output = input; break;
	109	case DESIRED_MODE: output = desired; break;
	110	case OUT_MODE: output = desired - output; break;
	111	case NOISE_MODE: output = input - output; break;
	112	case ERROR_MODE: break;
	113	}
	114	return output;
	115	}
	116
	117	static int fn(filter_channels)(AVFilterContext ctx, void arg, int jobnr, int nb_jobs)
	118	{
	119	AudioNLMSContext *s = ctx->priv;
	120	AVFrame *out = arg;
	121	const int start = (out->ch_layout.nb_channels * jobnr) / nb_jobs;
	122	const int end = (out->ch_layout.nb_channels * (jobnr+1)) / nb_jobs;
	123
	124	for (int c = start; c < end; c++) {
	125	const ftype input = (const ftype )s->frame[0]->extended_data[c];
	126	const ftype desired = (const ftype )s->frame[1]->extended_data[c];
	127	ftype delay = (ftype )s->delay->extended_data[c];
	128	ftype coeffs = (ftype )s->coeffs->extended_data[c];
	129	ftype tmp = (ftype )s->tmp->extended_data[c];
	130	int offset = (int )s->offset->extended_data[c];
	131	ftype output = (ftype )out->extended_data[c];
	132
	133	for (int n = 0; n < out->nb_samples; n++) {
	134	output[n] = fn(process_sample)(s, input[n], desired[n], delay, coeffs, tmp, offset);
	135	if (ctx->is_disabled)
	136	output[n] = input[n];
	137	}
	138	}
	139
	140	return 0;
	141	}

avfilter/af_anlms: add double sample format support

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code