FFmpeg coverage

Line	Branch	Exec	Source
1			/*
2			* Copyright (c) 2017 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 "config_components.h"
22
23			#include "libavutil/avstring.h"
24			#include "libavutil/imgutils.h"
25			#include "libavutil/mem.h"
26			#include "libavutil/opt.h"
27			#include "libavutil/pixdesc.h"
28
29			#include "avfilter.h"
30			#include "filters.h"
31			#include "formats.h"
32			#include "framesync.h"
33			#include "video.h"
34
35			typedef struct MixContext {
36			const AVClass *class;
37			const AVPixFmtDescriptor *desc;
38			char *weights_str;
39			int nb_inputs;
40			int nb_threads;
41			int duration;
42			float *weights;
43			float scale;
44			float wfactor;
45
46			int fast;
47			int tmix;
48			int nb_frames;
49			int nb_unique_frames;
50
51			int depth;
52			int max;
53			int planes;
54			int nb_planes;
55			int linesizes[4];
56			int height[4];
57
58			uint8_t *sum[4];
59
60			uint8_t **data;
61			int *linesize;
62
63			AVFrame **frames;
64			FFFrameSync fs;
65			} MixContext;
66
67		✗	static int query_formats(const AVFilterContext *ctx,
68			AVFilterFormatsConfig **cfg_in,
69			AVFilterFormatsConfig **cfg_out)
70			{
71		✗	unsigned reject_flags = AV_PIX_FMT_FLAG_BITSTREAM |
72			AV_PIX_FMT_FLAG_HWACCEL |
73			AV_PIX_FMT_FLAG_PAL;
74		✗	unsigned accept_flags = 0;
75
76			if (!HAVE_BIGENDIAN)
77		✗	reject_flags |= AV_PIX_FMT_FLAG_BE;
78			else
79			accept_flags |= AV_PIX_FMT_FLAG_BE;
80
81		✗	return ff_set_common_formats2(ctx, cfg_in, cfg_out,
82			ff_formats_pixdesc_filter(accept_flags, reject_flags));
83			}
84
85		✗	static int parse_weights(AVFilterContext *ctx)
86			{
87		✗	MixContext *s = ctx->priv;
88		✗	char *p, *arg, *saveptr = NULL;
89		✗	int i, last = 0;
90
91		✗	s->fast = 1;
92		✗	s->wfactor = 0.f;
93		✗	p = s->weights_str;
94		✗	for (i = 0; i < s->nb_inputs; i++) {
95		✗	if (!(arg = av_strtok(p, " |", &saveptr)))
96		✗	break;
97
98		✗	p = NULL;
99		✗	if (av_sscanf(arg, "%f", &s->weights[i]) != 1) {
100		✗	av_log(ctx, AV_LOG_ERROR, "Invalid syntax for weights[%d].\n", i);
101		✗	return AVERROR(EINVAL);
102			}
103		✗	s->wfactor += s->weights[i];
104		✗	if (i > 0)
105		✗	s->fast &= s->weights[i] == s->weights[0];
106		✗	last = i;
107			}
108
109		✗	for (; i < s->nb_inputs; i++) {
110		✗	s->weights[i] = s->weights[last];
111		✗	s->wfactor += s->weights[i];
112			}
113		✗	if (s->scale == 0) {
114		✗	s->wfactor = 1 / s->wfactor;
115			} else {
116		✗	if (s->scale != 1.f / s->wfactor)
117		✗	s->fast = 0;
118		✗	s->wfactor = s->scale;
119			}
120
121		✗	return 0;
122			}
123
124		✗	static av_cold int init(AVFilterContext *ctx)
125			{
126		✗	MixContext *s = ctx->priv;
127			int ret;
128
129		✗	s->tmix = !strcmp(ctx->filter->name, "tmix");
130
131		✗	s->frames = av_calloc(s->nb_inputs, sizeof(*s->frames));
132		✗	if (!s->frames)
133		✗	return AVERROR(ENOMEM);
134
135		✗	s->weights = av_calloc(s->nb_inputs, sizeof(*s->weights));
136		✗	if (!s->weights)
137		✗	return AVERROR(ENOMEM);
138
139		✗	if (!s->tmix) {
140		✗	for (int i = 0; i < s->nb_inputs; i++) {
141		✗	AVFilterPad pad = { 0 };
142
143		✗	pad.type = AVMEDIA_TYPE_VIDEO;
144		✗	pad.name = av_asprintf("input%d", i);
145		✗	if (!pad.name)
146		✗	return AVERROR(ENOMEM);
147
148		✗	if ((ret = ff_append_inpad_free_name(ctx, &pad)) < 0)
149		✗	return ret;
150			}
151			}
152
153		✗	return parse_weights(ctx);
154			}
155
156			typedef struct ThreadData {
157			AVFrame **in, *out;
158			} ThreadData;
159
160			#define FAST_TMIX_SLICE(type, stype, round) \
161			for (int p = 0; p < s->nb_planes; p++) { \
162			const int slice_start = (s->height[p] * jobnr) / nb_jobs; \
163			const int slice_end = (s->height[p] * (jobnr+1)) / nb_jobs; \
164			const int width = s->linesizes[p] / sizeof(type); \
165			stype *sum = (stype *)(s->sum[p] + slice_start * s->linesizes[p] * 2); \
166			type *dst = (type *)(out->data[p] + slice_start * out->linesize[p]); \
167			const ptrdiff_t sum_linesize = (s->linesizes[p] * 2) / sizeof(stype); \
168			const ptrdiff_t dst_linesize = out->linesize[p] / sizeof(type); \
169			const int idx = FFMAX(0, nb_inputs - nb_unique); \
170			const ptrdiff_t src_linesize[2] = { in[idx]->linesize[p], \
171			in[nb_inputs-1]->linesize[p] }; \
172			const type *src[2]; \
173			\
174			if (!((1 << p) & s->planes)) { \
175			av_image_copy_plane((uint8_t *)dst, out->linesize[p], \
176			in[0]->data[p] + slice_start * in[0]->linesize[p], \
177			in[0]->linesize[p], \
178			s->linesizes[p], slice_end - slice_start); \
179			continue; \
180			} \
181			\
182			src[0] = (const type *)(in[idx]->data[p] + slice_start*src_linesize[0]); \
183			src[1] = (const type *)(in[nb_inputs-1]->data[p] + slice_start * src_linesize[1]); \
184			\
185			for (int y = slice_start; y < slice_end; y++) { \
186			for (int x = 0; x < width; x++) { \
187			sum[x] += src[1][x] * (1 + (nb_inputs - 1) * (idx == (nb_inputs - 1))); \
188			dst[x] = (sum[x] + (round)) / nb_inputs; \
189			sum[x] -= src[0][x]; \
190			} \
191			\
192			dst += dst_linesize; \
193			sum += sum_linesize; \
194			src[0] += src_linesize[0] / sizeof(type); \
195			src[1] += src_linesize[1] / sizeof(type); \
196			} \
197			}
198
199			#define MIX_SLICE(type, fun, clip) \
200			for (int p = 0; p < s->nb_planes; p++) { \
201			const int slice_start = (s->height[p] * jobnr) / nb_jobs; \
202			const int slice_end = (s->height[p] * (jobnr+1)) / nb_jobs; \
203			const int width = s->linesizes[p] / sizeof(type); \
204			type *dst = (type *)(out->data[p] + slice_start * out->linesize[p]); \
205			const ptrdiff_t dst_linesize = out->linesize[p] / sizeof(type); \
206			\
207			if (!((1 << p) & s->planes)) { \
208			av_image_copy_plane((uint8_t *)dst, out->linesize[p], \
209			in[0]->data[p] + slice_start * in[0]->linesize[p], \
210			in[0]->linesize[p], \
211			s->linesizes[p], slice_end - slice_start); \
212			continue; \
213			} \
214			\
215			for (int i = 0; i < nb_inputs; i++) \
216			linesize[i] = in[i]->linesize[p]; \
217			\
218			for (int i = 0; i < nb_inputs; i++) \
219			srcf[i] = in[i]->data[p] + slice_start * linesize[i]; \
220			\
221			for (int y = slice_start; y < slice_end; y++) { \
222			for (int x = 0; x < width; x++) { \
223			float val = 0.f; \
224			\
225			for (int i = 0; i < nb_inputs; i++) { \
226			float src = *(type *)(srcf[i] + x * sizeof(type)); \
227			\
228			val += src * weights[i]; \
229			} \
230			\
231			dst[x] = clip(fun(val * wfactor), 0, max); \
232			} \
233			\
234			dst += dst_linesize; \
235			for (int i = 0; i < nb_inputs; i++) \
236			srcf[i] += linesize[i]; \
237			} \
238			}
239
240			#define CLIP8(x, min, max) av_clip_uint8(x)
241			#define CLIP16(x, min, max) av_clip(x, min, max)
242			#define CLIPF(x, min, max) (x)
243			#define NOP(x) (x)
244
245		✗	static int mix_frames(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
246			{
247		✗	MixContext *s = ctx->priv;
248		✗	ThreadData *td = arg;
249		✗	AVFrame **in = td->in;
250		✗	AVFrame *out = td->out;
251		✗	const float *weights = s->weights;
252		✗	uint8_t **srcf = s->data + jobnr * s->nb_inputs;
253		✗	int *linesize = s->linesize + jobnr * s->nb_inputs;
254		✗	const int nb_unique = s->nb_unique_frames;
255		✗	const int nb_inputs = s->nb_inputs;
256		✗	const float wfactor = s->wfactor;
257		✗	const int max = s->max;
258
259		✗	if (s->tmix && s->fast) {
260		✗	if (s->depth <= 8) {
261		✗	FAST_TMIX_SLICE(uint8_t, uint16_t, nb_inputs >> 1)
262		✗	} else if (s->depth <= 16) {
263		✗	FAST_TMIX_SLICE(uint16_t, uint32_t, nb_inputs >> 1)
264			} else {
265		✗	FAST_TMIX_SLICE(float, float, 0.f)
266			}
267
268		✗	return 0;
269			}
270
271		✗	if (s->depth <= 8) {
272		✗	MIX_SLICE(uint8_t, lrintf, CLIP8)
273		✗	} else if (s->depth <= 16) {
274		✗	MIX_SLICE(uint16_t, lrintf, CLIP16)
275			} else {
276		✗	MIX_SLICE(float, NOP, CLIPF)
277			}
278
279		✗	return 0;
280			}
281
282		✗	static int process_frame(FFFrameSync *fs)
283			{
284		✗	AVFilterContext *ctx = fs->parent;
285		✗	AVFilterLink *outlink = ctx->outputs[0];
286		✗	MixContext *s = fs->opaque;
287		✗	AVFrame **in = s->frames;
288			AVFrame *out;
289			ThreadData td;
290			int i, ret;
291
292		✗	for (i = 0; i < s->nb_inputs; i++) {
293		✗	if ((ret = ff_framesync_get_frame(&s->fs, i, &in[i], 0)) < 0)
294		✗	return ret;
295			}
296
297		✗	if (ctx->is_disabled) {
298		✗	out = av_frame_clone(s->frames[0]);
299		✗	if (!out)
300		✗	return AVERROR(ENOMEM);
301		✗	out->pts = av_rescale_q(s->fs.pts, s->fs.time_base, outlink->time_base);
302		✗	return ff_filter_frame(outlink, out);
303			}
304
305		✗	out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
306		✗	if (!out)
307		✗	return AVERROR(ENOMEM);
308		✗	out->pts = av_rescale_q(s->fs.pts, s->fs.time_base, outlink->time_base);
309
310		✗	td.in = in;
311		✗	td.out = out;
312		✗	ff_filter_execute(ctx, mix_frames, &td, NULL,
313		✗	FFMIN(s->height[1], s->nb_threads));
314
315		✗	return ff_filter_frame(outlink, out);
316			}
317
318		✗	static int config_output(AVFilterLink *outlink)
319			{
320		✗	AVFilterContext *ctx = outlink->src;
321		✗	MixContext *s = ctx->priv;
322		✗	FilterLink *il = ff_filter_link(ctx->inputs[0]);
323		✗	FilterLink *ol = ff_filter_link(outlink);
324		✗	AVRational sar = ctx->inputs[0]->sample_aspect_ratio;
325		✗	AVFilterLink *inlink = ctx->inputs[0];
326		✗	int height = ctx->inputs[0]->h;
327		✗	int width = ctx->inputs[0]->w;
328			FFFrameSyncIn *in;
329			int i, ret;
330
331		✗	if (!s->tmix) {
332		✗	for (i = 1; i < s->nb_inputs; i++) {
333		✗	if (ctx->inputs[i]->h != height || ctx->inputs[i]->w != width) {
334		✗	av_log(ctx, AV_LOG_ERROR, "Input %d size (%dx%d) does not match input %d size (%dx%d).\n", i, ctx->inputs[i]->w, ctx->inputs[i]->h, 0, width, height);
335		✗	return AVERROR(EINVAL);
336			}
337			}
338			}
339
340		✗	s->nb_threads = ff_filter_get_nb_threads(ctx);
341		✗	s->desc = av_pix_fmt_desc_get(outlink->format);
342		✗	if (!s->desc)
343		✗	return AVERROR_BUG;
344		✗	s->nb_planes = av_pix_fmt_count_planes(outlink->format);
345		✗	s->depth = s->desc->comp[0].depth;
346		✗	s->max = (1 << s->depth) - 1;
347
348		✗	if ((ret = av_image_fill_linesizes(s->linesizes, inlink->format, inlink->w)) < 0)
349		✗	return ret;
350
351		✗	s->height[1] = s->height[2] = AV_CEIL_RSHIFT(inlink->h, s->desc->log2_chroma_h);
352		✗	s->height[0] = s->height[3] = inlink->h;
353
354		✗	s->data = av_calloc(s->nb_threads * s->nb_inputs, sizeof(*s->data));
355		✗	if (!s->data)
356		✗	return AVERROR(ENOMEM);
357
358		✗	s->linesize = av_calloc(s->nb_threads * s->nb_inputs, sizeof(*s->linesize));
359		✗	if (!s->linesize)
360		✗	return AVERROR(ENOMEM);
361
362		✗	if (s->tmix) {
363		✗	for (int p = 0; p < s->nb_planes; p++) {
364		✗	s->sum[p] = av_calloc(s->linesizes[p], s->height[p] * sizeof(*s->sum) * 2);
365		✗	if (!s->sum[p])
366		✗	return AVERROR(ENOMEM);
367			}
368		✗	return 0;
369			}
370
371		✗	outlink->w = width;
372		✗	outlink->h = height;
373		✗	ol->frame_rate = il->frame_rate;
374		✗	outlink->sample_aspect_ratio = sar;
375
376		✗	if ((ret = ff_framesync_init(&s->fs, ctx, s->nb_inputs)) < 0)
377		✗	return ret;
378
379		✗	in = s->fs.in;
380		✗	s->fs.opaque = s;
381		✗	s->fs.on_event = process_frame;
382
383		✗	for (i = 0; i < s->nb_inputs; i++) {
384		✗	AVFilterLink *inlink = ctx->inputs[i];
385
386		✗	in[i].time_base = inlink->time_base;
387		✗	in[i].sync = 1;
388		✗	in[i].before = EXT_STOP;
389		✗	in[i].after = (s->duration == 1 || (s->duration == 2 && i == 0)) ? EXT_STOP : EXT_INFINITY;
390			}
391
392		✗	ret = ff_framesync_configure(&s->fs);
393		✗	outlink->time_base = s->fs.time_base;
394
395		✗	return ret;
396			}
397
398		✗	static av_cold void uninit(AVFilterContext *ctx)
399			{
400		✗	MixContext *s = ctx->priv;
401			int i;
402
403		✗	ff_framesync_uninit(&s->fs);
404		✗	av_freep(&s->weights);
405		✗	av_freep(&s->data);
406		✗	av_freep(&s->linesize);
407
408		✗	if (s->tmix) {
409		✗	for (i = 0; i < 4; i++)
410		✗	av_freep(&s->sum[i]);
411		✗	for (i = 0; i < s->nb_frames && s->frames; i++)
412		✗	av_frame_free(&s->frames[i]);
413			}
414		✗	av_freep(&s->frames);
415		✗	}
416
417		✗	static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
418			char *res, int res_len, int flags)
419			{
420			int ret;
421
422		✗	ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
423		✗	if (ret < 0)
424		✗	return ret;
425
426		✗	return parse_weights(ctx);
427			}
428
429		✗	static int activate(AVFilterContext *ctx)
430			{
431		✗	MixContext *s = ctx->priv;
432		✗	return ff_framesync_activate(&s->fs);
433			}
434
435			#define OFFSET(x) offsetof(MixContext, x)
436			#define FLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_FILTERING_PARAM
437			#define TFLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_RUNTIME_PARAM
438
439			static const AVOption mix_options[] = {
440			{ "inputs", "set number of inputs", OFFSET(nb_inputs), AV_OPT_TYPE_INT, {.i64=2}, 2, INT16_MAX, .flags = FLAGS },
441			{ "weights", "set weight for each input", OFFSET(weights_str), AV_OPT_TYPE_STRING, {.str="1 1"}, 0, 0, .flags = TFLAGS },
442			{ "scale", "set scale", OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=0}, 0, INT16_MAX, .flags = TFLAGS },
443			{ "planes", "set what planes to filter", OFFSET(planes), AV_OPT_TYPE_FLAGS, {.i64=15}, 0, 15, .flags = TFLAGS },
444			{ "duration", "how to determine end of stream", OFFSET(duration), AV_OPT_TYPE_INT, {.i64=0}, 0, 2, .flags = FLAGS, .unit = "duration" },
445			{ "longest", "Duration of longest input", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, FLAGS, .unit = "duration" },
446			{ "shortest", "Duration of shortest input", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, FLAGS, .unit = "duration" },
447			{ "first", "Duration of first input", 0, AV_OPT_TYPE_CONST, {.i64=2}, 0, 0, FLAGS, .unit = "duration" },
448			{ NULL },
449			};
450
451			static const AVFilterPad outputs[] = {
452			{
453			.name = "default",
454			.type = AVMEDIA_TYPE_VIDEO,
455			.config_props = config_output,
456			},
457			};
458
459			#if CONFIG_MIX_FILTER
460			AVFILTER_DEFINE_CLASS(mix);
461
462			const AVFilter ff_vf_mix = {
463			.name = "mix",
464			.description = NULL_IF_CONFIG_SMALL("Mix video inputs."),
465			.priv_size = sizeof(MixContext),
466			.priv_class = &mix_class,
467			FILTER_OUTPUTS(outputs),
468			FILTER_QUERY_FUNC2(query_formats),
469			.init = init,
470			.uninit = uninit,
471			.activate = activate,
472			.flags = AVFILTER_FLAG_DYNAMIC_INPUTS | AVFILTER_FLAG_SLICE_THREADS |
473			AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
474			.process_command = process_command,
475			};
476
477			#endif /* CONFIG_MIX_FILTER */
478
479			#if CONFIG_TMIX_FILTER
480		✗	static int tmix_filter_frame(AVFilterLink *inlink, AVFrame *in)
481			{
482		✗	AVFilterContext *ctx = inlink->dst;
483		✗	AVFilterLink *outlink = ctx->outputs[0];
484		✗	MixContext *s = ctx->priv;
485			ThreadData td;
486			AVFrame *out;
487
488		✗	if (s->nb_inputs == 1)
489		✗	return ff_filter_frame(outlink, in);
490
491		✗	if (s->nb_frames < s->nb_inputs) {
492		✗	s->frames[s->nb_frames] = in;
493		✗	s->nb_frames++;
494		✗	s->nb_unique_frames++;
495		✗	while (s->nb_frames < s->nb_inputs) {
496		✗	s->frames[s->nb_frames] = av_frame_clone(s->frames[s->nb_frames - 1]);
497		✗	if (!s->frames[s->nb_frames])
498		✗	return AVERROR(ENOMEM);
499		✗	s->nb_frames++;
500			}
501			} else {
502		✗	s->nb_unique_frames = FFMIN(s->nb_unique_frames + 1, s->nb_inputs);
503		✗	av_frame_free(&s->frames[0]);
504		✗	memmove(&s->frames[0], &s->frames[1], sizeof(*s->frames) * (s->nb_inputs - 1));
505		✗	s->frames[s->nb_inputs - 1] = in;
506			}
507
508		✗	if (ctx->is_disabled) {
509		✗	out = av_frame_clone(s->frames[0]);
510		✗	if (!out)
511		✗	return AVERROR(ENOMEM);
512		✗	return ff_filter_frame(outlink, out);
513			}
514
515		✗	out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
516		✗	if (!out)
517		✗	return AVERROR(ENOMEM);
518		✗	out->pts = s->frames[s->nb_frames - 1]->pts;
519
520		✗	td.out = out;
521		✗	td.in = s->frames;
522		✗	ff_filter_execute(ctx, mix_frames, &td, NULL,
523		✗	FFMIN(s->height[1], s->nb_threads));
524
525		✗	return ff_filter_frame(outlink, out);
526			}
527
528			static const AVOption tmix_options[] = {
529			{ "frames", "set number of successive frames to mix", OFFSET(nb_inputs), AV_OPT_TYPE_INT, {.i64=3}, 1, 1024, .flags = FLAGS },
530			{ "weights", "set weight for each frame", OFFSET(weights_str), AV_OPT_TYPE_STRING, {.str="1 1 1"}, 0, 0, .flags = TFLAGS },
531			{ "scale", "set scale", OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=0}, 0, INT16_MAX, .flags = TFLAGS },
532			{ "planes", "set what planes to filter", OFFSET(planes), AV_OPT_TYPE_FLAGS, {.i64=15}, 0, 15, .flags = TFLAGS },
533			{ NULL },
534			};
535
536			static const AVFilterPad inputs[] = {
537			{
538			.name = "default",
539			.type = AVMEDIA_TYPE_VIDEO,
540			.filter_frame = tmix_filter_frame,
541			},
542			};
543
544			AVFILTER_DEFINE_CLASS(tmix);
545
546			const AVFilter ff_vf_tmix = {
547			.name = "tmix",
548			.description = NULL_IF_CONFIG_SMALL("Mix successive video frames."),
549			.priv_size = sizeof(MixContext),
550			.priv_class = &tmix_class,
551			FILTER_OUTPUTS(outputs),
552			FILTER_INPUTS(inputs),
553			FILTER_QUERY_FUNC2(query_formats),
554			.init = init,
555			.uninit = uninit,
556			.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
557			.process_command = process_command,
558			};
559
560			#endif /* CONFIG_TMIX_FILTER */
561