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 "libavutil/avstring.h"
22 #include "libavutil/imgutils.h"
23 #include "libavutil/intreadwrite.h"
24 #include "libavutil/opt.h"
25 #include "libavutil/pixdesc.h"
26
27 #include "avfilter.h"
28 #include "formats.h"
29 #include "internal.h"
30 #include "framesync.h"
31 #include "video.h"
32
33 typedef struct MixContext {
34 const AVClass *class;
35 const AVPixFmtDescriptor *desc;
36 char *weights_str;
37 int nb_inputs;
38 int duration;
39 float *weights;
40 float scale;
41 float wfactor;
42
43 int tmix;
44 int nb_frames;
45
46 int depth;
47 int max;
48 int nb_planes;
49 int linesize[4];
50 int height[4];
51
52 AVFrame **frames;
53 FFFrameSync fs;
54 } MixContext;
55
query_formats(AVFilterContext * ctx)56 static int query_formats(AVFilterContext *ctx)
57 {
58 AVFilterFormats *formats = NULL;
59 int ret;
60
61 ret = ff_formats_pixdesc_filter(&formats, 0,
62 AV_PIX_FMT_FLAG_BITSTREAM |
63 AV_PIX_FMT_FLAG_PAL |
64 AV_PIX_FMT_FLAG_HWACCEL);
65 if (ret < 0)
66 return ret;
67 return ff_set_common_formats(ctx, formats);
68 }
69
parse_weights(AVFilterContext * ctx)70 static int parse_weights(AVFilterContext *ctx)
71 {
72 MixContext *s = ctx->priv;
73 char *p, *arg, *saveptr = NULL;
74 int i, last = 0;
75
76 s->wfactor = 0.f;
77 p = s->weights_str;
78 for (i = 0; i < s->nb_inputs; i++) {
79 if (!(arg = av_strtok(p, " |", &saveptr)))
80 break;
81
82 p = NULL;
83 if (av_sscanf(arg, "%f", &s->weights[i]) != 1) {
84 av_log(ctx, AV_LOG_ERROR, "Invalid syntax for weights[%d].\n", i);
85 return AVERROR(EINVAL);
86 }
87 s->wfactor += s->weights[i];
88 last = i;
89 }
90
91 for (; i < s->nb_inputs; i++) {
92 s->weights[i] = s->weights[last];
93 s->wfactor += s->weights[i];
94 }
95 if (s->scale == 0) {
96 s->wfactor = 1 / s->wfactor;
97 } else {
98 s->wfactor = s->scale;
99 }
100
101 return 0;
102 }
103
init(AVFilterContext * ctx)104 static av_cold int init(AVFilterContext *ctx)
105 {
106 MixContext *s = ctx->priv;
107 int ret;
108
109 s->tmix = !strcmp(ctx->filter->name, "tmix");
110
111 s->frames = av_calloc(s->nb_inputs, sizeof(*s->frames));
112 if (!s->frames)
113 return AVERROR(ENOMEM);
114
115 s->weights = av_calloc(s->nb_inputs, sizeof(*s->weights));
116 if (!s->weights)
117 return AVERROR(ENOMEM);
118
119 if (!s->tmix) {
120 for (int i = 0; i < s->nb_inputs; i++) {
121 AVFilterPad pad = { 0 };
122
123 pad.type = AVMEDIA_TYPE_VIDEO;
124 pad.name = av_asprintf("input%d", i);
125 if (!pad.name)
126 return AVERROR(ENOMEM);
127
128 if ((ret = ff_insert_inpad(ctx, i, &pad)) < 0) {
129 av_freep(&pad.name);
130 return ret;
131 }
132 }
133 }
134
135 return parse_weights(ctx);
136 }
137
138 typedef struct ThreadData {
139 AVFrame **in, *out;
140 } ThreadData;
141
mix_frames(AVFilterContext * ctx,void * arg,int jobnr,int nb_jobs)142 static int mix_frames(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
143 {
144 MixContext *s = ctx->priv;
145 ThreadData *td = arg;
146 AVFrame **in = td->in;
147 AVFrame *out = td->out;
148 int i, p, x, y;
149
150 if (s->depth <= 8) {
151 for (p = 0; p < s->nb_planes; p++) {
152 const int slice_start = (s->height[p] * jobnr) / nb_jobs;
153 const int slice_end = (s->height[p] * (jobnr+1)) / nb_jobs;
154 uint8_t *dst = out->data[p] + slice_start * out->linesize[p];
155
156 for (y = slice_start; y < slice_end; y++) {
157 for (x = 0; x < s->linesize[p]; x++) {
158 int val = 0;
159
160 for (i = 0; i < s->nb_inputs; i++) {
161 uint8_t src = in[i]->data[p][y * in[i]->linesize[p] + x];
162
163 val += src * s->weights[i];
164 }
165
166 dst[x] = av_clip_uint8(val * s->wfactor);
167 }
168
169 dst += out->linesize[p];
170 }
171 }
172 } else {
173 for (p = 0; p < s->nb_planes; p++) {
174 const int slice_start = (s->height[p] * jobnr) / nb_jobs;
175 const int slice_end = (s->height[p] * (jobnr+1)) / nb_jobs;
176 uint16_t *dst = (uint16_t *)(out->data[p] + slice_start * out->linesize[p]);
177
178 for (y = slice_start; y < slice_end; y++) {
179 for (x = 0; x < s->linesize[p] / 2; x++) {
180 int val = 0;
181
182 for (i = 0; i < s->nb_inputs; i++) {
183 uint16_t src = AV_RN16(in[i]->data[p] + y * in[i]->linesize[p] + x * 2);
184
185 val += src * s->weights[i];
186 }
187
188 dst[x] = av_clip(val * s->wfactor, 0, s->max);
189 }
190
191 dst += out->linesize[p] / 2;
192 }
193 }
194 }
195
196 return 0;
197 }
198
process_frame(FFFrameSync * fs)199 static int process_frame(FFFrameSync *fs)
200 {
201 AVFilterContext *ctx = fs->parent;
202 AVFilterLink *outlink = ctx->outputs[0];
203 MixContext *s = fs->opaque;
204 AVFrame **in = s->frames;
205 AVFrame *out;
206 ThreadData td;
207 int i, ret;
208
209 for (i = 0; i < s->nb_inputs; i++) {
210 if ((ret = ff_framesync_get_frame(&s->fs, i, &in[i], 0)) < 0)
211 return ret;
212 }
213
214 if (ctx->is_disabled) {
215 out = av_frame_clone(s->frames[0]);
216 if (!out)
217 return AVERROR(ENOMEM);
218 out->pts = av_rescale_q(s->fs.pts, s->fs.time_base, outlink->time_base);
219 return ff_filter_frame(outlink, out);
220 }
221
222 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
223 if (!out)
224 return AVERROR(ENOMEM);
225 out->pts = av_rescale_q(s->fs.pts, s->fs.time_base, outlink->time_base);
226
227 td.in = in;
228 td.out = out;
229 ctx->internal->execute(ctx, mix_frames, &td, NULL, FFMIN(s->height[0], ff_filter_get_nb_threads(ctx)));
230
231 return ff_filter_frame(outlink, out);
232 }
233
config_output(AVFilterLink * outlink)234 static int config_output(AVFilterLink *outlink)
235 {
236 AVFilterContext *ctx = outlink->src;
237 MixContext *s = ctx->priv;
238 AVRational frame_rate = ctx->inputs[0]->frame_rate;
239 AVRational sar = ctx->inputs[0]->sample_aspect_ratio;
240 AVFilterLink *inlink = ctx->inputs[0];
241 int height = ctx->inputs[0]->h;
242 int width = ctx->inputs[0]->w;
243 FFFrameSyncIn *in;
244 int i, ret;
245
246 if (!s->tmix) {
247 for (i = 1; i < s->nb_inputs; i++) {
248 if (ctx->inputs[i]->h != height || ctx->inputs[i]->w != width) {
249 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);
250 return AVERROR(EINVAL);
251 }
252 }
253 }
254
255 s->desc = av_pix_fmt_desc_get(outlink->format);
256 if (!s->desc)
257 return AVERROR_BUG;
258 s->nb_planes = av_pix_fmt_count_planes(outlink->format);
259 s->depth = s->desc->comp[0].depth;
260 s->max = (1 << s->depth) - 1;
261
262 if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
263 return ret;
264
265 s->height[1] = s->height[2] = AV_CEIL_RSHIFT(inlink->h, s->desc->log2_chroma_h);
266 s->height[0] = s->height[3] = inlink->h;
267
268 if (s->tmix)
269 return 0;
270
271 outlink->w = width;
272 outlink->h = height;
273 outlink->frame_rate = frame_rate;
274 outlink->sample_aspect_ratio = sar;
275
276 if ((ret = ff_framesync_init(&s->fs, ctx, s->nb_inputs)) < 0)
277 return ret;
278
279 in = s->fs.in;
280 s->fs.opaque = s;
281 s->fs.on_event = process_frame;
282
283 for (i = 0; i < s->nb_inputs; i++) {
284 AVFilterLink *inlink = ctx->inputs[i];
285
286 in[i].time_base = inlink->time_base;
287 in[i].sync = 1;
288 in[i].before = EXT_STOP;
289 in[i].after = (s->duration == 1 || (s->duration == 2 && i == 0)) ? EXT_STOP : EXT_INFINITY;
290 }
291
292 ret = ff_framesync_configure(&s->fs);
293 outlink->time_base = s->fs.time_base;
294
295 return ret;
296 }
297
uninit(AVFilterContext * ctx)298 static av_cold void uninit(AVFilterContext *ctx)
299 {
300 MixContext *s = ctx->priv;
301 int i;
302
303 ff_framesync_uninit(&s->fs);
304 av_freep(&s->weights);
305
306 if (!s->tmix) {
307 for (i = 0; i < ctx->nb_inputs; i++)
308 av_freep(&ctx->input_pads[i].name);
309 } else {
310 for (i = 0; i < s->nb_frames && s->frames; i++)
311 av_frame_free(&s->frames[i]);
312 }
313 av_freep(&s->frames);
314 }
315
process_command(AVFilterContext * ctx,const char * cmd,const char * args,char * res,int res_len,int flags)316 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
317 char *res, int res_len, int flags)
318 {
319 int ret;
320
321 ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
322 if (ret < 0)
323 return ret;
324
325 return parse_weights(ctx);
326 }
327
activate(AVFilterContext * ctx)328 static int activate(AVFilterContext *ctx)
329 {
330 MixContext *s = ctx->priv;
331 return ff_framesync_activate(&s->fs);
332 }
333
334 #define OFFSET(x) offsetof(MixContext, x)
335 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_FILTERING_PARAM
336 #define TFLAGS AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_RUNTIME_PARAM
337
338 static const AVOption mix_options[] = {
339 { "inputs", "set number of inputs", OFFSET(nb_inputs), AV_OPT_TYPE_INT, {.i64=2}, 2, INT16_MAX, .flags = FLAGS },
340 { "weights", "set weight for each input", OFFSET(weights_str), AV_OPT_TYPE_STRING, {.str="1 1"}, 0, 0, .flags = TFLAGS },
341 { "scale", "set scale", OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=0}, 0, INT16_MAX, .flags = TFLAGS },
342 { "duration", "how to determine end of stream", OFFSET(duration), AV_OPT_TYPE_INT, {.i64=0}, 0, 2, .flags = FLAGS, "duration" },
343 { "longest", "Duration of longest input", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, FLAGS, "duration" },
344 { "shortest", "Duration of shortest input", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, FLAGS, "duration" },
345 { "first", "Duration of first input", 0, AV_OPT_TYPE_CONST, {.i64=2}, 0, 0, FLAGS, "duration" },
346 { NULL },
347 };
348
349 static const AVFilterPad outputs[] = {
350 {
351 .name = "default",
352 .type = AVMEDIA_TYPE_VIDEO,
353 .config_props = config_output,
354 },
355 { NULL }
356 };
357
358 #if CONFIG_MIX_FILTER
359 AVFILTER_DEFINE_CLASS(mix);
360
361 AVFilter ff_vf_mix = {
362 .name = "mix",
363 .description = NULL_IF_CONFIG_SMALL("Mix video inputs."),
364 .priv_size = sizeof(MixContext),
365 .priv_class = &mix_class,
366 .query_formats = query_formats,
367 .outputs = outputs,
368 .init = init,
369 .uninit = uninit,
370 .activate = activate,
371 .flags = AVFILTER_FLAG_DYNAMIC_INPUTS | AVFILTER_FLAG_SLICE_THREADS |
372 AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
373 .process_command = process_command,
374 };
375
376 #endif /* CONFIG_MIX_FILTER */
377
378 #if CONFIG_TMIX_FILTER
tmix_filter_frame(AVFilterLink * inlink,AVFrame * in)379 static int tmix_filter_frame(AVFilterLink *inlink, AVFrame *in)
380 {
381 AVFilterContext *ctx = inlink->dst;
382 AVFilterLink *outlink = ctx->outputs[0];
383 MixContext *s = ctx->priv;
384 ThreadData td;
385 AVFrame *out;
386
387 if (s->nb_inputs == 1)
388 return ff_filter_frame(outlink, in);
389
390 if (s->nb_frames < s->nb_inputs) {
391 s->frames[s->nb_frames] = in;
392 s->nb_frames++;
393 if (s->nb_frames < s->nb_inputs)
394 return 0;
395 } else {
396 av_frame_free(&s->frames[0]);
397 memmove(&s->frames[0], &s->frames[1], sizeof(*s->frames) * (s->nb_inputs - 1));
398 s->frames[s->nb_inputs - 1] = in;
399 }
400
401 if (ctx->is_disabled) {
402 out = av_frame_clone(s->frames[0]);
403 if (!out)
404 return AVERROR(ENOMEM);
405 return ff_filter_frame(outlink, out);
406 }
407
408 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
409 if (!out)
410 return AVERROR(ENOMEM);
411 out->pts = s->frames[0]->pts;
412
413 td.out = out;
414 td.in = s->frames;
415 ctx->internal->execute(ctx, mix_frames, &td, NULL, FFMIN(s->height[0], ff_filter_get_nb_threads(ctx)));
416
417 return ff_filter_frame(outlink, out);
418 }
419
420 static const AVOption tmix_options[] = {
421 { "frames", "set number of successive frames to mix", OFFSET(nb_inputs), AV_OPT_TYPE_INT, {.i64=3}, 1, 128, .flags = FLAGS },
422 { "weights", "set weight for each frame", OFFSET(weights_str), AV_OPT_TYPE_STRING, {.str="1 1 1"}, 0, 0, .flags = TFLAGS },
423 { "scale", "set scale", OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=0}, 0, INT16_MAX, .flags = TFLAGS },
424 { NULL },
425 };
426
427 static const AVFilterPad inputs[] = {
428 {
429 .name = "default",
430 .type = AVMEDIA_TYPE_VIDEO,
431 .filter_frame = tmix_filter_frame,
432 },
433 { NULL }
434 };
435
436 AVFILTER_DEFINE_CLASS(tmix);
437
438 AVFilter ff_vf_tmix = {
439 .name = "tmix",
440 .description = NULL_IF_CONFIG_SMALL("Mix successive video frames."),
441 .priv_size = sizeof(MixContext),
442 .priv_class = &tmix_class,
443 .query_formats = query_formats,
444 .outputs = outputs,
445 .inputs = inputs,
446 .init = init,
447 .uninit = uninit,
448 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
449 .process_command = process_command,
450 };
451
452 #endif /* CONFIG_TMIX_FILTER */
453