summaryrefslogtreecommitdiff
path: root/libavfilter/vf_w3fdif.c (plain)
blob: b7872db341929ca2bfee512553337bb9969b30eb
1/*
2 * Copyright (C) 2012 British Broadcasting Corporation, All Rights Reserved
3 * Author of de-interlace algorithm: Jim Easterbrook for BBC R&D
4 * Based on the process described by Martin Weston for BBC R&D
5 * Author of FFmpeg filter: Mark Himsley for BBC Broadcast Systems Development
6 *
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24#include "libavutil/common.h"
25#include "libavutil/imgutils.h"
26#include "libavutil/opt.h"
27#include "libavutil/pixdesc.h"
28#include "avfilter.h"
29#include "formats.h"
30#include "internal.h"
31#include "video.h"
32#include "w3fdif.h"
33
34typedef struct W3FDIFContext {
35 const AVClass *class;
36 int filter; ///< 0 is simple, 1 is more complex
37 int deint; ///< which frames to deinterlace
38 int linesize[4]; ///< bytes of pixel data per line for each plane
39 int planeheight[4]; ///< height of each plane
40 int field; ///< which field are we on, 0 or 1
41 int eof;
42 int nb_planes;
43 AVFrame *prev, *cur, *next; ///< previous, current, next frames
44 int32_t **work_line; ///< lines we are calculating
45 int nb_threads;
46 int max;
47
48 W3FDIFDSPContext dsp;
49} W3FDIFContext;
50
51#define OFFSET(x) offsetof(W3FDIFContext, x)
52#define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
53#define CONST(name, help, val, unit) { name, help, 0, AV_OPT_TYPE_CONST, {.i64=val}, 0, 0, FLAGS, unit }
54
55static const AVOption w3fdif_options[] = {
56 { "filter", "specify the filter", OFFSET(filter), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, FLAGS, "filter" },
57 CONST("simple", NULL, 0, "filter"),
58 CONST("complex", NULL, 1, "filter"),
59 { "deint", "specify which frames to deinterlace", OFFSET(deint), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, FLAGS, "deint" },
60 CONST("all", "deinterlace all frames", 0, "deint"),
61 CONST("interlaced", "only deinterlace frames marked as interlaced", 1, "deint"),
62 { NULL }
63};
64
65AVFILTER_DEFINE_CLASS(w3fdif);
66
67static int query_formats(AVFilterContext *ctx)
68{
69 static const enum AVPixelFormat pix_fmts[] = {
70 AV_PIX_FMT_YUV410P, AV_PIX_FMT_YUV411P,
71 AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P,
72 AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUV444P,
73 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
74 AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
75 AV_PIX_FMT_YUVJ411P,
76 AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA444P,
77 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP,
78 AV_PIX_FMT_GRAY8,
79 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
80 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
81 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12,
82 AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
83 AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14,
84 AV_PIX_FMT_NONE
85 };
86
87 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
88 if (!fmts_list)
89 return AVERROR(ENOMEM);
90 return ff_set_common_formats(ctx, fmts_list);
91}
92
93static void filter_simple_low(int32_t *work_line,
94 uint8_t *in_lines_cur[2],
95 const int16_t *coef, int linesize)
96{
97 int i;
98
99 for (i = 0; i < linesize; i++) {
100 *work_line = *in_lines_cur[0]++ * coef[0];
101 *work_line++ += *in_lines_cur[1]++ * coef[1];
102 }
103}
104
105static void filter_complex_low(int32_t *work_line,
106 uint8_t *in_lines_cur[4],
107 const int16_t *coef, int linesize)
108{
109 int i;
110
111 for (i = 0; i < linesize; i++) {
112 *work_line = *in_lines_cur[0]++ * coef[0];
113 *work_line += *in_lines_cur[1]++ * coef[1];
114 *work_line += *in_lines_cur[2]++ * coef[2];
115 *work_line++ += *in_lines_cur[3]++ * coef[3];
116 }
117}
118
119static void filter_simple_high(int32_t *work_line,
120 uint8_t *in_lines_cur[3],
121 uint8_t *in_lines_adj[3],
122 const int16_t *coef, int linesize)
123{
124 int i;
125
126 for (i = 0; i < linesize; i++) {
127 *work_line += *in_lines_cur[0]++ * coef[0];
128 *work_line += *in_lines_adj[0]++ * coef[0];
129 *work_line += *in_lines_cur[1]++ * coef[1];
130 *work_line += *in_lines_adj[1]++ * coef[1];
131 *work_line += *in_lines_cur[2]++ * coef[2];
132 *work_line++ += *in_lines_adj[2]++ * coef[2];
133 }
134}
135
136static void filter_complex_high(int32_t *work_line,
137 uint8_t *in_lines_cur[5],
138 uint8_t *in_lines_adj[5],
139 const int16_t *coef, int linesize)
140{
141 int i;
142
143 for (i = 0; i < linesize; i++) {
144 *work_line += *in_lines_cur[0]++ * coef[0];
145 *work_line += *in_lines_adj[0]++ * coef[0];
146 *work_line += *in_lines_cur[1]++ * coef[1];
147 *work_line += *in_lines_adj[1]++ * coef[1];
148 *work_line += *in_lines_cur[2]++ * coef[2];
149 *work_line += *in_lines_adj[2]++ * coef[2];
150 *work_line += *in_lines_cur[3]++ * coef[3];
151 *work_line += *in_lines_adj[3]++ * coef[3];
152 *work_line += *in_lines_cur[4]++ * coef[4];
153 *work_line++ += *in_lines_adj[4]++ * coef[4];
154 }
155}
156
157static void filter_scale(uint8_t *out_pixel, const int32_t *work_pixel, int linesize, int max)
158{
159 int j;
160
161 for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
162 *out_pixel = av_clip(*work_pixel, 0, 255 * 256 * 128) >> 15;
163}
164
165static void filter16_simple_low(int32_t *work_line,
166 uint8_t *in_lines_cur8[2],
167 const int16_t *coef, int linesize)
168{
169 uint16_t *in_lines_cur[2] = { (uint16_t *)in_lines_cur8[0], (uint16_t *)in_lines_cur8[1] };
170 int i;
171
172 linesize /= 2;
173 for (i = 0; i < linesize; i++) {
174 *work_line = *in_lines_cur[0]++ * coef[0];
175 *work_line++ += *in_lines_cur[1]++ * coef[1];
176 }
177}
178
179static void filter16_complex_low(int32_t *work_line,
180 uint8_t *in_lines_cur8[4],
181 const int16_t *coef, int linesize)
182{
183 uint16_t *in_lines_cur[4] = { (uint16_t *)in_lines_cur8[0],
184 (uint16_t *)in_lines_cur8[1],
185 (uint16_t *)in_lines_cur8[2],
186 (uint16_t *)in_lines_cur8[3] };
187 int i;
188
189 linesize /= 2;
190 for (i = 0; i < linesize; i++) {
191 *work_line = *in_lines_cur[0]++ * coef[0];
192 *work_line += *in_lines_cur[1]++ * coef[1];
193 *work_line += *in_lines_cur[2]++ * coef[2];
194 *work_line++ += *in_lines_cur[3]++ * coef[3];
195 }
196}
197
198static void filter16_simple_high(int32_t *work_line,
199 uint8_t *in_lines_cur8[3],
200 uint8_t *in_lines_adj8[3],
201 const int16_t *coef, int linesize)
202{
203 uint16_t *in_lines_cur[3] = { (uint16_t *)in_lines_cur8[0],
204 (uint16_t *)in_lines_cur8[1],
205 (uint16_t *)in_lines_cur8[2] };
206 uint16_t *in_lines_adj[3] = { (uint16_t *)in_lines_adj8[0],
207 (uint16_t *)in_lines_adj8[1],
208 (uint16_t *)in_lines_adj8[2] };
209 int i;
210
211 linesize /= 2;
212 for (i = 0; i < linesize; i++) {
213 *work_line += *in_lines_cur[0]++ * coef[0];
214 *work_line += *in_lines_adj[0]++ * coef[0];
215 *work_line += *in_lines_cur[1]++ * coef[1];
216 *work_line += *in_lines_adj[1]++ * coef[1];
217 *work_line += *in_lines_cur[2]++ * coef[2];
218 *work_line++ += *in_lines_adj[2]++ * coef[2];
219 }
220}
221
222static void filter16_complex_high(int32_t *work_line,
223 uint8_t *in_lines_cur8[5],
224 uint8_t *in_lines_adj8[5],
225 const int16_t *coef, int linesize)
226{
227 uint16_t *in_lines_cur[5] = { (uint16_t *)in_lines_cur8[0],
228 (uint16_t *)in_lines_cur8[1],
229 (uint16_t *)in_lines_cur8[2],
230 (uint16_t *)in_lines_cur8[3],
231 (uint16_t *)in_lines_cur8[4] };
232 uint16_t *in_lines_adj[5] = { (uint16_t *)in_lines_adj8[0],
233 (uint16_t *)in_lines_adj8[1],
234 (uint16_t *)in_lines_adj8[2],
235 (uint16_t *)in_lines_adj8[3],
236 (uint16_t *)in_lines_adj8[4] };
237 int i;
238
239 linesize /= 2;
240 for (i = 0; i < linesize; i++) {
241 *work_line += *in_lines_cur[0]++ * coef[0];
242 *work_line += *in_lines_adj[0]++ * coef[0];
243 *work_line += *in_lines_cur[1]++ * coef[1];
244 *work_line += *in_lines_adj[1]++ * coef[1];
245 *work_line += *in_lines_cur[2]++ * coef[2];
246 *work_line += *in_lines_adj[2]++ * coef[2];
247 *work_line += *in_lines_cur[3]++ * coef[3];
248 *work_line += *in_lines_adj[3]++ * coef[3];
249 *work_line += *in_lines_cur[4]++ * coef[4];
250 *work_line++ += *in_lines_adj[4]++ * coef[4];
251 }
252}
253
254static void filter16_scale(uint8_t *out_pixel8, const int32_t *work_pixel, int linesize, int max)
255{
256 uint16_t *out_pixel = (uint16_t *)out_pixel8;
257 int j;
258
259 linesize /= 2;
260 for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
261 *out_pixel = av_clip(*work_pixel, 0, max) >> 15;
262}
263
264static int config_input(AVFilterLink *inlink)
265{
266 AVFilterContext *ctx = inlink->dst;
267 W3FDIFContext *s = ctx->priv;
268 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
269 int ret, i, depth;
270
271 if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
272 return ret;
273
274 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
275 s->planeheight[0] = s->planeheight[3] = inlink->h;
276
277 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
278 s->nb_threads = ff_filter_get_nb_threads(ctx);
279 s->work_line = av_calloc(s->nb_threads, sizeof(*s->work_line));
280 if (!s->work_line)
281 return AVERROR(ENOMEM);
282
283 for (i = 0; i < s->nb_threads; i++) {
284 s->work_line[i] = av_calloc(FFALIGN(s->linesize[0], 32), sizeof(*s->work_line[0]));
285 if (!s->work_line[i])
286 return AVERROR(ENOMEM);
287 }
288
289 depth = desc->comp[0].depth;
290 s->max = ((1 << depth) - 1) * 256 * 128;
291 if (depth <= 8) {
292 s->dsp.filter_simple_low = filter_simple_low;
293 s->dsp.filter_complex_low = filter_complex_low;
294 s->dsp.filter_simple_high = filter_simple_high;
295 s->dsp.filter_complex_high = filter_complex_high;
296 s->dsp.filter_scale = filter_scale;
297 } else {
298 s->dsp.filter_simple_low = filter16_simple_low;
299 s->dsp.filter_complex_low = filter16_complex_low;
300 s->dsp.filter_simple_high = filter16_simple_high;
301 s->dsp.filter_complex_high = filter16_complex_high;
302 s->dsp.filter_scale = filter16_scale;
303 }
304
305 if (ARCH_X86)
306 ff_w3fdif_init_x86(&s->dsp, depth);
307
308 return 0;
309}
310
311static int config_output(AVFilterLink *outlink)
312{
313 AVFilterLink *inlink = outlink->src->inputs[0];
314
315 outlink->time_base.num = inlink->time_base.num;
316 outlink->time_base.den = inlink->time_base.den * 2;
317 outlink->frame_rate.num = inlink->frame_rate.num * 2;
318 outlink->frame_rate.den = inlink->frame_rate.den;
319
320 return 0;
321}
322
323/*
324 * Filter coefficients from PH-2071, scaled by 256 * 128.
325 * Each set of coefficients has a set for low-frequencies and high-frequencies.
326 * n_coef_lf[] and n_coef_hf[] are the number of coefs for simple and more-complex.
327 * It is important for later that n_coef_lf[] is even and n_coef_hf[] is odd.
328 * coef_lf[][] and coef_hf[][] are the coefficients for low-frequencies
329 * and high-frequencies for simple and more-complex mode.
330 */
331static const int8_t n_coef_lf[2] = { 2, 4 };
332static const int16_t coef_lf[2][4] = {{ 16384, 16384, 0, 0},
333 { -852, 17236, 17236, -852}};
334static const int8_t n_coef_hf[2] = { 3, 5 };
335static const int16_t coef_hf[2][5] = {{ -2048, 4096, -2048, 0, 0},
336 { 1016, -3801, 5570, -3801, 1016}};
337
338typedef struct ThreadData {
339 AVFrame *out, *cur, *adj;
340 int plane;
341} ThreadData;
342
343static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
344{
345 W3FDIFContext *s = ctx->priv;
346 ThreadData *td = arg;
347 AVFrame *out = td->out;
348 AVFrame *cur = td->cur;
349 AVFrame *adj = td->adj;
350 const int plane = td->plane;
351 const int filter = s->filter;
352 uint8_t *in_line, *in_lines_cur[5], *in_lines_adj[5];
353 uint8_t *out_line, *out_pixel;
354 int32_t *work_line, *work_pixel;
355 uint8_t *cur_data = cur->data[plane];
356 uint8_t *adj_data = adj->data[plane];
357 uint8_t *dst_data = out->data[plane];
358 const int linesize = s->linesize[plane];
359 const int height = s->planeheight[plane];
360 const int cur_line_stride = cur->linesize[plane];
361 const int adj_line_stride = adj->linesize[plane];
362 const int dst_line_stride = out->linesize[plane];
363 const int start = (height * jobnr) / nb_jobs;
364 const int end = (height * (jobnr+1)) / nb_jobs;
365 const int max = s->max;
366 int j, y_in, y_out;
367
368 /* copy unchanged the lines of the field */
369 y_out = start + (s->field == cur->top_field_first) - (start & 1);
370
371 in_line = cur_data + (y_out * cur_line_stride);
372 out_line = dst_data + (y_out * dst_line_stride);
373
374 while (y_out < end) {
375 memcpy(out_line, in_line, linesize);
376 y_out += 2;
377 in_line += cur_line_stride * 2;
378 out_line += dst_line_stride * 2;
379 }
380
381 /* interpolate other lines of the field */
382 y_out = start + (s->field != cur->top_field_first) - (start & 1);
383
384 out_line = dst_data + (y_out * dst_line_stride);
385
386 while (y_out < end) {
387 /* get low vertical frequencies from current field */
388 for (j = 0; j < n_coef_lf[filter]; j++) {
389 y_in = (y_out + 1) + (j * 2) - n_coef_lf[filter];
390
391 while (y_in < 0)
392 y_in += 2;
393 while (y_in >= height)
394 y_in -= 2;
395
396 in_lines_cur[j] = cur_data + (y_in * cur_line_stride);
397 }
398
399 work_line = s->work_line[jobnr];
400 switch (n_coef_lf[filter]) {
401 case 2:
402 s->dsp.filter_simple_low(work_line, in_lines_cur,
403 coef_lf[filter], linesize);
404 break;
405 case 4:
406 s->dsp.filter_complex_low(work_line, in_lines_cur,
407 coef_lf[filter], linesize);
408 }
409
410 /* get high vertical frequencies from adjacent fields */
411 for (j = 0; j < n_coef_hf[filter]; j++) {
412 y_in = (y_out + 1) + (j * 2) - n_coef_hf[filter];
413
414 while (y_in < 0)
415 y_in += 2;
416 while (y_in >= height)
417 y_in -= 2;
418
419 in_lines_cur[j] = cur_data + (y_in * cur_line_stride);
420 in_lines_adj[j] = adj_data + (y_in * adj_line_stride);
421 }
422
423 work_line = s->work_line[jobnr];
424 switch (n_coef_hf[filter]) {
425 case 3:
426 s->dsp.filter_simple_high(work_line, in_lines_cur, in_lines_adj,
427 coef_hf[filter], linesize);
428 break;
429 case 5:
430 s->dsp.filter_complex_high(work_line, in_lines_cur, in_lines_adj,
431 coef_hf[filter], linesize);
432 }
433
434 /* save scaled result to the output frame, scaling down by 256 * 128 */
435 work_pixel = s->work_line[jobnr];
436 out_pixel = out_line;
437
438 s->dsp.filter_scale(out_pixel, work_pixel, linesize, max);
439
440 /* move on to next line */
441 y_out += 2;
442 out_line += dst_line_stride * 2;
443 }
444
445 return 0;
446}
447
448static int filter(AVFilterContext *ctx, int is_second)
449{
450 W3FDIFContext *s = ctx->priv;
451 AVFilterLink *outlink = ctx->outputs[0];
452 AVFrame *out, *adj;
453 ThreadData td;
454 int plane;
455
456 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
457 if (!out)
458 return AVERROR(ENOMEM);
459 av_frame_copy_props(out, s->cur);
460 out->interlaced_frame = 0;
461
462 if (!is_second) {
463 if (out->pts != AV_NOPTS_VALUE)
464 out->pts *= 2;
465 } else {
466 int64_t cur_pts = s->cur->pts;
467 int64_t next_pts = s->next->pts;
468
469 if (next_pts != AV_NOPTS_VALUE && cur_pts != AV_NOPTS_VALUE) {
470 out->pts = cur_pts + next_pts;
471 } else {
472 out->pts = AV_NOPTS_VALUE;
473 }
474 }
475
476 adj = s->field ? s->next : s->prev;
477 td.out = out; td.cur = s->cur; td.adj = adj;
478 for (plane = 0; plane < s->nb_planes; plane++) {
479 td.plane = plane;
480 ctx->internal->execute(ctx, deinterlace_slice, &td, NULL, FFMIN(s->planeheight[plane], s->nb_threads));
481 }
482
483 s->field = !s->field;
484
485 return ff_filter_frame(outlink, out);
486}
487
488static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
489{
490 AVFilterContext *ctx = inlink->dst;
491 W3FDIFContext *s = ctx->priv;
492 int ret;
493
494 av_frame_free(&s->prev);
495 s->prev = s->cur;
496 s->cur = s->next;
497 s->next = frame;
498
499 if (!s->cur) {
500 s->cur = av_frame_clone(s->next);
501 if (!s->cur)
502 return AVERROR(ENOMEM);
503 }
504
505 if ((s->deint && !s->cur->interlaced_frame) || ctx->is_disabled) {
506 AVFrame *out = av_frame_clone(s->cur);
507 if (!out)
508 return AVERROR(ENOMEM);
509
510 av_frame_free(&s->prev);
511 if (out->pts != AV_NOPTS_VALUE)
512 out->pts *= 2;
513 return ff_filter_frame(ctx->outputs[0], out);
514 }
515
516 if (!s->prev)
517 return 0;
518
519 ret = filter(ctx, 0);
520 if (ret < 0)
521 return ret;
522
523 return filter(ctx, 1);
524}
525
526static int request_frame(AVFilterLink *outlink)
527{
528 AVFilterContext *ctx = outlink->src;
529 W3FDIFContext *s = ctx->priv;
530 int ret;
531
532 if (s->eof)
533 return AVERROR_EOF;
534
535 ret = ff_request_frame(ctx->inputs[0]);
536
537 if (ret == AVERROR_EOF && s->cur) {
538 AVFrame *next = av_frame_clone(s->next);
539 if (!next)
540 return AVERROR(ENOMEM);
541 next->pts = s->next->pts * 2 - s->cur->pts;
542 filter_frame(ctx->inputs[0], next);
543 s->eof = 1;
544 } else if (ret < 0) {
545 return ret;
546 }
547
548 return 0;
549}
550
551static av_cold void uninit(AVFilterContext *ctx)
552{
553 W3FDIFContext *s = ctx->priv;
554 int i;
555
556 av_frame_free(&s->prev);
557 av_frame_free(&s->cur );
558 av_frame_free(&s->next);
559
560 for (i = 0; i < s->nb_threads; i++)
561 av_freep(&s->work_line[i]);
562
563 av_freep(&s->work_line);
564}
565
566static const AVFilterPad w3fdif_inputs[] = {
567 {
568 .name = "default",
569 .type = AVMEDIA_TYPE_VIDEO,
570 .filter_frame = filter_frame,
571 .config_props = config_input,
572 },
573 { NULL }
574};
575
576static const AVFilterPad w3fdif_outputs[] = {
577 {
578 .name = "default",
579 .type = AVMEDIA_TYPE_VIDEO,
580 .config_props = config_output,
581 .request_frame = request_frame,
582 },
583 { NULL }
584};
585
586AVFilter ff_vf_w3fdif = {
587 .name = "w3fdif",
588 .description = NULL_IF_CONFIG_SMALL("Apply Martin Weston three field deinterlace."),
589 .priv_size = sizeof(W3FDIFContext),
590 .priv_class = &w3fdif_class,
591 .uninit = uninit,
592 .query_formats = query_formats,
593 .inputs = w3fdif_inputs,
594 .outputs = w3fdif_outputs,
595 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
596};
597
generated by cgit at 2024-05-17 06:08:21 (GMT)