2 * Copyright (c) 2012-2013 Oka Motofumi (chikuzen.mo at gmail dot com)
3 * Copyright (c) 2015 Paul B Mahol
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "libavutil/avstring.h"
23 #include "libavutil/imgutils.h"
24 #include "libavutil/opt.h"
25 #include "libavutil/pixdesc.h"
31 typedef struct ConvolutionContext {
55 int (*filter[4])(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
58 #define OFFSET(x) offsetof(ConvolutionContext, x)
59 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
61 static const AVOption convolution_options[] = {
62 { "0m", "set matrix for 1st plane", OFFSET(matrix_str[0]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS },
63 { "1m", "set matrix for 2nd plane", OFFSET(matrix_str[1]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS },
64 { "2m", "set matrix for 3rd plane", OFFSET(matrix_str[2]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS },
65 { "3m", "set matrix for 4th plane", OFFSET(matrix_str[3]), AV_OPT_TYPE_STRING, {.str="0 0 0 0 1 0 0 0 0"}, 0, 0, FLAGS },
66 { "0rdiv", "set rdiv for 1st plane", OFFSET(rdiv[0]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS},
67 { "1rdiv", "set rdiv for 2nd plane", OFFSET(rdiv[1]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS},
68 { "2rdiv", "set rdiv for 3rd plane", OFFSET(rdiv[2]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS},
69 { "3rdiv", "set rdiv for 4th plane", OFFSET(rdiv[3]), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, INT_MAX, FLAGS},
70 { "0bias", "set bias for 1st plane", OFFSET(bias[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
71 { "1bias", "set bias for 2nd plane", OFFSET(bias[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
72 { "2bias", "set bias for 3rd plane", OFFSET(bias[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
73 { "3bias", "set bias for 4th plane", OFFSET(bias[3]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
77 AVFILTER_DEFINE_CLASS(convolution);
79 static const int same3x3[9] = {0, 0, 0,
83 static const int same5x5[25] = {0, 0, 0, 0, 0,
89 static int query_formats(AVFilterContext *ctx)
91 static const enum AVPixelFormat pix_fmts[] = {
92 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P,
93 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
94 AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUV420P,
95 AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
96 AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
97 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
98 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
99 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12,
100 AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
101 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
102 AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA444P9,
103 AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA444P10,
104 AV_PIX_FMT_YUVA420P16, AV_PIX_FMT_YUVA422P16, AV_PIX_FMT_YUVA444P16,
105 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
106 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
107 AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
108 AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY16,
112 return ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
115 static inline void line_copy8(uint8_t *line, const uint8_t *srcp, int width, int mergin)
119 memcpy(line, srcp, width);
121 for (i = mergin; i > 0; i--) {
123 line[width - 1 + i] = line[width - 1 - i];
127 static inline void line_copy16(uint16_t *line, const uint16_t *srcp, int width, int mergin)
131 memcpy(line, srcp, width * 2);
133 for (i = mergin; i > 0; i--) {
135 line[width - 1 + i] = line[width - 1 - i];
139 typedef struct ThreadData {
144 static int filter16_prewitt(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
146 ConvolutionContext *s = ctx->priv;
147 ThreadData *td = arg;
148 AVFrame *in = td->in;
149 AVFrame *out = td->out;
150 const int plane = td->plane;
151 const int peak = (1 << s->depth) - 1;
152 const int stride = in->linesize[plane] / 2;
153 const int bstride = s->bstride;
154 const int height = s->planeheight[plane];
155 const int width = s->planewidth[plane];
156 const int slice_start = (height * jobnr) / nb_jobs;
157 const int slice_end = (height * (jobnr+1)) / nb_jobs;
158 const uint16_t *src = (const uint16_t *)in->data[plane] + slice_start * stride;
159 uint16_t *dst = (uint16_t *)out->data[plane] + slice_start * (out->linesize[plane] / 2);
160 const float scale = s->scale;
161 const float delta = s->delta;
162 uint16_t *p0 = (uint16_t *)s->bptrs[jobnr] + 16;
163 uint16_t *p1 = p0 + bstride;
164 uint16_t *p2 = p1 + bstride;
165 uint16_t *orig = p0, *end = p2;
168 line_copy16(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
169 line_copy16(p1, src, width, 1);
171 for (y = slice_start; y < slice_end; y++) {
172 src += stride * (y < height - 1 ? 1 : -1);
173 line_copy16(p2, src, width, 1);
175 for (x = 0; x < width; x++) {
176 int suma = p0[x - 1] * -1 +
182 int sumb = p0[x - 1] * -1 +
189 dst[x] = av_clip(sqrt(suma*suma + sumb*sumb) * scale + delta, 0, peak);
194 p2 = (p2 == end) ? orig: p2 + bstride;
195 dst += out->linesize[plane] / 2;
201 static int filter16_roberts(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
203 ConvolutionContext *s = ctx->priv;
204 ThreadData *td = arg;
205 AVFrame *in = td->in;
206 AVFrame *out = td->out;
207 const int plane = td->plane;
208 const int peak = (1 << s->depth) - 1;
209 const int stride = in->linesize[plane] / 2;
210 const int bstride = s->bstride;
211 const int height = s->planeheight[plane];
212 const int width = s->planewidth[plane];
213 const int slice_start = (height * jobnr) / nb_jobs;
214 const int slice_end = (height * (jobnr+1)) / nb_jobs;
215 const uint16_t *src = (const uint16_t *)in->data[plane] + slice_start * stride;
216 uint16_t *dst = (uint16_t *)out->data[plane] + slice_start * (out->linesize[plane] / 2);
217 const float scale = s->scale;
218 const float delta = s->delta;
219 uint16_t *p0 = (uint16_t *)s->bptrs[jobnr] + 16;
220 uint16_t *p1 = p0 + bstride;
221 uint16_t *p2 = p1 + bstride;
222 uint16_t *orig = p0, *end = p2;
225 line_copy16(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
226 line_copy16(p1, src, width, 1);
228 for (y = slice_start; y < slice_end; y++) {
229 src += stride * (y < height - 1 ? 1 : -1);
230 line_copy16(p2, src, width, 1);
232 for (x = 0; x < width; x++) {
233 int suma = p0[x - 1] * 1 +
235 int sumb = p0[x ] * 1 +
238 dst[x] = av_clip(sqrt(suma*suma + sumb*sumb) * scale + delta, 0, peak);
243 p2 = (p2 == end) ? orig: p2 + bstride;
244 dst += out->linesize[plane] / 2;
250 static int filter16_sobel(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
252 ConvolutionContext *s = ctx->priv;
253 ThreadData *td = arg;
254 AVFrame *in = td->in;
255 AVFrame *out = td->out;
256 const int plane = td->plane;
257 const int peak = (1 << s->depth) - 1;
258 const int stride = in->linesize[plane] / 2;
259 const int bstride = s->bstride;
260 const int height = s->planeheight[plane];
261 const int width = s->planewidth[plane];
262 const int slice_start = (height * jobnr) / nb_jobs;
263 const int slice_end = (height * (jobnr+1)) / nb_jobs;
264 const uint16_t *src = (const uint16_t *)in->data[plane] + slice_start * stride;
265 uint16_t *dst = (uint16_t *)out->data[plane] + slice_start * (out->linesize[plane] / 2);
266 const float scale = s->scale;
267 const float delta = s->delta;
268 uint16_t *p0 = (uint16_t *)s->bptrs[jobnr] + 16;
269 uint16_t *p1 = p0 + bstride;
270 uint16_t *p2 = p1 + bstride;
271 uint16_t *orig = p0, *end = p2;
274 line_copy16(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
275 line_copy16(p1, src, width, 1);
277 for (y = slice_start; y < slice_end; y++) {
278 src += stride * (y < height - 1 ? 1 : -1);
279 line_copy16(p2, src, width, 1);
281 for (x = 0; x < width; x++) {
282 int suma = p0[x - 1] * -1 +
288 int sumb = p0[x - 1] * -1 +
295 dst[x] = av_clip(sqrt(suma*suma + sumb*sumb) * scale + delta, 0, peak);
300 p2 = (p2 == end) ? orig: p2 + bstride;
301 dst += out->linesize[plane] / 2;
307 static int filter_prewitt(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
309 ConvolutionContext *s = ctx->priv;
310 ThreadData *td = arg;
311 AVFrame *in = td->in;
312 AVFrame *out = td->out;
313 const int plane = td->plane;
314 const int stride = in->linesize[plane];
315 const int bstride = s->bstride;
316 const int height = s->planeheight[plane];
317 const int width = s->planewidth[plane];
318 const int slice_start = (height * jobnr) / nb_jobs;
319 const int slice_end = (height * (jobnr+1)) / nb_jobs;
320 const uint8_t *src = in->data[plane] + slice_start * stride;
321 uint8_t *dst = out->data[plane] + slice_start * out->linesize[plane];
322 const float scale = s->scale;
323 const float delta = s->delta;
324 uint8_t *p0 = s->bptrs[jobnr] + 16;
325 uint8_t *p1 = p0 + bstride;
326 uint8_t *p2 = p1 + bstride;
327 uint8_t *orig = p0, *end = p2;
330 line_copy8(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
331 line_copy8(p1, src, width, 1);
333 for (y = slice_start; y < slice_end; y++) {
334 src += stride * (y < height - 1 ? 1 : -1);
335 line_copy8(p2, src, width, 1);
337 for (x = 0; x < width; x++) {
338 int suma = p0[x - 1] * -1 +
344 int sumb = p0[x - 1] * -1 +
351 dst[x] = av_clip_uint8(sqrt(suma*suma + sumb*sumb) * scale + delta);
356 p2 = (p2 == end) ? orig: p2 + bstride;
357 dst += out->linesize[plane];
363 static int filter_roberts(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
365 ConvolutionContext *s = ctx->priv;
366 ThreadData *td = arg;
367 AVFrame *in = td->in;
368 AVFrame *out = td->out;
369 const int plane = td->plane;
370 const int stride = in->linesize[plane];
371 const int bstride = s->bstride;
372 const int height = s->planeheight[plane];
373 const int width = s->planewidth[plane];
374 const int slice_start = (height * jobnr) / nb_jobs;
375 const int slice_end = (height * (jobnr+1)) / nb_jobs;
376 const uint8_t *src = in->data[plane] + slice_start * stride;
377 uint8_t *dst = out->data[plane] + slice_start * out->linesize[plane];
378 const float scale = s->scale;
379 const float delta = s->delta;
380 uint8_t *p0 = s->bptrs[jobnr] + 16;
381 uint8_t *p1 = p0 + bstride;
382 uint8_t *p2 = p1 + bstride;
383 uint8_t *orig = p0, *end = p2;
386 line_copy8(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
387 line_copy8(p1, src, width, 1);
389 for (y = slice_start; y < slice_end; y++) {
390 src += stride * (y < height - 1 ? 1 : -1);
391 line_copy8(p2, src, width, 1);
393 for (x = 0; x < width; x++) {
394 int suma = p0[x - 1] * 1 +
396 int sumb = p0[x ] * 1 +
399 dst[x] = av_clip_uint8(sqrt(suma*suma + sumb*sumb) * scale + delta);
404 p2 = (p2 == end) ? orig: p2 + bstride;
405 dst += out->linesize[plane];
411 static int filter_sobel(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
413 ConvolutionContext *s = ctx->priv;
414 ThreadData *td = arg;
415 AVFrame *in = td->in;
416 AVFrame *out = td->out;
417 const int plane = td->plane;
418 const int stride = in->linesize[plane];
419 const int bstride = s->bstride;
420 const int height = s->planeheight[plane];
421 const int width = s->planewidth[plane];
422 const int slice_start = (height * jobnr) / nb_jobs;
423 const int slice_end = (height * (jobnr+1)) / nb_jobs;
424 const uint8_t *src = in->data[plane] + slice_start * stride;
425 uint8_t *dst = out->data[plane] + slice_start * out->linesize[plane];
426 const float scale = s->scale;
427 const float delta = s->delta;
428 uint8_t *p0 = s->bptrs[jobnr] + 16;
429 uint8_t *p1 = p0 + bstride;
430 uint8_t *p2 = p1 + bstride;
431 uint8_t *orig = p0, *end = p2;
434 line_copy8(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
435 line_copy8(p1, src, width, 1);
437 for (y = slice_start; y < slice_end; y++) {
438 src += stride * (y < height - 1 ? 1 : -1);
439 line_copy8(p2, src, width, 1);
441 for (x = 0; x < width; x++) {
442 int suma = p0[x - 1] * -1 +
448 int sumb = p0[x - 1] * -1 +
455 dst[x] = av_clip_uint8(sqrt(suma*suma + sumb*sumb) * scale + delta);
460 p2 = (p2 == end) ? orig: p2 + bstride;
461 dst += out->linesize[plane];
467 static int filter16_3x3(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
469 ConvolutionContext *s = ctx->priv;
470 ThreadData *td = arg;
471 AVFrame *in = td->in;
472 AVFrame *out = td->out;
473 const int plane = td->plane;
474 const int peak = (1 << s->depth) - 1;
475 const int stride = in->linesize[plane] / 2;
476 const int bstride = s->bstride;
477 const int height = s->planeheight[plane];
478 const int width = s->planewidth[plane];
479 const int slice_start = (height * jobnr) / nb_jobs;
480 const int slice_end = (height * (jobnr+1)) / nb_jobs;
481 const uint16_t *src = (const uint16_t *)in->data[plane] + slice_start * stride;
482 uint16_t *dst = (uint16_t *)out->data[plane] + slice_start * (out->linesize[plane] / 2);
483 uint16_t *p0 = (uint16_t *)s->bptrs[jobnr] + 16;
484 uint16_t *p1 = p0 + bstride;
485 uint16_t *p2 = p1 + bstride;
486 uint16_t *orig = p0, *end = p2;
487 const int *matrix = s->matrix[plane];
488 const float rdiv = s->rdiv[plane];
489 const float bias = s->bias[plane];
492 line_copy16(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
493 line_copy16(p1, src, width, 1);
495 for (y = slice_start; y < slice_end; y++) {
496 src += stride * (y < height - 1 ? 1 : -1);
497 line_copy16(p2, src, width, 1);
499 for (x = 0; x < width; x++) {
500 int sum = p0[x - 1] * matrix[0] +
502 p0[x + 1] * matrix[2] +
503 p1[x - 1] * matrix[3] +
505 p1[x + 1] * matrix[5] +
506 p2[x - 1] * matrix[6] +
508 p2[x + 1] * matrix[8];
509 sum = (int)(sum * rdiv + bias + 0.5f);
510 dst[x] = av_clip(sum, 0, peak);
515 p2 = (p2 == end) ? orig: p2 + bstride;
516 dst += out->linesize[plane] / 2;
522 static int filter16_5x5(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
524 ConvolutionContext *s = ctx->priv;
525 ThreadData *td = arg;
526 AVFrame *in = td->in;
527 AVFrame *out = td->out;
528 const int plane = td->plane;
529 const int peak = (1 << s->depth) - 1;
530 const int stride = in->linesize[plane] / 2;
531 const int bstride = s->bstride;
532 const int height = s->planeheight[plane];
533 const int width = s->planewidth[plane];
534 const int slice_start = (height * jobnr) / nb_jobs;
535 const int slice_end = (height * (jobnr+1)) / nb_jobs;
536 const uint16_t *src = (const uint16_t *)in->data[plane] + slice_start * stride;
537 uint16_t *dst = (uint16_t *)out->data[plane] + slice_start * (out->linesize[plane] / 2);
538 uint16_t *p0 = (uint16_t *)s->bptrs[jobnr] + 16;
539 uint16_t *p1 = p0 + bstride;
540 uint16_t *p2 = p1 + bstride;
541 uint16_t *p3 = p2 + bstride;
542 uint16_t *p4 = p3 + bstride;
543 uint16_t *orig = p0, *end = p4;
544 const int *matrix = s->matrix[plane];
545 float rdiv = s->rdiv[plane];
546 float bias = s->bias[plane];
549 line_copy16(p0, src + 2 * stride * (slice_start < 2 ? 1 : -1), width, 2);
550 line_copy16(p1, src + stride * (slice_start == 0 ? 1 : -1), width, 2);
551 line_copy16(p2, src, width, 2);
553 line_copy16(p3, src, width, 2);
555 for (y = slice_start; y < slice_end; y++) {
556 uint16_t *array[] = {
557 p0 - 2, p0 - 1, p0, p0 + 1, p0 + 2,
558 p1 - 2, p1 - 1, p1, p1 + 1, p1 + 2,
559 p2 - 2, p2 - 1, p2, p2 + 1, p2 + 2,
560 p3 - 2, p3 - 1, p3, p3 + 1, p3 + 2,
561 p4 - 2, p4 - 1, p4, p4 + 1, p4 + 2
564 src += stride * (y < height - 2 ? 1 : -1);
565 line_copy16(p4, src, width, 2);
567 for (x = 0; x < width; x++) {
570 for (i = 0; i < 25; i++) {
571 sum += *(array[i] + x) * matrix[i];
573 sum = (int)(sum * rdiv + bias + 0.5f);
574 dst[x] = av_clip(sum, 0, peak);
581 p4 = (p4 == end) ? orig: p4 + bstride;
582 dst += out->linesize[plane] / 2;
588 static int filter_3x3(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
590 ConvolutionContext *s = ctx->priv;
591 ThreadData *td = arg;
592 AVFrame *in = td->in;
593 AVFrame *out = td->out;
594 const int plane = td->plane;
595 const int stride = in->linesize[plane];
596 const int bstride = s->bstride;
597 const int height = s->planeheight[plane];
598 const int width = s->planewidth[plane];
599 const int slice_start = (height * jobnr) / nb_jobs;
600 const int slice_end = (height * (jobnr+1)) / nb_jobs;
601 const uint8_t *src = in->data[plane] + slice_start * stride;
602 uint8_t *dst = out->data[plane] + slice_start * out->linesize[plane];
603 uint8_t *p0 = s->bptrs[jobnr] + 16;
604 uint8_t *p1 = p0 + bstride;
605 uint8_t *p2 = p1 + bstride;
606 uint8_t *orig = p0, *end = p2;
607 const int *matrix = s->matrix[plane];
608 const float rdiv = s->rdiv[plane];
609 const float bias = s->bias[plane];
612 line_copy8(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
613 line_copy8(p1, src, width, 1);
615 for (y = slice_start; y < slice_end; y++) {
616 src += stride * (y < height - 1 ? 1 : -1);
617 line_copy8(p2, src, width, 1);
619 for (x = 0; x < width; x++) {
620 int sum = p0[x - 1] * matrix[0] +
622 p0[x + 1] * matrix[2] +
623 p1[x - 1] * matrix[3] +
625 p1[x + 1] * matrix[5] +
626 p2[x - 1] * matrix[6] +
628 p2[x + 1] * matrix[8];
629 sum = (int)(sum * rdiv + bias + 0.5f);
630 dst[x] = av_clip_uint8(sum);
635 p2 = (p2 == end) ? orig: p2 + bstride;
636 dst += out->linesize[plane];
642 static int filter_5x5(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
644 ConvolutionContext *s = ctx->priv;
645 ThreadData *td = arg;
646 AVFrame *in = td->in;
647 AVFrame *out = td->out;
648 const int plane = td->plane;
649 const int stride = in->linesize[plane];
650 const int bstride = s->bstride;
651 const int height = s->planeheight[plane];
652 const int width = s->planewidth[plane];
653 const int slice_start = (height * jobnr) / nb_jobs;
654 const int slice_end = (height * (jobnr+1)) / nb_jobs;
655 const uint8_t *src = in->data[plane] + slice_start * stride;
656 uint8_t *dst = out->data[plane] + slice_start * out->linesize[plane];
657 uint8_t *p0 = s->bptrs[jobnr] + 16;
658 uint8_t *p1 = p0 + bstride;
659 uint8_t *p2 = p1 + bstride;
660 uint8_t *p3 = p2 + bstride;
661 uint8_t *p4 = p3 + bstride;
662 uint8_t *orig = p0, *end = p4;
663 const int *matrix = s->matrix[plane];
664 float rdiv = s->rdiv[plane];
665 float bias = s->bias[plane];
668 line_copy8(p0, src + 2 * stride * (slice_start < 2 ? 1 : -1), width, 2);
669 line_copy8(p1, src + stride * (slice_start == 0 ? 1 : -1), width, 2);
670 line_copy8(p2, src, width, 2);
672 line_copy8(p3, src, width, 2);
675 for (y = slice_start; y < slice_end; y++) {
677 p0 - 2, p0 - 1, p0, p0 + 1, p0 + 2,
678 p1 - 2, p1 - 1, p1, p1 + 1, p1 + 2,
679 p2 - 2, p2 - 1, p2, p2 + 1, p2 + 2,
680 p3 - 2, p3 - 1, p3, p3 + 1, p3 + 2,
681 p4 - 2, p4 - 1, p4, p4 + 1, p4 + 2
684 src += stride * (y < height - 2 ? 1 : -1);
685 line_copy8(p4, src, width, 2);
687 for (x = 0; x < width; x++) {
690 for (i = 0; i < 25; i++) {
691 sum += *(array[i] + x) * matrix[i];
693 sum = (int)(sum * rdiv + bias + 0.5f);
694 dst[x] = av_clip_uint8(sum);
701 p4 = (p4 == end) ? orig: p4 + bstride;
702 dst += out->linesize[plane];
708 static int config_input(AVFilterLink *inlink)
710 AVFilterContext *ctx = inlink->dst;
711 ConvolutionContext *s = ctx->priv;
712 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
715 s->depth = desc->comp[0].depth;
717 s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
718 s->planewidth[0] = s->planewidth[3] = inlink->w;
719 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
720 s->planeheight[0] = s->planeheight[3] = inlink->h;
722 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
723 s->nb_threads = ff_filter_get_nb_threads(ctx);
724 s->bptrs = av_calloc(s->nb_threads, sizeof(*s->bptrs));
726 return AVERROR(ENOMEM);
728 s->bstride = s->planewidth[0] + 32;
729 s->bpc = (s->depth + 7) / 8;
730 s->buffer = av_malloc_array(5 * s->bstride * s->nb_threads, s->bpc);
732 return AVERROR(ENOMEM);
734 for (p = 0; p < s->nb_threads; p++) {
735 s->bptrs[p] = s->buffer + 5 * s->bstride * s->bpc * p;
738 if (!strcmp(ctx->filter->name, "convolution")) {
740 for (p = 0; p < s->nb_planes; p++) {
742 s->filter[p] = filter16_3x3;
743 else if (s->size[p] == 5)
744 s->filter[p] = filter16_5x5;
747 } else if (!strcmp(ctx->filter->name, "prewitt")) {
749 for (p = 0; p < s->nb_planes; p++)
750 s->filter[p] = filter16_prewitt;
751 } else if (!strcmp(ctx->filter->name, "roberts")) {
753 for (p = 0; p < s->nb_planes; p++)
754 s->filter[p] = filter16_roberts;
755 } else if (!strcmp(ctx->filter->name, "sobel")) {
757 for (p = 0; p < s->nb_planes; p++)
758 s->filter[p] = filter16_sobel;
764 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
766 AVFilterContext *ctx = inlink->dst;
767 ConvolutionContext *s = ctx->priv;
768 AVFilterLink *outlink = ctx->outputs[0];
772 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
775 return AVERROR(ENOMEM);
777 av_frame_copy_props(out, in);
779 for (plane = 0; plane < s->nb_planes; plane++) {
782 if (s->copy[plane]) {
783 av_image_copy_plane(out->data[plane], out->linesize[plane],
784 in->data[plane], in->linesize[plane],
785 s->planewidth[plane] * s->bpc,
786 s->planeheight[plane]);
793 ctx->internal->execute(ctx, s->filter[plane], &td, NULL, FFMIN(s->planeheight[plane], s->nb_threads));
797 return ff_filter_frame(outlink, out);
800 static av_cold int init(AVFilterContext *ctx)
802 ConvolutionContext *s = ctx->priv;
805 if (!strcmp(ctx->filter->name, "convolution")) {
806 for (i = 0; i < 4; i++) {
807 int *matrix = (int *)s->matrix[i];
808 char *p, *arg, *saveptr = NULL;
810 p = s->matrix_str[i];
811 while (s->matrix_length[i] < 25) {
812 if (!(arg = av_strtok(p, " ", &saveptr)))
816 sscanf(arg, "%d", &matrix[s->matrix_length[i]]);
817 s->matrix_length[i]++;
820 if (s->matrix_length[i] == 9) {
822 if (!memcmp(matrix, same3x3, sizeof(same3x3)))
825 s->filter[i] = filter_3x3;
826 } else if (s->matrix_length[i] == 25) {
828 if (!memcmp(matrix, same5x5, sizeof(same5x5)))
831 s->filter[i] = filter_5x5;
833 return AVERROR(EINVAL);
836 } else if (!strcmp(ctx->filter->name, "prewitt")) {
837 for (i = 0; i < 4; i++) {
838 if ((1 << i) & s->planes)
839 s->filter[i] = filter_prewitt;
843 } else if (!strcmp(ctx->filter->name, "roberts")) {
844 for (i = 0; i < 4; i++) {
845 if ((1 << i) & s->planes)
846 s->filter[i] = filter_roberts;
850 } else if (!strcmp(ctx->filter->name, "sobel")) {
851 for (i = 0; i < 4; i++) {
852 if ((1 << i) & s->planes)
853 s->filter[i] = filter_sobel;
862 static av_cold void uninit(AVFilterContext *ctx)
864 ConvolutionContext *s = ctx->priv;
867 av_freep(&s->buffer);
870 static const AVFilterPad convolution_inputs[] = {
873 .type = AVMEDIA_TYPE_VIDEO,
874 .config_props = config_input,
875 .filter_frame = filter_frame,
880 static const AVFilterPad convolution_outputs[] = {
883 .type = AVMEDIA_TYPE_VIDEO,
888 #if CONFIG_CONVOLUTION_FILTER
890 AVFilter ff_vf_convolution = {
891 .name = "convolution",
892 .description = NULL_IF_CONFIG_SMALL("Apply convolution filter."),
893 .priv_size = sizeof(ConvolutionContext),
894 .priv_class = &convolution_class,
897 .query_formats = query_formats,
898 .inputs = convolution_inputs,
899 .outputs = convolution_outputs,
900 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
903 #endif /* CONFIG_CONVOLUTION_FILTER */
905 #if CONFIG_PREWITT_FILTER
907 static const AVOption prewitt_options[] = {
908 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=15}, 0, 15, FLAGS},
909 { "scale", "set scale", OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, 65535, FLAGS},
910 { "delta", "set delta", OFFSET(delta), AV_OPT_TYPE_FLOAT, {.dbl=0}, -65535, 65535, FLAGS},
914 AVFILTER_DEFINE_CLASS(prewitt);
916 AVFilter ff_vf_prewitt = {
918 .description = NULL_IF_CONFIG_SMALL("Apply prewitt operator."),
919 .priv_size = sizeof(ConvolutionContext),
920 .priv_class = &prewitt_class,
923 .query_formats = query_formats,
924 .inputs = convolution_inputs,
925 .outputs = convolution_outputs,
926 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
929 #endif /* CONFIG_PREWITT_FILTER */
931 #if CONFIG_SOBEL_FILTER
933 static const AVOption sobel_options[] = {
934 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=15}, 0, 15, FLAGS},
935 { "scale", "set scale", OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, 65535, FLAGS},
936 { "delta", "set delta", OFFSET(delta), AV_OPT_TYPE_FLOAT, {.dbl=0}, -65535, 65535, FLAGS},
940 AVFILTER_DEFINE_CLASS(sobel);
942 AVFilter ff_vf_sobel = {
944 .description = NULL_IF_CONFIG_SMALL("Apply sobel operator."),
945 .priv_size = sizeof(ConvolutionContext),
946 .priv_class = &sobel_class,
949 .query_formats = query_formats,
950 .inputs = convolution_inputs,
951 .outputs = convolution_outputs,
952 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
955 #endif /* CONFIG_SOBEL_FILTER */
957 #if CONFIG_ROBERTS_FILTER
959 static const AVOption roberts_options[] = {
960 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=15}, 0, 15, FLAGS},
961 { "scale", "set scale", OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, 65535, FLAGS},
962 { "delta", "set delta", OFFSET(delta), AV_OPT_TYPE_FLOAT, {.dbl=0}, -65535, 65535, FLAGS},
966 AVFILTER_DEFINE_CLASS(roberts);
968 AVFilter ff_vf_roberts = {
970 .description = NULL_IF_CONFIG_SMALL("Apply roberts cross operator."),
971 .priv_size = sizeof(ConvolutionContext),
972 .priv_class = &roberts_class,
975 .query_formats = query_formats,
976 .inputs = convolution_inputs,
977 .outputs = convolution_outputs,
978 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
981 #endif /* CONFIG_ROBERTS_FILTER */