FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavfilter/vf_convolution.c
Date: 2026-07-18 19:31:15
Exec Total Coverage
Lines: 0 455 0.0%
Functions: 0 27 0.0%
Branches: 0 206 0.0%

Line Branch Exec Source
1 /*
2 * Copyright (c) 2012-2013 Oka Motofumi (chikuzen.mo at gmail dot com)
3 * Copyright (c) 2015 Paul B Mahol
4 *
5 * This file is part of FFmpeg.
6 *
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.
11 *
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.
16 *
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
20 */
21
22 #include "config_components.h"
23
24 #include "libavutil/avstring.h"
25 #include "libavutil/imgutils.h"
26 #include "libavutil/intreadwrite.h"
27 #include "libavutil/mem.h"
28 #include "libavutil/mem_internal.h"
29 #include "libavutil/opt.h"
30 #include "libavutil/pixdesc.h"
31 #include "avfilter.h"
32 #include "convolution.h"
33 #include "filters.h"
34 #include "video.h"
35
36 #define OFFSET(x) offsetof(ConvolutionContext, x)
37 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
38
39 static const AVOption convolution_options[] = {
40 { "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 },
41 { "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 },
42 { "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 },
43 { "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 },
44 { "0rdiv", "set rdiv for 1st plane", OFFSET(user_rdiv[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
45 { "1rdiv", "set rdiv for 2nd plane", OFFSET(user_rdiv[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
46 { "2rdiv", "set rdiv for 3rd plane", OFFSET(user_rdiv[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
47 { "3rdiv", "set rdiv for 4th plane", OFFSET(user_rdiv[3]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
48 { "0bias", "set bias for 1st plane", OFFSET(bias[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
49 { "1bias", "set bias for 2nd plane", OFFSET(bias[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
50 { "2bias", "set bias for 3rd plane", OFFSET(bias[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
51 { "3bias", "set bias for 4th plane", OFFSET(bias[3]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
52 { "0mode", "set matrix mode for 1st plane", OFFSET(mode[0]), AV_OPT_TYPE_INT, {.i64=MATRIX_SQUARE}, 0, MATRIX_NBMODES-1, FLAGS, .unit = "mode" },
53 { "1mode", "set matrix mode for 2nd plane", OFFSET(mode[1]), AV_OPT_TYPE_INT, {.i64=MATRIX_SQUARE}, 0, MATRIX_NBMODES-1, FLAGS, .unit = "mode" },
54 { "2mode", "set matrix mode for 3rd plane", OFFSET(mode[2]), AV_OPT_TYPE_INT, {.i64=MATRIX_SQUARE}, 0, MATRIX_NBMODES-1, FLAGS, .unit = "mode" },
55 { "3mode", "set matrix mode for 4th plane", OFFSET(mode[3]), AV_OPT_TYPE_INT, {.i64=MATRIX_SQUARE}, 0, MATRIX_NBMODES-1, FLAGS, .unit = "mode" },
56 { "square", "square matrix", 0, AV_OPT_TYPE_CONST, {.i64=MATRIX_SQUARE}, 0, 0, FLAGS, .unit = "mode" },
57 { "row", "single row matrix", 0, AV_OPT_TYPE_CONST, {.i64=MATRIX_ROW} , 0, 0, FLAGS, .unit = "mode" },
58 { "column", "single column matrix", 0, AV_OPT_TYPE_CONST, {.i64=MATRIX_COLUMN}, 0, 0, FLAGS, .unit = "mode" },
59 { NULL }
60 };
61
62 AVFILTER_DEFINE_CLASS(convolution);
63
64 static const int same3x3[9] = {0, 0, 0,
65 0, 1, 0,
66 0, 0, 0};
67
68 static const int same5x5[25] = {0, 0, 0, 0, 0,
69 0, 0, 0, 0, 0,
70 0, 0, 1, 0, 0,
71 0, 0, 0, 0, 0,
72 0, 0, 0, 0, 0};
73
74 static const int same7x7[49] = {0, 0, 0, 0, 0, 0, 0,
75 0, 0, 0, 0, 0, 0, 0,
76 0, 0, 0, 0, 0, 0, 0,
77 0, 0, 0, 1, 0, 0, 0,
78 0, 0, 0, 0, 0, 0, 0,
79 0, 0, 0, 0, 0, 0, 0,
80 0, 0, 0, 0, 0, 0, 0};
81
82 static const enum AVPixelFormat pix_fmts[] = {
83 AV_PIX_FMT_YUVA444P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV440P,
84 AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P,
85 AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUV420P,
86 AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVJ420P,
87 AV_PIX_FMT_YUVJ411P, AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV410P,
88 AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
89 AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10,
90 AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12,
91 AV_PIX_FMT_YUV420P14, AV_PIX_FMT_YUV422P14, AV_PIX_FMT_YUV444P14,
92 AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
93 AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA444P9,
94 AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_YUVA444P10,
95 AV_PIX_FMT_YUVA422P12, AV_PIX_FMT_YUVA444P12,
96 AV_PIX_FMT_YUVA420P16, AV_PIX_FMT_YUVA422P16, AV_PIX_FMT_YUVA444P16,
97 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10,
98 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, AV_PIX_FMT_GBRP16,
99 AV_PIX_FMT_GBRAP, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRAP16,
100 AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY9, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GRAY14, AV_PIX_FMT_GRAY16,
101 AV_PIX_FMT_NONE
102 };
103
104 typedef struct ThreadData {
105 AVFrame *in, *out;
106 } ThreadData;
107
108 static void filter16_prewitt(uint8_t *dstp, int width,
109 float scale, float delta, const int *const matrix,
110 const uint8_t *c[], int peak, int radius,
111 int dstride, int stride, int size)
112 {
113 uint16_t *dst = (uint16_t *)dstp;
114 int x;
115
116 for (x = 0; x < width; x++) {
117 float suma = AV_RN16A(&c[0][2 * x]) * -1 + AV_RN16A(&c[1][2 * x]) * -1 + AV_RN16A(&c[2][2 * x]) * -1 +
118 AV_RN16A(&c[6][2 * x]) * 1 + AV_RN16A(&c[7][2 * x]) * 1 + AV_RN16A(&c[8][2 * x]) * 1;
119 float sumb = AV_RN16A(&c[0][2 * x]) * -1 + AV_RN16A(&c[2][2 * x]) * 1 + AV_RN16A(&c[3][2 * x]) * -1 +
120 AV_RN16A(&c[5][2 * x]) * 1 + AV_RN16A(&c[6][2 * x]) * -1 + AV_RN16A(&c[8][2 * x]) * 1;
121
122 dst[x] = av_clip(sqrtf(suma*suma + sumb*sumb) * scale + delta, 0, peak);
123 }
124 }
125
126 static void filter16_roberts(uint8_t *dstp, int width,
127 float scale, float delta, const int *const matrix,
128 const uint8_t *c[], int peak, int radius,
129 int dstride, int stride, int size)
130 {
131 uint16_t *dst = (uint16_t *)dstp;
132 int x;
133
134 for (x = 0; x < width; x++) {
135 float suma = AV_RN16A(&c[0][2 * x]) * 1 + AV_RN16A(&c[1][2 * x]) * -1;
136 float sumb = AV_RN16A(&c[4][2 * x]) * 1 + AV_RN16A(&c[3][2 * x]) * -1;
137
138 dst[x] = av_clip(sqrtf(suma*suma + sumb*sumb) * scale + delta, 0, peak);
139 }
140 }
141
142 static void filter16_scharr(uint8_t *dstp, int width,
143 float scale, float delta, const int *const matrix,
144 const uint8_t *c[], int peak, int radius,
145 int dstride, int stride, int size)
146 {
147 uint16_t *dst = (uint16_t *)dstp;
148 int x;
149
150 for (x = 0; x < width; x++) {
151 float suma = AV_RN16A(&c[0][2 * x]) * -47 + AV_RN16A(&c[1][2 * x]) * -162 + AV_RN16A(&c[2][2 * x]) * -47 +
152 AV_RN16A(&c[6][2 * x]) * 47 + AV_RN16A(&c[7][2 * x]) * 162 + AV_RN16A(&c[8][2 * x]) * 47;
153 float sumb = AV_RN16A(&c[0][2 * x]) * -47 + AV_RN16A(&c[2][2 * x]) * 47 + AV_RN16A(&c[3][2 * x]) * -162 +
154 AV_RN16A(&c[5][2 * x]) * 162 + AV_RN16A(&c[6][2 * x]) * -47 + AV_RN16A(&c[8][2 * x]) * 47;
155
156 suma /= 256.f;
157 sumb /= 256.f;
158 dst[x] = av_clip(sqrtf(suma*suma + sumb*sumb) * scale + delta, 0, peak);
159 }
160 }
161
162 static void filter16_kirsch(uint8_t *dstp, int width,
163 float scale, float delta, const int *const matrix,
164 const uint8_t *c[], int peak, int radius,
165 int dstride, int stride, int size)
166 {
167 uint16_t *dst = (uint16_t *)dstp;
168 const uint16_t *c0 = (const uint16_t *)c[0], *c1 = (const uint16_t *)c[1], *c2 = (const uint16_t *)c[2];
169 const uint16_t *c3 = (const uint16_t *)c[3], *c5 = (const uint16_t *)c[5];
170 const uint16_t *c6 = (const uint16_t *)c[6], *c7 = (const uint16_t *)c[7], *c8 = (const uint16_t *)c[8];
171 int x;
172
173 for (x = 0; x < width; x++) {
174 int sum0 = c0[x] * 5 + c1[x] * 5 + c2[x] * 5 +
175 c3[x] * -3 + c5[x] * -3 +
176 c6[x] * -3 + c7[x] * -3 + c8[x] * -3;
177 int sum1 = c0[x] * -3 + c1[x] * 5 + c2[x] * 5 +
178 c3[x] * 5 + c5[x] * -3 +
179 c6[x] * -3 + c7[x] * -3 + c8[x] * -3;
180 int sum2 = c0[x] * -3 + c1[x] * -3 + c2[x] * 5 +
181 c3[x] * 5 + c5[x] * 5 +
182 c6[x] * -3 + c7[x] * -3 + c8[x] * -3;
183 int sum3 = c0[x] * -3 + c1[x] * -3 + c2[x] * -3 +
184 c3[x] * 5 + c5[x] * 5 +
185 c6[x] * 5 + c7[x] * -3 + c8[x] * -3;
186 int sum4 = c0[x] * -3 + c1[x] * -3 + c2[x] * -3 +
187 c3[x] * -3 + c5[x] * 5 +
188 c6[x] * 5 + c7[x] * 5 + c8[x] * -3;
189 int sum5 = c0[x] * -3 + c1[x] * -3 + c2[x] * -3 +
190 c3[x] * -3 + c5[x] * -3 +
191 c6[x] * 5 + c7[x] * 5 + c8[x] * 5;
192 int sum6 = c0[x] * 5 + c1[x] * -3 + c2[x] * -3 +
193 c3[x] * -3 + c5[x] * -3 +
194 c6[x] * -3 + c7[x] * 5 + c8[x] * 5;
195 int sum7 = c0[x] * 5 + c1[x] * 5 + c2[x] * -3 +
196 c3[x] * -3 + c5[x] * -3 +
197 c6[x] * -3 + c7[x] * -3 + c8[x] * 5;
198
199 sum0 = FFMAX(sum0, sum1);
200 sum2 = FFMAX(sum2, sum3);
201 sum4 = FFMAX(sum4, sum5);
202 sum6 = FFMAX(sum6, sum7);
203 sum0 = FFMAX(sum0, sum2);
204 sum4 = FFMAX(sum4, sum6);
205 sum0 = FFMAX(sum0, sum4);
206
207 dst[x] = av_clip(FFABS(sum0) * scale + delta, 0, peak);
208 }
209 }
210
211 static void filter_prewitt(uint8_t *dst, int width,
212 float scale, float delta, const int *const matrix,
213 const uint8_t *c[], int peak, int radius,
214 int dstride, int stride, int size)
215 {
216 const uint8_t *c0 = c[0], *c1 = c[1], *c2 = c[2];
217 const uint8_t *c3 = c[3], *c5 = c[5];
218 const uint8_t *c6 = c[6], *c7 = c[7], *c8 = c[8];
219 int x;
220
221 for (x = 0; x < width; x++) {
222 float suma = c0[x] * -1 + c1[x] * -1 + c2[x] * -1 +
223 c6[x] * 1 + c7[x] * 1 + c8[x] * 1;
224 float sumb = c0[x] * -1 + c2[x] * 1 + c3[x] * -1 +
225 c5[x] * 1 + c6[x] * -1 + c8[x] * 1;
226
227 dst[x] = av_clip_uint8(sqrtf(suma*suma + sumb*sumb) * scale + delta);
228 }
229 }
230
231 static void filter_roberts(uint8_t *dst, int width,
232 float scale, float delta, const int *const matrix,
233 const uint8_t *c[], int peak, int radius,
234 int dstride, int stride, int size)
235 {
236 int x;
237
238 for (x = 0; x < width; x++) {
239 float suma = c[0][x] * 1 + c[1][x] * -1;
240 float sumb = c[4][x] * 1 + c[3][x] * -1;
241
242 dst[x] = av_clip_uint8(sqrtf(suma*suma + sumb*sumb) * scale + delta);
243 }
244 }
245
246 static void filter_scharr(uint8_t *dst, int width,
247 float scale, float delta, const int *const matrix,
248 const uint8_t *c[], int peak, int radius,
249 int dstride, int stride, int size)
250 {
251 const uint8_t *c0 = c[0], *c1 = c[1], *c2 = c[2];
252 const uint8_t *c3 = c[3], *c5 = c[5];
253 const uint8_t *c6 = c[6], *c7 = c[7], *c8 = c[8];
254 int x;
255
256 for (x = 0; x < width; x++) {
257 float suma = c0[x] * -47 + c1[x] * -162 + c2[x] * -47 +
258 c6[x] * 47 + c7[x] * 162 + c8[x] * 47;
259 float sumb = c0[x] * -47 + c2[x] * 47 + c3[x] * -162 +
260 c5[x] * 162 + c6[x] * -47 + c8[x] * 47;
261
262 suma /= 256.f;
263 sumb /= 256.f;
264 dst[x] = av_clip_uint8(sqrtf(suma*suma + sumb*sumb) * scale + delta);
265 }
266 }
267
268 static void filter_kirsch(uint8_t *dst, int width,
269 float scale, float delta, const int *const matrix,
270 const uint8_t *c[], int peak, int radius,
271 int dstride, int stride, int size)
272 {
273 const uint8_t *c0 = c[0], *c1 = c[1], *c2 = c[2];
274 const uint8_t *c3 = c[3], *c5 = c[5];
275 const uint8_t *c6 = c[6], *c7 = c[7], *c8 = c[8];
276 int x;
277
278 for (x = 0; x < width; x++) {
279 int sum0 = c0[x] * 5 + c1[x] * 5 + c2[x] * 5 +
280 c3[x] * -3 + c5[x] * -3 +
281 c6[x] * -3 + c7[x] * -3 + c8[x] * -3;
282 int sum1 = c0[x] * -3 + c1[x] * 5 + c2[x] * 5 +
283 c3[x] * 5 + c5[x] * -3 +
284 c6[x] * -3 + c7[x] * -3 + c8[x] * -3;
285 int sum2 = c0[x] * -3 + c1[x] * -3 + c2[x] * 5 +
286 c3[x] * 5 + c5[x] * 5 +
287 c6[x] * -3 + c7[x] * -3 + c8[x] * -3;
288 int sum3 = c0[x] * -3 + c1[x] * -3 + c2[x] * -3 +
289 c3[x] * 5 + c5[x] * 5 +
290 c6[x] * 5 + c7[x] * -3 + c8[x] * -3;
291 int sum4 = c0[x] * -3 + c1[x] * -3 + c2[x] * -3 +
292 c3[x] * -3 + c5[x] * 5 +
293 c6[x] * 5 + c7[x] * 5 + c8[x] * -3;
294 int sum5 = c0[x] * -3 + c1[x] * -3 + c2[x] * -3 +
295 c3[x] * -3 + c5[x] * -3 +
296 c6[x] * 5 + c7[x] * 5 + c8[x] * 5;
297 int sum6 = c0[x] * 5 + c1[x] * -3 + c2[x] * -3 +
298 c3[x] * -3 + c5[x] * -3 +
299 c6[x] * -3 + c7[x] * 5 + c8[x] * 5;
300 int sum7 = c0[x] * 5 + c1[x] * 5 + c2[x] * -3 +
301 c3[x] * -3 + c5[x] * -3 +
302 c6[x] * -3 + c7[x] * -3 + c8[x] * 5;
303
304 sum0 = FFMAX(sum0, sum1);
305 sum2 = FFMAX(sum2, sum3);
306 sum4 = FFMAX(sum4, sum5);
307 sum6 = FFMAX(sum6, sum7);
308 sum0 = FFMAX(sum0, sum2);
309 sum4 = FFMAX(sum4, sum6);
310 sum0 = FFMAX(sum0, sum4);
311
312 dst[x] = av_clip_uint8(FFABS(sum0) * scale + delta);
313 }
314 }
315
316 static void filter16_3x3(uint8_t *dstp, int width,
317 float rdiv, float bias, const int *const matrix,
318 const uint8_t *c[], int peak, int radius,
319 int dstride, int stride, int size)
320 {
321 uint16_t *dst = (uint16_t *)dstp;
322 int x;
323
324 for (x = 0; x < width; x++) {
325 unsigned sum = (unsigned)AV_RN16A(&c[0][2 * x]) * matrix[0] +
326 (unsigned)AV_RN16A(&c[1][2 * x]) * matrix[1] +
327 (unsigned)AV_RN16A(&c[2][2 * x]) * matrix[2] +
328 (unsigned)AV_RN16A(&c[3][2 * x]) * matrix[3] +
329 (unsigned)AV_RN16A(&c[4][2 * x]) * matrix[4] +
330 (unsigned)AV_RN16A(&c[5][2 * x]) * matrix[5] +
331 (unsigned)AV_RN16A(&c[6][2 * x]) * matrix[6] +
332 (unsigned)AV_RN16A(&c[7][2 * x]) * matrix[7] +
333 (unsigned)AV_RN16A(&c[8][2 * x]) * matrix[8];
334 dst[x] = av_clip((int)sum * rdiv + bias + 0.5f, 0, peak);
335 }
336 }
337
338 static void filter16_5x5(uint8_t *dstp, int width,
339 float rdiv, float bias, const int *const matrix,
340 const uint8_t *c[], int peak, int radius,
341 int dstride, int stride, int size)
342 {
343 uint16_t *dst = (uint16_t *)dstp;
344 int x;
345
346 for (x = 0; x < width; x++) {
347 int i;
348 unsigned sum = 0;
349
350 for (i = 0; i < 25; i++)
351 sum += (unsigned)AV_RN16A(&c[i][2 * x]) * matrix[i];
352
353 dst[x] = av_clip((int)sum * rdiv + bias + 0.5f, 0, peak);
354 }
355 }
356
357 static void filter16_7x7(uint8_t *dstp, int width,
358 float rdiv, float bias, const int *const matrix,
359 const uint8_t *c[], int peak, int radius,
360 int dstride, int stride, int size)
361 {
362 uint16_t *dst = (uint16_t *)dstp;
363 int x;
364
365 for (x = 0; x < width; x++) {
366 int i;
367 unsigned sum = 0;
368
369 for (i = 0; i < 49; i++)
370 sum += (unsigned)AV_RN16A(&c[i][2 * x]) * matrix[i];
371
372 dst[x] = av_clip((int)sum * rdiv + bias + 0.5f, 0, peak);
373 }
374 }
375
376 static void filter16_row(uint8_t *dstp, int width,
377 float rdiv, float bias, const int *const matrix,
378 const uint8_t *c[], int peak, int radius,
379 int dstride, int stride, int size)
380 {
381 uint16_t *dst = (uint16_t *)dstp;
382 int x;
383
384 for (x = 0; x < width; x++) {
385 int i;
386 unsigned sum = 0;
387
388 for (i = 0; i < 2 * radius + 1; i++)
389 sum += (unsigned)AV_RN16A(&c[i][2 * x]) * matrix[i];
390
391 dst[x] = av_clip((int)sum * rdiv + bias + 0.5f, 0, peak);
392 }
393 }
394
395 static void filter16_column(uint8_t *dstp, int height,
396 float rdiv, float bias, const int *const matrix,
397 const uint8_t *c[], int peak, int radius,
398 int dstride, int stride, int size)
399 {
400 DECLARE_ALIGNED(64, unsigned, sum)[16];
401 uint16_t *dst = (uint16_t *)dstp;
402 const int width = FFMIN(16, size);
403
404 for (int y = 0; y < height; y++) {
405
406 memset(sum, 0, sizeof(sum));
407 for (int i = 0; i < 2 * radius + 1; i++) {
408 for (int off16 = 0; off16 < width; off16++)
409 sum[off16] += (unsigned)AV_RN16A(&c[i][0 + y * stride + off16 * 2]) * matrix[i];
410 }
411
412 for (int off16 = 0; off16 < width; off16++) {
413 dst[off16] = av_clip((int)sum[off16] * rdiv + bias + 0.5f, 0, peak);
414 }
415 dst += dstride / 2;
416 }
417 }
418
419 static void filter_7x7(uint8_t *dst, int width,
420 float rdiv, float bias, const int *const matrix,
421 const uint8_t *c[], int peak, int radius,
422 int dstride, int stride, int size)
423 {
424 int x;
425
426 for (x = 0; x < width; x++) {
427 int i;
428 unsigned sum = 0;
429
430 for (i = 0; i < 49; i++)
431 sum += (unsigned)c[i][x] * matrix[i];
432
433 dst[x] = av_clip_uint8((int)sum * rdiv + bias + 0.5f);
434 }
435 }
436
437 static void filter_5x5(uint8_t *dst, int width,
438 float rdiv, float bias, const int *const matrix,
439 const uint8_t *c[], int peak, int radius,
440 int dstride, int stride, int size)
441 {
442 int x;
443
444 for (x = 0; x < width; x++) {
445 int i;
446 unsigned sum = 0;
447
448 for (i = 0; i < 25; i++)
449 sum += (unsigned)c[i][x] * matrix[i];
450
451 dst[x] = av_clip_uint8((int)sum * rdiv + bias + 0.5f);
452 }
453 }
454
455 static void filter_3x3(uint8_t *dst, int width,
456 float rdiv, float bias, const int *const matrix,
457 const uint8_t *c[], int peak, int radius,
458 int dstride, int stride, int size)
459 {
460 const uint8_t *c0 = c[0], *c1 = c[1], *c2 = c[2];
461 const uint8_t *c3 = c[3], *c4 = c[4], *c5 = c[5];
462 const uint8_t *c6 = c[6], *c7 = c[7], *c8 = c[8];
463 int x;
464
465 for (x = 0; x < width; x++) {
466 unsigned sum = (unsigned)c0[x] * matrix[0] + (unsigned)c1[x] * matrix[1] + (unsigned)c2[x] * matrix[2] +
467 (unsigned)c3[x] * matrix[3] + (unsigned)c4[x] * matrix[4] + (unsigned)c5[x] * matrix[5] +
468 (unsigned)c6[x] * matrix[6] + (unsigned)c7[x] * matrix[7] + (unsigned)c8[x] * matrix[8];
469 dst[x] = av_clip_uint8((int)sum * rdiv + bias + 0.5f);
470 }
471 }
472
473 static void filter_row(uint8_t *dst, int width,
474 float rdiv, float bias, const int *const matrix,
475 const uint8_t *c[], int peak, int radius,
476 int dstride, int stride, int size)
477 {
478 int x;
479
480 for (x = 0; x < width; x++) {
481 int i;
482 unsigned sum = 0;
483
484 for (i = 0; i < 2 * radius + 1; i++)
485 sum += (unsigned)c[i][x] * matrix[i];
486
487 dst[x] = av_clip_uint8((int)sum * rdiv + bias + 0.5f);
488 }
489 }
490
491 static void filter_column(uint8_t *dst, int height,
492 float rdiv, float bias, const int *const matrix,
493 const uint8_t *c[], int peak, int radius,
494 int dstride, int stride, int size)
495 {
496 DECLARE_ALIGNED(64, unsigned, sum)[16];
497
498 for (int y = 0; y < height; y++) {
499 memset(sum, 0, sizeof(sum));
500
501 for (int i = 0; i < 2 * radius + 1; i++) {
502 for (int off16 = 0; off16 < 16; off16++)
503 sum[off16] += (unsigned)c[i][0 + y * stride + off16] * matrix[i];
504 }
505
506 for (int off16 = 0; off16 < 16; off16++) {
507 dst[off16] = av_clip_uint8((int)sum[off16] * rdiv + bias + 0.5f);
508 }
509 dst += dstride;
510 }
511 }
512
513 static void setup_5x5(int radius, const uint8_t *c[], const uint8_t *src, int stride,
514 int x, int w, int y, int h, int bpc)
515 {
516 int i;
517
518 for (i = 0; i < 25; i++) {
519 int xoff = avpriv_mirror(x + (i % 5) - 2, w - 1);
520 int yoff = avpriv_mirror(y + (i / 5) - 2, h - 1);
521
522 c[i] = src + xoff * bpc + yoff * stride;
523 }
524 }
525
526 static void setup_7x7(int radius, const uint8_t *c[], const uint8_t *src, int stride,
527 int x, int w, int y, int h, int bpc)
528 {
529 int i;
530
531 for (i = 0; i < 49; i++) {
532 int xoff = avpriv_mirror(x + (i % 7) - 3, w - 1);
533 int yoff = avpriv_mirror(y + (i / 7) - 3, h - 1);
534
535 c[i] = src + xoff * bpc + yoff * stride;
536 }
537 }
538
539 static void setup_row(int radius, const uint8_t *c[], const uint8_t *src, int stride,
540 int x, int w, int y, int h, int bpc)
541 {
542 int i;
543
544 for (i = 0; i < radius * 2 + 1; i++) {
545 int xoff = avpriv_mirror(x + i - radius, w - 1);
546
547 c[i] = src + xoff * bpc + y * stride;
548 }
549 }
550
551 static void setup_column(int radius, const uint8_t *c[], const uint8_t *src, int stride,
552 int x, int w, int y, int h, int bpc)
553 {
554 int i;
555
556 for (i = 0; i < radius * 2 + 1; i++) {
557 int xoff = avpriv_mirror(x + i - radius, h - 1);
558
559 c[i] = src + y * bpc + xoff * stride;
560 }
561 }
562
563 static int filter_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
564 {
565 ConvolutionContext *s = ctx->priv;
566 ThreadData *td = arg;
567 AVFrame *in = td->in;
568 AVFrame *out = td->out;
569 int plane;
570
571 for (plane = 0; plane < s->nb_planes; plane++) {
572 const int mode = s->mode[plane];
573 const int bpc = s->bpc;
574 const int radius = s->size[plane] / 2;
575 const int height = s->planeheight[plane];
576 const int width = s->planewidth[plane];
577 const int stride = in->linesize[plane];
578 const int dstride = out->linesize[plane];
579 const int sizeh = mode == MATRIX_COLUMN ? width : height;
580 const int sizew = mode == MATRIX_COLUMN ? height : width;
581 const int slice_start = ff_slice_pos(sizeh, jobnr, nb_jobs);
582 const int slice_end = ff_slice_pos(sizeh, jobnr + 1, nb_jobs);
583 const float rdiv = s->rdiv[plane];
584 const float bias = s->bias[plane];
585 const uint8_t *src = in->data[plane];
586 const int dst_pos = slice_start * (mode == MATRIX_COLUMN ? bpc : dstride);
587 uint8_t *dst = out->data[plane] + dst_pos;
588 const int *matrix = s->matrix[plane];
589 const int step = mode == MATRIX_COLUMN ? 16 : 1;
590 const uint8_t *c[49];
591 int y, x;
592
593 if (s->copy[plane]) {
594 if (mode == MATRIX_COLUMN)
595 av_image_copy_plane(dst, dstride, src + slice_start * bpc, stride,
596 (slice_end - slice_start) * bpc, height);
597 else
598 av_image_copy_plane(dst, dstride, src + slice_start * stride, stride,
599 width * bpc, slice_end - slice_start);
600 continue;
601 }
602 for (y = slice_start; y < slice_end; y += step) {
603 const int left = FFMIN(radius, sizew);
604 const int right = FFMAX(left, sizew - radius);
605 const int xoff = mode == MATRIX_COLUMN ? (y - slice_start) * bpc : left * bpc;
606 const int yoff = mode == MATRIX_COLUMN ? left * dstride : 0;
607
608 for (x = 0; x < left; x++) {
609 const int xoff = mode == MATRIX_COLUMN ? (y - slice_start) * bpc : x * bpc;
610 const int yoff = mode == MATRIX_COLUMN ? x * dstride : 0;
611
612 s->setup[plane](radius, c, src, stride, x, width, y, height, bpc);
613 s->filter[plane](dst + yoff + xoff, 1, rdiv,
614 bias, matrix, c, s->max, radius,
615 dstride, stride, slice_end - step);
616 }
617 s->setup[plane](radius, c, src, stride, left, width, y, height, bpc);
618 s->filter[plane](dst + yoff + xoff, right - left,
619 rdiv, bias, matrix, c, s->max, radius,
620 dstride, stride, slice_end - step);
621 for (x = right; x < sizew; x++) {
622 const int xoff = mode == MATRIX_COLUMN ? (y - slice_start) * bpc : x * bpc;
623 const int yoff = mode == MATRIX_COLUMN ? x * dstride : 0;
624
625 s->setup[plane](radius, c, src, stride, x, width, y, height, bpc);
626 s->filter[plane](dst + yoff + xoff, 1, rdiv,
627 bias, matrix, c, s->max, radius,
628 dstride, stride, slice_end - step);
629 }
630 if (mode != MATRIX_COLUMN)
631 dst += dstride;
632 }
633 }
634
635 return 0;
636 }
637
638 static int param_init(AVFilterContext *ctx)
639 {
640 ConvolutionContext *s = ctx->priv;
641 AVFilterLink *inlink = ctx->inputs[0];
642 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
643 int p, i;
644
645 s->depth = desc->comp[0].depth;
646 s->max = (1 << s->depth) - 1;
647
648 s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
649 s->planewidth[0] = s->planewidth[3] = inlink->w;
650 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
651 s->planeheight[0] = s->planeheight[3] = inlink->h;
652
653 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
654 s->nb_threads = ff_filter_get_nb_threads(ctx);
655 s->bpc = (s->depth + 7) / 8;
656
657 if (!strcmp(ctx->filter->name, "convolution")) {
658 for (i = 0; i < 4; i++) {
659 int *matrix = (int *)s->matrix[i];
660 char *orig, *p, *arg, *saveptr = NULL;
661 float sum = 1.f;
662
663 p = orig = av_strdup(s->matrix_str[i]);
664 if (p) {
665 s->matrix_length[i] = 0;
666 s->rdiv[i] = s->user_rdiv[i];
667 sum = 0.f;
668
669 while (s->matrix_length[i] < 49) {
670 if (!(arg = av_strtok(p, " |", &saveptr)))
671 break;
672
673 p = NULL;
674 sscanf(arg, "%d", &matrix[s->matrix_length[i]]);
675 sum += matrix[s->matrix_length[i]];
676 s->matrix_length[i]++;
677 }
678
679 av_freep(&orig);
680 if (!(s->matrix_length[i] & 1)) {
681 av_log(ctx, AV_LOG_ERROR, "number of matrix elements must be odd\n");
682 return AVERROR(EINVAL);
683 }
684 }
685
686 if (s->mode[i] == MATRIX_ROW) {
687 s->filter[i] = filter_row;
688 s->setup[i] = setup_row;
689 s->size[i] = s->matrix_length[i];
690 } else if (s->mode[i] == MATRIX_COLUMN) {
691 s->filter[i] = filter_column;
692 s->setup[i] = setup_column;
693 s->size[i] = s->matrix_length[i];
694 } else if (s->matrix_length[i] == 9) {
695 s->size[i] = 3;
696
697 if (!memcmp(matrix, same3x3, sizeof(same3x3))) {
698 s->copy[i] = 1;
699 } else {
700 s->filter[i] = filter_3x3;
701 s->copy[i] = 0;
702 }
703 s->setup[i] = setup_3x3;
704 } else if (s->matrix_length[i] == 25) {
705 s->size[i] = 5;
706 if (!memcmp(matrix, same5x5, sizeof(same5x5))) {
707 s->copy[i] = 1;
708 } else {
709 s->filter[i] = filter_5x5;
710 s->copy[i] = 0;
711 }
712 s->setup[i] = setup_5x5;
713 } else if (s->matrix_length[i] == 49) {
714 s->size[i] = 7;
715 if (!memcmp(matrix, same7x7, sizeof(same7x7))) {
716 s->copy[i] = 1;
717 } else {
718 s->filter[i] = filter_7x7;
719 s->copy[i] = 0;
720 }
721 s->setup[i] = setup_7x7;
722 } else {
723 return AVERROR(EINVAL);
724 }
725
726 if (sum == 0)
727 sum = 1;
728 if (s->rdiv[i] == 0)
729 s->rdiv[i] = 1. / sum;
730
731 if (s->copy[i] && (s->rdiv[i] != 1. || s->bias[i] != 0.))
732 s->copy[i] = 0;
733 }
734 } else if (!strcmp(ctx->filter->name, "prewitt")) {
735 for (i = 0; i < 4; i++) {
736 s->filter[i] = filter_prewitt;
737 s->copy[i] = !((1 << i) & s->planes);
738 s->size[i] = 3;
739 s->setup[i] = setup_3x3;
740 s->rdiv[i] = s->scale;
741 s->bias[i] = s->delta;
742 }
743 } else if (!strcmp(ctx->filter->name, "roberts")) {
744 for (i = 0; i < 4; i++) {
745 s->filter[i] = filter_roberts;
746 s->copy[i] = !((1 << i) & s->planes);
747 s->size[i] = 3;
748 s->setup[i] = setup_3x3;
749 s->rdiv[i] = s->scale;
750 s->bias[i] = s->delta;
751 }
752 #if CONFIG_SOBEL_FILTER
753 } else if (!strcmp(ctx->filter->name, "sobel")) {
754 ff_sobel_init(s, s->depth, s->nb_planes);
755 #endif
756 } else if (!strcmp(ctx->filter->name, "kirsch")) {
757 for (i = 0; i < 4; i++) {
758 s->filter[i] = filter_kirsch;
759 s->copy[i] = !((1 << i) & s->planes);
760 s->size[i] = 3;
761 s->setup[i] = setup_3x3;
762 s->rdiv[i] = s->scale;
763 s->bias[i] = s->delta;
764 }
765 } else if (!strcmp(ctx->filter->name, "scharr")) {
766 for (i = 0; i < 4; i++) {
767 s->filter[i] = filter_scharr;
768 s->copy[i] = !((1 << i) & s->planes);
769 s->size[i] = 3;
770 s->setup[i] = setup_3x3;
771 s->rdiv[i] = s->scale;
772 s->bias[i] = s->delta;
773 }
774 }
775
776 if (!strcmp(ctx->filter->name, "convolution")) {
777 if (s->depth > 8) {
778 for (p = 0; p < s->nb_planes; p++) {
779 if (s->mode[p] == MATRIX_ROW)
780 s->filter[p] = filter16_row;
781 else if (s->mode[p] == MATRIX_COLUMN)
782 s->filter[p] = filter16_column;
783 else if (s->size[p] == 3)
784 s->filter[p] = filter16_3x3;
785 else if (s->size[p] == 5)
786 s->filter[p] = filter16_5x5;
787 else if (s->size[p] == 7)
788 s->filter[p] = filter16_7x7;
789 }
790 }
791 #if CONFIG_CONVOLUTION_FILTER && ARCH_X86_64 && HAVE_X86ASM
792 ff_convolution_init_x86(s);
793 #endif
794 } else if (!strcmp(ctx->filter->name, "prewitt")) {
795 if (s->depth > 8)
796 for (p = 0; p < s->nb_planes; p++)
797 s->filter[p] = filter16_prewitt;
798 } else if (!strcmp(ctx->filter->name, "roberts")) {
799 if (s->depth > 8)
800 for (p = 0; p < s->nb_planes; p++)
801 s->filter[p] = filter16_roberts;
802 } else if (!strcmp(ctx->filter->name, "kirsch")) {
803 if (s->depth > 8)
804 for (p = 0; p < s->nb_planes; p++)
805 s->filter[p] = filter16_kirsch;
806 } else if (!strcmp(ctx->filter->name, "scharr")) {
807 if (s->depth > 8)
808 for (p = 0; p < s->nb_planes; p++)
809 s->filter[p] = filter16_scharr;
810 }
811
812 return 0;
813 }
814
815 static int config_input(AVFilterLink *inlink)
816 {
817 AVFilterContext *ctx = inlink->dst;
818 return param_init(ctx);
819 }
820
821 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
822 {
823 AVFilterContext *ctx = inlink->dst;
824 ConvolutionContext *s = ctx->priv;
825 AVFilterLink *outlink = ctx->outputs[0];
826 AVFrame *out;
827 ThreadData td;
828
829 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
830 if (!out) {
831 av_frame_free(&in);
832 return AVERROR(ENOMEM);
833 }
834 av_frame_copy_props(out, in);
835
836 td.in = in;
837 td.out = out;
838 ff_filter_execute(ctx, filter_slice, &td, NULL,
839 FFMIN3(s->planeheight[1], s->planewidth[1], s->nb_threads));
840
841 av_frame_free(&in);
842 return ff_filter_frame(outlink, out);
843 }
844
845 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
846 char *res, int res_len, int flags)
847 {
848 int ret;
849
850 ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
851 if (ret < 0)
852 return ret;
853
854 return param_init(ctx);
855 }
856
857 static const AVFilterPad convolution_inputs[] = {
858 {
859 .name = "default",
860 .type = AVMEDIA_TYPE_VIDEO,
861 .config_props = config_input,
862 .filter_frame = filter_frame,
863 },
864 };
865
866 #if CONFIG_CONVOLUTION_FILTER
867
868 const FFFilter ff_vf_convolution = {
869 .p.name = "convolution",
870 .p.description = NULL_IF_CONFIG_SMALL("Apply convolution filter."),
871 .p.priv_class = &convolution_class,
872 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
873 .priv_size = sizeof(ConvolutionContext),
874 FILTER_INPUTS(convolution_inputs),
875 FILTER_OUTPUTS(ff_video_default_filterpad),
876 FILTER_PIXFMTS_ARRAY(pix_fmts),
877 .process_command = process_command,
878 };
879
880 #endif /* CONFIG_CONVOLUTION_FILTER */
881
882 static const AVOption common_options[] = {
883 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=15}, 0, 15, FLAGS},
884 { "scale", "set scale", OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, 65535, FLAGS},
885 { "delta", "set delta", OFFSET(delta), AV_OPT_TYPE_FLOAT, {.dbl=0}, -65535, 65535, FLAGS},
886 { NULL }
887 };
888
889 AVFILTER_DEFINE_CLASS_EXT(common, "kirsch/prewitt/roberts/scharr/sobel",
890 common_options);
891
892 #if CONFIG_PREWITT_FILTER
893
894 const FFFilter ff_vf_prewitt = {
895 .p.name = "prewitt",
896 .p.description = NULL_IF_CONFIG_SMALL("Apply prewitt operator."),
897 .p.priv_class = &common_class,
898 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
899 .priv_size = sizeof(ConvolutionContext),
900 FILTER_INPUTS(convolution_inputs),
901 FILTER_OUTPUTS(ff_video_default_filterpad),
902 FILTER_PIXFMTS_ARRAY(pix_fmts),
903 .process_command = process_command,
904 };
905
906 #endif /* CONFIG_PREWITT_FILTER */
907
908 #if CONFIG_SOBEL_FILTER
909
910 const FFFilter ff_vf_sobel = {
911 .p.name = "sobel",
912 .p.description = NULL_IF_CONFIG_SMALL("Apply sobel operator."),
913 .p.priv_class = &common_class,
914 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
915 .priv_size = sizeof(ConvolutionContext),
916 FILTER_INPUTS(convolution_inputs),
917 FILTER_OUTPUTS(ff_video_default_filterpad),
918 FILTER_PIXFMTS_ARRAY(pix_fmts),
919 .process_command = process_command,
920 };
921
922 #endif /* CONFIG_SOBEL_FILTER */
923
924 #if CONFIG_ROBERTS_FILTER
925
926 const FFFilter ff_vf_roberts = {
927 .p.name = "roberts",
928 .p.description = NULL_IF_CONFIG_SMALL("Apply roberts cross operator."),
929 .p.priv_class = &common_class,
930 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
931 .priv_size = sizeof(ConvolutionContext),
932 FILTER_INPUTS(convolution_inputs),
933 FILTER_OUTPUTS(ff_video_default_filterpad),
934 FILTER_PIXFMTS_ARRAY(pix_fmts),
935 .process_command = process_command,
936 };
937
938 #endif /* CONFIG_ROBERTS_FILTER */
939
940 #if CONFIG_KIRSCH_FILTER
941
942 const FFFilter ff_vf_kirsch = {
943 .p.name = "kirsch",
944 .p.description = NULL_IF_CONFIG_SMALL("Apply kirsch operator."),
945 .p.priv_class = &common_class,
946 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
947 .priv_size = sizeof(ConvolutionContext),
948 FILTER_INPUTS(convolution_inputs),
949 FILTER_OUTPUTS(ff_video_default_filterpad),
950 FILTER_PIXFMTS_ARRAY(pix_fmts),
951 .process_command = process_command,
952 };
953
954 #endif /* CONFIG_KIRSCH_FILTER */
955
956 #if CONFIG_SCHARR_FILTER
957
958 const FFFilter ff_vf_scharr = {
959 .p.name = "scharr",
960 .p.description = NULL_IF_CONFIG_SMALL("Apply scharr operator."),
961 .p.priv_class = &common_class,
962 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
963 .priv_size = sizeof(ConvolutionContext),
964 FILTER_INPUTS(convolution_inputs),
965 FILTER_OUTPUTS(ff_video_default_filterpad),
966 FILTER_PIXFMTS_ARRAY(pix_fmts),
967 .process_command = process_command,
968 };
969
970 #endif /* CONFIG_SCHARR_FILTER */
971