FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavfilter/vf_convolution.c
Date: 2022-12-09 07:38:14
Exec Total Coverage
Lines: 0 447 0.0%
Functions: 0 27 0.0%
Branches: 0 222 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_internal.h"
28 #include "libavutil/opt.h"
29 #include "libavutil/pixdesc.h"
30 #include "avfilter.h"
31 #include "convolution.h"
32 #include "formats.h"
33 #include "internal.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(rdiv[0]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
45 { "1rdiv", "set rdiv for 2nd plane", OFFSET(rdiv[1]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
46 { "2rdiv", "set rdiv for 3rd plane", OFFSET(rdiv[2]), AV_OPT_TYPE_FLOAT, {.dbl=0.0}, 0.0, INT_MAX, FLAGS},
47 { "3rdiv", "set rdiv for 4th plane", OFFSET(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, "mode" },
53 { "1mode", "set matrix mode for 2nd plane", OFFSET(mode[1]), AV_OPT_TYPE_INT, {.i64=MATRIX_SQUARE}, 0, MATRIX_NBMODES-1, FLAGS, "mode" },
54 { "2mode", "set matrix mode for 3rd plane", OFFSET(mode[2]), AV_OPT_TYPE_INT, {.i64=MATRIX_SQUARE}, 0, MATRIX_NBMODES-1, FLAGS, "mode" },
55 { "3mode", "set matrix mode for 4th plane", OFFSET(mode[3]), AV_OPT_TYPE_INT, {.i64=MATRIX_SQUARE}, 0, MATRIX_NBMODES-1, FLAGS, "mode" },
56 { "square", "square matrix", 0, AV_OPT_TYPE_CONST, {.i64=MATRIX_SQUARE}, 0, 0, FLAGS, "mode" },
57 { "row", "single row matrix", 0, AV_OPT_TYPE_CONST, {.i64=MATRIX_ROW} , 0, 0, FLAGS, "mode" },
58 { "column", "single column matrix", 0, AV_OPT_TYPE_CONST, {.i64=MATRIX_COLUMN}, 0, 0, FLAGS, "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 int sum = AV_RN16A(&c[0][2 * x]) * matrix[0] +
326 AV_RN16A(&c[1][2 * x]) * matrix[1] +
327 AV_RN16A(&c[2][2 * x]) * matrix[2] +
328 AV_RN16A(&c[3][2 * x]) * matrix[3] +
329 AV_RN16A(&c[4][2 * x]) * matrix[4] +
330 AV_RN16A(&c[5][2 * x]) * matrix[5] +
331 AV_RN16A(&c[6][2 * x]) * matrix[6] +
332 AV_RN16A(&c[7][2 * x]) * matrix[7] +
333 AV_RN16A(&c[8][2 * x]) * matrix[8];
334 sum = (int)(sum * rdiv + bias + 0.5f);
335 dst[x] = av_clip(sum, 0, peak);
336 }
337 }
338
339 static void filter16_5x5(uint8_t *dstp, int width,
340 float rdiv, float bias, const int *const matrix,
341 const uint8_t *c[], int peak, int radius,
342 int dstride, int stride, int size)
343 {
344 uint16_t *dst = (uint16_t *)dstp;
345 int x;
346
347 for (x = 0; x < width; x++) {
348 int i, sum = 0;
349
350 for (i = 0; i < 25; i++)
351 sum += AV_RN16A(&c[i][2 * x]) * matrix[i];
352
353 sum = (int)(sum * rdiv + bias + 0.5f);
354 dst[x] = av_clip(sum, 0, peak);
355 }
356 }
357
358 static void filter16_7x7(uint8_t *dstp, int width,
359 float rdiv, float bias, const int *const matrix,
360 const uint8_t *c[], int peak, int radius,
361 int dstride, int stride, int size)
362 {
363 uint16_t *dst = (uint16_t *)dstp;
364 int x;
365
366 for (x = 0; x < width; x++) {
367 int i, sum = 0;
368
369 for (i = 0; i < 49; i++)
370 sum += AV_RN16A(&c[i][2 * x]) * matrix[i];
371
372 sum = (int)(sum * rdiv + bias + 0.5f);
373 dst[x] = av_clip(sum, 0, peak);
374 }
375 }
376
377 static void filter16_row(uint8_t *dstp, int width,
378 float rdiv, float bias, const int *const matrix,
379 const uint8_t *c[], int peak, int radius,
380 int dstride, int stride, int size)
381 {
382 uint16_t *dst = (uint16_t *)dstp;
383 int x;
384
385 for (x = 0; x < width; x++) {
386 int i, sum = 0;
387
388 for (i = 0; i < 2 * radius + 1; i++)
389 sum += AV_RN16A(&c[i][2 * x]) * matrix[i];
390
391 sum = (int)(sum * rdiv + bias + 0.5f);
392 dst[x] = av_clip(sum, 0, peak);
393 }
394 }
395
396 static void filter16_column(uint8_t *dstp, int height,
397 float rdiv, float bias, const int *const matrix,
398 const uint8_t *c[], int peak, int radius,
399 int dstride, int stride, int size)
400 {
401 DECLARE_ALIGNED(64, int, sum)[16];
402 uint16_t *dst = (uint16_t *)dstp;
403 const int width = FFMIN(16, size);
404
405 for (int y = 0; y < height; y++) {
406
407 memset(sum, 0, sizeof(sum));
408 for (int i = 0; i < 2 * radius + 1; i++) {
409 for (int off16 = 0; off16 < width; off16++)
410 sum[off16] += AV_RN16A(&c[i][0 + y * stride + off16 * 2]) * matrix[i];
411 }
412
413 for (int off16 = 0; off16 < width; off16++) {
414 sum[off16] = (int)(sum[off16] * rdiv + bias + 0.5f);
415 dst[off16] = av_clip(sum[off16], 0, peak);
416 }
417 dst += dstride / 2;
418 }
419 }
420
421 static void filter_7x7(uint8_t *dst, int width,
422 float rdiv, float bias, const int *const matrix,
423 const uint8_t *c[], int peak, int radius,
424 int dstride, int stride, int size)
425 {
426 int x;
427
428 for (x = 0; x < width; x++) {
429 int i, sum = 0;
430
431 for (i = 0; i < 49; i++)
432 sum += c[i][x] * matrix[i];
433
434 sum = (int)(sum * rdiv + bias + 0.5f);
435 dst[x] = av_clip_uint8(sum);
436 }
437 }
438
439 static void filter_5x5(uint8_t *dst, int width,
440 float rdiv, float bias, const int *const matrix,
441 const uint8_t *c[], int peak, int radius,
442 int dstride, int stride, int size)
443 {
444 int x;
445
446 for (x = 0; x < width; x++) {
447 int i, sum = 0;
448
449 for (i = 0; i < 25; i++)
450 sum += c[i][x] * matrix[i];
451
452 sum = (int)(sum * rdiv + bias + 0.5f);
453 dst[x] = av_clip_uint8(sum);
454 }
455 }
456
457 static void filter_3x3(uint8_t *dst, int width,
458 float rdiv, float bias, const int *const matrix,
459 const uint8_t *c[], int peak, int radius,
460 int dstride, int stride, int size)
461 {
462 const uint8_t *c0 = c[0], *c1 = c[1], *c2 = c[2];
463 const uint8_t *c3 = c[3], *c4 = c[4], *c5 = c[5];
464 const uint8_t *c6 = c[6], *c7 = c[7], *c8 = c[8];
465 int x;
466
467 for (x = 0; x < width; x++) {
468 int sum = c0[x] * matrix[0] + c1[x] * matrix[1] + c2[x] * matrix[2] +
469 c3[x] * matrix[3] + c4[x] * matrix[4] + c5[x] * matrix[5] +
470 c6[x] * matrix[6] + c7[x] * matrix[7] + c8[x] * matrix[8];
471 sum = (int)(sum * rdiv + bias + 0.5f);
472 dst[x] = av_clip_uint8(sum);
473 }
474 }
475
476 static void filter_row(uint8_t *dst, int width,
477 float rdiv, float bias, const int *const matrix,
478 const uint8_t *c[], int peak, int radius,
479 int dstride, int stride, int size)
480 {
481 int x;
482
483 for (x = 0; x < width; x++) {
484 int i, sum = 0;
485
486 for (i = 0; i < 2 * radius + 1; i++)
487 sum += c[i][x] * matrix[i];
488
489 sum = (int)(sum * rdiv + bias + 0.5f);
490 dst[x] = av_clip_uint8(sum);
491 }
492 }
493
494 static void filter_column(uint8_t *dst, int height,
495 float rdiv, float bias, const int *const matrix,
496 const uint8_t *c[], int peak, int radius,
497 int dstride, int stride, int size)
498 {
499 DECLARE_ALIGNED(64, int, sum)[16];
500
501 for (int y = 0; y < height; y++) {
502 memset(sum, 0, sizeof(sum));
503
504 for (int i = 0; i < 2 * radius + 1; i++) {
505 for (int off16 = 0; off16 < 16; off16++)
506 sum[off16] += c[i][0 + y * stride + off16] * matrix[i];
507 }
508
509 for (int off16 = 0; off16 < 16; off16++) {
510 sum[off16] = (int)(sum[off16] * rdiv + bias + 0.5f);
511 dst[off16] = av_clip_uint8(sum[off16]);
512 }
513 dst += dstride;
514 }
515 }
516
517 static void setup_5x5(int radius, const uint8_t *c[], const uint8_t *src, int stride,
518 int x, int w, int y, int h, int bpc)
519 {
520 int i;
521
522 for (i = 0; i < 25; i++) {
523 int xoff = FFABS(x + ((i % 5) - 2));
524 int yoff = FFABS(y + (i / 5) - 2);
525
526 xoff = xoff >= w ? 2 * w - 1 - xoff : xoff;
527 yoff = yoff >= h ? 2 * h - 1 - yoff : yoff;
528
529 c[i] = src + xoff * bpc + yoff * stride;
530 }
531 }
532
533 static void setup_7x7(int radius, const uint8_t *c[], const uint8_t *src, int stride,
534 int x, int w, int y, int h, int bpc)
535 {
536 int i;
537
538 for (i = 0; i < 49; i++) {
539 int xoff = FFABS(x + ((i % 7) - 3));
540 int yoff = FFABS(y + (i / 7) - 3);
541
542 xoff = xoff >= w ? 2 * w - 1 - xoff : xoff;
543 yoff = yoff >= h ? 2 * h - 1 - yoff : yoff;
544
545 c[i] = src + xoff * bpc + yoff * stride;
546 }
547 }
548
549 static void setup_row(int radius, const uint8_t *c[], const uint8_t *src, int stride,
550 int x, int w, int y, int h, int bpc)
551 {
552 int i;
553
554 for (i = 0; i < radius * 2 + 1; i++) {
555 int xoff = FFABS(x + i - radius);
556
557 xoff = xoff >= w ? 2 * w - 1 - xoff : xoff;
558
559 c[i] = src + xoff * bpc + y * stride;
560 }
561 }
562
563 static void setup_column(int radius, const uint8_t *c[], const uint8_t *src, int stride,
564 int x, int w, int y, int h, int bpc)
565 {
566 int i;
567
568 for (i = 0; i < radius * 2 + 1; i++) {
569 int xoff = FFABS(x + i - radius);
570
571 xoff = xoff >= h ? 2 * h - 1 - xoff : xoff;
572
573 c[i] = src + y * bpc + xoff * stride;
574 }
575 }
576
577 static int filter_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
578 {
579 ConvolutionContext *s = ctx->priv;
580 ThreadData *td = arg;
581 AVFrame *in = td->in;
582 AVFrame *out = td->out;
583 int plane;
584
585 for (plane = 0; plane < s->nb_planes; plane++) {
586 const int mode = s->mode[plane];
587 const int bpc = s->bpc;
588 const int radius = s->size[plane] / 2;
589 const int height = s->planeheight[plane];
590 const int width = s->planewidth[plane];
591 const int stride = in->linesize[plane];
592 const int dstride = out->linesize[plane];
593 const int sizeh = mode == MATRIX_COLUMN ? width : height;
594 const int sizew = mode == MATRIX_COLUMN ? height : width;
595 const int slice_start = (sizeh * jobnr) / nb_jobs;
596 const int slice_end = (sizeh * (jobnr+1)) / nb_jobs;
597 const float rdiv = s->rdiv[plane];
598 const float bias = s->bias[plane];
599 const uint8_t *src = in->data[plane];
600 const int dst_pos = slice_start * (mode == MATRIX_COLUMN ? bpc : dstride);
601 uint8_t *dst = out->data[plane] + dst_pos;
602 const int *matrix = s->matrix[plane];
603 const int step = mode == MATRIX_COLUMN ? 16 : 1;
604 const uint8_t *c[49];
605 int y, x;
606
607 if (s->copy[plane]) {
608 if (mode == MATRIX_COLUMN)
609 av_image_copy_plane(dst, dstride, src + slice_start * bpc, stride,
610 (slice_end - slice_start) * bpc, height);
611 else
612 av_image_copy_plane(dst, dstride, src + slice_start * stride, stride,
613 width * bpc, slice_end - slice_start);
614 continue;
615 }
616 for (y = slice_start; y < slice_end; y += step) {
617 const int xoff = mode == MATRIX_COLUMN ? (y - slice_start) * bpc : radius * bpc;
618 const int yoff = mode == MATRIX_COLUMN ? radius * dstride : 0;
619
620 for (x = 0; x < radius; x++) {
621 const int xoff = mode == MATRIX_COLUMN ? (y - slice_start) * bpc : x * bpc;
622 const int yoff = mode == MATRIX_COLUMN ? x * dstride : 0;
623
624 s->setup[plane](radius, c, src, stride, x, width, y, height, bpc);
625 s->filter[plane](dst + yoff + xoff, 1, rdiv,
626 bias, matrix, c, s->max, radius,
627 dstride, stride, slice_end - step);
628 }
629 s->setup[plane](radius, c, src, stride, radius, width, y, height, bpc);
630 s->filter[plane](dst + yoff + xoff, sizew - 2 * radius,
631 rdiv, bias, matrix, c, s->max, radius,
632 dstride, stride, slice_end - step);
633 for (x = sizew - radius; x < sizew; x++) {
634 const int xoff = mode == MATRIX_COLUMN ? (y - slice_start) * bpc : x * bpc;
635 const int yoff = mode == MATRIX_COLUMN ? x * dstride : 0;
636
637 s->setup[plane](radius, c, src, stride, x, width, y, height, bpc);
638 s->filter[plane](dst + yoff + xoff, 1, rdiv,
639 bias, matrix, c, s->max, radius,
640 dstride, stride, slice_end - step);
641 }
642 if (mode != MATRIX_COLUMN)
643 dst += dstride;
644 }
645 }
646
647 return 0;
648 }
649
650 static int param_init(AVFilterContext *ctx)
651 {
652 ConvolutionContext *s = ctx->priv;
653 AVFilterLink *inlink = ctx->inputs[0];
654 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
655 int p, i;
656
657 s->depth = desc->comp[0].depth;
658 s->max = (1 << s->depth) - 1;
659
660 s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
661 s->planewidth[0] = s->planewidth[3] = inlink->w;
662 s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
663 s->planeheight[0] = s->planeheight[3] = inlink->h;
664
665 s->nb_planes = av_pix_fmt_count_planes(inlink->format);
666 s->nb_threads = ff_filter_get_nb_threads(ctx);
667 s->bpc = (s->depth + 7) / 8;
668
669 if (!strcmp(ctx->filter->name, "convolution")) {
670 for (i = 0; i < 4; i++) {
671 int *matrix = (int *)s->matrix[i];
672 char *orig, *p, *arg, *saveptr = NULL;
673 float sum = 1.f;
674
675 p = orig = av_strdup(s->matrix_str[i]);
676 if (p) {
677 s->matrix_length[i] = 0;
678 s->rdiv[i] = 0.f;
679 sum = 0.f;
680
681 while (s->matrix_length[i] < 49) {
682 if (!(arg = av_strtok(p, " |", &saveptr)))
683 break;
684
685 p = NULL;
686 sscanf(arg, "%d", &matrix[s->matrix_length[i]]);
687 sum += matrix[s->matrix_length[i]];
688 s->matrix_length[i]++;
689 }
690
691 av_freep(&orig);
692 if (!(s->matrix_length[i] & 1)) {
693 av_log(ctx, AV_LOG_ERROR, "number of matrix elements must be odd\n");
694 return AVERROR(EINVAL);
695 }
696 }
697
698 if (s->mode[i] == MATRIX_ROW) {
699 s->filter[i] = filter_row;
700 s->setup[i] = setup_row;
701 s->size[i] = s->matrix_length[i];
702 } else if (s->mode[i] == MATRIX_COLUMN) {
703 s->filter[i] = filter_column;
704 s->setup[i] = setup_column;
705 s->size[i] = s->matrix_length[i];
706 } else if (s->matrix_length[i] == 9) {
707 s->size[i] = 3;
708
709 if (!memcmp(matrix, same3x3, sizeof(same3x3))) {
710 s->copy[i] = 1;
711 } else {
712 s->filter[i] = filter_3x3;
713 s->copy[i] = 0;
714 }
715 s->setup[i] = setup_3x3;
716 } else if (s->matrix_length[i] == 25) {
717 s->size[i] = 5;
718 if (!memcmp(matrix, same5x5, sizeof(same5x5))) {
719 s->copy[i] = 1;
720 } else {
721 s->filter[i] = filter_5x5;
722 s->copy[i] = 0;
723 }
724 s->setup[i] = setup_5x5;
725 } else if (s->matrix_length[i] == 49) {
726 s->size[i] = 7;
727 if (!memcmp(matrix, same7x7, sizeof(same7x7))) {
728 s->copy[i] = 1;
729 } else {
730 s->filter[i] = filter_7x7;
731 s->copy[i] = 0;
732 }
733 s->setup[i] = setup_7x7;
734 } else {
735 return AVERROR(EINVAL);
736 }
737
738 if (sum == 0)
739 sum = 1;
740 if (s->rdiv[i] == 0)
741 s->rdiv[i] = 1. / sum;
742
743 if (s->copy[i] && (s->rdiv[i] != 1. || s->bias[i] != 0.))
744 s->copy[i] = 0;
745 }
746 } else if (!strcmp(ctx->filter->name, "prewitt")) {
747 for (i = 0; i < 4; i++) {
748 s->filter[i] = filter_prewitt;
749 s->copy[i] = !((1 << i) & s->planes);
750 s->size[i] = 3;
751 s->setup[i] = setup_3x3;
752 s->rdiv[i] = s->scale;
753 s->bias[i] = s->delta;
754 }
755 } else if (!strcmp(ctx->filter->name, "roberts")) {
756 for (i = 0; i < 4; i++) {
757 s->filter[i] = filter_roberts;
758 s->copy[i] = !((1 << i) & s->planes);
759 s->size[i] = 3;
760 s->setup[i] = setup_3x3;
761 s->rdiv[i] = s->scale;
762 s->bias[i] = s->delta;
763 }
764 } else if (!strcmp(ctx->filter->name, "sobel")) {
765 ff_sobel_init(s, s->depth, s->nb_planes);
766 } else if (!strcmp(ctx->filter->name, "kirsch")) {
767 for (i = 0; i < 4; i++) {
768 s->filter[i] = filter_kirsch;
769 s->copy[i] = !((1 << i) & s->planes);
770 s->size[i] = 3;
771 s->setup[i] = setup_3x3;
772 s->rdiv[i] = s->scale;
773 s->bias[i] = s->delta;
774 }
775 } else if (!strcmp(ctx->filter->name, "scharr")) {
776 for (i = 0; i < 4; i++) {
777 s->filter[i] = filter_scharr;
778 s->copy[i] = !((1 << i) & s->planes);
779 s->size[i] = 3;
780 s->setup[i] = setup_3x3;
781 s->rdiv[i] = s->scale;
782 s->bias[i] = s->delta;
783 }
784 }
785
786 if (!strcmp(ctx->filter->name, "convolution")) {
787 if (s->depth > 8) {
788 for (p = 0; p < s->nb_planes; p++) {
789 if (s->mode[p] == MATRIX_ROW)
790 s->filter[p] = filter16_row;
791 else if (s->mode[p] == MATRIX_COLUMN)
792 s->filter[p] = filter16_column;
793 else if (s->size[p] == 3)
794 s->filter[p] = filter16_3x3;
795 else if (s->size[p] == 5)
796 s->filter[p] = filter16_5x5;
797 else if (s->size[p] == 7)
798 s->filter[p] = filter16_7x7;
799 }
800 }
801 #if CONFIG_CONVOLUTION_FILTER && ARCH_X86_64
802 ff_convolution_init_x86(s);
803 #endif
804 } else if (!strcmp(ctx->filter->name, "prewitt")) {
805 if (s->depth > 8)
806 for (p = 0; p < s->nb_planes; p++)
807 s->filter[p] = filter16_prewitt;
808 } else if (!strcmp(ctx->filter->name, "roberts")) {
809 if (s->depth > 8)
810 for (p = 0; p < s->nb_planes; p++)
811 s->filter[p] = filter16_roberts;
812 } else if (!strcmp(ctx->filter->name, "kirsch")) {
813 if (s->depth > 8)
814 for (p = 0; p < s->nb_planes; p++)
815 s->filter[p] = filter16_kirsch;
816 } else if (!strcmp(ctx->filter->name, "scharr")) {
817 if (s->depth > 8)
818 for (p = 0; p < s->nb_planes; p++)
819 s->filter[p] = filter16_scharr;
820 }
821
822 return 0;
823 }
824
825 static int config_input(AVFilterLink *inlink)
826 {
827 AVFilterContext *ctx = inlink->dst;
828 return param_init(ctx);
829 }
830
831 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
832 {
833 AVFilterContext *ctx = inlink->dst;
834 ConvolutionContext *s = ctx->priv;
835 AVFilterLink *outlink = ctx->outputs[0];
836 AVFrame *out;
837 ThreadData td;
838
839 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
840 if (!out) {
841 av_frame_free(&in);
842 return AVERROR(ENOMEM);
843 }
844 av_frame_copy_props(out, in);
845
846 td.in = in;
847 td.out = out;
848 ff_filter_execute(ctx, filter_slice, &td, NULL,
849 FFMIN3(s->planeheight[1], s->planewidth[1], s->nb_threads));
850
851 av_frame_free(&in);
852 return ff_filter_frame(outlink, out);
853 }
854
855 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
856 char *res, int res_len, int flags)
857 {
858 int ret;
859
860 ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
861 if (ret < 0)
862 return ret;
863
864 return param_init(ctx);
865 }
866
867 static const AVFilterPad convolution_inputs[] = {
868 {
869 .name = "default",
870 .type = AVMEDIA_TYPE_VIDEO,
871 .config_props = config_input,
872 .filter_frame = filter_frame,
873 },
874 };
875
876 static const AVFilterPad convolution_outputs[] = {
877 {
878 .name = "default",
879 .type = AVMEDIA_TYPE_VIDEO,
880 },
881 };
882
883 #if CONFIG_CONVOLUTION_FILTER
884
885 const AVFilter ff_vf_convolution = {
886 .name = "convolution",
887 .description = NULL_IF_CONFIG_SMALL("Apply convolution filter."),
888 .priv_size = sizeof(ConvolutionContext),
889 .priv_class = &convolution_class,
890 FILTER_INPUTS(convolution_inputs),
891 FILTER_OUTPUTS(convolution_outputs),
892 FILTER_PIXFMTS_ARRAY(pix_fmts),
893 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
894 .process_command = process_command,
895 };
896
897 #endif /* CONFIG_CONVOLUTION_FILTER */
898
899 static const AVOption common_options[] = {
900 { "planes", "set planes to filter", OFFSET(planes), AV_OPT_TYPE_INT, {.i64=15}, 0, 15, FLAGS},
901 { "scale", "set scale", OFFSET(scale), AV_OPT_TYPE_FLOAT, {.dbl=1.0}, 0.0, 65535, FLAGS},
902 { "delta", "set delta", OFFSET(delta), AV_OPT_TYPE_FLOAT, {.dbl=0}, -65535, 65535, FLAGS},
903 { NULL }
904 };
905
906 AVFILTER_DEFINE_CLASS_EXT(common, "kirsch/prewitt/roberts/scharr/sobel",
907 common_options);
908
909 #if CONFIG_PREWITT_FILTER
910
911 const AVFilter ff_vf_prewitt = {
912 .name = "prewitt",
913 .description = NULL_IF_CONFIG_SMALL("Apply prewitt operator."),
914 .priv_size = sizeof(ConvolutionContext),
915 .priv_class = &common_class,
916 FILTER_INPUTS(convolution_inputs),
917 FILTER_OUTPUTS(convolution_outputs),
918 FILTER_PIXFMTS_ARRAY(pix_fmts),
919 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
920 .process_command = process_command,
921 };
922
923 #endif /* CONFIG_PREWITT_FILTER */
924
925 #if CONFIG_SOBEL_FILTER
926
927 const AVFilter ff_vf_sobel = {
928 .name = "sobel",
929 .description = NULL_IF_CONFIG_SMALL("Apply sobel operator."),
930 .priv_size = sizeof(ConvolutionContext),
931 .priv_class = &common_class,
932 FILTER_INPUTS(convolution_inputs),
933 FILTER_OUTPUTS(convolution_outputs),
934 FILTER_PIXFMTS_ARRAY(pix_fmts),
935 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
936 .process_command = process_command,
937 };
938
939 #endif /* CONFIG_SOBEL_FILTER */
940
941 #if CONFIG_ROBERTS_FILTER
942
943 const AVFilter ff_vf_roberts = {
944 .name = "roberts",
945 .description = NULL_IF_CONFIG_SMALL("Apply roberts cross operator."),
946 .priv_size = sizeof(ConvolutionContext),
947 .priv_class = &common_class,
948 FILTER_INPUTS(convolution_inputs),
949 FILTER_OUTPUTS(convolution_outputs),
950 FILTER_PIXFMTS_ARRAY(pix_fmts),
951 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
952 .process_command = process_command,
953 };
954
955 #endif /* CONFIG_ROBERTS_FILTER */
956
957 #if CONFIG_KIRSCH_FILTER
958
959 const AVFilter ff_vf_kirsch = {
960 .name = "kirsch",
961 .description = NULL_IF_CONFIG_SMALL("Apply kirsch operator."),
962 .priv_size = sizeof(ConvolutionContext),
963 .priv_class = &common_class,
964 FILTER_INPUTS(convolution_inputs),
965 FILTER_OUTPUTS(convolution_outputs),
966 FILTER_PIXFMTS_ARRAY(pix_fmts),
967 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
968 .process_command = process_command,
969 };
970
971 #endif /* CONFIG_KIRSCH_FILTER */
972
973 #if CONFIG_SCHARR_FILTER
974
975 const AVFilter ff_vf_scharr = {
976 .name = "scharr",
977 .description = NULL_IF_CONFIG_SMALL("Apply scharr operator."),
978 .priv_size = sizeof(ConvolutionContext),
979 .priv_class = &common_class,
980 FILTER_INPUTS(convolution_inputs),
981 FILTER_OUTPUTS(convolution_outputs),
982 FILTER_PIXFMTS_ARRAY(pix_fmts),
983 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
984 .process_command = process_command,
985 };
986
987 #endif /* CONFIG_SCHARR_FILTER */
988