blob: 7a294b07610f3c45184e8c60e774a12e2b0d4122
1 | /* |
2 | * Copyright (c) 2013 Clément Bœsch |
3 | * |
4 | * This file is part of FFmpeg. |
5 | * |
6 | * FFmpeg is free software; you can redistribute it and/or |
7 | * modify it under the terms of the GNU Lesser General Public |
8 | * License as published by the Free Software Foundation; either |
9 | * version 2.1 of the License, or (at your option) any later version. |
10 | * |
11 | * FFmpeg is distributed in the hope that it will be useful, |
12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
14 | * Lesser General Public License for more details. |
15 | * |
16 | * You should have received a copy of the GNU Lesser General Public |
17 | * License along with FFmpeg; if not, write to the Free Software |
18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
19 | */ |
20 | |
21 | /** |
22 | * @file |
23 | * 3D Lookup table filter |
24 | */ |
25 | |
26 | #include "libavutil/opt.h" |
27 | #include "libavutil/file.h" |
28 | #include "libavutil/intreadwrite.h" |
29 | #include "libavutil/avassert.h" |
30 | #include "libavutil/pixdesc.h" |
31 | #include "libavutil/avstring.h" |
32 | #include "avfilter.h" |
33 | #include "drawutils.h" |
34 | #include "dualinput.h" |
35 | #include "formats.h" |
36 | #include "internal.h" |
37 | #include "video.h" |
38 | |
39 | #define R 0 |
40 | #define G 1 |
41 | #define B 2 |
42 | #define A 3 |
43 | |
44 | enum interp_mode { |
45 | INTERPOLATE_NEAREST, |
46 | INTERPOLATE_TRILINEAR, |
47 | INTERPOLATE_TETRAHEDRAL, |
48 | NB_INTERP_MODE |
49 | }; |
50 | |
51 | struct rgbvec { |
52 | float r, g, b; |
53 | }; |
54 | |
55 | /* 3D LUT don't often go up to level 32, but it is common to have a Hald CLUT |
56 | * of 512x512 (64x64x64) */ |
57 | #define MAX_LEVEL 64 |
58 | |
59 | typedef struct LUT3DContext { |
60 | const AVClass *class; |
61 | int interpolation; ///<interp_mode |
62 | char *file; |
63 | uint8_t rgba_map[4]; |
64 | int step; |
65 | avfilter_action_func *interp; |
66 | struct rgbvec lut[MAX_LEVEL][MAX_LEVEL][MAX_LEVEL]; |
67 | int lutsize; |
68 | #if CONFIG_HALDCLUT_FILTER |
69 | uint8_t clut_rgba_map[4]; |
70 | int clut_step; |
71 | int clut_is16bit; |
72 | int clut_width; |
73 | FFDualInputContext dinput; |
74 | #endif |
75 | } LUT3DContext; |
76 | |
77 | typedef struct ThreadData { |
78 | AVFrame *in, *out; |
79 | } ThreadData; |
80 | |
81 | #define OFFSET(x) offsetof(LUT3DContext, x) |
82 | #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM |
83 | #define COMMON_OPTIONS \ |
84 | { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, NB_INTERP_MODE-1, FLAGS, "interp_mode" }, \ |
85 | { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_NEAREST}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \ |
86 | { "trilinear", "interpolate values using the 8 points defining a cube", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TRILINEAR}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \ |
87 | { "tetrahedral", "interpolate values using a tetrahedron", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TETRAHEDRAL}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \ |
88 | { NULL } |
89 | |
90 | static inline float lerpf(float v0, float v1, float f) |
91 | { |
92 | return v0 + (v1 - v0) * f; |
93 | } |
94 | |
95 | static inline struct rgbvec lerp(const struct rgbvec *v0, const struct rgbvec *v1, float f) |
96 | { |
97 | struct rgbvec v = { |
98 | lerpf(v0->r, v1->r, f), lerpf(v0->g, v1->g, f), lerpf(v0->b, v1->b, f) |
99 | }; |
100 | return v; |
101 | } |
102 | |
103 | #define NEAR(x) ((int)((x) + .5)) |
104 | #define PREV(x) ((int)(x)) |
105 | #define NEXT(x) (FFMIN((int)(x) + 1, lut3d->lutsize - 1)) |
106 | |
107 | /** |
108 | * Get the nearest defined point |
109 | */ |
110 | static inline struct rgbvec interp_nearest(const LUT3DContext *lut3d, |
111 | const struct rgbvec *s) |
112 | { |
113 | return lut3d->lut[NEAR(s->r)][NEAR(s->g)][NEAR(s->b)]; |
114 | } |
115 | |
116 | /** |
117 | * Interpolate using the 8 vertices of a cube |
118 | * @see https://en.wikipedia.org/wiki/Trilinear_interpolation |
119 | */ |
120 | static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d, |
121 | const struct rgbvec *s) |
122 | { |
123 | const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)}; |
124 | const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)}; |
125 | const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]}; |
126 | const struct rgbvec c000 = lut3d->lut[prev[0]][prev[1]][prev[2]]; |
127 | const struct rgbvec c001 = lut3d->lut[prev[0]][prev[1]][next[2]]; |
128 | const struct rgbvec c010 = lut3d->lut[prev[0]][next[1]][prev[2]]; |
129 | const struct rgbvec c011 = lut3d->lut[prev[0]][next[1]][next[2]]; |
130 | const struct rgbvec c100 = lut3d->lut[next[0]][prev[1]][prev[2]]; |
131 | const struct rgbvec c101 = lut3d->lut[next[0]][prev[1]][next[2]]; |
132 | const struct rgbvec c110 = lut3d->lut[next[0]][next[1]][prev[2]]; |
133 | const struct rgbvec c111 = lut3d->lut[next[0]][next[1]][next[2]]; |
134 | const struct rgbvec c00 = lerp(&c000, &c100, d.r); |
135 | const struct rgbvec c10 = lerp(&c010, &c110, d.r); |
136 | const struct rgbvec c01 = lerp(&c001, &c101, d.r); |
137 | const struct rgbvec c11 = lerp(&c011, &c111, d.r); |
138 | const struct rgbvec c0 = lerp(&c00, &c10, d.g); |
139 | const struct rgbvec c1 = lerp(&c01, &c11, d.g); |
140 | const struct rgbvec c = lerp(&c0, &c1, d.b); |
141 | return c; |
142 | } |
143 | |
144 | /** |
145 | * Tetrahedral interpolation. Based on code found in Truelight Software Library paper. |
146 | * @see http://www.filmlight.ltd.uk/pdf/whitepapers/FL-TL-TN-0057-SoftwareLib.pdf |
147 | */ |
148 | static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d, |
149 | const struct rgbvec *s) |
150 | { |
151 | const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)}; |
152 | const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)}; |
153 | const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]}; |
154 | const struct rgbvec c000 = lut3d->lut[prev[0]][prev[1]][prev[2]]; |
155 | const struct rgbvec c111 = lut3d->lut[next[0]][next[1]][next[2]]; |
156 | struct rgbvec c; |
157 | if (d.r > d.g) { |
158 | if (d.g > d.b) { |
159 | const struct rgbvec c100 = lut3d->lut[next[0]][prev[1]][prev[2]]; |
160 | const struct rgbvec c110 = lut3d->lut[next[0]][next[1]][prev[2]]; |
161 | c.r = (1-d.r) * c000.r + (d.r-d.g) * c100.r + (d.g-d.b) * c110.r + (d.b) * c111.r; |
162 | c.g = (1-d.r) * c000.g + (d.r-d.g) * c100.g + (d.g-d.b) * c110.g + (d.b) * c111.g; |
163 | c.b = (1-d.r) * c000.b + (d.r-d.g) * c100.b + (d.g-d.b) * c110.b + (d.b) * c111.b; |
164 | } else if (d.r > d.b) { |
165 | const struct rgbvec c100 = lut3d->lut[next[0]][prev[1]][prev[2]]; |
166 | const struct rgbvec c101 = lut3d->lut[next[0]][prev[1]][next[2]]; |
167 | c.r = (1-d.r) * c000.r + (d.r-d.b) * c100.r + (d.b-d.g) * c101.r + (d.g) * c111.r; |
168 | c.g = (1-d.r) * c000.g + (d.r-d.b) * c100.g + (d.b-d.g) * c101.g + (d.g) * c111.g; |
169 | c.b = (1-d.r) * c000.b + (d.r-d.b) * c100.b + (d.b-d.g) * c101.b + (d.g) * c111.b; |
170 | } else { |
171 | const struct rgbvec c001 = lut3d->lut[prev[0]][prev[1]][next[2]]; |
172 | const struct rgbvec c101 = lut3d->lut[next[0]][prev[1]][next[2]]; |
173 | c.r = (1-d.b) * c000.r + (d.b-d.r) * c001.r + (d.r-d.g) * c101.r + (d.g) * c111.r; |
174 | c.g = (1-d.b) * c000.g + (d.b-d.r) * c001.g + (d.r-d.g) * c101.g + (d.g) * c111.g; |
175 | c.b = (1-d.b) * c000.b + (d.b-d.r) * c001.b + (d.r-d.g) * c101.b + (d.g) * c111.b; |
176 | } |
177 | } else { |
178 | if (d.b > d.g) { |
179 | const struct rgbvec c001 = lut3d->lut[prev[0]][prev[1]][next[2]]; |
180 | const struct rgbvec c011 = lut3d->lut[prev[0]][next[1]][next[2]]; |
181 | c.r = (1-d.b) * c000.r + (d.b-d.g) * c001.r + (d.g-d.r) * c011.r + (d.r) * c111.r; |
182 | c.g = (1-d.b) * c000.g + (d.b-d.g) * c001.g + (d.g-d.r) * c011.g + (d.r) * c111.g; |
183 | c.b = (1-d.b) * c000.b + (d.b-d.g) * c001.b + (d.g-d.r) * c011.b + (d.r) * c111.b; |
184 | } else if (d.b > d.r) { |
185 | const struct rgbvec c010 = lut3d->lut[prev[0]][next[1]][prev[2]]; |
186 | const struct rgbvec c011 = lut3d->lut[prev[0]][next[1]][next[2]]; |
187 | c.r = (1-d.g) * c000.r + (d.g-d.b) * c010.r + (d.b-d.r) * c011.r + (d.r) * c111.r; |
188 | c.g = (1-d.g) * c000.g + (d.g-d.b) * c010.g + (d.b-d.r) * c011.g + (d.r) * c111.g; |
189 | c.b = (1-d.g) * c000.b + (d.g-d.b) * c010.b + (d.b-d.r) * c011.b + (d.r) * c111.b; |
190 | } else { |
191 | const struct rgbvec c010 = lut3d->lut[prev[0]][next[1]][prev[2]]; |
192 | const struct rgbvec c110 = lut3d->lut[next[0]][next[1]][prev[2]]; |
193 | c.r = (1-d.g) * c000.r + (d.g-d.r) * c010.r + (d.r-d.b) * c110.r + (d.b) * c111.r; |
194 | c.g = (1-d.g) * c000.g + (d.g-d.r) * c010.g + (d.r-d.b) * c110.g + (d.b) * c111.g; |
195 | c.b = (1-d.g) * c000.b + (d.g-d.r) * c010.b + (d.r-d.b) * c110.b + (d.b) * c111.b; |
196 | } |
197 | } |
198 | return c; |
199 | } |
200 | |
201 | #define DEFINE_INTERP_FUNC(name, nbits) \ |
202 | static int interp_##nbits##_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \ |
203 | { \ |
204 | int x, y; \ |
205 | const LUT3DContext *lut3d = ctx->priv; \ |
206 | const ThreadData *td = arg; \ |
207 | const AVFrame *in = td->in; \ |
208 | const AVFrame *out = td->out; \ |
209 | const int direct = out == in; \ |
210 | const int step = lut3d->step; \ |
211 | const uint8_t r = lut3d->rgba_map[R]; \ |
212 | const uint8_t g = lut3d->rgba_map[G]; \ |
213 | const uint8_t b = lut3d->rgba_map[B]; \ |
214 | const uint8_t a = lut3d->rgba_map[A]; \ |
215 | const int slice_start = (in->height * jobnr ) / nb_jobs; \ |
216 | const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \ |
217 | uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \ |
218 | const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \ |
219 | const float scale = (1. / ((1<<nbits) - 1)) * (lut3d->lutsize - 1); \ |
220 | \ |
221 | for (y = slice_start; y < slice_end; y++) { \ |
222 | uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \ |
223 | const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \ |
224 | for (x = 0; x < in->width * step; x += step) { \ |
225 | const struct rgbvec scaled_rgb = {src[x + r] * scale, \ |
226 | src[x + g] * scale, \ |
227 | src[x + b] * scale}; \ |
228 | struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \ |
229 | dst[x + r] = av_clip_uint##nbits(vec.r * (float)((1<<nbits) - 1)); \ |
230 | dst[x + g] = av_clip_uint##nbits(vec.g * (float)((1<<nbits) - 1)); \ |
231 | dst[x + b] = av_clip_uint##nbits(vec.b * (float)((1<<nbits) - 1)); \ |
232 | if (!direct && step == 4) \ |
233 | dst[x + a] = src[x + a]; \ |
234 | } \ |
235 | dstrow += out->linesize[0]; \ |
236 | srcrow += in ->linesize[0]; \ |
237 | } \ |
238 | return 0; \ |
239 | } |
240 | |
241 | DEFINE_INTERP_FUNC(nearest, 8) |
242 | DEFINE_INTERP_FUNC(trilinear, 8) |
243 | DEFINE_INTERP_FUNC(tetrahedral, 8) |
244 | |
245 | DEFINE_INTERP_FUNC(nearest, 16) |
246 | DEFINE_INTERP_FUNC(trilinear, 16) |
247 | DEFINE_INTERP_FUNC(tetrahedral, 16) |
248 | |
249 | #define MAX_LINE_SIZE 512 |
250 | |
251 | static int skip_line(const char *p) |
252 | { |
253 | while (*p && av_isspace(*p)) |
254 | p++; |
255 | return !*p || *p == '#'; |
256 | } |
257 | |
258 | #define NEXT_LINE(loop_cond) do { \ |
259 | if (!fgets(line, sizeof(line), f)) { \ |
260 | av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \ |
261 | return AVERROR_INVALIDDATA; \ |
262 | } \ |
263 | } while (loop_cond) |
264 | |
265 | /* Basically r g and b float values on each line, with a facultative 3DLUTSIZE |
266 | * directive; seems to be generated by Davinci */ |
267 | static int parse_dat(AVFilterContext *ctx, FILE *f) |
268 | { |
269 | LUT3DContext *lut3d = ctx->priv; |
270 | char line[MAX_LINE_SIZE]; |
271 | int i, j, k, size; |
272 | |
273 | lut3d->lutsize = size = 33; |
274 | |
275 | NEXT_LINE(skip_line(line)); |
276 | if (!strncmp(line, "3DLUTSIZE ", 10)) { |
277 | size = strtol(line + 10, NULL, 0); |
278 | if (size < 2 || size > MAX_LEVEL) { |
279 | av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n"); |
280 | return AVERROR(EINVAL); |
281 | } |
282 | lut3d->lutsize = size; |
283 | NEXT_LINE(skip_line(line)); |
284 | } |
285 | for (k = 0; k < size; k++) { |
286 | for (j = 0; j < size; j++) { |
287 | for (i = 0; i < size; i++) { |
288 | struct rgbvec *vec = &lut3d->lut[k][j][i]; |
289 | if (k != 0 || j != 0 || i != 0) |
290 | NEXT_LINE(skip_line(line)); |
291 | if (sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) |
292 | return AVERROR_INVALIDDATA; |
293 | } |
294 | } |
295 | } |
296 | return 0; |
297 | } |
298 | |
299 | /* Iridas format */ |
300 | static int parse_cube(AVFilterContext *ctx, FILE *f) |
301 | { |
302 | LUT3DContext *lut3d = ctx->priv; |
303 | char line[MAX_LINE_SIZE]; |
304 | float min[3] = {0.0, 0.0, 0.0}; |
305 | float max[3] = {1.0, 1.0, 1.0}; |
306 | |
307 | while (fgets(line, sizeof(line), f)) { |
308 | if (!strncmp(line, "LUT_3D_SIZE ", 12)) { |
309 | int i, j, k; |
310 | const int size = strtol(line + 12, NULL, 0); |
311 | |
312 | if (size < 2 || size > MAX_LEVEL) { |
313 | av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n"); |
314 | return AVERROR(EINVAL); |
315 | } |
316 | lut3d->lutsize = size; |
317 | for (k = 0; k < size; k++) { |
318 | for (j = 0; j < size; j++) { |
319 | for (i = 0; i < size; i++) { |
320 | struct rgbvec *vec = &lut3d->lut[i][j][k]; |
321 | |
322 | do { |
323 | try_again: |
324 | NEXT_LINE(0); |
325 | if (!strncmp(line, "DOMAIN_", 7)) { |
326 | float *vals = NULL; |
327 | if (!strncmp(line + 7, "MIN ", 4)) vals = min; |
328 | else if (!strncmp(line + 7, "MAX ", 4)) vals = max; |
329 | if (!vals) |
330 | return AVERROR_INVALIDDATA; |
331 | sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2); |
332 | av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n", |
333 | min[0], min[1], min[2], max[0], max[1], max[2]); |
334 | goto try_again; |
335 | } |
336 | } while (skip_line(line)); |
337 | if (sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) |
338 | return AVERROR_INVALIDDATA; |
339 | vec->r *= max[0] - min[0]; |
340 | vec->g *= max[1] - min[1]; |
341 | vec->b *= max[2] - min[2]; |
342 | } |
343 | } |
344 | } |
345 | break; |
346 | } |
347 | } |
348 | return 0; |
349 | } |
350 | |
351 | /* Assume 17x17x17 LUT with a 16-bit depth |
352 | * FIXME: it seems there are various 3dl formats */ |
353 | static int parse_3dl(AVFilterContext *ctx, FILE *f) |
354 | { |
355 | char line[MAX_LINE_SIZE]; |
356 | LUT3DContext *lut3d = ctx->priv; |
357 | int i, j, k; |
358 | const int size = 17; |
359 | const float scale = 16*16*16; |
360 | |
361 | lut3d->lutsize = size; |
362 | NEXT_LINE(skip_line(line)); |
363 | for (k = 0; k < size; k++) { |
364 | for (j = 0; j < size; j++) { |
365 | for (i = 0; i < size; i++) { |
366 | int r, g, b; |
367 | struct rgbvec *vec = &lut3d->lut[k][j][i]; |
368 | |
369 | NEXT_LINE(skip_line(line)); |
370 | if (sscanf(line, "%d %d %d", &r, &g, &b) != 3) |
371 | return AVERROR_INVALIDDATA; |
372 | vec->r = r / scale; |
373 | vec->g = g / scale; |
374 | vec->b = b / scale; |
375 | } |
376 | } |
377 | } |
378 | return 0; |
379 | } |
380 | |
381 | /* Pandora format */ |
382 | static int parse_m3d(AVFilterContext *ctx, FILE *f) |
383 | { |
384 | LUT3DContext *lut3d = ctx->priv; |
385 | float scale; |
386 | int i, j, k, size, in = -1, out = -1; |
387 | char line[MAX_LINE_SIZE]; |
388 | uint8_t rgb_map[3] = {0, 1, 2}; |
389 | |
390 | while (fgets(line, sizeof(line), f)) { |
391 | if (!strncmp(line, "in", 2)) in = strtol(line + 2, NULL, 0); |
392 | else if (!strncmp(line, "out", 3)) out = strtol(line + 3, NULL, 0); |
393 | else if (!strncmp(line, "values", 6)) { |
394 | const char *p = line + 6; |
395 | #define SET_COLOR(id) do { \ |
396 | while (av_isspace(*p)) \ |
397 | p++; \ |
398 | switch (*p) { \ |
399 | case 'r': rgb_map[id] = 0; break; \ |
400 | case 'g': rgb_map[id] = 1; break; \ |
401 | case 'b': rgb_map[id] = 2; break; \ |
402 | } \ |
403 | while (*p && !av_isspace(*p)) \ |
404 | p++; \ |
405 | } while (0) |
406 | SET_COLOR(0); |
407 | SET_COLOR(1); |
408 | SET_COLOR(2); |
409 | break; |
410 | } |
411 | } |
412 | |
413 | if (in == -1 || out == -1) { |
414 | av_log(ctx, AV_LOG_ERROR, "in and out must be defined\n"); |
415 | return AVERROR_INVALIDDATA; |
416 | } |
417 | if (in < 2 || out < 2 || |
418 | in > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL || |
419 | out > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL) { |
420 | av_log(ctx, AV_LOG_ERROR, "invalid in (%d) or out (%d)\n", in, out); |
421 | return AVERROR_INVALIDDATA; |
422 | } |
423 | for (size = 1; size*size*size < in; size++); |
424 | lut3d->lutsize = size; |
425 | scale = 1. / (out - 1); |
426 | |
427 | for (k = 0; k < size; k++) { |
428 | for (j = 0; j < size; j++) { |
429 | for (i = 0; i < size; i++) { |
430 | struct rgbvec *vec = &lut3d->lut[k][j][i]; |
431 | float val[3]; |
432 | |
433 | NEXT_LINE(0); |
434 | if (sscanf(line, "%f %f %f", val, val + 1, val + 2) != 3) |
435 | return AVERROR_INVALIDDATA; |
436 | vec->r = val[rgb_map[0]] * scale; |
437 | vec->g = val[rgb_map[1]] * scale; |
438 | vec->b = val[rgb_map[2]] * scale; |
439 | } |
440 | } |
441 | } |
442 | return 0; |
443 | } |
444 | |
445 | static void set_identity_matrix(LUT3DContext *lut3d, int size) |
446 | { |
447 | int i, j, k; |
448 | const float c = 1. / (size - 1); |
449 | |
450 | lut3d->lutsize = size; |
451 | for (k = 0; k < size; k++) { |
452 | for (j = 0; j < size; j++) { |
453 | for (i = 0; i < size; i++) { |
454 | struct rgbvec *vec = &lut3d->lut[k][j][i]; |
455 | vec->r = k * c; |
456 | vec->g = j * c; |
457 | vec->b = i * c; |
458 | } |
459 | } |
460 | } |
461 | } |
462 | |
463 | static int query_formats(AVFilterContext *ctx) |
464 | { |
465 | static const enum AVPixelFormat pix_fmts[] = { |
466 | AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24, |
467 | AV_PIX_FMT_RGBA, AV_PIX_FMT_BGRA, |
468 | AV_PIX_FMT_ARGB, AV_PIX_FMT_ABGR, |
469 | AV_PIX_FMT_0RGB, AV_PIX_FMT_0BGR, |
470 | AV_PIX_FMT_RGB0, AV_PIX_FMT_BGR0, |
471 | AV_PIX_FMT_RGB48, AV_PIX_FMT_BGR48, |
472 | AV_PIX_FMT_RGBA64, AV_PIX_FMT_BGRA64, |
473 | AV_PIX_FMT_NONE |
474 | }; |
475 | AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts); |
476 | if (!fmts_list) |
477 | return AVERROR(ENOMEM); |
478 | return ff_set_common_formats(ctx, fmts_list); |
479 | } |
480 | |
481 | static int config_input(AVFilterLink *inlink) |
482 | { |
483 | int is16bit = 0; |
484 | LUT3DContext *lut3d = inlink->dst->priv; |
485 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); |
486 | |
487 | switch (inlink->format) { |
488 | case AV_PIX_FMT_RGB48: |
489 | case AV_PIX_FMT_BGR48: |
490 | case AV_PIX_FMT_RGBA64: |
491 | case AV_PIX_FMT_BGRA64: |
492 | is16bit = 1; |
493 | } |
494 | |
495 | ff_fill_rgba_map(lut3d->rgba_map, inlink->format); |
496 | lut3d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit); |
497 | |
498 | #define SET_FUNC(name) do { \ |
499 | if (is16bit) lut3d->interp = interp_16_##name; \ |
500 | else lut3d->interp = interp_8_##name; \ |
501 | } while (0) |
502 | |
503 | switch (lut3d->interpolation) { |
504 | case INTERPOLATE_NEAREST: SET_FUNC(nearest); break; |
505 | case INTERPOLATE_TRILINEAR: SET_FUNC(trilinear); break; |
506 | case INTERPOLATE_TETRAHEDRAL: SET_FUNC(tetrahedral); break; |
507 | default: |
508 | av_assert0(0); |
509 | } |
510 | |
511 | return 0; |
512 | } |
513 | |
514 | static AVFrame *apply_lut(AVFilterLink *inlink, AVFrame *in) |
515 | { |
516 | AVFilterContext *ctx = inlink->dst; |
517 | LUT3DContext *lut3d = ctx->priv; |
518 | AVFilterLink *outlink = inlink->dst->outputs[0]; |
519 | AVFrame *out; |
520 | ThreadData td; |
521 | |
522 | if (av_frame_is_writable(in)) { |
523 | out = in; |
524 | } else { |
525 | out = ff_get_video_buffer(outlink, outlink->w, outlink->h); |
526 | if (!out) { |
527 | av_frame_free(&in); |
528 | return NULL; |
529 | } |
530 | av_frame_copy_props(out, in); |
531 | } |
532 | |
533 | td.in = in; |
534 | td.out = out; |
535 | ctx->internal->execute(ctx, lut3d->interp, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx))); |
536 | |
537 | if (out != in) |
538 | av_frame_free(&in); |
539 | |
540 | return out; |
541 | } |
542 | |
543 | static int filter_frame(AVFilterLink *inlink, AVFrame *in) |
544 | { |
545 | AVFilterLink *outlink = inlink->dst->outputs[0]; |
546 | AVFrame *out = apply_lut(inlink, in); |
547 | if (!out) |
548 | return AVERROR(ENOMEM); |
549 | return ff_filter_frame(outlink, out); |
550 | } |
551 | |
552 | #if CONFIG_LUT3D_FILTER |
553 | static const AVOption lut3d_options[] = { |
554 | { "file", "set 3D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, |
555 | COMMON_OPTIONS |
556 | }; |
557 | |
558 | AVFILTER_DEFINE_CLASS(lut3d); |
559 | |
560 | static av_cold int lut3d_init(AVFilterContext *ctx) |
561 | { |
562 | int ret; |
563 | FILE *f; |
564 | const char *ext; |
565 | LUT3DContext *lut3d = ctx->priv; |
566 | |
567 | if (!lut3d->file) { |
568 | set_identity_matrix(lut3d, 32); |
569 | return 0; |
570 | } |
571 | |
572 | f = fopen(lut3d->file, "r"); |
573 | if (!f) { |
574 | ret = AVERROR(errno); |
575 | av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret)); |
576 | return ret; |
577 | } |
578 | |
579 | ext = strrchr(lut3d->file, '.'); |
580 | if (!ext) { |
581 | av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n"); |
582 | ret = AVERROR_INVALIDDATA; |
583 | goto end; |
584 | } |
585 | ext++; |
586 | |
587 | if (!av_strcasecmp(ext, "dat")) { |
588 | ret = parse_dat(ctx, f); |
589 | } else if (!av_strcasecmp(ext, "3dl")) { |
590 | ret = parse_3dl(ctx, f); |
591 | } else if (!av_strcasecmp(ext, "cube")) { |
592 | ret = parse_cube(ctx, f); |
593 | } else if (!av_strcasecmp(ext, "m3d")) { |
594 | ret = parse_m3d(ctx, f); |
595 | } else { |
596 | av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext); |
597 | ret = AVERROR(EINVAL); |
598 | } |
599 | |
600 | if (!ret && !lut3d->lutsize) { |
601 | av_log(ctx, AV_LOG_ERROR, "3D LUT is empty\n"); |
602 | ret = AVERROR_INVALIDDATA; |
603 | } |
604 | |
605 | end: |
606 | fclose(f); |
607 | return ret; |
608 | } |
609 | |
610 | static const AVFilterPad lut3d_inputs[] = { |
611 | { |
612 | .name = "default", |
613 | .type = AVMEDIA_TYPE_VIDEO, |
614 | .filter_frame = filter_frame, |
615 | .config_props = config_input, |
616 | }, |
617 | { NULL } |
618 | }; |
619 | |
620 | static const AVFilterPad lut3d_outputs[] = { |
621 | { |
622 | .name = "default", |
623 | .type = AVMEDIA_TYPE_VIDEO, |
624 | }, |
625 | { NULL } |
626 | }; |
627 | |
628 | AVFilter ff_vf_lut3d = { |
629 | .name = "lut3d", |
630 | .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."), |
631 | .priv_size = sizeof(LUT3DContext), |
632 | .init = lut3d_init, |
633 | .query_formats = query_formats, |
634 | .inputs = lut3d_inputs, |
635 | .outputs = lut3d_outputs, |
636 | .priv_class = &lut3d_class, |
637 | .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS, |
638 | }; |
639 | #endif |
640 | |
641 | #if CONFIG_HALDCLUT_FILTER |
642 | |
643 | static void update_clut(LUT3DContext *lut3d, const AVFrame *frame) |
644 | { |
645 | const uint8_t *data = frame->data[0]; |
646 | const int linesize = frame->linesize[0]; |
647 | const int w = lut3d->clut_width; |
648 | const int step = lut3d->clut_step; |
649 | const uint8_t *rgba_map = lut3d->clut_rgba_map; |
650 | const int level = lut3d->lutsize; |
651 | |
652 | #define LOAD_CLUT(nbits) do { \ |
653 | int i, j, k, x = 0, y = 0; \ |
654 | \ |
655 | for (k = 0; k < level; k++) { \ |
656 | for (j = 0; j < level; j++) { \ |
657 | for (i = 0; i < level; i++) { \ |
658 | const uint##nbits##_t *src = (const uint##nbits##_t *) \ |
659 | (data + y*linesize + x*step); \ |
660 | struct rgbvec *vec = &lut3d->lut[i][j][k]; \ |
661 | vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \ |
662 | vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \ |
663 | vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \ |
664 | if (++x == w) { \ |
665 | x = 0; \ |
666 | y++; \ |
667 | } \ |
668 | } \ |
669 | } \ |
670 | } \ |
671 | } while (0) |
672 | |
673 | if (!lut3d->clut_is16bit) LOAD_CLUT(8); |
674 | else LOAD_CLUT(16); |
675 | } |
676 | |
677 | |
678 | static int config_output(AVFilterLink *outlink) |
679 | { |
680 | AVFilterContext *ctx = outlink->src; |
681 | LUT3DContext *lut3d = ctx->priv; |
682 | int ret; |
683 | |
684 | outlink->w = ctx->inputs[0]->w; |
685 | outlink->h = ctx->inputs[0]->h; |
686 | outlink->time_base = ctx->inputs[0]->time_base; |
687 | if ((ret = ff_dualinput_init(ctx, &lut3d->dinput)) < 0) |
688 | return ret; |
689 | return 0; |
690 | } |
691 | |
692 | static int filter_frame_hald(AVFilterLink *inlink, AVFrame *inpicref) |
693 | { |
694 | LUT3DContext *s = inlink->dst->priv; |
695 | return ff_dualinput_filter_frame(&s->dinput, inlink, inpicref); |
696 | } |
697 | |
698 | static int request_frame(AVFilterLink *outlink) |
699 | { |
700 | LUT3DContext *s = outlink->src->priv; |
701 | return ff_dualinput_request_frame(&s->dinput, outlink); |
702 | } |
703 | |
704 | static int config_clut(AVFilterLink *inlink) |
705 | { |
706 | int size, level, w, h; |
707 | AVFilterContext *ctx = inlink->dst; |
708 | LUT3DContext *lut3d = ctx->priv; |
709 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); |
710 | |
711 | av_assert0(desc); |
712 | |
713 | lut3d->clut_is16bit = 0; |
714 | switch (inlink->format) { |
715 | case AV_PIX_FMT_RGB48: |
716 | case AV_PIX_FMT_BGR48: |
717 | case AV_PIX_FMT_RGBA64: |
718 | case AV_PIX_FMT_BGRA64: |
719 | lut3d->clut_is16bit = 1; |
720 | } |
721 | |
722 | lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3; |
723 | ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format); |
724 | |
725 | if (inlink->w > inlink->h) |
726 | av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the " |
727 | "Hald CLUT will be ignored\n", inlink->w - inlink->h); |
728 | else if (inlink->w < inlink->h) |
729 | av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the " |
730 | "Hald CLUT will be ignored\n", inlink->h - inlink->w); |
731 | lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h); |
732 | |
733 | for (level = 1; level*level*level < w; level++); |
734 | size = level*level*level; |
735 | if (size != w) { |
736 | av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n"); |
737 | return AVERROR_INVALIDDATA; |
738 | } |
739 | av_assert0(w == h && w == size); |
740 | level *= level; |
741 | if (level > MAX_LEVEL) { |
742 | const int max_clut_level = sqrt(MAX_LEVEL); |
743 | const int max_clut_size = max_clut_level*max_clut_level*max_clut_level; |
744 | av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT " |
745 | "(maximum level is %d, or %dx%d CLUT)\n", |
746 | max_clut_level, max_clut_size, max_clut_size); |
747 | return AVERROR(EINVAL); |
748 | } |
749 | lut3d->lutsize = level; |
750 | |
751 | return 0; |
752 | } |
753 | |
754 | static AVFrame *update_apply_clut(AVFilterContext *ctx, AVFrame *main, |
755 | const AVFrame *second) |
756 | { |
757 | AVFilterLink *inlink = ctx->inputs[0]; |
758 | update_clut(ctx->priv, second); |
759 | return apply_lut(inlink, main); |
760 | } |
761 | |
762 | static av_cold int haldclut_init(AVFilterContext *ctx) |
763 | { |
764 | LUT3DContext *lut3d = ctx->priv; |
765 | lut3d->dinput.process = update_apply_clut; |
766 | return 0; |
767 | } |
768 | |
769 | static av_cold void haldclut_uninit(AVFilterContext *ctx) |
770 | { |
771 | LUT3DContext *lut3d = ctx->priv; |
772 | ff_dualinput_uninit(&lut3d->dinput); |
773 | } |
774 | |
775 | static const AVOption haldclut_options[] = { |
776 | { "shortest", "force termination when the shortest input terminates", OFFSET(dinput.shortest), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS }, |
777 | { "repeatlast", "continue applying the last clut after eos", OFFSET(dinput.repeatlast), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, FLAGS }, |
778 | COMMON_OPTIONS |
779 | }; |
780 | |
781 | AVFILTER_DEFINE_CLASS(haldclut); |
782 | |
783 | static const AVFilterPad haldclut_inputs[] = { |
784 | { |
785 | .name = "main", |
786 | .type = AVMEDIA_TYPE_VIDEO, |
787 | .filter_frame = filter_frame_hald, |
788 | .config_props = config_input, |
789 | },{ |
790 | .name = "clut", |
791 | .type = AVMEDIA_TYPE_VIDEO, |
792 | .filter_frame = filter_frame_hald, |
793 | .config_props = config_clut, |
794 | }, |
795 | { NULL } |
796 | }; |
797 | |
798 | static const AVFilterPad haldclut_outputs[] = { |
799 | { |
800 | .name = "default", |
801 | .type = AVMEDIA_TYPE_VIDEO, |
802 | .request_frame = request_frame, |
803 | .config_props = config_output, |
804 | }, |
805 | { NULL } |
806 | }; |
807 | |
808 | AVFilter ff_vf_haldclut = { |
809 | .name = "haldclut", |
810 | .description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."), |
811 | .priv_size = sizeof(LUT3DContext), |
812 | .init = haldclut_init, |
813 | .uninit = haldclut_uninit, |
814 | .query_formats = query_formats, |
815 | .inputs = haldclut_inputs, |
816 | .outputs = haldclut_outputs, |
817 | .priv_class = &haldclut_class, |
818 | .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS, |
819 | }; |
820 | #endif |
821 |