2 * Copyright (c) 2013 Clément Bœsch
3 * Copyright (c) 2018 Paul B Mahol
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * 3D Lookup table filter
27 #include "libavutil/opt.h"
28 #include "libavutil/file.h"
29 #include "libavutil/intreadwrite.h"
30 #include "libavutil/avassert.h"
31 #include "libavutil/pixdesc.h"
32 #include "libavutil/avstring.h"
34 #include "drawutils.h"
36 #include "framesync.h"
47 INTERPOLATE_TRILINEAR,
48 INTERPOLATE_TETRAHEDRAL,
56 /* 3D LUT don't often go up to level 32, but it is common to have a Hald CLUT
57 * of 512x512 (64x64x64) */
60 typedef struct LUT3DContext {
62 int interpolation; ///<interp_mode
66 avfilter_action_func *interp;
67 struct rgbvec lut[MAX_LEVEL][MAX_LEVEL][MAX_LEVEL];
69 #if CONFIG_HALDCLUT_FILTER
70 uint8_t clut_rgba_map[4];
79 typedef struct ThreadData {
83 #define OFFSET(x) offsetof(LUT3DContext, x)
84 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
85 #define COMMON_OPTIONS \
86 { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, NB_INTERP_MODE-1, FLAGS, "interp_mode" }, \
87 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_NEAREST}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \
88 { "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" }, \
89 { "tetrahedral", "interpolate values using a tetrahedron", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TETRAHEDRAL}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \
92 static inline float lerpf(float v0, float v1, float f)
94 return v0 + (v1 - v0) * f;
97 static inline struct rgbvec lerp(const struct rgbvec *v0, const struct rgbvec *v1, float f)
100 lerpf(v0->r, v1->r, f), lerpf(v0->g, v1->g, f), lerpf(v0->b, v1->b, f)
105 #define NEAR(x) ((int)((x) + .5))
106 #define PREV(x) ((int)(x))
107 #define NEXT(x) (FFMIN((int)(x) + 1, lut3d->lutsize - 1))
110 * Get the nearest defined point
112 static inline struct rgbvec interp_nearest(const LUT3DContext *lut3d,
113 const struct rgbvec *s)
115 return lut3d->lut[NEAR(s->r)][NEAR(s->g)][NEAR(s->b)];
119 * Interpolate using the 8 vertices of a cube
120 * @see https://en.wikipedia.org/wiki/Trilinear_interpolation
122 static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d,
123 const struct rgbvec *s)
125 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
126 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
127 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
128 const struct rgbvec c000 = lut3d->lut[prev[0]][prev[1]][prev[2]];
129 const struct rgbvec c001 = lut3d->lut[prev[0]][prev[1]][next[2]];
130 const struct rgbvec c010 = lut3d->lut[prev[0]][next[1]][prev[2]];
131 const struct rgbvec c011 = lut3d->lut[prev[0]][next[1]][next[2]];
132 const struct rgbvec c100 = lut3d->lut[next[0]][prev[1]][prev[2]];
133 const struct rgbvec c101 = lut3d->lut[next[0]][prev[1]][next[2]];
134 const struct rgbvec c110 = lut3d->lut[next[0]][next[1]][prev[2]];
135 const struct rgbvec c111 = lut3d->lut[next[0]][next[1]][next[2]];
136 const struct rgbvec c00 = lerp(&c000, &c100, d.r);
137 const struct rgbvec c10 = lerp(&c010, &c110, d.r);
138 const struct rgbvec c01 = lerp(&c001, &c101, d.r);
139 const struct rgbvec c11 = lerp(&c011, &c111, d.r);
140 const struct rgbvec c0 = lerp(&c00, &c10, d.g);
141 const struct rgbvec c1 = lerp(&c01, &c11, d.g);
142 const struct rgbvec c = lerp(&c0, &c1, d.b);
147 * Tetrahedral interpolation. Based on code found in Truelight Software Library paper.
148 * @see http://www.filmlight.ltd.uk/pdf/whitepapers/FL-TL-TN-0057-SoftwareLib.pdf
150 static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d,
151 const struct rgbvec *s)
153 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
154 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
155 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
156 const struct rgbvec c000 = lut3d->lut[prev[0]][prev[1]][prev[2]];
157 const struct rgbvec c111 = lut3d->lut[next[0]][next[1]][next[2]];
161 const struct rgbvec c100 = lut3d->lut[next[0]][prev[1]][prev[2]];
162 const struct rgbvec c110 = lut3d->lut[next[0]][next[1]][prev[2]];
163 c.r = (1-d.r) * c000.r + (d.r-d.g) * c100.r + (d.g-d.b) * c110.r + (d.b) * c111.r;
164 c.g = (1-d.r) * c000.g + (d.r-d.g) * c100.g + (d.g-d.b) * c110.g + (d.b) * c111.g;
165 c.b = (1-d.r) * c000.b + (d.r-d.g) * c100.b + (d.g-d.b) * c110.b + (d.b) * c111.b;
166 } else if (d.r > d.b) {
167 const struct rgbvec c100 = lut3d->lut[next[0]][prev[1]][prev[2]];
168 const struct rgbvec c101 = lut3d->lut[next[0]][prev[1]][next[2]];
169 c.r = (1-d.r) * c000.r + (d.r-d.b) * c100.r + (d.b-d.g) * c101.r + (d.g) * c111.r;
170 c.g = (1-d.r) * c000.g + (d.r-d.b) * c100.g + (d.b-d.g) * c101.g + (d.g) * c111.g;
171 c.b = (1-d.r) * c000.b + (d.r-d.b) * c100.b + (d.b-d.g) * c101.b + (d.g) * c111.b;
173 const struct rgbvec c001 = lut3d->lut[prev[0]][prev[1]][next[2]];
174 const struct rgbvec c101 = lut3d->lut[next[0]][prev[1]][next[2]];
175 c.r = (1-d.b) * c000.r + (d.b-d.r) * c001.r + (d.r-d.g) * c101.r + (d.g) * c111.r;
176 c.g = (1-d.b) * c000.g + (d.b-d.r) * c001.g + (d.r-d.g) * c101.g + (d.g) * c111.g;
177 c.b = (1-d.b) * c000.b + (d.b-d.r) * c001.b + (d.r-d.g) * c101.b + (d.g) * c111.b;
181 const struct rgbvec c001 = lut3d->lut[prev[0]][prev[1]][next[2]];
182 const struct rgbvec c011 = lut3d->lut[prev[0]][next[1]][next[2]];
183 c.r = (1-d.b) * c000.r + (d.b-d.g) * c001.r + (d.g-d.r) * c011.r + (d.r) * c111.r;
184 c.g = (1-d.b) * c000.g + (d.b-d.g) * c001.g + (d.g-d.r) * c011.g + (d.r) * c111.g;
185 c.b = (1-d.b) * c000.b + (d.b-d.g) * c001.b + (d.g-d.r) * c011.b + (d.r) * c111.b;
186 } else if (d.b > d.r) {
187 const struct rgbvec c010 = lut3d->lut[prev[0]][next[1]][prev[2]];
188 const struct rgbvec c011 = lut3d->lut[prev[0]][next[1]][next[2]];
189 c.r = (1-d.g) * c000.r + (d.g-d.b) * c010.r + (d.b-d.r) * c011.r + (d.r) * c111.r;
190 c.g = (1-d.g) * c000.g + (d.g-d.b) * c010.g + (d.b-d.r) * c011.g + (d.r) * c111.g;
191 c.b = (1-d.g) * c000.b + (d.g-d.b) * c010.b + (d.b-d.r) * c011.b + (d.r) * c111.b;
193 const struct rgbvec c010 = lut3d->lut[prev[0]][next[1]][prev[2]];
194 const struct rgbvec c110 = lut3d->lut[next[0]][next[1]][prev[2]];
195 c.r = (1-d.g) * c000.r + (d.g-d.r) * c010.r + (d.r-d.b) * c110.r + (d.b) * c111.r;
196 c.g = (1-d.g) * c000.g + (d.g-d.r) * c010.g + (d.r-d.b) * c110.g + (d.b) * c111.g;
197 c.b = (1-d.g) * c000.b + (d.g-d.r) * c010.b + (d.r-d.b) * c110.b + (d.b) * c111.b;
203 #define DEFINE_INTERP_FUNC_PLANAR(name, nbits, depth) \
204 static int interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
207 const LUT3DContext *lut3d = ctx->priv; \
208 const ThreadData *td = arg; \
209 const AVFrame *in = td->in; \
210 const AVFrame *out = td->out; \
211 const int direct = out == in; \
212 const int slice_start = (in->height * jobnr ) / nb_jobs; \
213 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
214 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
215 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
216 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
217 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
218 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
219 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
220 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
221 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
222 const float scale = (1. / ((1<<depth) - 1)) * (lut3d->lutsize - 1); \
224 for (y = slice_start; y < slice_end; y++) { \
225 uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
226 uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
227 uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
228 uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
229 const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
230 const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
231 const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
232 const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
233 for (x = 0; x < in->width; x++) { \
234 const struct rgbvec scaled_rgb = {srcr[x] * scale, \
237 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
238 dstr[x] = av_clip_uintp2(vec.r * (float)((1<<depth) - 1), depth); \
239 dstg[x] = av_clip_uintp2(vec.g * (float)((1<<depth) - 1), depth); \
240 dstb[x] = av_clip_uintp2(vec.b * (float)((1<<depth) - 1), depth); \
241 if (!direct && in->linesize[3]) \
244 grow += out->linesize[0]; \
245 brow += out->linesize[1]; \
246 rrow += out->linesize[2]; \
247 arow += out->linesize[3]; \
248 srcgrow += in->linesize[0]; \
249 srcbrow += in->linesize[1]; \
250 srcrrow += in->linesize[2]; \
251 srcarow += in->linesize[3]; \
256 DEFINE_INTERP_FUNC_PLANAR(nearest, 8, 8)
257 DEFINE_INTERP_FUNC_PLANAR(trilinear, 8, 8)
258 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 8, 8)
260 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 9)
261 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 9)
262 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 9)
264 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 10)
265 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 10)
266 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 10)
268 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 12)
269 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 12)
270 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 12)
272 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 14)
273 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 14)
274 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 14)
276 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 16)
277 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 16)
278 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 16)
280 #define DEFINE_INTERP_FUNC(name, nbits) \
281 static int interp_##nbits##_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
284 const LUT3DContext *lut3d = ctx->priv; \
285 const ThreadData *td = arg; \
286 const AVFrame *in = td->in; \
287 const AVFrame *out = td->out; \
288 const int direct = out == in; \
289 const int step = lut3d->step; \
290 const uint8_t r = lut3d->rgba_map[R]; \
291 const uint8_t g = lut3d->rgba_map[G]; \
292 const uint8_t b = lut3d->rgba_map[B]; \
293 const uint8_t a = lut3d->rgba_map[A]; \
294 const int slice_start = (in->height * jobnr ) / nb_jobs; \
295 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
296 uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
297 const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
298 const float scale = (1. / ((1<<nbits) - 1)) * (lut3d->lutsize - 1); \
300 for (y = slice_start; y < slice_end; y++) { \
301 uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
302 const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
303 for (x = 0; x < in->width * step; x += step) { \
304 const struct rgbvec scaled_rgb = {src[x + r] * scale, \
305 src[x + g] * scale, \
306 src[x + b] * scale}; \
307 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
308 dst[x + r] = av_clip_uint##nbits(vec.r * (float)((1<<nbits) - 1)); \
309 dst[x + g] = av_clip_uint##nbits(vec.g * (float)((1<<nbits) - 1)); \
310 dst[x + b] = av_clip_uint##nbits(vec.b * (float)((1<<nbits) - 1)); \
311 if (!direct && step == 4) \
312 dst[x + a] = src[x + a]; \
314 dstrow += out->linesize[0]; \
315 srcrow += in ->linesize[0]; \
320 DEFINE_INTERP_FUNC(nearest, 8)
321 DEFINE_INTERP_FUNC(trilinear, 8)
322 DEFINE_INTERP_FUNC(tetrahedral, 8)
324 DEFINE_INTERP_FUNC(nearest, 16)
325 DEFINE_INTERP_FUNC(trilinear, 16)
326 DEFINE_INTERP_FUNC(tetrahedral, 16)
328 #define MAX_LINE_SIZE 512
330 static int skip_line(const char *p)
332 while (*p && av_isspace(*p))
334 return !*p || *p == '#';
337 #define NEXT_LINE(loop_cond) do { \
338 if (!fgets(line, sizeof(line), f)) { \
339 av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \
340 return AVERROR_INVALIDDATA; \
344 /* Basically r g and b float values on each line, with a facultative 3DLUTSIZE
345 * directive; seems to be generated by Davinci */
346 static int parse_dat(AVFilterContext *ctx, FILE *f)
348 LUT3DContext *lut3d = ctx->priv;
349 char line[MAX_LINE_SIZE];
352 lut3d->lutsize = size = 33;
354 NEXT_LINE(skip_line(line));
355 if (!strncmp(line, "3DLUTSIZE ", 10)) {
356 size = strtol(line + 10, NULL, 0);
357 if (size < 2 || size > MAX_LEVEL) {
358 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n");
359 return AVERROR(EINVAL);
361 lut3d->lutsize = size;
362 NEXT_LINE(skip_line(line));
364 for (k = 0; k < size; k++) {
365 for (j = 0; j < size; j++) {
366 for (i = 0; i < size; i++) {
367 struct rgbvec *vec = &lut3d->lut[k][j][i];
368 if (k != 0 || j != 0 || i != 0)
369 NEXT_LINE(skip_line(line));
370 if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
371 return AVERROR_INVALIDDATA;
379 static int parse_cube(AVFilterContext *ctx, FILE *f)
381 LUT3DContext *lut3d = ctx->priv;
382 char line[MAX_LINE_SIZE];
383 float min[3] = {0.0, 0.0, 0.0};
384 float max[3] = {1.0, 1.0, 1.0};
386 while (fgets(line, sizeof(line), f)) {
387 if (!strncmp(line, "LUT_3D_SIZE", 11)) {
389 const int size = strtol(line + 12, NULL, 0);
391 if (size < 2 || size > MAX_LEVEL) {
392 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n");
393 return AVERROR(EINVAL);
395 lut3d->lutsize = size;
396 for (k = 0; k < size; k++) {
397 for (j = 0; j < size; j++) {
398 for (i = 0; i < size; i++) {
399 struct rgbvec *vec = &lut3d->lut[i][j][k];
404 if (!strncmp(line, "DOMAIN_", 7)) {
406 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
407 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
409 return AVERROR_INVALIDDATA;
410 av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
411 av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
412 min[0], min[1], min[2], max[0], max[1], max[2]);
414 } else if (!strncmp(line, "TITLE", 5)) {
417 } while (skip_line(line));
418 if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
419 return AVERROR_INVALIDDATA;
420 vec->r *= max[0] - min[0];
421 vec->g *= max[1] - min[1];
422 vec->b *= max[2] - min[2];
432 /* Assume 17x17x17 LUT with a 16-bit depth
433 * FIXME: it seems there are various 3dl formats */
434 static int parse_3dl(AVFilterContext *ctx, FILE *f)
436 char line[MAX_LINE_SIZE];
437 LUT3DContext *lut3d = ctx->priv;
440 const float scale = 16*16*16;
442 lut3d->lutsize = size;
443 NEXT_LINE(skip_line(line));
444 for (k = 0; k < size; k++) {
445 for (j = 0; j < size; j++) {
446 for (i = 0; i < size; i++) {
448 struct rgbvec *vec = &lut3d->lut[k][j][i];
450 NEXT_LINE(skip_line(line));
451 if (av_sscanf(line, "%d %d %d", &r, &g, &b) != 3)
452 return AVERROR_INVALIDDATA;
463 static int parse_m3d(AVFilterContext *ctx, FILE *f)
465 LUT3DContext *lut3d = ctx->priv;
467 int i, j, k, size, in = -1, out = -1;
468 char line[MAX_LINE_SIZE];
469 uint8_t rgb_map[3] = {0, 1, 2};
471 while (fgets(line, sizeof(line), f)) {
472 if (!strncmp(line, "in", 2)) in = strtol(line + 2, NULL, 0);
473 else if (!strncmp(line, "out", 3)) out = strtol(line + 3, NULL, 0);
474 else if (!strncmp(line, "values", 6)) {
475 const char *p = line + 6;
476 #define SET_COLOR(id) do { \
477 while (av_isspace(*p)) \
480 case 'r': rgb_map[id] = 0; break; \
481 case 'g': rgb_map[id] = 1; break; \
482 case 'b': rgb_map[id] = 2; break; \
484 while (*p && !av_isspace(*p)) \
494 if (in == -1 || out == -1) {
495 av_log(ctx, AV_LOG_ERROR, "in and out must be defined\n");
496 return AVERROR_INVALIDDATA;
498 if (in < 2 || out < 2 ||
499 in > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL ||
500 out > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL) {
501 av_log(ctx, AV_LOG_ERROR, "invalid in (%d) or out (%d)\n", in, out);
502 return AVERROR_INVALIDDATA;
504 for (size = 1; size*size*size < in; size++);
505 lut3d->lutsize = size;
506 scale = 1. / (out - 1);
508 for (k = 0; k < size; k++) {
509 for (j = 0; j < size; j++) {
510 for (i = 0; i < size; i++) {
511 struct rgbvec *vec = &lut3d->lut[k][j][i];
515 if (av_sscanf(line, "%f %f %f", val, val + 1, val + 2) != 3)
516 return AVERROR_INVALIDDATA;
517 vec->r = val[rgb_map[0]] * scale;
518 vec->g = val[rgb_map[1]] * scale;
519 vec->b = val[rgb_map[2]] * scale;
526 static void set_identity_matrix(LUT3DContext *lut3d, int size)
529 const float c = 1. / (size - 1);
531 lut3d->lutsize = size;
532 for (k = 0; k < size; k++) {
533 for (j = 0; j < size; j++) {
534 for (i = 0; i < size; i++) {
535 struct rgbvec *vec = &lut3d->lut[k][j][i];
544 static int query_formats(AVFilterContext *ctx)
546 static const enum AVPixelFormat pix_fmts[] = {
547 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
548 AV_PIX_FMT_RGBA, AV_PIX_FMT_BGRA,
549 AV_PIX_FMT_ARGB, AV_PIX_FMT_ABGR,
550 AV_PIX_FMT_0RGB, AV_PIX_FMT_0BGR,
551 AV_PIX_FMT_RGB0, AV_PIX_FMT_BGR0,
552 AV_PIX_FMT_RGB48, AV_PIX_FMT_BGR48,
553 AV_PIX_FMT_RGBA64, AV_PIX_FMT_BGRA64,
554 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP,
556 AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRAP10,
557 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRAP12,
559 AV_PIX_FMT_GBRP16, AV_PIX_FMT_GBRAP16,
562 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
564 return AVERROR(ENOMEM);
565 return ff_set_common_formats(ctx, fmts_list);
568 static int config_input(AVFilterLink *inlink)
570 int depth, is16bit = 0, planar = 0;
571 LUT3DContext *lut3d = inlink->dst->priv;
572 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
574 depth = desc->comp[0].depth;
576 switch (inlink->format) {
577 case AV_PIX_FMT_RGB48:
578 case AV_PIX_FMT_BGR48:
579 case AV_PIX_FMT_RGBA64:
580 case AV_PIX_FMT_BGRA64:
583 case AV_PIX_FMT_GBRP9:
584 case AV_PIX_FMT_GBRP10:
585 case AV_PIX_FMT_GBRP12:
586 case AV_PIX_FMT_GBRP14:
587 case AV_PIX_FMT_GBRP16:
588 case AV_PIX_FMT_GBRAP10:
589 case AV_PIX_FMT_GBRAP12:
590 case AV_PIX_FMT_GBRAP16:
592 case AV_PIX_FMT_GBRP:
593 case AV_PIX_FMT_GBRAP:
598 ff_fill_rgba_map(lut3d->rgba_map, inlink->format);
599 lut3d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
601 #define SET_FUNC(name) do { \
604 case 8: lut3d->interp = interp_8_##name##_p8; break; \
605 case 9: lut3d->interp = interp_16_##name##_p9; break; \
606 case 10: lut3d->interp = interp_16_##name##_p10; break; \
607 case 12: lut3d->interp = interp_16_##name##_p12; break; \
608 case 14: lut3d->interp = interp_16_##name##_p14; break; \
609 case 16: lut3d->interp = interp_16_##name##_p16; break; \
611 } else if (is16bit) { lut3d->interp = interp_16_##name; \
612 } else { lut3d->interp = interp_8_##name; } \
615 switch (lut3d->interpolation) {
616 case INTERPOLATE_NEAREST: SET_FUNC(nearest); break;
617 case INTERPOLATE_TRILINEAR: SET_FUNC(trilinear); break;
618 case INTERPOLATE_TETRAHEDRAL: SET_FUNC(tetrahedral); break;
626 static AVFrame *apply_lut(AVFilterLink *inlink, AVFrame *in)
628 AVFilterContext *ctx = inlink->dst;
629 LUT3DContext *lut3d = ctx->priv;
630 AVFilterLink *outlink = inlink->dst->outputs[0];
634 if (av_frame_is_writable(in)) {
637 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
642 av_frame_copy_props(out, in);
647 ctx->internal->execute(ctx, lut3d->interp, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
655 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
657 AVFilterLink *outlink = inlink->dst->outputs[0];
658 AVFrame *out = apply_lut(inlink, in);
660 return AVERROR(ENOMEM);
661 return ff_filter_frame(outlink, out);
664 #if CONFIG_LUT3D_FILTER
665 static const AVOption lut3d_options[] = {
666 { "file", "set 3D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
670 AVFILTER_DEFINE_CLASS(lut3d);
672 static av_cold int lut3d_init(AVFilterContext *ctx)
677 LUT3DContext *lut3d = ctx->priv;
680 set_identity_matrix(lut3d, 32);
684 f = fopen(lut3d->file, "r");
686 ret = AVERROR(errno);
687 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret));
691 ext = strrchr(lut3d->file, '.');
693 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
694 ret = AVERROR_INVALIDDATA;
699 if (!av_strcasecmp(ext, "dat")) {
700 ret = parse_dat(ctx, f);
701 } else if (!av_strcasecmp(ext, "3dl")) {
702 ret = parse_3dl(ctx, f);
703 } else if (!av_strcasecmp(ext, "cube")) {
704 ret = parse_cube(ctx, f);
705 } else if (!av_strcasecmp(ext, "m3d")) {
706 ret = parse_m3d(ctx, f);
708 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
709 ret = AVERROR(EINVAL);
712 if (!ret && !lut3d->lutsize) {
713 av_log(ctx, AV_LOG_ERROR, "3D LUT is empty\n");
714 ret = AVERROR_INVALIDDATA;
722 static const AVFilterPad lut3d_inputs[] = {
725 .type = AVMEDIA_TYPE_VIDEO,
726 .filter_frame = filter_frame,
727 .config_props = config_input,
732 static const AVFilterPad lut3d_outputs[] = {
735 .type = AVMEDIA_TYPE_VIDEO,
740 AVFilter ff_vf_lut3d = {
742 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
743 .priv_size = sizeof(LUT3DContext),
745 .query_formats = query_formats,
746 .inputs = lut3d_inputs,
747 .outputs = lut3d_outputs,
748 .priv_class = &lut3d_class,
749 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
753 #if CONFIG_HALDCLUT_FILTER
755 static void update_clut_packed(LUT3DContext *lut3d, const AVFrame *frame)
757 const uint8_t *data = frame->data[0];
758 const int linesize = frame->linesize[0];
759 const int w = lut3d->clut_width;
760 const int step = lut3d->clut_step;
761 const uint8_t *rgba_map = lut3d->clut_rgba_map;
762 const int level = lut3d->lutsize;
764 #define LOAD_CLUT(nbits) do { \
765 int i, j, k, x = 0, y = 0; \
767 for (k = 0; k < level; k++) { \
768 for (j = 0; j < level; j++) { \
769 for (i = 0; i < level; i++) { \
770 const uint##nbits##_t *src = (const uint##nbits##_t *) \
771 (data + y*linesize + x*step); \
772 struct rgbvec *vec = &lut3d->lut[i][j][k]; \
773 vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \
774 vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \
775 vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \
785 switch (lut3d->clut_bits) {
786 case 8: LOAD_CLUT(8); break;
787 case 16: LOAD_CLUT(16); break;
791 static void update_clut_planar(LUT3DContext *lut3d, const AVFrame *frame)
793 const uint8_t *datag = frame->data[0];
794 const uint8_t *datab = frame->data[1];
795 const uint8_t *datar = frame->data[2];
796 const int glinesize = frame->linesize[0];
797 const int blinesize = frame->linesize[1];
798 const int rlinesize = frame->linesize[2];
799 const int w = lut3d->clut_width;
800 const int level = lut3d->lutsize;
802 #define LOAD_CLUT_PLANAR(nbits, depth) do { \
803 int i, j, k, x = 0, y = 0; \
805 for (k = 0; k < level; k++) { \
806 for (j = 0; j < level; j++) { \
807 for (i = 0; i < level; i++) { \
808 const uint##nbits##_t *gsrc = (const uint##nbits##_t *) \
809 (datag + y*glinesize); \
810 const uint##nbits##_t *bsrc = (const uint##nbits##_t *) \
811 (datab + y*blinesize); \
812 const uint##nbits##_t *rsrc = (const uint##nbits##_t *) \
813 (datar + y*rlinesize); \
814 struct rgbvec *vec = &lut3d->lut[i][j][k]; \
815 vec->r = gsrc[x] / (float)((1<<(depth)) - 1); \
816 vec->g = bsrc[x] / (float)((1<<(depth)) - 1); \
817 vec->b = rsrc[x] / (float)((1<<(depth)) - 1); \
827 switch (lut3d->clut_bits) {
828 case 8: LOAD_CLUT_PLANAR(8, 8); break;
829 case 9: LOAD_CLUT_PLANAR(16, 9); break;
830 case 10: LOAD_CLUT_PLANAR(16, 10); break;
831 case 12: LOAD_CLUT_PLANAR(16, 12); break;
832 case 14: LOAD_CLUT_PLANAR(16, 14); break;
833 case 16: LOAD_CLUT_PLANAR(16, 16); break;
837 static int config_output(AVFilterLink *outlink)
839 AVFilterContext *ctx = outlink->src;
840 LUT3DContext *lut3d = ctx->priv;
843 ret = ff_framesync_init_dualinput(&lut3d->fs, ctx);
846 outlink->w = ctx->inputs[0]->w;
847 outlink->h = ctx->inputs[0]->h;
848 outlink->time_base = ctx->inputs[0]->time_base;
849 if ((ret = ff_framesync_configure(&lut3d->fs)) < 0)
854 static int activate(AVFilterContext *ctx)
856 LUT3DContext *s = ctx->priv;
857 return ff_framesync_activate(&s->fs);
860 static int config_clut(AVFilterLink *inlink)
862 int size, level, w, h;
863 AVFilterContext *ctx = inlink->dst;
864 LUT3DContext *lut3d = ctx->priv;
865 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
869 lut3d->clut_bits = desc->comp[0].depth;
870 lut3d->clut_planar = av_pix_fmt_count_planes(inlink->format) > 1;
872 lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3;
873 ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format);
875 if (inlink->w > inlink->h)
876 av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the "
877 "Hald CLUT will be ignored\n", inlink->w - inlink->h);
878 else if (inlink->w < inlink->h)
879 av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the "
880 "Hald CLUT will be ignored\n", inlink->h - inlink->w);
881 lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h);
883 for (level = 1; level*level*level < w; level++);
884 size = level*level*level;
886 av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n");
887 return AVERROR_INVALIDDATA;
889 av_assert0(w == h && w == size);
891 if (level > MAX_LEVEL) {
892 const int max_clut_level = sqrt(MAX_LEVEL);
893 const int max_clut_size = max_clut_level*max_clut_level*max_clut_level;
894 av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT "
895 "(maximum level is %d, or %dx%d CLUT)\n",
896 max_clut_level, max_clut_size, max_clut_size);
897 return AVERROR(EINVAL);
899 lut3d->lutsize = level;
904 static int update_apply_clut(FFFrameSync *fs)
906 AVFilterContext *ctx = fs->parent;
907 LUT3DContext *lut3d = ctx->priv;
908 AVFilterLink *inlink = ctx->inputs[0];
909 AVFrame *master, *second, *out;
912 ret = ff_framesync_dualinput_get(fs, &master, &second);
916 return ff_filter_frame(ctx->outputs[0], master);
917 if (lut3d->clut_planar)
918 update_clut_planar(ctx->priv, second);
920 update_clut_packed(ctx->priv, second);
921 out = apply_lut(inlink, master);
922 return ff_filter_frame(ctx->outputs[0], out);
925 static av_cold int haldclut_init(AVFilterContext *ctx)
927 LUT3DContext *lut3d = ctx->priv;
928 lut3d->fs.on_event = update_apply_clut;
932 static av_cold void haldclut_uninit(AVFilterContext *ctx)
934 LUT3DContext *lut3d = ctx->priv;
935 ff_framesync_uninit(&lut3d->fs);
938 static const AVOption haldclut_options[] = {
942 FRAMESYNC_DEFINE_CLASS(haldclut, LUT3DContext, fs);
944 static const AVFilterPad haldclut_inputs[] = {
947 .type = AVMEDIA_TYPE_VIDEO,
948 .config_props = config_input,
951 .type = AVMEDIA_TYPE_VIDEO,
952 .config_props = config_clut,
957 static const AVFilterPad haldclut_outputs[] = {
960 .type = AVMEDIA_TYPE_VIDEO,
961 .config_props = config_output,
966 AVFilter ff_vf_haldclut = {
968 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
969 .priv_size = sizeof(LUT3DContext),
970 .preinit = haldclut_framesync_preinit,
971 .init = haldclut_init,
972 .uninit = haldclut_uninit,
973 .query_formats = query_formats,
974 .activate = activate,
975 .inputs = haldclut_inputs,
976 .outputs = haldclut_outputs,
977 .priv_class = &haldclut_class,
978 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
982 #if CONFIG_LUT1D_FILTER
984 enum interp_1d_mode {
985 INTERPOLATE_1D_NEAREST,
986 INTERPOLATE_1D_LINEAR,
987 INTERPOLATE_1D_CUBIC,
988 INTERPOLATE_1D_COSINE,
989 INTERPOLATE_1D_SPLINE,
993 #define MAX_1D_LEVEL 65536
995 typedef struct LUT1DContext {
996 const AVClass *class;
998 int interpolation; ///<interp_1d_mode
1001 float lut[3][MAX_1D_LEVEL];
1003 avfilter_action_func *interp;
1007 #define OFFSET(x) offsetof(LUT1DContext, x)
1009 static void set_identity_matrix_1d(LUT1DContext *lut1d, int size)
1011 const float c = 1. / (size - 1);
1014 lut1d->lutsize = size;
1015 for (i = 0; i < size; i++) {
1016 lut1d->lut[0][i] = i * c;
1017 lut1d->lut[1][i] = i * c;
1018 lut1d->lut[2][i] = i * c;
1022 static int parse_cube_1d(AVFilterContext *ctx, FILE *f)
1024 LUT1DContext *lut1d = ctx->priv;
1025 char line[MAX_LINE_SIZE];
1026 float min[3] = {0.0, 0.0, 0.0};
1027 float max[3] = {1.0, 1.0, 1.0};
1029 while (fgets(line, sizeof(line), f)) {
1030 if (!strncmp(line, "LUT_1D_SIZE", 11)) {
1031 const int size = strtol(line + 12, NULL, 0);
1034 if (size < 2 || size > MAX_1D_LEVEL) {
1035 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1036 return AVERROR(EINVAL);
1038 lut1d->lutsize = size;
1039 for (i = 0; i < size; i++) {
1043 if (!strncmp(line, "DOMAIN_", 7)) {
1045 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
1046 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
1048 return AVERROR_INVALIDDATA;
1049 av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
1050 av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
1051 min[0], min[1], min[2], max[0], max[1], max[2]);
1053 } else if (!strncmp(line, "LUT_1D_INPUT_RANGE ", 19)) {
1054 av_sscanf(line + 19, "%f %f", min, max);
1055 min[1] = min[2] = min[0];
1056 max[1] = max[2] = max[0];
1058 } else if (!strncmp(line, "TITLE", 5)) {
1061 } while (skip_line(line));
1062 if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1063 return AVERROR_INVALIDDATA;
1064 lut1d->lut[0][i] *= max[0] - min[0];
1065 lut1d->lut[1][i] *= max[1] - min[1];
1066 lut1d->lut[2][i] *= max[2] - min[2];
1074 static const AVOption lut1d_options[] = {
1075 { "file", "set 1D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
1076 { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_1D_LINEAR}, 0, NB_INTERP_1D_MODE-1, FLAGS, "interp_mode" },
1077 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_NEAREST}, INT_MIN, INT_MAX, FLAGS, "interp_mode" },
1078 { "linear", "use values from the linear interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_LINEAR}, INT_MIN, INT_MAX, FLAGS, "interp_mode" },
1079 { "cosine", "use values from the cosine interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_COSINE}, INT_MIN, INT_MAX, FLAGS, "interp_mode" },
1080 { "cubic", "use values from the cubic interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_CUBIC}, INT_MIN, INT_MAX, FLAGS, "interp_mode" },
1081 { "spline", "use values from the spline interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_SPLINE}, INT_MIN, INT_MAX, FLAGS, "interp_mode" },
1085 AVFILTER_DEFINE_CLASS(lut1d);
1087 static inline float interp_1d_nearest(const LUT1DContext *lut1d,
1088 int idx, const float s)
1090 return lut1d->lut[idx][NEAR(s)];
1093 #define NEXT1D(x) (FFMIN((int)(x) + 1, lut1d->lutsize - 1))
1095 static inline float interp_1d_linear(const LUT1DContext *lut1d,
1096 int idx, const float s)
1098 const int prev = PREV(s);
1099 const int next = NEXT1D(s);
1100 const float d = s - prev;
1101 const float p = lut1d->lut[idx][prev];
1102 const float n = lut1d->lut[idx][next];
1104 return lerpf(p, n, d);
1107 static inline float interp_1d_cosine(const LUT1DContext *lut1d,
1108 int idx, const float s)
1110 const int prev = PREV(s);
1111 const int next = NEXT1D(s);
1112 const float d = s - prev;
1113 const float p = lut1d->lut[idx][prev];
1114 const float n = lut1d->lut[idx][next];
1115 const float m = (1.f - cosf(d * M_PI)) * .5f;
1117 return lerpf(p, n, m);
1120 static inline float interp_1d_cubic(const LUT1DContext *lut1d,
1121 int idx, const float s)
1123 const int prev = PREV(s);
1124 const int next = NEXT1D(s);
1125 const float mu = s - prev;
1126 float a0, a1, a2, a3, mu2;
1128 float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1129 float y1 = lut1d->lut[idx][prev];
1130 float y2 = lut1d->lut[idx][next];
1131 float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1135 a0 = y3 - y2 - y0 + y1;
1140 return a0 * mu * mu2 + a1 * mu2 + a2 * mu + a3;
1143 static inline float interp_1d_spline(const LUT1DContext *lut1d,
1144 int idx, const float s)
1146 const int prev = PREV(s);
1147 const int next = NEXT1D(s);
1148 const float x = s - prev;
1149 float c0, c1, c2, c3;
1151 float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1152 float y1 = lut1d->lut[idx][prev];
1153 float y2 = lut1d->lut[idx][next];
1154 float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1157 c1 = .5f * (y2 - y0);
1158 c2 = y0 - 2.5f * y1 + 2.f * y2 - .5f * y3;
1159 c3 = .5f * (y3 - y0) + 1.5f * (y1 - y2);
1161 return ((c3 * x + c2) * x + c1) * x + c0;
1164 #define DEFINE_INTERP_FUNC_PLANAR_1D(name, nbits, depth) \
1165 static int interp_1d_##nbits##_##name##_p##depth(AVFilterContext *ctx, \
1166 void *arg, int jobnr, \
1170 const LUT1DContext *lut1d = ctx->priv; \
1171 const ThreadData *td = arg; \
1172 const AVFrame *in = td->in; \
1173 const AVFrame *out = td->out; \
1174 const int direct = out == in; \
1175 const int slice_start = (in->height * jobnr ) / nb_jobs; \
1176 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
1177 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
1178 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
1179 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
1180 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
1181 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
1182 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
1183 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
1184 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
1185 const float factor = (1 << depth) - 1; \
1186 const float scale = (1. / factor) * (lut1d->lutsize - 1); \
1188 for (y = slice_start; y < slice_end; y++) { \
1189 uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
1190 uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
1191 uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
1192 uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
1193 const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
1194 const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
1195 const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
1196 const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
1197 for (x = 0; x < in->width; x++) { \
1198 float r = srcr[x] * scale; \
1199 float g = srcg[x] * scale; \
1200 float b = srcb[x] * scale; \
1201 r = interp_1d_##name(lut1d, 0, r); \
1202 g = interp_1d_##name(lut1d, 1, g); \
1203 b = interp_1d_##name(lut1d, 2, b); \
1204 dstr[x] = av_clip_uintp2(r * factor, depth); \
1205 dstg[x] = av_clip_uintp2(g * factor, depth); \
1206 dstb[x] = av_clip_uintp2(b * factor, depth); \
1207 if (!direct && in->linesize[3]) \
1208 dsta[x] = srca[x]; \
1210 grow += out->linesize[0]; \
1211 brow += out->linesize[1]; \
1212 rrow += out->linesize[2]; \
1213 arow += out->linesize[3]; \
1214 srcgrow += in->linesize[0]; \
1215 srcbrow += in->linesize[1]; \
1216 srcrrow += in->linesize[2]; \
1217 srcarow += in->linesize[3]; \
1222 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 8, 8)
1223 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 8, 8)
1224 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 8, 8)
1225 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 8, 8)
1226 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 8, 8)
1228 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 9)
1229 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 9)
1230 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 9)
1231 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 9)
1232 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 9)
1234 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 10)
1235 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 10)
1236 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 10)
1237 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 10)
1238 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 10)
1240 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 12)
1241 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 12)
1242 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 12)
1243 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 12)
1244 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 12)
1246 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 14)
1247 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 14)
1248 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 14)
1249 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 14)
1250 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 14)
1252 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 16)
1253 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 16)
1254 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 16)
1255 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 16)
1256 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 16)
1258 #define DEFINE_INTERP_FUNC_1D(name, nbits) \
1259 static int interp_1d_##nbits##_##name(AVFilterContext *ctx, void *arg, \
1260 int jobnr, int nb_jobs) \
1263 const LUT1DContext *lut1d = ctx->priv; \
1264 const ThreadData *td = arg; \
1265 const AVFrame *in = td->in; \
1266 const AVFrame *out = td->out; \
1267 const int direct = out == in; \
1268 const int step = lut1d->step; \
1269 const uint8_t r = lut1d->rgba_map[R]; \
1270 const uint8_t g = lut1d->rgba_map[G]; \
1271 const uint8_t b = lut1d->rgba_map[B]; \
1272 const uint8_t a = lut1d->rgba_map[A]; \
1273 const int slice_start = (in->height * jobnr ) / nb_jobs; \
1274 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
1275 uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
1276 const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
1277 const float factor = (1 << nbits) - 1; \
1278 const float scale = (1. / factor) * (lut1d->lutsize - 1); \
1280 for (y = slice_start; y < slice_end; y++) { \
1281 uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
1282 const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
1283 for (x = 0; x < in->width * step; x += step) { \
1284 float rr = src[x + r] * scale; \
1285 float gg = src[x + g] * scale; \
1286 float bb = src[x + b] * scale; \
1287 rr = interp_1d_##name(lut1d, 0, rr); \
1288 gg = interp_1d_##name(lut1d, 1, gg); \
1289 bb = interp_1d_##name(lut1d, 2, bb); \
1290 dst[x + r] = av_clip_uint##nbits(rr * factor); \
1291 dst[x + g] = av_clip_uint##nbits(gg * factor); \
1292 dst[x + b] = av_clip_uint##nbits(bb * factor); \
1293 if (!direct && step == 4) \
1294 dst[x + a] = src[x + a]; \
1296 dstrow += out->linesize[0]; \
1297 srcrow += in ->linesize[0]; \
1302 DEFINE_INTERP_FUNC_1D(nearest, 8)
1303 DEFINE_INTERP_FUNC_1D(linear, 8)
1304 DEFINE_INTERP_FUNC_1D(cosine, 8)
1305 DEFINE_INTERP_FUNC_1D(cubic, 8)
1306 DEFINE_INTERP_FUNC_1D(spline, 8)
1308 DEFINE_INTERP_FUNC_1D(nearest, 16)
1309 DEFINE_INTERP_FUNC_1D(linear, 16)
1310 DEFINE_INTERP_FUNC_1D(cosine, 16)
1311 DEFINE_INTERP_FUNC_1D(cubic, 16)
1312 DEFINE_INTERP_FUNC_1D(spline, 16)
1314 static int config_input_1d(AVFilterLink *inlink)
1316 int depth, is16bit = 0, planar = 0;
1317 LUT1DContext *lut1d = inlink->dst->priv;
1318 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1320 depth = desc->comp[0].depth;
1322 switch (inlink->format) {
1323 case AV_PIX_FMT_RGB48:
1324 case AV_PIX_FMT_BGR48:
1325 case AV_PIX_FMT_RGBA64:
1326 case AV_PIX_FMT_BGRA64:
1329 case AV_PIX_FMT_GBRP9:
1330 case AV_PIX_FMT_GBRP10:
1331 case AV_PIX_FMT_GBRP12:
1332 case AV_PIX_FMT_GBRP14:
1333 case AV_PIX_FMT_GBRP16:
1334 case AV_PIX_FMT_GBRAP10:
1335 case AV_PIX_FMT_GBRAP12:
1336 case AV_PIX_FMT_GBRAP16:
1338 case AV_PIX_FMT_GBRP:
1339 case AV_PIX_FMT_GBRAP:
1344 ff_fill_rgba_map(lut1d->rgba_map, inlink->format);
1345 lut1d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
1347 #define SET_FUNC_1D(name) do { \
1350 case 8: lut1d->interp = interp_1d_8_##name##_p8; break; \
1351 case 9: lut1d->interp = interp_1d_16_##name##_p9; break; \
1352 case 10: lut1d->interp = interp_1d_16_##name##_p10; break; \
1353 case 12: lut1d->interp = interp_1d_16_##name##_p12; break; \
1354 case 14: lut1d->interp = interp_1d_16_##name##_p14; break; \
1355 case 16: lut1d->interp = interp_1d_16_##name##_p16; break; \
1357 } else if (is16bit) { lut1d->interp = interp_1d_16_##name; \
1358 } else { lut1d->interp = interp_1d_8_##name; } \
1361 switch (lut1d->interpolation) {
1362 case INTERPOLATE_1D_NEAREST: SET_FUNC_1D(nearest); break;
1363 case INTERPOLATE_1D_LINEAR: SET_FUNC_1D(linear); break;
1364 case INTERPOLATE_1D_COSINE: SET_FUNC_1D(cosine); break;
1365 case INTERPOLATE_1D_CUBIC: SET_FUNC_1D(cubic); break;
1366 case INTERPOLATE_1D_SPLINE: SET_FUNC_1D(spline); break;
1374 static av_cold int lut1d_init(AVFilterContext *ctx)
1379 LUT1DContext *lut1d = ctx->priv;
1382 set_identity_matrix_1d(lut1d, 32);
1386 f = fopen(lut1d->file, "r");
1388 ret = AVERROR(errno);
1389 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut1d->file, av_err2str(ret));
1393 ext = strrchr(lut1d->file, '.');
1395 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
1396 ret = AVERROR_INVALIDDATA;
1401 if (!av_strcasecmp(ext, "cube") || !av_strcasecmp(ext, "1dlut")) {
1402 ret = parse_cube_1d(ctx, f);
1404 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
1405 ret = AVERROR(EINVAL);
1408 if (!ret && !lut1d->lutsize) {
1409 av_log(ctx, AV_LOG_ERROR, "1D LUT is empty\n");
1410 ret = AVERROR_INVALIDDATA;
1418 static AVFrame *apply_1d_lut(AVFilterLink *inlink, AVFrame *in)
1420 AVFilterContext *ctx = inlink->dst;
1421 LUT1DContext *lut1d = ctx->priv;
1422 AVFilterLink *outlink = inlink->dst->outputs[0];
1426 if (av_frame_is_writable(in)) {
1429 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
1434 av_frame_copy_props(out, in);
1439 ctx->internal->execute(ctx, lut1d->interp, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
1447 static int filter_frame_1d(AVFilterLink *inlink, AVFrame *in)
1449 AVFilterLink *outlink = inlink->dst->outputs[0];
1450 AVFrame *out = apply_1d_lut(inlink, in);
1452 return AVERROR(ENOMEM);
1453 return ff_filter_frame(outlink, out);
1456 static const AVFilterPad lut1d_inputs[] = {
1459 .type = AVMEDIA_TYPE_VIDEO,
1460 .filter_frame = filter_frame_1d,
1461 .config_props = config_input_1d,
1466 static const AVFilterPad lut1d_outputs[] = {
1469 .type = AVMEDIA_TYPE_VIDEO,
1474 AVFilter ff_vf_lut1d = {
1476 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 1D LUT."),
1477 .priv_size = sizeof(LUT1DContext),
1479 .query_formats = query_formats,
1480 .inputs = lut1d_inputs,
1481 .outputs = lut1d_outputs,
1482 .priv_class = &lut1d_class,
1483 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,