2 * Copyright (c) 2013 Clément Bœsch
4 * This file is part of FFmpeg.
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.
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.
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
23 * 3D Lookup table filter
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"
33 #include "drawutils.h"
35 #include "framesync.h"
46 INTERPOLATE_TRILINEAR,
47 INTERPOLATE_TETRAHEDRAL,
55 /* 3D LUT don't often go up to level 32, but it is common to have a Hald CLUT
56 * of 512x512 (64x64x64) */
59 typedef struct LUT3DContext {
61 int interpolation; ///<interp_mode
65 avfilter_action_func *interp;
66 struct rgbvec lut[MAX_LEVEL][MAX_LEVEL][MAX_LEVEL];
68 #if CONFIG_HALDCLUT_FILTER
69 uint8_t clut_rgba_map[4];
77 typedef struct ThreadData {
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" }, \
90 static inline float lerpf(float v0, float v1, float f)
92 return v0 + (v1 - v0) * f;
95 static inline struct rgbvec lerp(const struct rgbvec *v0, const struct rgbvec *v1, float f)
98 lerpf(v0->r, v1->r, f), lerpf(v0->g, v1->g, f), lerpf(v0->b, v1->b, f)
103 #define NEAR(x) ((int)((x) + .5))
104 #define PREV(x) ((int)(x))
105 #define NEXT(x) (FFMIN((int)(x) + 1, lut3d->lutsize - 1))
108 * Get the nearest defined point
110 static inline struct rgbvec interp_nearest(const LUT3DContext *lut3d,
111 const struct rgbvec *s)
113 return lut3d->lut[NEAR(s->r)][NEAR(s->g)][NEAR(s->b)];
117 * Interpolate using the 8 vertices of a cube
118 * @see https://en.wikipedia.org/wiki/Trilinear_interpolation
120 static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d,
121 const struct rgbvec *s)
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);
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
148 static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d,
149 const struct rgbvec *s)
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]];
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;
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;
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;
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;
201 #define DEFINE_INTERP_FUNC(name, nbits) \
202 static int interp_##nbits##_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
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); \
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]; \
235 dstrow += out->linesize[0]; \
236 srcrow += in ->linesize[0]; \
241 DEFINE_INTERP_FUNC(nearest, 8)
242 DEFINE_INTERP_FUNC(trilinear, 8)
243 DEFINE_INTERP_FUNC(tetrahedral, 8)
245 DEFINE_INTERP_FUNC(nearest, 16)
246 DEFINE_INTERP_FUNC(trilinear, 16)
247 DEFINE_INTERP_FUNC(tetrahedral, 16)
249 #define MAX_LINE_SIZE 512
251 static int skip_line(const char *p)
253 while (*p && av_isspace(*p))
255 return !*p || *p == '#';
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; \
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)
269 LUT3DContext *lut3d = ctx->priv;
270 char line[MAX_LINE_SIZE];
273 lut3d->lutsize = size = 33;
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);
282 lut3d->lutsize = size;
283 NEXT_LINE(skip_line(line));
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;
300 static int parse_cube(AVFilterContext *ctx, FILE *f)
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};
307 while (fgets(line, sizeof(line), f)) {
308 if (!strncmp(line, "LUT_3D_SIZE ", 12)) {
310 const int size = strtol(line + 12, NULL, 0);
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);
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];
325 if (!strncmp(line, "DOMAIN_", 7)) {
327 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
328 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
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]);
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];
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)
355 char line[MAX_LINE_SIZE];
356 LUT3DContext *lut3d = ctx->priv;
359 const float scale = 16*16*16;
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++) {
367 struct rgbvec *vec = &lut3d->lut[k][j][i];
369 NEXT_LINE(skip_line(line));
370 if (sscanf(line, "%d %d %d", &r, &g, &b) != 3)
371 return AVERROR_INVALIDDATA;
382 static int parse_m3d(AVFilterContext *ctx, FILE *f)
384 LUT3DContext *lut3d = ctx->priv;
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};
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)) \
399 case 'r': rgb_map[id] = 0; break; \
400 case 'g': rgb_map[id] = 1; break; \
401 case 'b': rgb_map[id] = 2; break; \
403 while (*p && !av_isspace(*p)) \
413 if (in == -1 || out == -1) {
414 av_log(ctx, AV_LOG_ERROR, "in and out must be defined\n");
415 return AVERROR_INVALIDDATA;
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;
423 for (size = 1; size*size*size < in; size++);
424 lut3d->lutsize = size;
425 scale = 1. / (out - 1);
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];
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;
445 static void set_identity_matrix(LUT3DContext *lut3d, int size)
448 const float c = 1. / (size - 1);
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];
463 static int query_formats(AVFilterContext *ctx)
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,
475 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
477 return AVERROR(ENOMEM);
478 return ff_set_common_formats(ctx, fmts_list);
481 static int config_input(AVFilterLink *inlink)
484 LUT3DContext *lut3d = inlink->dst->priv;
485 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
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:
495 ff_fill_rgba_map(lut3d->rgba_map, inlink->format);
496 lut3d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
498 #define SET_FUNC(name) do { \
499 if (is16bit) lut3d->interp = interp_16_##name; \
500 else lut3d->interp = interp_8_##name; \
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;
514 static AVFrame *apply_lut(AVFilterLink *inlink, AVFrame *in)
516 AVFilterContext *ctx = inlink->dst;
517 LUT3DContext *lut3d = ctx->priv;
518 AVFilterLink *outlink = inlink->dst->outputs[0];
522 if (av_frame_is_writable(in)) {
525 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
530 av_frame_copy_props(out, in);
535 ctx->internal->execute(ctx, lut3d->interp, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
543 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
545 AVFilterLink *outlink = inlink->dst->outputs[0];
546 AVFrame *out = apply_lut(inlink, in);
548 return AVERROR(ENOMEM);
549 return ff_filter_frame(outlink, out);
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 },
558 AVFILTER_DEFINE_CLASS(lut3d);
560 static av_cold int lut3d_init(AVFilterContext *ctx)
565 LUT3DContext *lut3d = ctx->priv;
568 set_identity_matrix(lut3d, 32);
572 f = fopen(lut3d->file, "r");
574 ret = AVERROR(errno);
575 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret));
579 ext = strrchr(lut3d->file, '.');
581 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
582 ret = AVERROR_INVALIDDATA;
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);
596 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
597 ret = AVERROR(EINVAL);
600 if (!ret && !lut3d->lutsize) {
601 av_log(ctx, AV_LOG_ERROR, "3D LUT is empty\n");
602 ret = AVERROR_INVALIDDATA;
610 static const AVFilterPad lut3d_inputs[] = {
613 .type = AVMEDIA_TYPE_VIDEO,
614 .filter_frame = filter_frame,
615 .config_props = config_input,
620 static const AVFilterPad lut3d_outputs[] = {
623 .type = AVMEDIA_TYPE_VIDEO,
628 AVFilter ff_vf_lut3d = {
630 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
631 .priv_size = sizeof(LUT3DContext),
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,
641 #if CONFIG_HALDCLUT_FILTER
643 static void update_clut(LUT3DContext *lut3d, const AVFrame *frame)
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;
652 #define LOAD_CLUT(nbits) do { \
653 int i, j, k, x = 0, y = 0; \
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); \
673 if (!lut3d->clut_is16bit) LOAD_CLUT(8);
678 static int config_output(AVFilterLink *outlink)
680 AVFilterContext *ctx = outlink->src;
681 LUT3DContext *lut3d = ctx->priv;
684 ret = ff_framesync_init_dualinput(&lut3d->fs, ctx);
687 outlink->w = ctx->inputs[0]->w;
688 outlink->h = ctx->inputs[0]->h;
689 outlink->time_base = ctx->inputs[0]->time_base;
690 if ((ret = ff_framesync_configure(&lut3d->fs)) < 0)
695 static int activate(AVFilterContext *ctx)
697 LUT3DContext *s = ctx->priv;
698 return ff_framesync_activate(&s->fs);
701 static int config_clut(AVFilterLink *inlink)
703 int size, level, w, h;
704 AVFilterContext *ctx = inlink->dst;
705 LUT3DContext *lut3d = ctx->priv;
706 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
710 lut3d->clut_is16bit = 0;
711 switch (inlink->format) {
712 case AV_PIX_FMT_RGB48:
713 case AV_PIX_FMT_BGR48:
714 case AV_PIX_FMT_RGBA64:
715 case AV_PIX_FMT_BGRA64:
716 lut3d->clut_is16bit = 1;
719 lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3;
720 ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format);
722 if (inlink->w > inlink->h)
723 av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the "
724 "Hald CLUT will be ignored\n", inlink->w - inlink->h);
725 else if (inlink->w < inlink->h)
726 av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the "
727 "Hald CLUT will be ignored\n", inlink->h - inlink->w);
728 lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h);
730 for (level = 1; level*level*level < w; level++);
731 size = level*level*level;
733 av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n");
734 return AVERROR_INVALIDDATA;
736 av_assert0(w == h && w == size);
738 if (level > MAX_LEVEL) {
739 const int max_clut_level = sqrt(MAX_LEVEL);
740 const int max_clut_size = max_clut_level*max_clut_level*max_clut_level;
741 av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT "
742 "(maximum level is %d, or %dx%d CLUT)\n",
743 max_clut_level, max_clut_size, max_clut_size);
744 return AVERROR(EINVAL);
746 lut3d->lutsize = level;
751 static int update_apply_clut(FFFrameSync *fs)
753 AVFilterContext *ctx = fs->parent;
754 AVFilterLink *inlink = ctx->inputs[0];
755 AVFrame *master, *second, *out;
758 ret = ff_framesync_dualinput_get(fs, &master, &second);
762 return ff_filter_frame(ctx->outputs[0], master);
763 update_clut(ctx->priv, second);
764 out = apply_lut(inlink, master);
765 return ff_filter_frame(ctx->outputs[0], out);
768 static av_cold int haldclut_init(AVFilterContext *ctx)
770 LUT3DContext *lut3d = ctx->priv;
771 lut3d->fs.on_event = update_apply_clut;
775 static av_cold void haldclut_uninit(AVFilterContext *ctx)
777 LUT3DContext *lut3d = ctx->priv;
778 ff_framesync_uninit(&lut3d->fs);
781 static const AVOption haldclut_options[] = {
785 FRAMESYNC_DEFINE_CLASS(haldclut, LUT3DContext, fs);
787 static const AVFilterPad haldclut_inputs[] = {
790 .type = AVMEDIA_TYPE_VIDEO,
791 .config_props = config_input,
794 .type = AVMEDIA_TYPE_VIDEO,
795 .config_props = config_clut,
800 static const AVFilterPad haldclut_outputs[] = {
803 .type = AVMEDIA_TYPE_VIDEO,
804 .config_props = config_output,
809 AVFilter ff_vf_haldclut = {
811 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
812 .priv_size = sizeof(LUT3DContext),
813 .preinit = haldclut_framesync_preinit,
814 .init = haldclut_init,
815 .uninit = haldclut_uninit,
816 .query_formats = query_formats,
817 .activate = activate,
818 .inputs = haldclut_inputs,
819 .outputs = haldclut_outputs,
820 .priv_class = &haldclut_class,
821 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,