Imported Debian version 2.4.3~trusty1
[deb_ffmpeg.git] / ffmpeg / libavfilter / vf_lut3d.c
CommitLineData
2ba45a60
DM
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
44enum interp_mode {
45 INTERPOLATE_NEAREST,
46 INTERPOLATE_TRILINEAR,
47 INTERPOLATE_TETRAHEDRAL,
48 NB_INTERP_MODE
49};
50
51struct 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
59typedef struct LUT3DContext {
60 const AVClass *class;
61 enum interp_mode interpolation;
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
77typedef 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
90static inline float lerpf(float v0, float v1, float f)
91{
92 return v0 + (v1 - v0) * f;
93}
94
95static 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 */
110static 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 */
120static 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 */
148static 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) \
202static 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
241DEFINE_INTERP_FUNC(nearest, 8)
242DEFINE_INTERP_FUNC(trilinear, 8)
243DEFINE_INTERP_FUNC(tetrahedral, 8)
244
245DEFINE_INTERP_FUNC(nearest, 16)
246DEFINE_INTERP_FUNC(trilinear, 16)
247DEFINE_INTERP_FUNC(tetrahedral, 16)
248
249#define MAX_LINE_SIZE 512
250
251static 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 */
267static 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 */
300static 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 NEXT_LINE(0);
324 if (!strncmp(line, "DOMAIN_", 7)) {
325 float *vals = NULL;
326 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
327 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
328 if (!vals)
329 return AVERROR_INVALIDDATA;
330 sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
331 av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
332 min[0], min[1], min[2], max[0], max[1], max[2]);
333 continue;
334 }
335 } while (skip_line(line));
336 if (sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
337 return AVERROR_INVALIDDATA;
338 vec->r *= max[0] - min[0];
339 vec->g *= max[1] - min[1];
340 vec->b *= max[2] - min[2];
341 }
342 }
343 }
344 break;
345 }
346 }
347 return 0;
348}
349
350/* Assume 17x17x17 LUT with a 16-bit depth
351 * FIXME: it seems there are various 3dl formats */
352static int parse_3dl(AVFilterContext *ctx, FILE *f)
353{
354 char line[MAX_LINE_SIZE];
355 LUT3DContext *lut3d = ctx->priv;
356 int i, j, k;
357 const int size = 17;
358 const float scale = 16*16*16;
359
360 lut3d->lutsize = size;
361 NEXT_LINE(skip_line(line));
362 for (k = 0; k < size; k++) {
363 for (j = 0; j < size; j++) {
364 for (i = 0; i < size; i++) {
365 int r, g, b;
366 struct rgbvec *vec = &lut3d->lut[k][j][i];
367
368 NEXT_LINE(skip_line(line));
369 if (sscanf(line, "%d %d %d", &r, &g, &b) != 3)
370 return AVERROR_INVALIDDATA;
371 vec->r = r / scale;
372 vec->g = g / scale;
373 vec->b = b / scale;
374 }
375 }
376 }
377 return 0;
378}
379
380/* Pandora format */
381static int parse_m3d(AVFilterContext *ctx, FILE *f)
382{
383 LUT3DContext *lut3d = ctx->priv;
384 float scale;
385 int i, j, k, size, in = -1, out = -1;
386 char line[MAX_LINE_SIZE];
387 uint8_t rgb_map[3] = {0, 1, 2};
388
389 while (fgets(line, sizeof(line), f)) {
390 if (!strncmp(line, "in", 2)) in = strtol(line + 2, NULL, 0);
391 else if (!strncmp(line, "out", 3)) out = strtol(line + 3, NULL, 0);
392 else if (!strncmp(line, "values", 6)) {
393 const char *p = line + 6;
394#define SET_COLOR(id) do { \
395 while (av_isspace(*p)) \
396 p++; \
397 switch (*p) { \
398 case 'r': rgb_map[id] = 0; break; \
399 case 'g': rgb_map[id] = 1; break; \
400 case 'b': rgb_map[id] = 2; break; \
401 } \
402 while (*p && !av_isspace(*p)) \
403 p++; \
404} while (0)
405 SET_COLOR(0);
406 SET_COLOR(1);
407 SET_COLOR(2);
408 break;
409 }
410 }
411
412 if (in == -1 || out == -1) {
413 av_log(ctx, AV_LOG_ERROR, "in and out must be defined\n");
414 return AVERROR_INVALIDDATA;
415 }
416 if (in < 2 || out < 2 ||
417 in > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL ||
418 out > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL) {
419 av_log(ctx, AV_LOG_ERROR, "invalid in (%d) or out (%d)\n", in, out);
420 return AVERROR_INVALIDDATA;
421 }
422 for (size = 1; size*size*size < in; size++);
423 lut3d->lutsize = size;
424 scale = 1. / (out - 1);
425
426 for (k = 0; k < size; k++) {
427 for (j = 0; j < size; j++) {
428 for (i = 0; i < size; i++) {
429 struct rgbvec *vec = &lut3d->lut[k][j][i];
430 float val[3];
431
432 NEXT_LINE(0);
433 if (sscanf(line, "%f %f %f", val, val + 1, val + 2) != 3)
434 return AVERROR_INVALIDDATA;
435 vec->r = val[rgb_map[0]] * scale;
436 vec->g = val[rgb_map[1]] * scale;
437 vec->b = val[rgb_map[2]] * scale;
438 }
439 }
440 }
441 return 0;
442}
443
444static void set_identity_matrix(LUT3DContext *lut3d, int size)
445{
446 int i, j, k;
447 const float c = 1. / (size - 1);
448
449 lut3d->lutsize = size;
450 for (k = 0; k < size; k++) {
451 for (j = 0; j < size; j++) {
452 for (i = 0; i < size; i++) {
453 struct rgbvec *vec = &lut3d->lut[k][j][i];
454 vec->r = k * c;
455 vec->g = j * c;
456 vec->b = i * c;
457 }
458 }
459 }
460}
461
462static int query_formats(AVFilterContext *ctx)
463{
464 static const enum AVPixelFormat pix_fmts[] = {
465 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
466 AV_PIX_FMT_RGBA, AV_PIX_FMT_BGRA,
467 AV_PIX_FMT_ARGB, AV_PIX_FMT_ABGR,
468 AV_PIX_FMT_0RGB, AV_PIX_FMT_0BGR,
469 AV_PIX_FMT_RGB0, AV_PIX_FMT_BGR0,
470 AV_PIX_FMT_RGB48, AV_PIX_FMT_BGR48,
471 AV_PIX_FMT_RGBA64, AV_PIX_FMT_BGRA64,
472 AV_PIX_FMT_NONE
473 };
474 ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
475 return 0;
476}
477
478static int config_input(AVFilterLink *inlink)
479{
480 int is16bit = 0;
481 LUT3DContext *lut3d = inlink->dst->priv;
482 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
483
484 switch (inlink->format) {
485 case AV_PIX_FMT_RGB48:
486 case AV_PIX_FMT_BGR48:
487 case AV_PIX_FMT_RGBA64:
488 case AV_PIX_FMT_BGRA64:
489 is16bit = 1;
490 }
491
492 ff_fill_rgba_map(lut3d->rgba_map, inlink->format);
493 lut3d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
494
495#define SET_FUNC(name) do { \
496 if (is16bit) lut3d->interp = interp_16_##name; \
497 else lut3d->interp = interp_8_##name; \
498} while (0)
499
500 switch (lut3d->interpolation) {
501 case INTERPOLATE_NEAREST: SET_FUNC(nearest); break;
502 case INTERPOLATE_TRILINEAR: SET_FUNC(trilinear); break;
503 case INTERPOLATE_TETRAHEDRAL: SET_FUNC(tetrahedral); break;
504 default:
505 av_assert0(0);
506 }
507
508 return 0;
509}
510
511static AVFrame *apply_lut(AVFilterLink *inlink, AVFrame *in)
512{
513 AVFilterContext *ctx = inlink->dst;
514 LUT3DContext *lut3d = ctx->priv;
515 AVFilterLink *outlink = inlink->dst->outputs[0];
516 AVFrame *out;
517 ThreadData td;
518
519 if (av_frame_is_writable(in)) {
520 out = in;
521 } else {
522 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
523 if (!out) {
524 av_frame_free(&in);
525 return NULL;
526 }
527 av_frame_copy_props(out, in);
528 }
529
530 td.in = in;
531 td.out = out;
532 ctx->internal->execute(ctx, lut3d->interp, &td, NULL, FFMIN(outlink->h, ctx->graph->nb_threads));
533
534 if (out != in)
535 av_frame_free(&in);
536
537 return out;
538}
539
540static int filter_frame(AVFilterLink *inlink, AVFrame *in)
541{
542 AVFilterLink *outlink = inlink->dst->outputs[0];
543 AVFrame *out = apply_lut(inlink, in);
544 if (!out)
545 return AVERROR(ENOMEM);
546 return ff_filter_frame(outlink, out);
547}
548
549#if CONFIG_LUT3D_FILTER
550static const AVOption lut3d_options[] = {
551 { "file", "set 3D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
552 COMMON_OPTIONS
553};
554
555AVFILTER_DEFINE_CLASS(lut3d);
556
557static av_cold int lut3d_init(AVFilterContext *ctx)
558{
559 int ret;
560 FILE *f;
561 const char *ext;
562 LUT3DContext *lut3d = ctx->priv;
563
564 if (!lut3d->file) {
565 set_identity_matrix(lut3d, 32);
566 return 0;
567 }
568
569 f = fopen(lut3d->file, "r");
570 if (!f) {
571 ret = AVERROR(errno);
572 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret));
573 return ret;
574 }
575
576 ext = strrchr(lut3d->file, '.');
577 if (!ext) {
578 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
579 ret = AVERROR_INVALIDDATA;
580 goto end;
581 }
582 ext++;
583
584 if (!av_strcasecmp(ext, "dat")) {
585 ret = parse_dat(ctx, f);
586 } else if (!av_strcasecmp(ext, "3dl")) {
587 ret = parse_3dl(ctx, f);
588 } else if (!av_strcasecmp(ext, "cube")) {
589 ret = parse_cube(ctx, f);
590 } else if (!av_strcasecmp(ext, "m3d")) {
591 ret = parse_m3d(ctx, f);
592 } else {
593 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
594 ret = AVERROR(EINVAL);
595 }
596
597 if (!ret && !lut3d->lutsize) {
598 av_log(ctx, AV_LOG_ERROR, "3D LUT is empty\n");
599 ret = AVERROR_INVALIDDATA;
600 }
601
602end:
603 fclose(f);
604 return ret;
605}
606
607static const AVFilterPad lut3d_inputs[] = {
608 {
609 .name = "default",
610 .type = AVMEDIA_TYPE_VIDEO,
611 .filter_frame = filter_frame,
612 .config_props = config_input,
613 },
614 { NULL }
615};
616
617static const AVFilterPad lut3d_outputs[] = {
618 {
619 .name = "default",
620 .type = AVMEDIA_TYPE_VIDEO,
621 },
622 { NULL }
623};
624
625AVFilter ff_vf_lut3d = {
626 .name = "lut3d",
627 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
628 .priv_size = sizeof(LUT3DContext),
629 .init = lut3d_init,
630 .query_formats = query_formats,
631 .inputs = lut3d_inputs,
632 .outputs = lut3d_outputs,
633 .priv_class = &lut3d_class,
634 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
635};
636#endif
637
638#if CONFIG_HALDCLUT_FILTER
639
640static void update_clut(LUT3DContext *lut3d, const AVFrame *frame)
641{
642 const uint8_t *data = frame->data[0];
643 const int linesize = frame->linesize[0];
644 const int w = lut3d->clut_width;
645 const int step = lut3d->clut_step;
646 const uint8_t *rgba_map = lut3d->clut_rgba_map;
647 const int level = lut3d->lutsize;
648
649#define LOAD_CLUT(nbits) do { \
650 int i, j, k, x = 0, y = 0; \
651 \
652 for (k = 0; k < level; k++) { \
653 for (j = 0; j < level; j++) { \
654 for (i = 0; i < level; i++) { \
655 const uint##nbits##_t *src = (const uint##nbits##_t *) \
656 (data + y*linesize + x*step); \
657 struct rgbvec *vec = &lut3d->lut[i][j][k]; \
658 vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \
659 vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \
660 vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \
661 if (++x == w) { \
662 x = 0; \
663 y++; \
664 } \
665 } \
666 } \
667 } \
668} while (0)
669
670 if (!lut3d->clut_is16bit) LOAD_CLUT(8);
671 else LOAD_CLUT(16);
672}
673
674
675static int config_output(AVFilterLink *outlink)
676{
677 AVFilterContext *ctx = outlink->src;
678 LUT3DContext *lut3d = ctx->priv;
679 int ret;
680
681 outlink->w = ctx->inputs[0]->w;
682 outlink->h = ctx->inputs[0]->h;
683 outlink->time_base = ctx->inputs[0]->time_base;
684 if ((ret = ff_dualinput_init(ctx, &lut3d->dinput)) < 0)
685 return ret;
686 return 0;
687}
688
689static int filter_frame_hald(AVFilterLink *inlink, AVFrame *inpicref)
690{
691 LUT3DContext *s = inlink->dst->priv;
692 return ff_dualinput_filter_frame(&s->dinput, inlink, inpicref);
693}
694
695static int request_frame(AVFilterLink *outlink)
696{
697 LUT3DContext *s = outlink->src->priv;
698 return ff_dualinput_request_frame(&s->dinput, outlink);
699}
700
701static int config_clut(AVFilterLink *inlink)
702{
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);
707
708 lut3d->clut_is16bit = 0;
709 switch (inlink->format) {
710 case AV_PIX_FMT_RGB48:
711 case AV_PIX_FMT_BGR48:
712 case AV_PIX_FMT_RGBA64:
713 case AV_PIX_FMT_BGRA64:
714 lut3d->clut_is16bit = 1;
715 }
716
717 lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3;
718 ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format);
719
720 if (inlink->w > inlink->h)
721 av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the "
722 "Hald CLUT will be ignored\n", inlink->w - inlink->h);
723 else if (inlink->w < inlink->h)
724 av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the "
725 "Hald CLUT will be ignored\n", inlink->h - inlink->w);
726 lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h);
727
728 for (level = 1; level*level*level < w; level++);
729 size = level*level*level;
730 if (size != w) {
731 av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n");
732 return AVERROR_INVALIDDATA;
733 }
734 av_assert0(w == h && w == size);
735 level *= level;
736 if (level > MAX_LEVEL) {
737 const int max_clut_level = sqrt(MAX_LEVEL);
738 const int max_clut_size = max_clut_level*max_clut_level*max_clut_level;
739 av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT "
740 "(maximum level is %d, or %dx%d CLUT)\n",
741 max_clut_level, max_clut_size, max_clut_size);
742 return AVERROR(EINVAL);
743 }
744 lut3d->lutsize = level;
745
746 return 0;
747}
748
749static AVFrame *update_apply_clut(AVFilterContext *ctx, AVFrame *main,
750 const AVFrame *second)
751{
752 AVFilterLink *inlink = ctx->inputs[0];
753 update_clut(ctx->priv, second);
754 return apply_lut(inlink, main);
755}
756
757static av_cold int haldclut_init(AVFilterContext *ctx)
758{
759 LUT3DContext *lut3d = ctx->priv;
760 lut3d->dinput.process = update_apply_clut;
761 return 0;
762}
763
764static av_cold void haldclut_uninit(AVFilterContext *ctx)
765{
766 LUT3DContext *lut3d = ctx->priv;
767 ff_dualinput_uninit(&lut3d->dinput);
768}
769
770static const AVOption haldclut_options[] = {
771 { "shortest", "force termination when the shortest input terminates", OFFSET(dinput.shortest), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS },
772 { "repeatlast", "continue applying the last clut after eos", OFFSET(dinput.repeatlast), AV_OPT_TYPE_INT, { .i64 = 1 }, 0, 1, FLAGS },
773 COMMON_OPTIONS
774};
775
776AVFILTER_DEFINE_CLASS(haldclut);
777
778static const AVFilterPad haldclut_inputs[] = {
779 {
780 .name = "main",
781 .type = AVMEDIA_TYPE_VIDEO,
782 .filter_frame = filter_frame_hald,
783 .config_props = config_input,
784 },{
785 .name = "clut",
786 .type = AVMEDIA_TYPE_VIDEO,
787 .filter_frame = filter_frame_hald,
788 .config_props = config_clut,
789 },
790 { NULL }
791};
792
793static const AVFilterPad haldclut_outputs[] = {
794 {
795 .name = "default",
796 .type = AVMEDIA_TYPE_VIDEO,
797 .request_frame = request_frame,
798 .config_props = config_output,
799 },
800 { NULL }
801};
802
803AVFilter ff_vf_haldclut = {
804 .name = "haldclut",
805 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
806 .priv_size = sizeof(LUT3DContext),
807 .init = haldclut_init,
808 .uninit = haldclut_uninit,
809 .query_formats = query_formats,
810 .inputs = haldclut_inputs,
811 .outputs = haldclut_outputs,
812 .priv_class = &haldclut_class,
813 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
814};
815#endif