FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
vf_lut3d.c
Go to the documentation of this file.
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 
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;
62  char *file;
64  int step;
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  \
220  for (y = slice_start; y < slice_end; y++) { \
221  uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
222  const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
223  for (x = 0; x < in->width * step; x += step) { \
224  const float scale = (1. / ((1<<nbits) - 1)) * (lut3d->lutsize - 1); \
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; seems to be generated by
266  * Davinci */
267 static int parse_dat(AVFilterContext *ctx, FILE *f)
268 {
269  LUT3DContext *lut3d = ctx->priv;
270  const int size = lut3d->lutsize;
271  int i, j, k;
272 
273  for (k = 0; k < size; k++) {
274  for (j = 0; j < size; j++) {
275  for (i = 0; i < size; i++) {
276  char line[MAX_LINE_SIZE];
277  struct rgbvec *vec = &lut3d->lut[k][j][i];
278  NEXT_LINE(skip_line(line));
279  sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b);
280  }
281  }
282  }
283  return 0;
284 }
285 
286 /* Iridas format */
287 static int parse_cube(AVFilterContext *ctx, FILE *f)
288 {
289  LUT3DContext *lut3d = ctx->priv;
290  char line[MAX_LINE_SIZE];
291  float min[3] = {0.0, 0.0, 0.0};
292  float max[3] = {1.0, 1.0, 1.0};
293 
294  while (fgets(line, sizeof(line), f)) {
295  if (!strncmp(line, "LUT_3D_SIZE ", 12)) {
296  int i, j, k;
297  const int size = strtol(line + 12, NULL, 0);
298 
300  av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n");
301  return AVERROR(EINVAL);
302  }
303  lut3d->lutsize = size;
304  for (k = 0; k < size; k++) {
305  for (j = 0; j < size; j++) {
306  for (i = 0; i < size; i++) {
307  struct rgbvec *vec = &lut3d->lut[i][j][k];
308 
309  do {
310  NEXT_LINE(0);
311  if (!strncmp(line, "DOMAIN_", 7)) {
312  float *vals = NULL;
313  if (!strncmp(line + 7, "MIN ", 4)) vals = min;
314  else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
315  if (!vals)
316  return AVERROR_INVALIDDATA;
317  sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
318  av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
319  min[0], min[1], min[2], max[0], max[1], max[2]);
320  continue;
321  }
322  } while (skip_line(line));
323  if (sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
324  return AVERROR_INVALIDDATA;
325  vec->r *= max[0] - min[0];
326  vec->g *= max[1] - min[1];
327  vec->b *= max[2] - min[2];
328  }
329  }
330  }
331  break;
332  }
333  }
334  return 0;
335 }
336 
337 /* Assume 17x17x17 LUT with a 16-bit depth
338  * FIXME: it seems there are various 3dl formats */
339 static int parse_3dl(AVFilterContext *ctx, FILE *f)
340 {
341  char line[MAX_LINE_SIZE];
342  LUT3DContext *lut3d = ctx->priv;
343  int i, j, k;
344  const int size = 17;
345  const float scale = 16*16*16;
346 
347  lut3d->lutsize = size;
348  NEXT_LINE(skip_line(line));
349  for (k = 0; k < size; k++) {
350  for (j = 0; j < size; j++) {
351  for (i = 0; i < size; i++) {
352  int r, g, b;
353  struct rgbvec *vec = &lut3d->lut[k][j][i];
354 
355  NEXT_LINE(skip_line(line));
356  if (sscanf(line, "%d %d %d", &r, &g, &b) != 3)
357  return AVERROR_INVALIDDATA;
358  vec->r = r / scale;
359  vec->g = g / scale;
360  vec->b = b / scale;
361  }
362  }
363  }
364  return 0;
365 }
366 
367 /* Pandora format */
368 static int parse_m3d(AVFilterContext *ctx, FILE *f)
369 {
370  LUT3DContext *lut3d = ctx->priv;
371  float scale;
372  int i, j, k, size, in = -1, out = -1;
373  char line[MAX_LINE_SIZE];
374  uint8_t rgb_map[3] = {0, 1, 2};
375 
376  while (fgets(line, sizeof(line), f)) {
377  if (!strncmp(line, "in", 2)) in = strtol(line + 2, NULL, 0);
378  else if (!strncmp(line, "out", 3)) out = strtol(line + 3, NULL, 0);
379  else if (!strncmp(line, "values", 6)) {
380  const char *p = line + 6;
381 #define SET_COLOR(id) do { \
382  while (av_isspace(*p)) \
383  p++; \
384  switch (*p) { \
385  case 'r': rgb_map[id] = 0; break; \
386  case 'g': rgb_map[id] = 1; break; \
387  case 'b': rgb_map[id] = 2; break; \
388  } \
389  while (*p && !av_isspace(*p)) \
390  p++; \
391 } while (0)
392  SET_COLOR(0);
393  SET_COLOR(1);
394  SET_COLOR(2);
395  break;
396  }
397  }
398 
399  if (in == -1 || out == -1) {
400  av_log(ctx, AV_LOG_ERROR, "in and out must be defined\n");
401  return AVERROR_INVALIDDATA;
402  }
403  if (in < 2 || out < 2 ||
406  av_log(ctx, AV_LOG_ERROR, "invalid in (%d) or out (%d)\n", in, out);
407  return AVERROR_INVALIDDATA;
408  }
409  for (size = 1; size*size*size < in; size++);
410  lut3d->lutsize = size;
411  scale = 1. / (out - 1);
412 
413  for (k = 0; k < size; k++) {
414  for (j = 0; j < size; j++) {
415  for (i = 0; i < size; i++) {
416  struct rgbvec *vec = &lut3d->lut[k][j][i];
417  float val[3];
418 
419  NEXT_LINE(0);
420  if (sscanf(line, "%f %f %f", val, val + 1, val + 2) != 3)
421  return AVERROR_INVALIDDATA;
422  vec->r = val[rgb_map[0]] * scale;
423  vec->g = val[rgb_map[1]] * scale;
424  vec->b = val[rgb_map[2]] * scale;
425  }
426  }
427  }
428  return 0;
429 }
430 
431 static void set_identity_matrix(LUT3DContext *lut3d, int size)
432 {
433  int i, j, k;
434  const float c = 1. / (size - 1);
435 
436  lut3d->lutsize = size;
437  for (k = 0; k < size; k++) {
438  for (j = 0; j < size; j++) {
439  for (i = 0; i < size; i++) {
440  struct rgbvec *vec = &lut3d->lut[k][j][i];
441  vec->r = k * c;
442  vec->g = j * c;
443  vec->b = i * c;
444  }
445  }
446  }
447 }
448 
450 {
451  static const enum AVPixelFormat pix_fmts[] = {
460  };
462  return 0;
463 }
464 
465 static int config_input(AVFilterLink *inlink)
466 {
467  int is16bit = 0;
468  LUT3DContext *lut3d = inlink->dst->priv;
469  const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
470 
471  switch (inlink->format) {
472  case AV_PIX_FMT_RGB48:
473  case AV_PIX_FMT_BGR48:
474  case AV_PIX_FMT_RGBA64:
475  case AV_PIX_FMT_BGRA64:
476  is16bit = 1;
477  }
478 
479  ff_fill_rgba_map(lut3d->rgba_map, inlink->format);
480  lut3d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
481 
482 #define SET_FUNC(name) do { \
483  if (is16bit) lut3d->interp = interp_16_##name; \
484  else lut3d->interp = interp_8_##name; \
485 } while (0)
486 
487  switch (lut3d->interpolation) {
488  case INTERPOLATE_NEAREST: SET_FUNC(nearest); break;
489  case INTERPOLATE_TRILINEAR: SET_FUNC(trilinear); break;
490  case INTERPOLATE_TETRAHEDRAL: SET_FUNC(tetrahedral); break;
491  default:
492  av_assert0(0);
493  }
494 
495  return 0;
496 }
497 
499 {
500  AVFilterContext *ctx = inlink->dst;
501  LUT3DContext *lut3d = ctx->priv;
502  AVFilterLink *outlink = inlink->dst->outputs[0];
503  AVFrame *out;
504  ThreadData td;
505 
506  if (av_frame_is_writable(in)) {
507  out = in;
508  } else {
509  out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
510  if (!out) {
511  av_frame_free(&in);
512  return NULL;
513  }
514  av_frame_copy_props(out, in);
515  }
516 
517  td.in = in;
518  td.out = out;
519  ctx->internal->execute(ctx, lut3d->interp, &td, NULL, FFMIN(outlink->h, ctx->graph->nb_threads));
520 
521  if (out != in)
522  av_frame_free(&in);
523 
524  return out;
525 }
526 
527 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
528 {
529  AVFilterLink *outlink = inlink->dst->outputs[0];
530  AVFrame *out = apply_lut(inlink, in);
531  if (!out)
532  return AVERROR(ENOMEM);
533  return ff_filter_frame(outlink, out);
534 }
535 
536 #if CONFIG_LUT3D_FILTER
537 static const AVOption lut3d_options[] = {
538  { "file", "set 3D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
540 };
541 
542 AVFILTER_DEFINE_CLASS(lut3d);
543 
544 static av_cold int lut3d_init(AVFilterContext *ctx)
545 {
546  int ret;
547  FILE *f;
548  const char *ext;
549  LUT3DContext *lut3d = ctx->priv;
550 
551  if (!lut3d->file) {
552  set_identity_matrix(lut3d, 32);
553  return 0;
554  }
555 
556  f = fopen(lut3d->file, "r");
557  if (!f) {
558  ret = AVERROR(errno);
559  av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret));
560  return ret;
561  }
562 
563  ext = strrchr(lut3d->file, '.');
564  if (!ext) {
565  av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
566  ret = AVERROR_INVALIDDATA;
567  goto end;
568  }
569  ext++;
570 
571  if (!av_strcasecmp(ext, "dat")) {
572  lut3d->lutsize = 33;
573  ret = parse_dat(ctx, f);
574  } else if (!av_strcasecmp(ext, "3dl")) {
575  ret = parse_3dl(ctx, f);
576  } else if (!av_strcasecmp(ext, "cube")) {
577  ret = parse_cube(ctx, f);
578  } else if (!av_strcasecmp(ext, "m3d")) {
579  ret = parse_m3d(ctx, f);
580  } else {
581  av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
582  ret = AVERROR(EINVAL);
583  }
584 
585  if (!ret && !lut3d->lutsize) {
586  av_log(ctx, AV_LOG_ERROR, "3D LUT is empty\n");
587  ret = AVERROR_INVALIDDATA;
588  }
589 
590 end:
591  fclose(f);
592  return ret;
593 }
594 
595 static const AVFilterPad lut3d_inputs[] = {
596  {
597  .name = "default",
598  .type = AVMEDIA_TYPE_VIDEO,
599  .filter_frame = filter_frame,
600  .config_props = config_input,
601  },
602  { NULL }
603 };
604 
605 static const AVFilterPad lut3d_outputs[] = {
606  {
607  .name = "default",
608  .type = AVMEDIA_TYPE_VIDEO,
609  },
610  { NULL }
611 };
612 
613 AVFilter ff_vf_lut3d = {
614  .name = "lut3d",
615  .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
616  .priv_size = sizeof(LUT3DContext),
617  .init = lut3d_init,
619  .inputs = lut3d_inputs,
620  .outputs = lut3d_outputs,
621  .priv_class = &lut3d_class,
623 };
624 #endif
625 
626 #if CONFIG_HALDCLUT_FILTER
627 
628 static void update_clut(LUT3DContext *lut3d, const AVFrame *frame)
629 {
630  const uint8_t *data = frame->data[0];
631  const int linesize = frame->linesize[0];
632  const int w = lut3d->clut_width;
633  const int step = lut3d->clut_step;
634  const uint8_t *rgba_map = lut3d->clut_rgba_map;
635  const int level = lut3d->lutsize;
636 
637 #define LOAD_CLUT(nbits) do { \
638  int i, j, k, x = 0, y = 0; \
639  \
640  for (k = 0; k < level; k++) { \
641  for (j = 0; j < level; j++) { \
642  for (i = 0; i < level; i++) { \
643  const uint##nbits##_t *src = (const uint##nbits##_t *) \
644  (data + y*linesize + x*step); \
645  struct rgbvec *vec = &lut3d->lut[k][j][i]; \
646  vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \
647  vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \
648  vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \
649  if (++x == w) { \
650  x = 0; \
651  y++; \
652  } \
653  } \
654  } \
655  } \
656 } while (0)
657 
658  if (!lut3d->clut_is16bit) LOAD_CLUT(8);
659  else LOAD_CLUT(16);
660 }
661 
662 
663 static int config_output(AVFilterLink *outlink)
664 {
665  AVFilterContext *ctx = outlink->src;
666  LUT3DContext *lut3d = ctx->priv;
667  int ret;
668 
669  outlink->w = ctx->inputs[0]->w;
670  outlink->h = ctx->inputs[0]->h;
671  outlink->time_base = ctx->inputs[0]->time_base;
672  if ((ret = ff_dualinput_init(ctx, &lut3d->dinput)) < 0)
673  return ret;
674  return 0;
675 }
676 
677 static int filter_frame_hald(AVFilterLink *inlink, AVFrame *inpicref)
678 {
679  LUT3DContext *s = inlink->dst->priv;
680  return ff_dualinput_filter_frame(&s->dinput, inlink, inpicref);
681 }
682 
683 static int request_frame(AVFilterLink *outlink)
684 {
685  LUT3DContext *s = outlink->src->priv;
686  return ff_dualinput_request_frame(&s->dinput, outlink);
687 }
688 
689 static int config_clut(AVFilterLink *inlink)
690 {
691  int size, level, w, h;
692  AVFilterContext *ctx = inlink->dst;
693  LUT3DContext *lut3d = ctx->priv;
694  const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
695 
696  lut3d->clut_is16bit = 0;
697  switch (inlink->format) {
698  case AV_PIX_FMT_RGB48:
699  case AV_PIX_FMT_BGR48:
700  case AV_PIX_FMT_RGBA64:
701  case AV_PIX_FMT_BGRA64:
702  lut3d->clut_is16bit = 1;
703  }
704 
705  lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3;
706  ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format);
707 
708  if (inlink->w > inlink->h)
709  av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the "
710  "Hald CLUT will be ignored\n", inlink->w - inlink->h);
711  else if (inlink->w < inlink->h)
712  av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the "
713  "Hald CLUT will be ignored\n", inlink->h - inlink->w);
714  lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h);
715 
716  for (level = 1; level*level*level < w; level++);
717  size = level*level*level;
718  if (size != w) {
719  av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n");
720  return AVERROR_INVALIDDATA;
721  }
722  av_assert0(w == h && w == size);
723  level *= level;
724  if (level > MAX_LEVEL) {
725  const int max_clut_level = sqrt(MAX_LEVEL);
726  const int max_clut_size = max_clut_level*max_clut_level*max_clut_level;
727  av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT "
728  "(maximum level is %d, or %dx%d CLUT)\n",
729  max_clut_level, max_clut_size, max_clut_size);
730  return AVERROR(EINVAL);
731  }
732  lut3d->lutsize = level;
733 
734  return 0;
735 }
736 
737 static AVFrame *update_apply_clut(AVFilterContext *ctx, AVFrame *main,
738  const AVFrame *second)
739 {
740  AVFilterLink *inlink = ctx->inputs[0];
741  update_clut(ctx->priv, second);
742  return apply_lut(inlink, main);
743 }
744 
745 static av_cold int haldclut_init(AVFilterContext *ctx)
746 {
747  LUT3DContext *lut3d = ctx->priv;
748  lut3d->dinput.process = update_apply_clut;
749  return 0;
750 }
751 
752 static av_cold void haldclut_uninit(AVFilterContext *ctx)
753 {
754  LUT3DContext *lut3d = ctx->priv;
755  ff_dualinput_uninit(&lut3d->dinput);
756 }
757 
758 static const AVOption haldclut_options[] = {
759  { "shortest", "force termination when the shortest input terminates", OFFSET(dinput.shortest), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, FLAGS },
760  { "repeatlast", "continue applying the last clut after eos", OFFSET(dinput.repeatlast), AV_OPT_TYPE_INT, { .i64 = 1 }, 0, 1, FLAGS },
762 };
763 
764 AVFILTER_DEFINE_CLASS(haldclut);
765 
766 static const AVFilterPad haldclut_inputs[] = {
767  {
768  .name = "main",
769  .type = AVMEDIA_TYPE_VIDEO,
770  .filter_frame = filter_frame_hald,
771  .config_props = config_input,
772  },{
773  .name = "clut",
774  .type = AVMEDIA_TYPE_VIDEO,
775  .filter_frame = filter_frame_hald,
776  .config_props = config_clut,
777  },
778  { NULL }
779 };
780 
781 static const AVFilterPad haldclut_outputs[] = {
782  {
783  .name = "default",
784  .type = AVMEDIA_TYPE_VIDEO,
785  .request_frame = request_frame,
786  .config_props = config_output,
787  },
788  { NULL }
789 };
790 
791 AVFilter ff_vf_haldclut = {
792  .name = "haldclut",
793  .description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
794  .priv_size = sizeof(LUT3DContext),
795  .init = haldclut_init,
796  .uninit = haldclut_uninit,
798  .inputs = haldclut_inputs,
799  .outputs = haldclut_outputs,
800  .priv_class = &haldclut_class,
802 };
803 #endif