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
29 #include "libavutil/opt.h"
30 #include "libavutil/file.h"
31 #include "libavutil/intreadwrite.h"
32 #include "libavutil/intfloat.h"
33 #include "libavutil/avassert.h"
34 #include "libavutil/pixdesc.h"
35 #include "libavutil/avstring.h"
37 #include "drawutils.h"
39 #include "framesync.h"
50 INTERPOLATE_TRILINEAR,
51 INTERPOLATE_TETRAHEDRAL,
61 /* 3D LUT don't often go up to level 32, but it is common to have a Hald CLUT
62 * of 512x512 (64x64x64) */
64 #define PRELUT_SIZE 65536
66 typedef struct Lut3DPreLut {
74 typedef struct LUT3DContext {
76 int interpolation; ///<interp_mode
80 avfilter_action_func *interp;
86 #if CONFIG_HALDCLUT_FILTER
87 uint8_t clut_rgba_map[4];
97 typedef struct ThreadData {
101 #define OFFSET(x) offsetof(LUT3DContext, x)
102 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
103 #define COMMON_OPTIONS \
104 { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, NB_INTERP_MODE-1, FLAGS, "interp_mode" }, \
105 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_NEAREST}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \
106 { "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" }, \
107 { "tetrahedral", "interpolate values using a tetrahedron", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TETRAHEDRAL}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \
108 { "pyramid", "interpolate values using a pyramid", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PYRAMID}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \
109 { "prism", "interpolate values using a prism", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PRISM}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \
112 #define EXPONENT_MASK 0x7F800000
113 #define MANTISSA_MASK 0x007FFFFF
114 #define SIGN_MASK 0x80000000
116 static inline float sanitizef(float f)
118 union av_intfloat32 t;
121 if ((t.i & EXPONENT_MASK) == EXPONENT_MASK) {
122 if ((t.i & MANTISSA_MASK) != 0) {
125 } else if (t.i & SIGN_MASK) {
136 static inline float lerpf(float v0, float v1, float f)
138 return v0 + (v1 - v0) * f;
141 static inline struct rgbvec lerp(const struct rgbvec *v0, const struct rgbvec *v1, float f)
144 lerpf(v0->r, v1->r, f), lerpf(v0->g, v1->g, f), lerpf(v0->b, v1->b, f)
149 #define NEAR(x) ((int)((x) + .5))
150 #define PREV(x) ((int)(x))
151 #define NEXT(x) (FFMIN((int)(x) + 1, lut3d->lutsize - 1))
154 * Get the nearest defined point
156 static inline struct rgbvec interp_nearest(const LUT3DContext *lut3d,
157 const struct rgbvec *s)
159 return lut3d->lut[NEAR(s->r) * lut3d->lutsize2 + NEAR(s->g) * lut3d->lutsize + NEAR(s->b)];
163 * Interpolate using the 8 vertices of a cube
164 * @see https://en.wikipedia.org/wiki/Trilinear_interpolation
166 static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d,
167 const struct rgbvec *s)
169 const int lutsize2 = lut3d->lutsize2;
170 const int lutsize = lut3d->lutsize;
171 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
172 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
173 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
174 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
175 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
176 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
177 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
178 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
179 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
180 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
181 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
182 const struct rgbvec c00 = lerp(&c000, &c100, d.r);
183 const struct rgbvec c10 = lerp(&c010, &c110, d.r);
184 const struct rgbvec c01 = lerp(&c001, &c101, d.r);
185 const struct rgbvec c11 = lerp(&c011, &c111, d.r);
186 const struct rgbvec c0 = lerp(&c00, &c10, d.g);
187 const struct rgbvec c1 = lerp(&c01, &c11, d.g);
188 const struct rgbvec c = lerp(&c0, &c1, d.b);
192 static inline struct rgbvec interp_pyramid(const LUT3DContext *lut3d,
193 const struct rgbvec *s)
195 const int lutsize2 = lut3d->lutsize2;
196 const int lutsize = lut3d->lutsize;
197 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
198 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
199 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
200 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
201 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
204 if (d.g > d.r && d.b > d.r) {
205 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
206 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
207 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
209 c.r = c000.r + (c111.r - c011.r) * d.r + (c010.r - c000.r) * d.g + (c001.r - c000.r) * d.b +
210 (c011.r - c001.r - c010.r + c000.r) * d.g * d.b;
211 c.g = c000.g + (c111.g - c011.g) * d.r + (c010.g - c000.g) * d.g + (c001.g - c000.g) * d.b +
212 (c011.g - c001.g - c010.g + c000.g) * d.g * d.b;
213 c.b = c000.b + (c111.b - c011.b) * d.r + (c010.b - c000.b) * d.g + (c001.b - c000.b) * d.b +
214 (c011.b - c001.b - c010.b + c000.b) * d.g * d.b;
215 } else if (d.r > d.g && d.b > d.g) {
216 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
217 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
218 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
220 c.r = c000.r + (c100.r - c000.r) * d.r + (c111.r - c101.r) * d.g + (c001.r - c000.r) * d.b +
221 (c101.r - c001.r - c100.r + c000.r) * d.r * d.b;
222 c.g = c000.g + (c100.g - c000.g) * d.r + (c111.g - c101.g) * d.g + (c001.g - c000.g) * d.b +
223 (c101.g - c001.g - c100.g + c000.g) * d.r * d.b;
224 c.b = c000.b + (c100.b - c000.b) * d.r + (c111.b - c101.b) * d.g + (c001.b - c000.b) * d.b +
225 (c101.b - c001.b - c100.b + c000.b) * d.r * d.b;
227 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
228 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
229 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
231 c.r = c000.r + (c100.r - c000.r) * d.r + (c010.r - c000.r) * d.g + (c111.r - c110.r) * d.b +
232 (c110.r - c100.r - c010.r + c000.r) * d.r * d.g;
233 c.g = c000.g + (c100.g - c000.g) * d.r + (c010.g - c000.g) * d.g + (c111.g - c110.g) * d.b +
234 (c110.g - c100.g - c010.g + c000.g) * d.r * d.g;
235 c.b = c000.b + (c100.b - c000.b) * d.r + (c010.b - c000.b) * d.g + (c111.b - c110.b) * d.b +
236 (c110.b - c100.b - c010.b + c000.b) * d.r * d.g;
242 static inline struct rgbvec interp_prism(const LUT3DContext *lut3d,
243 const struct rgbvec *s)
245 const int lutsize2 = lut3d->lutsize2;
246 const int lutsize = lut3d->lutsize;
247 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
248 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
249 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
250 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
251 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
252 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
253 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
257 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
258 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
260 c.r = c000.r + (c001.r - c000.r) * d.b + (c101.r - c001.r) * d.r + (c010.r - c000.r) * d.g +
261 (c000.r - c010.r - c001.r + c011.r) * d.b * d.g +
262 (c001.r - c011.r - c101.r + c111.r) * d.r * d.g;
263 c.g = c000.g + (c001.g - c000.g) * d.b + (c101.g - c001.g) * d.r + (c010.g - c000.g) * d.g +
264 (c000.g - c010.g - c001.g + c011.g) * d.b * d.g +
265 (c001.g - c011.g - c101.g + c111.g) * d.r * d.g;
266 c.b = c000.b + (c001.b - c000.b) * d.b + (c101.b - c001.b) * d.r + (c010.b - c000.b) * d.g +
267 (c000.b - c010.b - c001.b + c011.b) * d.b * d.g +
268 (c001.b - c011.b - c101.b + c111.b) * d.r * d.g;
270 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
271 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
273 c.r = c000.r + (c101.r - c100.r) * d.b + (c100.r - c000.r) * d.r + (c010.r - c000.r) * d.g +
274 (c100.r - c110.r - c101.r + c111.r) * d.b * d.g +
275 (c000.r - c010.r - c100.r + c110.r) * d.r * d.g;
276 c.g = c000.g + (c101.g - c100.g) * d.b + (c100.g - c000.g) * d.r + (c010.g - c000.g) * d.g +
277 (c100.g - c110.g - c101.g + c111.g) * d.b * d.g +
278 (c000.g - c010.g - c100.g + c110.g) * d.r * d.g;
279 c.b = c000.b + (c101.b - c100.b) * d.b + (c100.b - c000.b) * d.r + (c010.b - c000.b) * d.g +
280 (c100.b - c110.b - c101.b + c111.b) * d.b * d.g +
281 (c000.b - c010.b - c100.b + c110.b) * d.r * d.g;
288 * Tetrahedral interpolation. Based on code found in Truelight Software Library paper.
289 * @see http://www.filmlight.ltd.uk/pdf/whitepapers/FL-TL-TN-0057-SoftwareLib.pdf
291 static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d,
292 const struct rgbvec *s)
294 const int lutsize2 = lut3d->lutsize2;
295 const int lutsize = lut3d->lutsize;
296 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
297 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
298 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
299 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
300 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
304 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
305 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
306 c.r = (1-d.r) * c000.r + (d.r-d.g) * c100.r + (d.g-d.b) * c110.r + (d.b) * c111.r;
307 c.g = (1-d.r) * c000.g + (d.r-d.g) * c100.g + (d.g-d.b) * c110.g + (d.b) * c111.g;
308 c.b = (1-d.r) * c000.b + (d.r-d.g) * c100.b + (d.g-d.b) * c110.b + (d.b) * c111.b;
309 } else if (d.r > d.b) {
310 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
311 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
312 c.r = (1-d.r) * c000.r + (d.r-d.b) * c100.r + (d.b-d.g) * c101.r + (d.g) * c111.r;
313 c.g = (1-d.r) * c000.g + (d.r-d.b) * c100.g + (d.b-d.g) * c101.g + (d.g) * c111.g;
314 c.b = (1-d.r) * c000.b + (d.r-d.b) * c100.b + (d.b-d.g) * c101.b + (d.g) * c111.b;
316 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
317 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
318 c.r = (1-d.b) * c000.r + (d.b-d.r) * c001.r + (d.r-d.g) * c101.r + (d.g) * c111.r;
319 c.g = (1-d.b) * c000.g + (d.b-d.r) * c001.g + (d.r-d.g) * c101.g + (d.g) * c111.g;
320 c.b = (1-d.b) * c000.b + (d.b-d.r) * c001.b + (d.r-d.g) * c101.b + (d.g) * c111.b;
324 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
325 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
326 c.r = (1-d.b) * c000.r + (d.b-d.g) * c001.r + (d.g-d.r) * c011.r + (d.r) * c111.r;
327 c.g = (1-d.b) * c000.g + (d.b-d.g) * c001.g + (d.g-d.r) * c011.g + (d.r) * c111.g;
328 c.b = (1-d.b) * c000.b + (d.b-d.g) * c001.b + (d.g-d.r) * c011.b + (d.r) * c111.b;
329 } else if (d.b > d.r) {
330 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
331 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
332 c.r = (1-d.g) * c000.r + (d.g-d.b) * c010.r + (d.b-d.r) * c011.r + (d.r) * c111.r;
333 c.g = (1-d.g) * c000.g + (d.g-d.b) * c010.g + (d.b-d.r) * c011.g + (d.r) * c111.g;
334 c.b = (1-d.g) * c000.b + (d.g-d.b) * c010.b + (d.b-d.r) * c011.b + (d.r) * c111.b;
336 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
337 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
338 c.r = (1-d.g) * c000.r + (d.g-d.r) * c010.r + (d.r-d.b) * c110.r + (d.b) * c111.r;
339 c.g = (1-d.g) * c000.g + (d.g-d.r) * c010.g + (d.r-d.b) * c110.g + (d.b) * c111.g;
340 c.b = (1-d.g) * c000.b + (d.g-d.r) * c010.b + (d.r-d.b) * c110.b + (d.b) * c111.b;
346 static inline float prelut_interp_1d_linear(const Lut3DPreLut *prelut,
347 int idx, const float s)
349 const int lut_max = prelut->size - 1;
350 const float scaled = (s - prelut->min[idx]) * prelut->scale[idx];
351 const float x = av_clipf(scaled, 0.0f, lut_max);
352 const int prev = PREV(x);
353 const int next = FFMIN((int)(x) + 1, lut_max);
354 const float p = prelut->lut[idx][prev];
355 const float n = prelut->lut[idx][next];
356 const float d = x - (float)prev;
357 return lerpf(p, n, d);
360 static inline struct rgbvec apply_prelut(const Lut3DPreLut *prelut,
361 const struct rgbvec *s)
365 if (prelut->size <= 0)
368 c.r = prelut_interp_1d_linear(prelut, 0, s->r);
369 c.g = prelut_interp_1d_linear(prelut, 1, s->g);
370 c.b = prelut_interp_1d_linear(prelut, 2, s->b);
374 #define DEFINE_INTERP_FUNC_PLANAR(name, nbits, depth) \
375 static int interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
378 const LUT3DContext *lut3d = ctx->priv; \
379 const Lut3DPreLut *prelut = &lut3d->prelut; \
380 const ThreadData *td = arg; \
381 const AVFrame *in = td->in; \
382 const AVFrame *out = td->out; \
383 const int direct = out == in; \
384 const int slice_start = (in->height * jobnr ) / nb_jobs; \
385 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
386 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
387 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
388 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
389 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
390 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
391 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
392 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
393 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
394 const float lut_max = lut3d->lutsize - 1; \
395 const float scale_f = 1.0f / ((1<<depth) - 1); \
396 const float scale_r = lut3d->scale.r * lut_max; \
397 const float scale_g = lut3d->scale.g * lut_max; \
398 const float scale_b = lut3d->scale.b * lut_max; \
400 for (y = slice_start; y < slice_end; y++) { \
401 uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
402 uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
403 uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
404 uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
405 const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
406 const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
407 const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
408 const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
409 for (x = 0; x < in->width; x++) { \
410 const struct rgbvec rgb = {srcr[x] * scale_f, \
412 srcb[x] * scale_f}; \
413 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
414 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
415 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
416 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
417 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
418 dstr[x] = av_clip_uintp2(vec.r * (float)((1<<depth) - 1), depth); \
419 dstg[x] = av_clip_uintp2(vec.g * (float)((1<<depth) - 1), depth); \
420 dstb[x] = av_clip_uintp2(vec.b * (float)((1<<depth) - 1), depth); \
421 if (!direct && in->linesize[3]) \
424 grow += out->linesize[0]; \
425 brow += out->linesize[1]; \
426 rrow += out->linesize[2]; \
427 arow += out->linesize[3]; \
428 srcgrow += in->linesize[0]; \
429 srcbrow += in->linesize[1]; \
430 srcrrow += in->linesize[2]; \
431 srcarow += in->linesize[3]; \
436 DEFINE_INTERP_FUNC_PLANAR(nearest, 8, 8)
437 DEFINE_INTERP_FUNC_PLANAR(trilinear, 8, 8)
438 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 8, 8)
439 DEFINE_INTERP_FUNC_PLANAR(pyramid, 8, 8)
440 DEFINE_INTERP_FUNC_PLANAR(prism, 8, 8)
442 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 9)
443 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 9)
444 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 9)
445 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 9)
446 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 9)
448 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 10)
449 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 10)
450 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 10)
451 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 10)
452 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 10)
454 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 12)
455 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 12)
456 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 12)
457 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 12)
458 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 12)
460 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 14)
461 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 14)
462 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 14)
463 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 14)
464 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 14)
466 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 16)
467 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 16)
468 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 16)
469 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 16)
470 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 16)
472 #define DEFINE_INTERP_FUNC_PLANAR_FLOAT(name, depth) \
473 static int interp_##name##_pf##depth(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
476 const LUT3DContext *lut3d = ctx->priv; \
477 const Lut3DPreLut *prelut = &lut3d->prelut; \
478 const ThreadData *td = arg; \
479 const AVFrame *in = td->in; \
480 const AVFrame *out = td->out; \
481 const int direct = out == in; \
482 const int slice_start = (in->height * jobnr ) / nb_jobs; \
483 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
484 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
485 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
486 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
487 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
488 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
489 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
490 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
491 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
492 const float lut_max = lut3d->lutsize - 1; \
493 const float scale_r = lut3d->scale.r * lut_max; \
494 const float scale_g = lut3d->scale.g * lut_max; \
495 const float scale_b = lut3d->scale.b * lut_max; \
497 for (y = slice_start; y < slice_end; y++) { \
498 float *dstg = (float *)grow; \
499 float *dstb = (float *)brow; \
500 float *dstr = (float *)rrow; \
501 float *dsta = (float *)arow; \
502 const float *srcg = (const float *)srcgrow; \
503 const float *srcb = (const float *)srcbrow; \
504 const float *srcr = (const float *)srcrrow; \
505 const float *srca = (const float *)srcarow; \
506 for (x = 0; x < in->width; x++) { \
507 const struct rgbvec rgb = {sanitizef(srcr[x]), \
508 sanitizef(srcg[x]), \
509 sanitizef(srcb[x])}; \
510 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
511 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
512 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
513 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
514 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
518 if (!direct && in->linesize[3]) \
521 grow += out->linesize[0]; \
522 brow += out->linesize[1]; \
523 rrow += out->linesize[2]; \
524 arow += out->linesize[3]; \
525 srcgrow += in->linesize[0]; \
526 srcbrow += in->linesize[1]; \
527 srcrrow += in->linesize[2]; \
528 srcarow += in->linesize[3]; \
533 DEFINE_INTERP_FUNC_PLANAR_FLOAT(nearest, 32)
534 DEFINE_INTERP_FUNC_PLANAR_FLOAT(trilinear, 32)
535 DEFINE_INTERP_FUNC_PLANAR_FLOAT(tetrahedral, 32)
536 DEFINE_INTERP_FUNC_PLANAR_FLOAT(pyramid, 32)
537 DEFINE_INTERP_FUNC_PLANAR_FLOAT(prism, 32)
539 #define DEFINE_INTERP_FUNC(name, nbits) \
540 static int interp_##nbits##_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
543 const LUT3DContext *lut3d = ctx->priv; \
544 const Lut3DPreLut *prelut = &lut3d->prelut; \
545 const ThreadData *td = arg; \
546 const AVFrame *in = td->in; \
547 const AVFrame *out = td->out; \
548 const int direct = out == in; \
549 const int step = lut3d->step; \
550 const uint8_t r = lut3d->rgba_map[R]; \
551 const uint8_t g = lut3d->rgba_map[G]; \
552 const uint8_t b = lut3d->rgba_map[B]; \
553 const uint8_t a = lut3d->rgba_map[A]; \
554 const int slice_start = (in->height * jobnr ) / nb_jobs; \
555 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
556 uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
557 const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
558 const float lut_max = lut3d->lutsize - 1; \
559 const float scale_f = 1.0f / ((1<<nbits) - 1); \
560 const float scale_r = lut3d->scale.r * lut_max; \
561 const float scale_g = lut3d->scale.g * lut_max; \
562 const float scale_b = lut3d->scale.b * lut_max; \
564 for (y = slice_start; y < slice_end; y++) { \
565 uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
566 const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
567 for (x = 0; x < in->width * step; x += step) { \
568 const struct rgbvec rgb = {src[x + r] * scale_f, \
569 src[x + g] * scale_f, \
570 src[x + b] * scale_f}; \
571 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
572 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
573 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
574 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
575 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
576 dst[x + r] = av_clip_uint##nbits(vec.r * (float)((1<<nbits) - 1)); \
577 dst[x + g] = av_clip_uint##nbits(vec.g * (float)((1<<nbits) - 1)); \
578 dst[x + b] = av_clip_uint##nbits(vec.b * (float)((1<<nbits) - 1)); \
579 if (!direct && step == 4) \
580 dst[x + a] = src[x + a]; \
582 dstrow += out->linesize[0]; \
583 srcrow += in ->linesize[0]; \
588 DEFINE_INTERP_FUNC(nearest, 8)
589 DEFINE_INTERP_FUNC(trilinear, 8)
590 DEFINE_INTERP_FUNC(tetrahedral, 8)
591 DEFINE_INTERP_FUNC(pyramid, 8)
592 DEFINE_INTERP_FUNC(prism, 8)
594 DEFINE_INTERP_FUNC(nearest, 16)
595 DEFINE_INTERP_FUNC(trilinear, 16)
596 DEFINE_INTERP_FUNC(tetrahedral, 16)
597 DEFINE_INTERP_FUNC(pyramid, 16)
598 DEFINE_INTERP_FUNC(prism, 16)
600 #define MAX_LINE_SIZE 512
602 static int skip_line(const char *p)
604 while (*p && av_isspace(*p))
606 return !*p || *p == '#';
609 static char* fget_next_word(char* dst, int max, FILE* f)
616 /* skip until next non whitespace char */
617 while ((c = fgetc(f)) != EOF) {
626 /* get max bytes or up until next whitespace char */
627 for (; max > 0; max--) {
628 if ((c = fgetc(f)) == EOF)
643 #define NEXT_LINE(loop_cond) do { \
644 if (!fgets(line, sizeof(line), f)) { \
645 av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \
646 return AVERROR_INVALIDDATA; \
650 #define NEXT_LINE_OR_GOTO(loop_cond, label) do { \
651 if (!fgets(line, sizeof(line), f)) { \
652 av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \
653 ret = AVERROR_INVALIDDATA; \
658 static int allocate_3dlut(AVFilterContext *ctx, int lutsize, int prelut)
660 LUT3DContext *lut3d = ctx->priv;
662 if (lutsize < 2 || lutsize > MAX_LEVEL) {
663 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n");
664 return AVERROR(EINVAL);
667 av_freep(&lut3d->lut);
668 lut3d->lut = av_malloc_array(lutsize * lutsize * lutsize, sizeof(*lut3d->lut));
670 return AVERROR(ENOMEM);
673 lut3d->prelut.size = PRELUT_SIZE;
674 for (i = 0; i < 3; i++) {
675 av_freep(&lut3d->prelut.lut[i]);
676 lut3d->prelut.lut[i] = av_malloc_array(PRELUT_SIZE, sizeof(*lut3d->prelut.lut[0]));
677 if (!lut3d->prelut.lut[i])
678 return AVERROR(ENOMEM);
681 lut3d->prelut.size = 0;
682 for (i = 0; i < 3; i++) {
683 av_freep(&lut3d->prelut.lut[i]);
686 lut3d->lutsize = lutsize;
687 lut3d->lutsize2 = lutsize * lutsize;
691 /* Basically r g and b float values on each line, with a facultative 3DLUTSIZE
692 * directive; seems to be generated by Davinci */
693 static int parse_dat(AVFilterContext *ctx, FILE *f)
695 LUT3DContext *lut3d = ctx->priv;
696 char line[MAX_LINE_SIZE];
697 int ret, i, j, k, size, size2;
699 lut3d->lutsize = size = 33;
702 NEXT_LINE(skip_line(line));
703 if (!strncmp(line, "3DLUTSIZE ", 10)) {
704 size = strtol(line + 10, NULL, 0);
706 NEXT_LINE(skip_line(line));
709 ret = allocate_3dlut(ctx, size, 0);
713 for (k = 0; k < size; k++) {
714 for (j = 0; j < size; j++) {
715 for (i = 0; i < size; i++) {
716 struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
717 if (k != 0 || j != 0 || i != 0)
718 NEXT_LINE(skip_line(line));
719 if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
720 return AVERROR_INVALIDDATA;
728 static int parse_cube(AVFilterContext *ctx, FILE *f)
730 LUT3DContext *lut3d = ctx->priv;
731 char line[MAX_LINE_SIZE];
732 float min[3] = {0.0, 0.0, 0.0};
733 float max[3] = {1.0, 1.0, 1.0};
735 while (fgets(line, sizeof(line), f)) {
736 if (!strncmp(line, "LUT_3D_SIZE", 11)) {
738 const int size = strtol(line + 12, NULL, 0);
739 const int size2 = size * size;
741 ret = allocate_3dlut(ctx, size, 0);
745 for (k = 0; k < size; k++) {
746 for (j = 0; j < size; j++) {
747 for (i = 0; i < size; i++) {
748 struct rgbvec *vec = &lut3d->lut[i * size2 + j * size + k];
753 if (!strncmp(line, "DOMAIN_", 7)) {
755 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
756 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
758 return AVERROR_INVALIDDATA;
759 av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
760 av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
761 min[0], min[1], min[2], max[0], max[1], max[2]);
763 } else if (!strncmp(line, "TITLE", 5)) {
766 } while (skip_line(line));
767 if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
768 return AVERROR_INVALIDDATA;
776 lut3d->scale.r = av_clipf(1. / (max[0] - min[0]), 0.f, 1.f);
777 lut3d->scale.g = av_clipf(1. / (max[1] - min[1]), 0.f, 1.f);
778 lut3d->scale.b = av_clipf(1. / (max[2] - min[2]), 0.f, 1.f);
783 /* Assume 17x17x17 LUT with a 16-bit depth
784 * FIXME: it seems there are various 3dl formats */
785 static int parse_3dl(AVFilterContext *ctx, FILE *f)
787 char line[MAX_LINE_SIZE];
788 LUT3DContext *lut3d = ctx->priv;
791 const int size2 = 17 * 17;
792 const float scale = 16*16*16;
794 lut3d->lutsize = size;
796 ret = allocate_3dlut(ctx, size, 0);
800 NEXT_LINE(skip_line(line));
801 for (k = 0; k < size; k++) {
802 for (j = 0; j < size; j++) {
803 for (i = 0; i < size; i++) {
805 struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
807 NEXT_LINE(skip_line(line));
808 if (av_sscanf(line, "%d %d %d", &r, &g, &b) != 3)
809 return AVERROR_INVALIDDATA;
820 static int parse_m3d(AVFilterContext *ctx, FILE *f)
822 LUT3DContext *lut3d = ctx->priv;
824 int ret, i, j, k, size, size2, in = -1, out = -1;
825 char line[MAX_LINE_SIZE];
826 uint8_t rgb_map[3] = {0, 1, 2};
828 while (fgets(line, sizeof(line), f)) {
829 if (!strncmp(line, "in", 2)) in = strtol(line + 2, NULL, 0);
830 else if (!strncmp(line, "out", 3)) out = strtol(line + 3, NULL, 0);
831 else if (!strncmp(line, "values", 6)) {
832 const char *p = line + 6;
833 #define SET_COLOR(id) do { \
834 while (av_isspace(*p)) \
837 case 'r': rgb_map[id] = 0; break; \
838 case 'g': rgb_map[id] = 1; break; \
839 case 'b': rgb_map[id] = 2; break; \
841 while (*p && !av_isspace(*p)) \
851 if (in == -1 || out == -1) {
852 av_log(ctx, AV_LOG_ERROR, "in and out must be defined\n");
853 return AVERROR_INVALIDDATA;
855 if (in < 2 || out < 2 ||
856 in > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL ||
857 out > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL) {
858 av_log(ctx, AV_LOG_ERROR, "invalid in (%d) or out (%d)\n", in, out);
859 return AVERROR_INVALIDDATA;
861 for (size = 1; size*size*size < in; size++);
862 lut3d->lutsize = size;
865 ret = allocate_3dlut(ctx, size, 0);
869 scale = 1. / (out - 1);
871 for (k = 0; k < size; k++) {
872 for (j = 0; j < size; j++) {
873 for (i = 0; i < size; i++) {
874 struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
878 if (av_sscanf(line, "%f %f %f", val, val + 1, val + 2) != 3)
879 return AVERROR_INVALIDDATA;
880 vec->r = val[rgb_map[0]] * scale;
881 vec->g = val[rgb_map[1]] * scale;
882 vec->b = val[rgb_map[2]] * scale;
889 static int nearest_sample_index(float *data, float x, int low, int hi)
899 av_assert0(x >= data[low]);
900 av_assert0(x <= data[hi]);
901 av_assert0((hi-low) > 0);
906 mid = (low + hi) / 2;
917 #define NEXT_FLOAT_OR_GOTO(value, label) \
918 if (!fget_next_word(line, sizeof(line) ,f)) { \
919 ret = AVERROR_INVALIDDATA; \
922 if (av_sscanf(line, "%f", &value) != 1) { \
923 ret = AVERROR_INVALIDDATA; \
927 static int parse_cinespace(AVFilterContext *ctx, FILE *f)
929 LUT3DContext *lut3d = ctx->priv;
930 char line[MAX_LINE_SIZE];
931 float in_min[3] = {0.0, 0.0, 0.0};
932 float in_max[3] = {1.0, 1.0, 1.0};
933 float out_min[3] = {0.0, 0.0, 0.0};
934 float out_max[3] = {1.0, 1.0, 1.0};
935 int inside_metadata = 0, size, size2;
939 int prelut_sizes[3] = {0, 0, 0};
940 float *in_prelut[3] = {NULL, NULL, NULL};
941 float *out_prelut[3] = {NULL, NULL, NULL};
943 NEXT_LINE_OR_GOTO(skip_line(line), end);
944 if (strncmp(line, "CSPLUTV100", 10)) {
945 av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n");
946 ret = AVERROR(EINVAL);
950 NEXT_LINE_OR_GOTO(skip_line(line), end);
951 if (strncmp(line, "3D", 2)) {
952 av_log(ctx, AV_LOG_ERROR, "Not 3D LUT format\n");
953 ret = AVERROR(EINVAL);
958 NEXT_LINE_OR_GOTO(skip_line(line), end);
960 if (!strncmp(line, "BEGIN METADATA", 14)) {
964 if (!strncmp(line, "END METADATA", 12)) {
968 if (inside_metadata == 0) {
969 int size_r, size_g, size_b;
971 for (int i = 0; i < 3; i++) {
972 int npoints = strtol(line, NULL, 0);
977 if (npoints > PRELUT_SIZE) {
978 av_log(ctx, AV_LOG_ERROR, "Prelut size too large.\n");
979 ret = AVERROR_INVALIDDATA;
983 if (in_prelut[i] || out_prelut[i]) {
984 av_log(ctx, AV_LOG_ERROR, "Invalid file has multiple preluts.\n");
985 ret = AVERROR_INVALIDDATA;
989 in_prelut[i] = (float*)av_malloc(npoints * sizeof(float));
990 out_prelut[i] = (float*)av_malloc(npoints * sizeof(float));
991 if (!in_prelut[i] || !out_prelut[i]) {
992 ret = AVERROR(ENOMEM);
996 prelut_sizes[i] = npoints;
998 in_max[i] = -FLT_MAX;
999 out_min[i] = FLT_MAX;
1000 out_max[i] = -FLT_MAX;
1002 for (int j = 0; j < npoints; j++) {
1003 NEXT_FLOAT_OR_GOTO(v, end)
1004 in_min[i] = FFMIN(in_min[i], v);
1005 in_max[i] = FFMAX(in_max[i], v);
1006 in_prelut[i][j] = v;
1007 if (j > 0 && v < last) {
1008 av_log(ctx, AV_LOG_ERROR, "Invalid file, non increasing prelut.\n");
1009 ret = AVERROR(ENOMEM);
1015 for (int j = 0; j < npoints; j++) {
1016 NEXT_FLOAT_OR_GOTO(v, end)
1017 out_min[i] = FFMIN(out_min[i], v);
1018 out_max[i] = FFMAX(out_max[i], v);
1019 out_prelut[i][j] = v;
1022 } else if (npoints == 2) {
1023 NEXT_LINE_OR_GOTO(skip_line(line), end);
1024 if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2) {
1025 ret = AVERROR_INVALIDDATA;
1028 NEXT_LINE_OR_GOTO(skip_line(line), end);
1029 if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2) {
1030 ret = AVERROR_INVALIDDATA;
1035 av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n");
1036 ret = AVERROR_PATCHWELCOME;
1040 NEXT_LINE_OR_GOTO(skip_line(line), end);
1043 if (av_sscanf(line, "%d %d %d", &size_r, &size_g, &size_b) != 3) {
1044 ret = AVERROR(EINVAL);
1047 if (size_r != size_g || size_r != size_b) {
1048 av_log(ctx, AV_LOG_ERROR, "Unsupported size combination: %dx%dx%d.\n", size_r, size_g, size_b);
1049 ret = AVERROR_PATCHWELCOME;
1054 size2 = size * size;
1056 if (prelut_sizes[0] && prelut_sizes[1] && prelut_sizes[2])
1059 ret = allocate_3dlut(ctx, size, prelut);
1063 for (int k = 0; k < size; k++) {
1064 for (int j = 0; j < size; j++) {
1065 for (int i = 0; i < size; i++) {
1066 struct rgbvec *vec = &lut3d->lut[i * size2 + j * size + k];
1068 NEXT_LINE_OR_GOTO(skip_line(line), end);
1069 if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) {
1070 ret = AVERROR_INVALIDDATA;
1074 vec->r *= out_max[0] - out_min[0];
1075 vec->g *= out_max[1] - out_min[1];
1076 vec->b *= out_max[2] - out_min[2];
1086 for (int c = 0; c < 3; c++) {
1088 lut3d->prelut.min[c] = in_min[c];
1089 lut3d->prelut.max[c] = in_max[c];
1090 lut3d->prelut.scale[c] = (1.0f / (float)(in_max[c] - in_min[c])) * (lut3d->prelut.size - 1);
1092 for (int i = 0; i < lut3d->prelut.size; ++i) {
1093 float mix = (float) i / (float)(lut3d->prelut.size - 1);
1094 float x = lerpf(in_min[c], in_max[c], mix), a, b;
1096 int idx = nearest_sample_index(in_prelut[c], x, 0, prelut_sizes[c]-1);
1097 av_assert0(idx + 1 < prelut_sizes[c]);
1099 a = out_prelut[c][idx + 0];
1100 b = out_prelut[c][idx + 1];
1101 mix = x - in_prelut[c][idx];
1103 lut3d->prelut.lut[c][i] = sanitizef(lerpf(a, b, mix));
1106 lut3d->scale.r = 1.00f;
1107 lut3d->scale.g = 1.00f;
1108 lut3d->scale.b = 1.00f;
1111 lut3d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f);
1112 lut3d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f);
1113 lut3d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f);
1117 for (int c = 0; c < 3; c++) {
1118 av_freep(&in_prelut[c]);
1119 av_freep(&out_prelut[c]);
1124 static int set_identity_matrix(AVFilterContext *ctx, int size)
1126 LUT3DContext *lut3d = ctx->priv;
1128 const int size2 = size * size;
1129 const float c = 1. / (size - 1);
1131 ret = allocate_3dlut(ctx, size, 0);
1135 for (k = 0; k < size; k++) {
1136 for (j = 0; j < size; j++) {
1137 for (i = 0; i < size; i++) {
1138 struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
1149 static int query_formats(AVFilterContext *ctx)
1151 static const enum AVPixelFormat pix_fmts[] = {
1152 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
1153 AV_PIX_FMT_RGBA, AV_PIX_FMT_BGRA,
1154 AV_PIX_FMT_ARGB, AV_PIX_FMT_ABGR,
1155 AV_PIX_FMT_0RGB, AV_PIX_FMT_0BGR,
1156 AV_PIX_FMT_RGB0, AV_PIX_FMT_BGR0,
1157 AV_PIX_FMT_RGB48, AV_PIX_FMT_BGR48,
1158 AV_PIX_FMT_RGBA64, AV_PIX_FMT_BGRA64,
1159 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP,
1161 AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRAP10,
1162 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRAP12,
1164 AV_PIX_FMT_GBRP16, AV_PIX_FMT_GBRAP16,
1165 AV_PIX_FMT_GBRPF32, AV_PIX_FMT_GBRAPF32,
1168 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
1170 return AVERROR(ENOMEM);
1171 return ff_set_common_formats(ctx, fmts_list);
1174 static int config_input(AVFilterLink *inlink)
1176 int depth, is16bit, isfloat, planar;
1177 LUT3DContext *lut3d = inlink->dst->priv;
1178 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1180 depth = desc->comp[0].depth;
1181 is16bit = desc->comp[0].depth > 8;
1182 planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR;
1183 isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
1184 ff_fill_rgba_map(lut3d->rgba_map, inlink->format);
1185 lut3d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
1187 #define SET_FUNC(name) do { \
1188 if (planar && !isfloat) { \
1190 case 8: lut3d->interp = interp_8_##name##_p8; break; \
1191 case 9: lut3d->interp = interp_16_##name##_p9; break; \
1192 case 10: lut3d->interp = interp_16_##name##_p10; break; \
1193 case 12: lut3d->interp = interp_16_##name##_p12; break; \
1194 case 14: lut3d->interp = interp_16_##name##_p14; break; \
1195 case 16: lut3d->interp = interp_16_##name##_p16; break; \
1197 } else if (isfloat) { lut3d->interp = interp_##name##_pf32; \
1198 } else if (is16bit) { lut3d->interp = interp_16_##name; \
1199 } else { lut3d->interp = interp_8_##name; } \
1202 switch (lut3d->interpolation) {
1203 case INTERPOLATE_NEAREST: SET_FUNC(nearest); break;
1204 case INTERPOLATE_TRILINEAR: SET_FUNC(trilinear); break;
1205 case INTERPOLATE_TETRAHEDRAL: SET_FUNC(tetrahedral); break;
1206 case INTERPOLATE_PYRAMID: SET_FUNC(pyramid); break;
1207 case INTERPOLATE_PRISM: SET_FUNC(prism); break;
1215 static AVFrame *apply_lut(AVFilterLink *inlink, AVFrame *in)
1217 AVFilterContext *ctx = inlink->dst;
1218 LUT3DContext *lut3d = ctx->priv;
1219 AVFilterLink *outlink = inlink->dst->outputs[0];
1223 if (av_frame_is_writable(in)) {
1226 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
1231 av_frame_copy_props(out, in);
1236 ctx->internal->execute(ctx, lut3d->interp, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
1244 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
1246 AVFilterLink *outlink = inlink->dst->outputs[0];
1247 AVFrame *out = apply_lut(inlink, in);
1249 return AVERROR(ENOMEM);
1250 return ff_filter_frame(outlink, out);
1253 #if CONFIG_LUT3D_FILTER
1254 static const AVOption lut3d_options[] = {
1255 { "file", "set 3D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
1259 AVFILTER_DEFINE_CLASS(lut3d);
1261 static av_cold int lut3d_init(AVFilterContext *ctx)
1266 LUT3DContext *lut3d = ctx->priv;
1268 lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1271 return set_identity_matrix(ctx, 32);
1274 f = av_fopen_utf8(lut3d->file, "r");
1276 ret = AVERROR(errno);
1277 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret));
1281 ext = strrchr(lut3d->file, '.');
1283 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
1284 ret = AVERROR_INVALIDDATA;
1289 if (!av_strcasecmp(ext, "dat")) {
1290 ret = parse_dat(ctx, f);
1291 } else if (!av_strcasecmp(ext, "3dl")) {
1292 ret = parse_3dl(ctx, f);
1293 } else if (!av_strcasecmp(ext, "cube")) {
1294 ret = parse_cube(ctx, f);
1295 } else if (!av_strcasecmp(ext, "m3d")) {
1296 ret = parse_m3d(ctx, f);
1297 } else if (!av_strcasecmp(ext, "csp")) {
1298 ret = parse_cinespace(ctx, f);
1300 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
1301 ret = AVERROR(EINVAL);
1304 if (!ret && !lut3d->lutsize) {
1305 av_log(ctx, AV_LOG_ERROR, "3D LUT is empty\n");
1306 ret = AVERROR_INVALIDDATA;
1314 static av_cold void lut3d_uninit(AVFilterContext *ctx)
1316 LUT3DContext *lut3d = ctx->priv;
1318 av_freep(&lut3d->lut);
1320 for (i = 0; i < 3; i++) {
1321 av_freep(&lut3d->prelut.lut[i]);
1325 static const AVFilterPad lut3d_inputs[] = {
1328 .type = AVMEDIA_TYPE_VIDEO,
1329 .filter_frame = filter_frame,
1330 .config_props = config_input,
1335 static const AVFilterPad lut3d_outputs[] = {
1338 .type = AVMEDIA_TYPE_VIDEO,
1343 AVFilter ff_vf_lut3d = {
1345 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
1346 .priv_size = sizeof(LUT3DContext),
1348 .uninit = lut3d_uninit,
1349 .query_formats = query_formats,
1350 .inputs = lut3d_inputs,
1351 .outputs = lut3d_outputs,
1352 .priv_class = &lut3d_class,
1353 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
1357 #if CONFIG_HALDCLUT_FILTER
1359 static void update_clut_packed(LUT3DContext *lut3d, const AVFrame *frame)
1361 const uint8_t *data = frame->data[0];
1362 const int linesize = frame->linesize[0];
1363 const int w = lut3d->clut_width;
1364 const int step = lut3d->clut_step;
1365 const uint8_t *rgba_map = lut3d->clut_rgba_map;
1366 const int level = lut3d->lutsize;
1367 const int level2 = lut3d->lutsize2;
1369 #define LOAD_CLUT(nbits) do { \
1370 int i, j, k, x = 0, y = 0; \
1372 for (k = 0; k < level; k++) { \
1373 for (j = 0; j < level; j++) { \
1374 for (i = 0; i < level; i++) { \
1375 const uint##nbits##_t *src = (const uint##nbits##_t *) \
1376 (data + y*linesize + x*step); \
1377 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1378 vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \
1379 vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \
1380 vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \
1390 switch (lut3d->clut_bits) {
1391 case 8: LOAD_CLUT(8); break;
1392 case 16: LOAD_CLUT(16); break;
1396 static void update_clut_planar(LUT3DContext *lut3d, const AVFrame *frame)
1398 const uint8_t *datag = frame->data[0];
1399 const uint8_t *datab = frame->data[1];
1400 const uint8_t *datar = frame->data[2];
1401 const int glinesize = frame->linesize[0];
1402 const int blinesize = frame->linesize[1];
1403 const int rlinesize = frame->linesize[2];
1404 const int w = lut3d->clut_width;
1405 const int level = lut3d->lutsize;
1406 const int level2 = lut3d->lutsize2;
1408 #define LOAD_CLUT_PLANAR(nbits, depth) do { \
1409 int i, j, k, x = 0, y = 0; \
1411 for (k = 0; k < level; k++) { \
1412 for (j = 0; j < level; j++) { \
1413 for (i = 0; i < level; i++) { \
1414 const uint##nbits##_t *gsrc = (const uint##nbits##_t *) \
1415 (datag + y*glinesize); \
1416 const uint##nbits##_t *bsrc = (const uint##nbits##_t *) \
1417 (datab + y*blinesize); \
1418 const uint##nbits##_t *rsrc = (const uint##nbits##_t *) \
1419 (datar + y*rlinesize); \
1420 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1421 vec->r = gsrc[x] / (float)((1<<(depth)) - 1); \
1422 vec->g = bsrc[x] / (float)((1<<(depth)) - 1); \
1423 vec->b = rsrc[x] / (float)((1<<(depth)) - 1); \
1433 switch (lut3d->clut_bits) {
1434 case 8: LOAD_CLUT_PLANAR(8, 8); break;
1435 case 9: LOAD_CLUT_PLANAR(16, 9); break;
1436 case 10: LOAD_CLUT_PLANAR(16, 10); break;
1437 case 12: LOAD_CLUT_PLANAR(16, 12); break;
1438 case 14: LOAD_CLUT_PLANAR(16, 14); break;
1439 case 16: LOAD_CLUT_PLANAR(16, 16); break;
1443 static void update_clut_float(LUT3DContext *lut3d, const AVFrame *frame)
1445 const uint8_t *datag = frame->data[0];
1446 const uint8_t *datab = frame->data[1];
1447 const uint8_t *datar = frame->data[2];
1448 const int glinesize = frame->linesize[0];
1449 const int blinesize = frame->linesize[1];
1450 const int rlinesize = frame->linesize[2];
1451 const int w = lut3d->clut_width;
1452 const int level = lut3d->lutsize;
1453 const int level2 = lut3d->lutsize2;
1455 int i, j, k, x = 0, y = 0;
1457 for (k = 0; k < level; k++) {
1458 for (j = 0; j < level; j++) {
1459 for (i = 0; i < level; i++) {
1460 const float *gsrc = (const float *)(datag + y*glinesize);
1461 const float *bsrc = (const float *)(datab + y*blinesize);
1462 const float *rsrc = (const float *)(datar + y*rlinesize);
1463 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k];
1476 static int config_output(AVFilterLink *outlink)
1478 AVFilterContext *ctx = outlink->src;
1479 LUT3DContext *lut3d = ctx->priv;
1482 ret = ff_framesync_init_dualinput(&lut3d->fs, ctx);
1485 outlink->w = ctx->inputs[0]->w;
1486 outlink->h = ctx->inputs[0]->h;
1487 outlink->time_base = ctx->inputs[0]->time_base;
1488 if ((ret = ff_framesync_configure(&lut3d->fs)) < 0)
1493 static int activate(AVFilterContext *ctx)
1495 LUT3DContext *s = ctx->priv;
1496 return ff_framesync_activate(&s->fs);
1499 static int config_clut(AVFilterLink *inlink)
1501 int size, level, w, h;
1502 AVFilterContext *ctx = inlink->dst;
1503 LUT3DContext *lut3d = ctx->priv;
1504 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1508 lut3d->clut_bits = desc->comp[0].depth;
1509 lut3d->clut_planar = av_pix_fmt_count_planes(inlink->format) > 1;
1510 lut3d->clut_float = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
1512 lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3;
1513 ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format);
1515 if (inlink->w > inlink->h)
1516 av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the "
1517 "Hald CLUT will be ignored\n", inlink->w - inlink->h);
1518 else if (inlink->w < inlink->h)
1519 av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the "
1520 "Hald CLUT will be ignored\n", inlink->h - inlink->w);
1521 lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h);
1523 for (level = 1; level*level*level < w; level++);
1524 size = level*level*level;
1526 av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n");
1527 return AVERROR_INVALIDDATA;
1529 av_assert0(w == h && w == size);
1531 if (level > MAX_LEVEL) {
1532 const int max_clut_level = sqrt(MAX_LEVEL);
1533 const int max_clut_size = max_clut_level*max_clut_level*max_clut_level;
1534 av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT "
1535 "(maximum level is %d, or %dx%d CLUT)\n",
1536 max_clut_level, max_clut_size, max_clut_size);
1537 return AVERROR(EINVAL);
1540 return allocate_3dlut(ctx, level, 0);
1543 static int update_apply_clut(FFFrameSync *fs)
1545 AVFilterContext *ctx = fs->parent;
1546 LUT3DContext *lut3d = ctx->priv;
1547 AVFilterLink *inlink = ctx->inputs[0];
1548 AVFrame *master, *second, *out;
1551 ret = ff_framesync_dualinput_get(fs, &master, &second);
1555 return ff_filter_frame(ctx->outputs[0], master);
1556 if (lut3d->clut_float)
1557 update_clut_float(ctx->priv, second);
1558 else if (lut3d->clut_planar)
1559 update_clut_planar(ctx->priv, second);
1561 update_clut_packed(ctx->priv, second);
1562 out = apply_lut(inlink, master);
1563 return ff_filter_frame(ctx->outputs[0], out);
1566 static av_cold int haldclut_init(AVFilterContext *ctx)
1568 LUT3DContext *lut3d = ctx->priv;
1569 lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1570 lut3d->fs.on_event = update_apply_clut;
1574 static av_cold void haldclut_uninit(AVFilterContext *ctx)
1576 LUT3DContext *lut3d = ctx->priv;
1577 ff_framesync_uninit(&lut3d->fs);
1578 av_freep(&lut3d->lut);
1581 static const AVOption haldclut_options[] = {
1585 FRAMESYNC_DEFINE_CLASS(haldclut, LUT3DContext, fs);
1587 static const AVFilterPad haldclut_inputs[] = {
1590 .type = AVMEDIA_TYPE_VIDEO,
1591 .config_props = config_input,
1594 .type = AVMEDIA_TYPE_VIDEO,
1595 .config_props = config_clut,
1600 static const AVFilterPad haldclut_outputs[] = {
1603 .type = AVMEDIA_TYPE_VIDEO,
1604 .config_props = config_output,
1609 AVFilter ff_vf_haldclut = {
1611 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
1612 .priv_size = sizeof(LUT3DContext),
1613 .preinit = haldclut_framesync_preinit,
1614 .init = haldclut_init,
1615 .uninit = haldclut_uninit,
1616 .query_formats = query_formats,
1617 .activate = activate,
1618 .inputs = haldclut_inputs,
1619 .outputs = haldclut_outputs,
1620 .priv_class = &haldclut_class,
1621 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
1625 #if CONFIG_LUT1D_FILTER
1627 enum interp_1d_mode {
1628 INTERPOLATE_1D_NEAREST,
1629 INTERPOLATE_1D_LINEAR,
1630 INTERPOLATE_1D_CUBIC,
1631 INTERPOLATE_1D_COSINE,
1632 INTERPOLATE_1D_SPLINE,
1636 #define MAX_1D_LEVEL 65536
1638 typedef struct LUT1DContext {
1639 const AVClass *class;
1641 int interpolation; ///<interp_1d_mode
1642 struct rgbvec scale;
1643 uint8_t rgba_map[4];
1645 float lut[3][MAX_1D_LEVEL];
1647 avfilter_action_func *interp;
1651 #define OFFSET(x) offsetof(LUT1DContext, x)
1653 static void set_identity_matrix_1d(LUT1DContext *lut1d, int size)
1655 const float c = 1. / (size - 1);
1658 lut1d->lutsize = size;
1659 for (i = 0; i < size; i++) {
1660 lut1d->lut[0][i] = i * c;
1661 lut1d->lut[1][i] = i * c;
1662 lut1d->lut[2][i] = i * c;
1666 static int parse_cinespace_1d(AVFilterContext *ctx, FILE *f)
1668 LUT1DContext *lut1d = ctx->priv;
1669 char line[MAX_LINE_SIZE];
1670 float in_min[3] = {0.0, 0.0, 0.0};
1671 float in_max[3] = {1.0, 1.0, 1.0};
1672 float out_min[3] = {0.0, 0.0, 0.0};
1673 float out_max[3] = {1.0, 1.0, 1.0};
1674 int inside_metadata = 0, size;
1676 NEXT_LINE(skip_line(line));
1677 if (strncmp(line, "CSPLUTV100", 10)) {
1678 av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n");
1679 return AVERROR(EINVAL);
1682 NEXT_LINE(skip_line(line));
1683 if (strncmp(line, "1D", 2)) {
1684 av_log(ctx, AV_LOG_ERROR, "Not 1D LUT format\n");
1685 return AVERROR(EINVAL);
1689 NEXT_LINE(skip_line(line));
1691 if (!strncmp(line, "BEGIN METADATA", 14)) {
1692 inside_metadata = 1;
1695 if (!strncmp(line, "END METADATA", 12)) {
1696 inside_metadata = 0;
1699 if (inside_metadata == 0) {
1700 for (int i = 0; i < 3; i++) {
1701 int npoints = strtol(line, NULL, 0);
1704 av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n");
1705 return AVERROR_PATCHWELCOME;
1708 NEXT_LINE(skip_line(line));
1709 if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2)
1710 return AVERROR_INVALIDDATA;
1711 NEXT_LINE(skip_line(line));
1712 if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2)
1713 return AVERROR_INVALIDDATA;
1714 NEXT_LINE(skip_line(line));
1717 size = strtol(line, NULL, 0);
1719 if (size < 2 || size > MAX_1D_LEVEL) {
1720 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1721 return AVERROR(EINVAL);
1724 lut1d->lutsize = size;
1726 for (int i = 0; i < size; i++) {
1727 NEXT_LINE(skip_line(line));
1728 if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1729 return AVERROR_INVALIDDATA;
1730 lut1d->lut[0][i] *= out_max[0] - out_min[0];
1731 lut1d->lut[1][i] *= out_max[1] - out_min[1];
1732 lut1d->lut[2][i] *= out_max[2] - out_min[2];
1739 lut1d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f);
1740 lut1d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f);
1741 lut1d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f);
1746 static int parse_cube_1d(AVFilterContext *ctx, FILE *f)
1748 LUT1DContext *lut1d = ctx->priv;
1749 char line[MAX_LINE_SIZE];
1750 float min[3] = {0.0, 0.0, 0.0};
1751 float max[3] = {1.0, 1.0, 1.0};
1753 while (fgets(line, sizeof(line), f)) {
1754 if (!strncmp(line, "LUT_1D_SIZE", 11)) {
1755 const int size = strtol(line + 12, NULL, 0);
1758 if (size < 2 || size > MAX_1D_LEVEL) {
1759 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1760 return AVERROR(EINVAL);
1762 lut1d->lutsize = size;
1763 for (i = 0; i < size; i++) {
1767 if (!strncmp(line, "DOMAIN_", 7)) {
1769 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
1770 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
1772 return AVERROR_INVALIDDATA;
1773 av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
1774 av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
1775 min[0], min[1], min[2], max[0], max[1], max[2]);
1777 } else if (!strncmp(line, "LUT_1D_INPUT_RANGE ", 19)) {
1778 av_sscanf(line + 19, "%f %f", min, max);
1779 min[1] = min[2] = min[0];
1780 max[1] = max[2] = max[0];
1782 } else if (!strncmp(line, "TITLE", 5)) {
1785 } while (skip_line(line));
1786 if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1787 return AVERROR_INVALIDDATA;
1793 lut1d->scale.r = av_clipf(1. / (max[0] - min[0]), 0.f, 1.f);
1794 lut1d->scale.g = av_clipf(1. / (max[1] - min[1]), 0.f, 1.f);
1795 lut1d->scale.b = av_clipf(1. / (max[2] - min[2]), 0.f, 1.f);
1800 static const AVOption lut1d_options[] = {
1801 { "file", "set 1D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
1802 { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_1D_LINEAR}, 0, NB_INTERP_1D_MODE-1, FLAGS, "interp_mode" },
1803 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_NEAREST}, INT_MIN, INT_MAX, FLAGS, "interp_mode" },
1804 { "linear", "use values from the linear interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_LINEAR}, INT_MIN, INT_MAX, FLAGS, "interp_mode" },
1805 { "cosine", "use values from the cosine interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_COSINE}, INT_MIN, INT_MAX, FLAGS, "interp_mode" },
1806 { "cubic", "use values from the cubic interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_CUBIC}, INT_MIN, INT_MAX, FLAGS, "interp_mode" },
1807 { "spline", "use values from the spline interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_SPLINE}, INT_MIN, INT_MAX, FLAGS, "interp_mode" },
1811 AVFILTER_DEFINE_CLASS(lut1d);
1813 static inline float interp_1d_nearest(const LUT1DContext *lut1d,
1814 int idx, const float s)
1816 return lut1d->lut[idx][NEAR(s)];
1819 #define NEXT1D(x) (FFMIN((int)(x) + 1, lut1d->lutsize - 1))
1821 static inline float interp_1d_linear(const LUT1DContext *lut1d,
1822 int idx, const float s)
1824 const int prev = PREV(s);
1825 const int next = NEXT1D(s);
1826 const float d = s - prev;
1827 const float p = lut1d->lut[idx][prev];
1828 const float n = lut1d->lut[idx][next];
1830 return lerpf(p, n, d);
1833 static inline float interp_1d_cosine(const LUT1DContext *lut1d,
1834 int idx, const float s)
1836 const int prev = PREV(s);
1837 const int next = NEXT1D(s);
1838 const float d = s - prev;
1839 const float p = lut1d->lut[idx][prev];
1840 const float n = lut1d->lut[idx][next];
1841 const float m = (1.f - cosf(d * M_PI)) * .5f;
1843 return lerpf(p, n, m);
1846 static inline float interp_1d_cubic(const LUT1DContext *lut1d,
1847 int idx, const float s)
1849 const int prev = PREV(s);
1850 const int next = NEXT1D(s);
1851 const float mu = s - prev;
1852 float a0, a1, a2, a3, mu2;
1854 float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1855 float y1 = lut1d->lut[idx][prev];
1856 float y2 = lut1d->lut[idx][next];
1857 float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1861 a0 = y3 - y2 - y0 + y1;
1866 return a0 * mu * mu2 + a1 * mu2 + a2 * mu + a3;
1869 static inline float interp_1d_spline(const LUT1DContext *lut1d,
1870 int idx, const float s)
1872 const int prev = PREV(s);
1873 const int next = NEXT1D(s);
1874 const float x = s - prev;
1875 float c0, c1, c2, c3;
1877 float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1878 float y1 = lut1d->lut[idx][prev];
1879 float y2 = lut1d->lut[idx][next];
1880 float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1883 c1 = .5f * (y2 - y0);
1884 c2 = y0 - 2.5f * y1 + 2.f * y2 - .5f * y3;
1885 c3 = .5f * (y3 - y0) + 1.5f * (y1 - y2);
1887 return ((c3 * x + c2) * x + c1) * x + c0;
1890 #define DEFINE_INTERP_FUNC_PLANAR_1D(name, nbits, depth) \
1891 static int interp_1d_##nbits##_##name##_p##depth(AVFilterContext *ctx, \
1892 void *arg, int jobnr, \
1896 const LUT1DContext *lut1d = ctx->priv; \
1897 const ThreadData *td = arg; \
1898 const AVFrame *in = td->in; \
1899 const AVFrame *out = td->out; \
1900 const int direct = out == in; \
1901 const int slice_start = (in->height * jobnr ) / nb_jobs; \
1902 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
1903 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
1904 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
1905 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
1906 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
1907 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
1908 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
1909 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
1910 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
1911 const float factor = (1 << depth) - 1; \
1912 const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \
1913 const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \
1914 const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \
1916 for (y = slice_start; y < slice_end; y++) { \
1917 uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
1918 uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
1919 uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
1920 uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
1921 const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
1922 const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
1923 const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
1924 const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
1925 for (x = 0; x < in->width; x++) { \
1926 float r = srcr[x] * scale_r; \
1927 float g = srcg[x] * scale_g; \
1928 float b = srcb[x] * scale_b; \
1929 r = interp_1d_##name(lut1d, 0, r); \
1930 g = interp_1d_##name(lut1d, 1, g); \
1931 b = interp_1d_##name(lut1d, 2, b); \
1932 dstr[x] = av_clip_uintp2(r * factor, depth); \
1933 dstg[x] = av_clip_uintp2(g * factor, depth); \
1934 dstb[x] = av_clip_uintp2(b * factor, depth); \
1935 if (!direct && in->linesize[3]) \
1936 dsta[x] = srca[x]; \
1938 grow += out->linesize[0]; \
1939 brow += out->linesize[1]; \
1940 rrow += out->linesize[2]; \
1941 arow += out->linesize[3]; \
1942 srcgrow += in->linesize[0]; \
1943 srcbrow += in->linesize[1]; \
1944 srcrrow += in->linesize[2]; \
1945 srcarow += in->linesize[3]; \
1950 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 8, 8)
1951 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 8, 8)
1952 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 8, 8)
1953 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 8, 8)
1954 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 8, 8)
1956 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 9)
1957 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 9)
1958 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 9)
1959 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 9)
1960 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 9)
1962 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 10)
1963 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 10)
1964 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 10)
1965 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 10)
1966 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 10)
1968 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 12)
1969 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 12)
1970 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 12)
1971 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 12)
1972 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 12)
1974 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 14)
1975 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 14)
1976 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 14)
1977 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 14)
1978 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 14)
1980 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 16)
1981 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 16)
1982 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 16)
1983 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 16)
1984 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 16)
1986 #define DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(name, depth) \
1987 static int interp_1d_##name##_pf##depth(AVFilterContext *ctx, \
1988 void *arg, int jobnr, \
1992 const LUT1DContext *lut1d = ctx->priv; \
1993 const ThreadData *td = arg; \
1994 const AVFrame *in = td->in; \
1995 const AVFrame *out = td->out; \
1996 const int direct = out == in; \
1997 const int slice_start = (in->height * jobnr ) / nb_jobs; \
1998 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
1999 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
2000 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
2001 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
2002 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
2003 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
2004 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
2005 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
2006 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
2007 const float lutsize = lut1d->lutsize - 1; \
2008 const float scale_r = lut1d->scale.r * lutsize; \
2009 const float scale_g = lut1d->scale.g * lutsize; \
2010 const float scale_b = lut1d->scale.b * lutsize; \
2012 for (y = slice_start; y < slice_end; y++) { \
2013 float *dstg = (float *)grow; \
2014 float *dstb = (float *)brow; \
2015 float *dstr = (float *)rrow; \
2016 float *dsta = (float *)arow; \
2017 const float *srcg = (const float *)srcgrow; \
2018 const float *srcb = (const float *)srcbrow; \
2019 const float *srcr = (const float *)srcrrow; \
2020 const float *srca = (const float *)srcarow; \
2021 for (x = 0; x < in->width; x++) { \
2022 float r = av_clipf(sanitizef(srcr[x]) * scale_r, 0.0f, lutsize); \
2023 float g = av_clipf(sanitizef(srcg[x]) * scale_g, 0.0f, lutsize); \
2024 float b = av_clipf(sanitizef(srcb[x]) * scale_b, 0.0f, lutsize); \
2025 r = interp_1d_##name(lut1d, 0, r); \
2026 g = interp_1d_##name(lut1d, 1, g); \
2027 b = interp_1d_##name(lut1d, 2, b); \
2031 if (!direct && in->linesize[3]) \
2032 dsta[x] = srca[x]; \
2034 grow += out->linesize[0]; \
2035 brow += out->linesize[1]; \
2036 rrow += out->linesize[2]; \
2037 arow += out->linesize[3]; \
2038 srcgrow += in->linesize[0]; \
2039 srcbrow += in->linesize[1]; \
2040 srcrrow += in->linesize[2]; \
2041 srcarow += in->linesize[3]; \
2046 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(nearest, 32)
2047 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(linear, 32)
2048 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cosine, 32)
2049 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cubic, 32)
2050 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(spline, 32)
2052 #define DEFINE_INTERP_FUNC_1D(name, nbits) \
2053 static int interp_1d_##nbits##_##name(AVFilterContext *ctx, void *arg, \
2054 int jobnr, int nb_jobs) \
2057 const LUT1DContext *lut1d = ctx->priv; \
2058 const ThreadData *td = arg; \
2059 const AVFrame *in = td->in; \
2060 const AVFrame *out = td->out; \
2061 const int direct = out == in; \
2062 const int step = lut1d->step; \
2063 const uint8_t r = lut1d->rgba_map[R]; \
2064 const uint8_t g = lut1d->rgba_map[G]; \
2065 const uint8_t b = lut1d->rgba_map[B]; \
2066 const uint8_t a = lut1d->rgba_map[A]; \
2067 const int slice_start = (in->height * jobnr ) / nb_jobs; \
2068 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
2069 uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
2070 const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
2071 const float factor = (1 << nbits) - 1; \
2072 const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \
2073 const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \
2074 const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \
2076 for (y = slice_start; y < slice_end; y++) { \
2077 uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
2078 const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
2079 for (x = 0; x < in->width * step; x += step) { \
2080 float rr = src[x + r] * scale_r; \
2081 float gg = src[x + g] * scale_g; \
2082 float bb = src[x + b] * scale_b; \
2083 rr = interp_1d_##name(lut1d, 0, rr); \
2084 gg = interp_1d_##name(lut1d, 1, gg); \
2085 bb = interp_1d_##name(lut1d, 2, bb); \
2086 dst[x + r] = av_clip_uint##nbits(rr * factor); \
2087 dst[x + g] = av_clip_uint##nbits(gg * factor); \
2088 dst[x + b] = av_clip_uint##nbits(bb * factor); \
2089 if (!direct && step == 4) \
2090 dst[x + a] = src[x + a]; \
2092 dstrow += out->linesize[0]; \
2093 srcrow += in ->linesize[0]; \
2098 DEFINE_INTERP_FUNC_1D(nearest, 8)
2099 DEFINE_INTERP_FUNC_1D(linear, 8)
2100 DEFINE_INTERP_FUNC_1D(cosine, 8)
2101 DEFINE_INTERP_FUNC_1D(cubic, 8)
2102 DEFINE_INTERP_FUNC_1D(spline, 8)
2104 DEFINE_INTERP_FUNC_1D(nearest, 16)
2105 DEFINE_INTERP_FUNC_1D(linear, 16)
2106 DEFINE_INTERP_FUNC_1D(cosine, 16)
2107 DEFINE_INTERP_FUNC_1D(cubic, 16)
2108 DEFINE_INTERP_FUNC_1D(spline, 16)
2110 static int config_input_1d(AVFilterLink *inlink)
2112 int depth, is16bit, isfloat, planar;
2113 LUT1DContext *lut1d = inlink->dst->priv;
2114 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
2116 depth = desc->comp[0].depth;
2117 is16bit = desc->comp[0].depth > 8;
2118 planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR;
2119 isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
2120 ff_fill_rgba_map(lut1d->rgba_map, inlink->format);
2121 lut1d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
2123 #define SET_FUNC_1D(name) do { \
2124 if (planar && !isfloat) { \
2126 case 8: lut1d->interp = interp_1d_8_##name##_p8; break; \
2127 case 9: lut1d->interp = interp_1d_16_##name##_p9; break; \
2128 case 10: lut1d->interp = interp_1d_16_##name##_p10; break; \
2129 case 12: lut1d->interp = interp_1d_16_##name##_p12; break; \
2130 case 14: lut1d->interp = interp_1d_16_##name##_p14; break; \
2131 case 16: lut1d->interp = interp_1d_16_##name##_p16; break; \
2133 } else if (isfloat) { lut1d->interp = interp_1d_##name##_pf32; \
2134 } else if (is16bit) { lut1d->interp = interp_1d_16_##name; \
2135 } else { lut1d->interp = interp_1d_8_##name; } \
2138 switch (lut1d->interpolation) {
2139 case INTERPOLATE_1D_NEAREST: SET_FUNC_1D(nearest); break;
2140 case INTERPOLATE_1D_LINEAR: SET_FUNC_1D(linear); break;
2141 case INTERPOLATE_1D_COSINE: SET_FUNC_1D(cosine); break;
2142 case INTERPOLATE_1D_CUBIC: SET_FUNC_1D(cubic); break;
2143 case INTERPOLATE_1D_SPLINE: SET_FUNC_1D(spline); break;
2151 static av_cold int lut1d_init(AVFilterContext *ctx)
2156 LUT1DContext *lut1d = ctx->priv;
2158 lut1d->scale.r = lut1d->scale.g = lut1d->scale.b = 1.f;
2161 set_identity_matrix_1d(lut1d, 32);
2165 f = av_fopen_utf8(lut1d->file, "r");
2167 ret = AVERROR(errno);
2168 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut1d->file, av_err2str(ret));
2172 ext = strrchr(lut1d->file, '.');
2174 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
2175 ret = AVERROR_INVALIDDATA;
2180 if (!av_strcasecmp(ext, "cube") || !av_strcasecmp(ext, "1dlut")) {
2181 ret = parse_cube_1d(ctx, f);
2182 } else if (!av_strcasecmp(ext, "csp")) {
2183 ret = parse_cinespace_1d(ctx, f);
2185 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
2186 ret = AVERROR(EINVAL);
2189 if (!ret && !lut1d->lutsize) {
2190 av_log(ctx, AV_LOG_ERROR, "1D LUT is empty\n");
2191 ret = AVERROR_INVALIDDATA;
2199 static AVFrame *apply_1d_lut(AVFilterLink *inlink, AVFrame *in)
2201 AVFilterContext *ctx = inlink->dst;
2202 LUT1DContext *lut1d = ctx->priv;
2203 AVFilterLink *outlink = inlink->dst->outputs[0];
2207 if (av_frame_is_writable(in)) {
2210 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
2215 av_frame_copy_props(out, in);
2220 ctx->internal->execute(ctx, lut1d->interp, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
2228 static int filter_frame_1d(AVFilterLink *inlink, AVFrame *in)
2230 AVFilterLink *outlink = inlink->dst->outputs[0];
2231 AVFrame *out = apply_1d_lut(inlink, in);
2233 return AVERROR(ENOMEM);
2234 return ff_filter_frame(outlink, out);
2237 static const AVFilterPad lut1d_inputs[] = {
2240 .type = AVMEDIA_TYPE_VIDEO,
2241 .filter_frame = filter_frame_1d,
2242 .config_props = config_input_1d,
2247 static const AVFilterPad lut1d_outputs[] = {
2250 .type = AVMEDIA_TYPE_VIDEO,
2255 AVFilter ff_vf_lut1d = {
2257 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 1D LUT."),
2258 .priv_size = sizeof(LUT1DContext),
2260 .query_formats = query_formats,
2261 .inputs = lut1d_inputs,
2262 .outputs = lut1d_outputs,
2263 .priv_class = &lut1d_class,
2264 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,