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 TFLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
104 #define COMMON_OPTIONS \
105 { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, NB_INTERP_MODE-1, FLAGS, "interp_mode" }, \
106 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_NEAREST}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \
107 { "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" }, \
108 { "tetrahedral", "interpolate values using a tetrahedron", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TETRAHEDRAL}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \
109 { "pyramid", "interpolate values using a pyramid", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PYRAMID}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \
110 { "prism", "interpolate values using a prism", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PRISM}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \
113 #define EXPONENT_MASK 0x7F800000
114 #define MANTISSA_MASK 0x007FFFFF
115 #define SIGN_MASK 0x80000000
117 static inline float sanitizef(float f)
119 union av_intfloat32 t;
122 if ((t.i & EXPONENT_MASK) == EXPONENT_MASK) {
123 if ((t.i & MANTISSA_MASK) != 0) {
126 } else if (t.i & SIGN_MASK) {
137 static inline float lerpf(float v0, float v1, float f)
139 return v0 + (v1 - v0) * f;
142 static inline struct rgbvec lerp(const struct rgbvec *v0, const struct rgbvec *v1, float f)
145 lerpf(v0->r, v1->r, f), lerpf(v0->g, v1->g, f), lerpf(v0->b, v1->b, f)
150 #define NEAR(x) ((int)((x) + .5))
151 #define PREV(x) ((int)(x))
152 #define NEXT(x) (FFMIN((int)(x) + 1, lut3d->lutsize - 1))
155 * Get the nearest defined point
157 static inline struct rgbvec interp_nearest(const LUT3DContext *lut3d,
158 const struct rgbvec *s)
160 return lut3d->lut[NEAR(s->r) * lut3d->lutsize2 + NEAR(s->g) * lut3d->lutsize + NEAR(s->b)];
164 * Interpolate using the 8 vertices of a cube
165 * @see https://en.wikipedia.org/wiki/Trilinear_interpolation
167 static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d,
168 const struct rgbvec *s)
170 const int lutsize2 = lut3d->lutsize2;
171 const int lutsize = lut3d->lutsize;
172 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
173 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
174 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
175 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
176 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
177 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
178 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
179 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
180 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
181 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
182 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
183 const struct rgbvec c00 = lerp(&c000, &c100, d.r);
184 const struct rgbvec c10 = lerp(&c010, &c110, d.r);
185 const struct rgbvec c01 = lerp(&c001, &c101, d.r);
186 const struct rgbvec c11 = lerp(&c011, &c111, d.r);
187 const struct rgbvec c0 = lerp(&c00, &c10, d.g);
188 const struct rgbvec c1 = lerp(&c01, &c11, d.g);
189 const struct rgbvec c = lerp(&c0, &c1, d.b);
193 static inline struct rgbvec interp_pyramid(const LUT3DContext *lut3d,
194 const struct rgbvec *s)
196 const int lutsize2 = lut3d->lutsize2;
197 const int lutsize = lut3d->lutsize;
198 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
199 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
200 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
201 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
202 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
205 if (d.g > d.r && d.b > d.r) {
206 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
207 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
208 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
210 c.r = c000.r + (c111.r - c011.r) * d.r + (c010.r - c000.r) * d.g + (c001.r - c000.r) * d.b +
211 (c011.r - c001.r - c010.r + c000.r) * d.g * d.b;
212 c.g = c000.g + (c111.g - c011.g) * d.r + (c010.g - c000.g) * d.g + (c001.g - c000.g) * d.b +
213 (c011.g - c001.g - c010.g + c000.g) * d.g * d.b;
214 c.b = c000.b + (c111.b - c011.b) * d.r + (c010.b - c000.b) * d.g + (c001.b - c000.b) * d.b +
215 (c011.b - c001.b - c010.b + c000.b) * d.g * d.b;
216 } else if (d.r > d.g && d.b > d.g) {
217 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
218 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
219 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
221 c.r = c000.r + (c100.r - c000.r) * d.r + (c111.r - c101.r) * d.g + (c001.r - c000.r) * d.b +
222 (c101.r - c001.r - c100.r + c000.r) * d.r * d.b;
223 c.g = c000.g + (c100.g - c000.g) * d.r + (c111.g - c101.g) * d.g + (c001.g - c000.g) * d.b +
224 (c101.g - c001.g - c100.g + c000.g) * d.r * d.b;
225 c.b = c000.b + (c100.b - c000.b) * d.r + (c111.b - c101.b) * d.g + (c001.b - c000.b) * d.b +
226 (c101.b - c001.b - c100.b + c000.b) * d.r * d.b;
228 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
229 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
230 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
232 c.r = c000.r + (c100.r - c000.r) * d.r + (c010.r - c000.r) * d.g + (c111.r - c110.r) * d.b +
233 (c110.r - c100.r - c010.r + c000.r) * d.r * d.g;
234 c.g = c000.g + (c100.g - c000.g) * d.r + (c010.g - c000.g) * d.g + (c111.g - c110.g) * d.b +
235 (c110.g - c100.g - c010.g + c000.g) * d.r * d.g;
236 c.b = c000.b + (c100.b - c000.b) * d.r + (c010.b - c000.b) * d.g + (c111.b - c110.b) * d.b +
237 (c110.b - c100.b - c010.b + c000.b) * d.r * d.g;
243 static inline struct rgbvec interp_prism(const LUT3DContext *lut3d,
244 const struct rgbvec *s)
246 const int lutsize2 = lut3d->lutsize2;
247 const int lutsize = lut3d->lutsize;
248 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
249 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
250 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
251 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
252 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
253 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
254 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
258 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
259 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
261 c.r = c000.r + (c001.r - c000.r) * d.b + (c101.r - c001.r) * d.r + (c010.r - c000.r) * d.g +
262 (c000.r - c010.r - c001.r + c011.r) * d.b * d.g +
263 (c001.r - c011.r - c101.r + c111.r) * d.r * d.g;
264 c.g = c000.g + (c001.g - c000.g) * d.b + (c101.g - c001.g) * d.r + (c010.g - c000.g) * d.g +
265 (c000.g - c010.g - c001.g + c011.g) * d.b * d.g +
266 (c001.g - c011.g - c101.g + c111.g) * d.r * d.g;
267 c.b = c000.b + (c001.b - c000.b) * d.b + (c101.b - c001.b) * d.r + (c010.b - c000.b) * d.g +
268 (c000.b - c010.b - c001.b + c011.b) * d.b * d.g +
269 (c001.b - c011.b - c101.b + c111.b) * d.r * d.g;
271 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
272 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
274 c.r = c000.r + (c101.r - c100.r) * d.b + (c100.r - c000.r) * d.r + (c010.r - c000.r) * d.g +
275 (c100.r - c110.r - c101.r + c111.r) * d.b * d.g +
276 (c000.r - c010.r - c100.r + c110.r) * d.r * d.g;
277 c.g = c000.g + (c101.g - c100.g) * d.b + (c100.g - c000.g) * d.r + (c010.g - c000.g) * d.g +
278 (c100.g - c110.g - c101.g + c111.g) * d.b * d.g +
279 (c000.g - c010.g - c100.g + c110.g) * d.r * d.g;
280 c.b = c000.b + (c101.b - c100.b) * d.b + (c100.b - c000.b) * d.r + (c010.b - c000.b) * d.g +
281 (c100.b - c110.b - c101.b + c111.b) * d.b * d.g +
282 (c000.b - c010.b - c100.b + c110.b) * d.r * d.g;
289 * Tetrahedral interpolation. Based on code found in Truelight Software Library paper.
290 * @see http://www.filmlight.ltd.uk/pdf/whitepapers/FL-TL-TN-0057-SoftwareLib.pdf
292 static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d,
293 const struct rgbvec *s)
295 const int lutsize2 = lut3d->lutsize2;
296 const int lutsize = lut3d->lutsize;
297 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
298 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
299 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
300 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
301 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
305 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
306 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
307 c.r = (1-d.r) * c000.r + (d.r-d.g) * c100.r + (d.g-d.b) * c110.r + (d.b) * c111.r;
308 c.g = (1-d.r) * c000.g + (d.r-d.g) * c100.g + (d.g-d.b) * c110.g + (d.b) * c111.g;
309 c.b = (1-d.r) * c000.b + (d.r-d.g) * c100.b + (d.g-d.b) * c110.b + (d.b) * c111.b;
310 } else if (d.r > d.b) {
311 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
312 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
313 c.r = (1-d.r) * c000.r + (d.r-d.b) * c100.r + (d.b-d.g) * c101.r + (d.g) * c111.r;
314 c.g = (1-d.r) * c000.g + (d.r-d.b) * c100.g + (d.b-d.g) * c101.g + (d.g) * c111.g;
315 c.b = (1-d.r) * c000.b + (d.r-d.b) * c100.b + (d.b-d.g) * c101.b + (d.g) * c111.b;
317 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
318 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
319 c.r = (1-d.b) * c000.r + (d.b-d.r) * c001.r + (d.r-d.g) * c101.r + (d.g) * c111.r;
320 c.g = (1-d.b) * c000.g + (d.b-d.r) * c001.g + (d.r-d.g) * c101.g + (d.g) * c111.g;
321 c.b = (1-d.b) * c000.b + (d.b-d.r) * c001.b + (d.r-d.g) * c101.b + (d.g) * c111.b;
325 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
326 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
327 c.r = (1-d.b) * c000.r + (d.b-d.g) * c001.r + (d.g-d.r) * c011.r + (d.r) * c111.r;
328 c.g = (1-d.b) * c000.g + (d.b-d.g) * c001.g + (d.g-d.r) * c011.g + (d.r) * c111.g;
329 c.b = (1-d.b) * c000.b + (d.b-d.g) * c001.b + (d.g-d.r) * c011.b + (d.r) * c111.b;
330 } else if (d.b > d.r) {
331 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
332 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
333 c.r = (1-d.g) * c000.r + (d.g-d.b) * c010.r + (d.b-d.r) * c011.r + (d.r) * c111.r;
334 c.g = (1-d.g) * c000.g + (d.g-d.b) * c010.g + (d.b-d.r) * c011.g + (d.r) * c111.g;
335 c.b = (1-d.g) * c000.b + (d.g-d.b) * c010.b + (d.b-d.r) * c011.b + (d.r) * c111.b;
337 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
338 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
339 c.r = (1-d.g) * c000.r + (d.g-d.r) * c010.r + (d.r-d.b) * c110.r + (d.b) * c111.r;
340 c.g = (1-d.g) * c000.g + (d.g-d.r) * c010.g + (d.r-d.b) * c110.g + (d.b) * c111.g;
341 c.b = (1-d.g) * c000.b + (d.g-d.r) * c010.b + (d.r-d.b) * c110.b + (d.b) * c111.b;
347 static inline float prelut_interp_1d_linear(const Lut3DPreLut *prelut,
348 int idx, const float s)
350 const int lut_max = prelut->size - 1;
351 const float scaled = (s - prelut->min[idx]) * prelut->scale[idx];
352 const float x = av_clipf(scaled, 0.0f, lut_max);
353 const int prev = PREV(x);
354 const int next = FFMIN((int)(x) + 1, lut_max);
355 const float p = prelut->lut[idx][prev];
356 const float n = prelut->lut[idx][next];
357 const float d = x - (float)prev;
358 return lerpf(p, n, d);
361 static inline struct rgbvec apply_prelut(const Lut3DPreLut *prelut,
362 const struct rgbvec *s)
366 if (prelut->size <= 0)
369 c.r = prelut_interp_1d_linear(prelut, 0, s->r);
370 c.g = prelut_interp_1d_linear(prelut, 1, s->g);
371 c.b = prelut_interp_1d_linear(prelut, 2, s->b);
375 #define DEFINE_INTERP_FUNC_PLANAR(name, nbits, depth) \
376 static int interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
379 const LUT3DContext *lut3d = ctx->priv; \
380 const Lut3DPreLut *prelut = &lut3d->prelut; \
381 const ThreadData *td = arg; \
382 const AVFrame *in = td->in; \
383 const AVFrame *out = td->out; \
384 const int direct = out == in; \
385 const int slice_start = (in->height * jobnr ) / nb_jobs; \
386 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
387 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
388 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
389 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
390 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
391 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
392 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
393 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
394 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
395 const float lut_max = lut3d->lutsize - 1; \
396 const float scale_f = 1.0f / ((1<<depth) - 1); \
397 const float scale_r = lut3d->scale.r * lut_max; \
398 const float scale_g = lut3d->scale.g * lut_max; \
399 const float scale_b = lut3d->scale.b * lut_max; \
401 for (y = slice_start; y < slice_end; y++) { \
402 uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
403 uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
404 uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
405 uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
406 const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
407 const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
408 const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
409 const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
410 for (x = 0; x < in->width; x++) { \
411 const struct rgbvec rgb = {srcr[x] * scale_f, \
413 srcb[x] * scale_f}; \
414 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
415 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
416 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
417 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
418 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
419 dstr[x] = av_clip_uintp2(vec.r * (float)((1<<depth) - 1), depth); \
420 dstg[x] = av_clip_uintp2(vec.g * (float)((1<<depth) - 1), depth); \
421 dstb[x] = av_clip_uintp2(vec.b * (float)((1<<depth) - 1), depth); \
422 if (!direct && in->linesize[3]) \
425 grow += out->linesize[0]; \
426 brow += out->linesize[1]; \
427 rrow += out->linesize[2]; \
428 arow += out->linesize[3]; \
429 srcgrow += in->linesize[0]; \
430 srcbrow += in->linesize[1]; \
431 srcrrow += in->linesize[2]; \
432 srcarow += in->linesize[3]; \
437 DEFINE_INTERP_FUNC_PLANAR(nearest, 8, 8)
438 DEFINE_INTERP_FUNC_PLANAR(trilinear, 8, 8)
439 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 8, 8)
440 DEFINE_INTERP_FUNC_PLANAR(pyramid, 8, 8)
441 DEFINE_INTERP_FUNC_PLANAR(prism, 8, 8)
443 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 9)
444 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 9)
445 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 9)
446 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 9)
447 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 9)
449 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 10)
450 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 10)
451 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 10)
452 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 10)
453 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 10)
455 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 12)
456 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 12)
457 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 12)
458 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 12)
459 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 12)
461 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 14)
462 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 14)
463 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 14)
464 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 14)
465 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 14)
467 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 16)
468 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 16)
469 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 16)
470 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 16)
471 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 16)
473 #define DEFINE_INTERP_FUNC_PLANAR_FLOAT(name, depth) \
474 static int interp_##name##_pf##depth(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
477 const LUT3DContext *lut3d = ctx->priv; \
478 const Lut3DPreLut *prelut = &lut3d->prelut; \
479 const ThreadData *td = arg; \
480 const AVFrame *in = td->in; \
481 const AVFrame *out = td->out; \
482 const int direct = out == in; \
483 const int slice_start = (in->height * jobnr ) / nb_jobs; \
484 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
485 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
486 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
487 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
488 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
489 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
490 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
491 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
492 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
493 const float lut_max = lut3d->lutsize - 1; \
494 const float scale_r = lut3d->scale.r * lut_max; \
495 const float scale_g = lut3d->scale.g * lut_max; \
496 const float scale_b = lut3d->scale.b * lut_max; \
498 for (y = slice_start; y < slice_end; y++) { \
499 float *dstg = (float *)grow; \
500 float *dstb = (float *)brow; \
501 float *dstr = (float *)rrow; \
502 float *dsta = (float *)arow; \
503 const float *srcg = (const float *)srcgrow; \
504 const float *srcb = (const float *)srcbrow; \
505 const float *srcr = (const float *)srcrrow; \
506 const float *srca = (const float *)srcarow; \
507 for (x = 0; x < in->width; x++) { \
508 const struct rgbvec rgb = {sanitizef(srcr[x]), \
509 sanitizef(srcg[x]), \
510 sanitizef(srcb[x])}; \
511 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
512 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
513 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
514 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
515 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
519 if (!direct && in->linesize[3]) \
522 grow += out->linesize[0]; \
523 brow += out->linesize[1]; \
524 rrow += out->linesize[2]; \
525 arow += out->linesize[3]; \
526 srcgrow += in->linesize[0]; \
527 srcbrow += in->linesize[1]; \
528 srcrrow += in->linesize[2]; \
529 srcarow += in->linesize[3]; \
534 DEFINE_INTERP_FUNC_PLANAR_FLOAT(nearest, 32)
535 DEFINE_INTERP_FUNC_PLANAR_FLOAT(trilinear, 32)
536 DEFINE_INTERP_FUNC_PLANAR_FLOAT(tetrahedral, 32)
537 DEFINE_INTERP_FUNC_PLANAR_FLOAT(pyramid, 32)
538 DEFINE_INTERP_FUNC_PLANAR_FLOAT(prism, 32)
540 #define DEFINE_INTERP_FUNC(name, nbits) \
541 static int interp_##nbits##_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
544 const LUT3DContext *lut3d = ctx->priv; \
545 const Lut3DPreLut *prelut = &lut3d->prelut; \
546 const ThreadData *td = arg; \
547 const AVFrame *in = td->in; \
548 const AVFrame *out = td->out; \
549 const int direct = out == in; \
550 const int step = lut3d->step; \
551 const uint8_t r = lut3d->rgba_map[R]; \
552 const uint8_t g = lut3d->rgba_map[G]; \
553 const uint8_t b = lut3d->rgba_map[B]; \
554 const uint8_t a = lut3d->rgba_map[A]; \
555 const int slice_start = (in->height * jobnr ) / nb_jobs; \
556 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
557 uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
558 const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
559 const float lut_max = lut3d->lutsize - 1; \
560 const float scale_f = 1.0f / ((1<<nbits) - 1); \
561 const float scale_r = lut3d->scale.r * lut_max; \
562 const float scale_g = lut3d->scale.g * lut_max; \
563 const float scale_b = lut3d->scale.b * lut_max; \
565 for (y = slice_start; y < slice_end; y++) { \
566 uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
567 const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
568 for (x = 0; x < in->width * step; x += step) { \
569 const struct rgbvec rgb = {src[x + r] * scale_f, \
570 src[x + g] * scale_f, \
571 src[x + b] * scale_f}; \
572 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
573 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
574 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
575 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
576 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
577 dst[x + r] = av_clip_uint##nbits(vec.r * (float)((1<<nbits) - 1)); \
578 dst[x + g] = av_clip_uint##nbits(vec.g * (float)((1<<nbits) - 1)); \
579 dst[x + b] = av_clip_uint##nbits(vec.b * (float)((1<<nbits) - 1)); \
580 if (!direct && step == 4) \
581 dst[x + a] = src[x + a]; \
583 dstrow += out->linesize[0]; \
584 srcrow += in ->linesize[0]; \
589 DEFINE_INTERP_FUNC(nearest, 8)
590 DEFINE_INTERP_FUNC(trilinear, 8)
591 DEFINE_INTERP_FUNC(tetrahedral, 8)
592 DEFINE_INTERP_FUNC(pyramid, 8)
593 DEFINE_INTERP_FUNC(prism, 8)
595 DEFINE_INTERP_FUNC(nearest, 16)
596 DEFINE_INTERP_FUNC(trilinear, 16)
597 DEFINE_INTERP_FUNC(tetrahedral, 16)
598 DEFINE_INTERP_FUNC(pyramid, 16)
599 DEFINE_INTERP_FUNC(prism, 16)
601 #define MAX_LINE_SIZE 512
603 static int skip_line(const char *p)
605 while (*p && av_isspace(*p))
607 return !*p || *p == '#';
610 static char* fget_next_word(char* dst, int max, FILE* f)
617 /* skip until next non whitespace char */
618 while ((c = fgetc(f)) != EOF) {
627 /* get max bytes or up until next whitespace char */
628 for (; max > 0; max--) {
629 if ((c = fgetc(f)) == EOF)
644 #define NEXT_LINE(loop_cond) do { \
645 if (!fgets(line, sizeof(line), f)) { \
646 av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \
647 return AVERROR_INVALIDDATA; \
651 #define NEXT_LINE_OR_GOTO(loop_cond, label) do { \
652 if (!fgets(line, sizeof(line), f)) { \
653 av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \
654 ret = AVERROR_INVALIDDATA; \
659 static int allocate_3dlut(AVFilterContext *ctx, int lutsize, int prelut)
661 LUT3DContext *lut3d = ctx->priv;
663 if (lutsize < 2 || lutsize > MAX_LEVEL) {
664 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n");
665 return AVERROR(EINVAL);
668 av_freep(&lut3d->lut);
669 lut3d->lut = av_malloc_array(lutsize * lutsize * lutsize, sizeof(*lut3d->lut));
671 return AVERROR(ENOMEM);
674 lut3d->prelut.size = PRELUT_SIZE;
675 for (i = 0; i < 3; i++) {
676 av_freep(&lut3d->prelut.lut[i]);
677 lut3d->prelut.lut[i] = av_malloc_array(PRELUT_SIZE, sizeof(*lut3d->prelut.lut[0]));
678 if (!lut3d->prelut.lut[i])
679 return AVERROR(ENOMEM);
682 lut3d->prelut.size = 0;
683 for (i = 0; i < 3; i++) {
684 av_freep(&lut3d->prelut.lut[i]);
687 lut3d->lutsize = lutsize;
688 lut3d->lutsize2 = lutsize * lutsize;
692 /* Basically r g and b float values on each line, with a facultative 3DLUTSIZE
693 * directive; seems to be generated by Davinci */
694 static int parse_dat(AVFilterContext *ctx, FILE *f)
696 LUT3DContext *lut3d = ctx->priv;
697 char line[MAX_LINE_SIZE];
698 int ret, i, j, k, size, size2;
700 lut3d->lutsize = size = 33;
703 NEXT_LINE(skip_line(line));
704 if (!strncmp(line, "3DLUTSIZE ", 10)) {
705 size = strtol(line + 10, NULL, 0);
707 NEXT_LINE(skip_line(line));
710 ret = allocate_3dlut(ctx, size, 0);
714 for (k = 0; k < size; k++) {
715 for (j = 0; j < size; j++) {
716 for (i = 0; i < size; i++) {
717 struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
718 if (k != 0 || j != 0 || i != 0)
719 NEXT_LINE(skip_line(line));
720 if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
721 return AVERROR_INVALIDDATA;
729 static int parse_cube(AVFilterContext *ctx, FILE *f)
731 LUT3DContext *lut3d = ctx->priv;
732 char line[MAX_LINE_SIZE];
733 float min[3] = {0.0, 0.0, 0.0};
734 float max[3] = {1.0, 1.0, 1.0};
736 while (fgets(line, sizeof(line), f)) {
737 if (!strncmp(line, "LUT_3D_SIZE", 11)) {
739 const int size = strtol(line + 12, NULL, 0);
740 const int size2 = size * size;
742 ret = allocate_3dlut(ctx, size, 0);
746 for (k = 0; k < size; k++) {
747 for (j = 0; j < size; j++) {
748 for (i = 0; i < size; i++) {
749 struct rgbvec *vec = &lut3d->lut[i * size2 + j * size + k];
754 if (!strncmp(line, "DOMAIN_", 7)) {
756 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
757 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
759 return AVERROR_INVALIDDATA;
760 av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
761 av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
762 min[0], min[1], min[2], max[0], max[1], max[2]);
764 } else if (!strncmp(line, "TITLE", 5)) {
767 } while (skip_line(line));
768 if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
769 return AVERROR_INVALIDDATA;
777 lut3d->scale.r = av_clipf(1. / (max[0] - min[0]), 0.f, 1.f);
778 lut3d->scale.g = av_clipf(1. / (max[1] - min[1]), 0.f, 1.f);
779 lut3d->scale.b = av_clipf(1. / (max[2] - min[2]), 0.f, 1.f);
784 /* Assume 17x17x17 LUT with a 16-bit depth
785 * FIXME: it seems there are various 3dl formats */
786 static int parse_3dl(AVFilterContext *ctx, FILE *f)
788 char line[MAX_LINE_SIZE];
789 LUT3DContext *lut3d = ctx->priv;
792 const int size2 = 17 * 17;
793 const float scale = 16*16*16;
795 lut3d->lutsize = size;
797 ret = allocate_3dlut(ctx, size, 0);
801 NEXT_LINE(skip_line(line));
802 for (k = 0; k < size; k++) {
803 for (j = 0; j < size; j++) {
804 for (i = 0; i < size; i++) {
806 struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
808 NEXT_LINE(skip_line(line));
809 if (av_sscanf(line, "%d %d %d", &r, &g, &b) != 3)
810 return AVERROR_INVALIDDATA;
821 static int parse_m3d(AVFilterContext *ctx, FILE *f)
823 LUT3DContext *lut3d = ctx->priv;
825 int ret, i, j, k, size, size2, in = -1, out = -1;
826 char line[MAX_LINE_SIZE];
827 uint8_t rgb_map[3] = {0, 1, 2};
829 while (fgets(line, sizeof(line), f)) {
830 if (!strncmp(line, "in", 2)) in = strtol(line + 2, NULL, 0);
831 else if (!strncmp(line, "out", 3)) out = strtol(line + 3, NULL, 0);
832 else if (!strncmp(line, "values", 6)) {
833 const char *p = line + 6;
834 #define SET_COLOR(id) do { \
835 while (av_isspace(*p)) \
838 case 'r': rgb_map[id] = 0; break; \
839 case 'g': rgb_map[id] = 1; break; \
840 case 'b': rgb_map[id] = 2; break; \
842 while (*p && !av_isspace(*p)) \
852 if (in == -1 || out == -1) {
853 av_log(ctx, AV_LOG_ERROR, "in and out must be defined\n");
854 return AVERROR_INVALIDDATA;
856 if (in < 2 || out < 2 ||
857 in > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL ||
858 out > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL) {
859 av_log(ctx, AV_LOG_ERROR, "invalid in (%d) or out (%d)\n", in, out);
860 return AVERROR_INVALIDDATA;
862 for (size = 1; size*size*size < in; size++);
863 lut3d->lutsize = size;
866 ret = allocate_3dlut(ctx, size, 0);
870 scale = 1. / (out - 1);
872 for (k = 0; k < size; k++) {
873 for (j = 0; j < size; j++) {
874 for (i = 0; i < size; i++) {
875 struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
879 if (av_sscanf(line, "%f %f %f", val, val + 1, val + 2) != 3)
880 return AVERROR_INVALIDDATA;
881 vec->r = val[rgb_map[0]] * scale;
882 vec->g = val[rgb_map[1]] * scale;
883 vec->b = val[rgb_map[2]] * scale;
890 static int nearest_sample_index(float *data, float x, int low, int hi)
900 av_assert0(x >= data[low]);
901 av_assert0(x <= data[hi]);
902 av_assert0((hi-low) > 0);
907 mid = (low + hi) / 2;
918 #define NEXT_FLOAT_OR_GOTO(value, label) \
919 if (!fget_next_word(line, sizeof(line) ,f)) { \
920 ret = AVERROR_INVALIDDATA; \
923 if (av_sscanf(line, "%f", &value) != 1) { \
924 ret = AVERROR_INVALIDDATA; \
928 static int parse_cinespace(AVFilterContext *ctx, FILE *f)
930 LUT3DContext *lut3d = ctx->priv;
931 char line[MAX_LINE_SIZE];
932 float in_min[3] = {0.0, 0.0, 0.0};
933 float in_max[3] = {1.0, 1.0, 1.0};
934 float out_min[3] = {0.0, 0.0, 0.0};
935 float out_max[3] = {1.0, 1.0, 1.0};
936 int inside_metadata = 0, size, size2;
940 int prelut_sizes[3] = {0, 0, 0};
941 float *in_prelut[3] = {NULL, NULL, NULL};
942 float *out_prelut[3] = {NULL, NULL, NULL};
944 NEXT_LINE_OR_GOTO(skip_line(line), end);
945 if (strncmp(line, "CSPLUTV100", 10)) {
946 av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n");
947 ret = AVERROR(EINVAL);
951 NEXT_LINE_OR_GOTO(skip_line(line), end);
952 if (strncmp(line, "3D", 2)) {
953 av_log(ctx, AV_LOG_ERROR, "Not 3D LUT format\n");
954 ret = AVERROR(EINVAL);
959 NEXT_LINE_OR_GOTO(skip_line(line), end);
961 if (!strncmp(line, "BEGIN METADATA", 14)) {
965 if (!strncmp(line, "END METADATA", 12)) {
969 if (inside_metadata == 0) {
970 int size_r, size_g, size_b;
972 for (int i = 0; i < 3; i++) {
973 int npoints = strtol(line, NULL, 0);
978 if (npoints > PRELUT_SIZE) {
979 av_log(ctx, AV_LOG_ERROR, "Prelut size too large.\n");
980 ret = AVERROR_INVALIDDATA;
984 if (in_prelut[i] || out_prelut[i]) {
985 av_log(ctx, AV_LOG_ERROR, "Invalid file has multiple preluts.\n");
986 ret = AVERROR_INVALIDDATA;
990 in_prelut[i] = (float*)av_malloc(npoints * sizeof(float));
991 out_prelut[i] = (float*)av_malloc(npoints * sizeof(float));
992 if (!in_prelut[i] || !out_prelut[i]) {
993 ret = AVERROR(ENOMEM);
997 prelut_sizes[i] = npoints;
999 in_max[i] = -FLT_MAX;
1000 out_min[i] = FLT_MAX;
1001 out_max[i] = -FLT_MAX;
1003 for (int j = 0; j < npoints; j++) {
1004 NEXT_FLOAT_OR_GOTO(v, end)
1005 in_min[i] = FFMIN(in_min[i], v);
1006 in_max[i] = FFMAX(in_max[i], v);
1007 in_prelut[i][j] = v;
1008 if (j > 0 && v < last) {
1009 av_log(ctx, AV_LOG_ERROR, "Invalid file, non increasing prelut.\n");
1010 ret = AVERROR(ENOMEM);
1016 for (int j = 0; j < npoints; j++) {
1017 NEXT_FLOAT_OR_GOTO(v, end)
1018 out_min[i] = FFMIN(out_min[i], v);
1019 out_max[i] = FFMAX(out_max[i], v);
1020 out_prelut[i][j] = v;
1023 } else if (npoints == 2) {
1024 NEXT_LINE_OR_GOTO(skip_line(line), end);
1025 if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2) {
1026 ret = AVERROR_INVALIDDATA;
1029 NEXT_LINE_OR_GOTO(skip_line(line), end);
1030 if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2) {
1031 ret = AVERROR_INVALIDDATA;
1036 av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n");
1037 ret = AVERROR_PATCHWELCOME;
1041 NEXT_LINE_OR_GOTO(skip_line(line), end);
1044 if (av_sscanf(line, "%d %d %d", &size_r, &size_g, &size_b) != 3) {
1045 ret = AVERROR(EINVAL);
1048 if (size_r != size_g || size_r != size_b) {
1049 av_log(ctx, AV_LOG_ERROR, "Unsupported size combination: %dx%dx%d.\n", size_r, size_g, size_b);
1050 ret = AVERROR_PATCHWELCOME;
1055 size2 = size * size;
1057 if (prelut_sizes[0] && prelut_sizes[1] && prelut_sizes[2])
1060 ret = allocate_3dlut(ctx, size, prelut);
1064 for (int k = 0; k < size; k++) {
1065 for (int j = 0; j < size; j++) {
1066 for (int i = 0; i < size; i++) {
1067 struct rgbvec *vec = &lut3d->lut[i * size2 + j * size + k];
1069 NEXT_LINE_OR_GOTO(skip_line(line), end);
1070 if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) {
1071 ret = AVERROR_INVALIDDATA;
1075 vec->r *= out_max[0] - out_min[0];
1076 vec->g *= out_max[1] - out_min[1];
1077 vec->b *= out_max[2] - out_min[2];
1087 for (int c = 0; c < 3; c++) {
1089 lut3d->prelut.min[c] = in_min[c];
1090 lut3d->prelut.max[c] = in_max[c];
1091 lut3d->prelut.scale[c] = (1.0f / (float)(in_max[c] - in_min[c])) * (lut3d->prelut.size - 1);
1093 for (int i = 0; i < lut3d->prelut.size; ++i) {
1094 float mix = (float) i / (float)(lut3d->prelut.size - 1);
1095 float x = lerpf(in_min[c], in_max[c], mix), a, b;
1097 int idx = nearest_sample_index(in_prelut[c], x, 0, prelut_sizes[c]-1);
1098 av_assert0(idx + 1 < prelut_sizes[c]);
1100 a = out_prelut[c][idx + 0];
1101 b = out_prelut[c][idx + 1];
1102 mix = x - in_prelut[c][idx];
1104 lut3d->prelut.lut[c][i] = sanitizef(lerpf(a, b, mix));
1107 lut3d->scale.r = 1.00f;
1108 lut3d->scale.g = 1.00f;
1109 lut3d->scale.b = 1.00f;
1112 lut3d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f);
1113 lut3d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f);
1114 lut3d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f);
1118 for (int c = 0; c < 3; c++) {
1119 av_freep(&in_prelut[c]);
1120 av_freep(&out_prelut[c]);
1125 static int set_identity_matrix(AVFilterContext *ctx, int size)
1127 LUT3DContext *lut3d = ctx->priv;
1129 const int size2 = size * size;
1130 const float c = 1. / (size - 1);
1132 ret = allocate_3dlut(ctx, size, 0);
1136 for (k = 0; k < size; k++) {
1137 for (j = 0; j < size; j++) {
1138 for (i = 0; i < size; i++) {
1139 struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
1150 static int query_formats(AVFilterContext *ctx)
1152 static const enum AVPixelFormat pix_fmts[] = {
1153 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
1154 AV_PIX_FMT_RGBA, AV_PIX_FMT_BGRA,
1155 AV_PIX_FMT_ARGB, AV_PIX_FMT_ABGR,
1156 AV_PIX_FMT_0RGB, AV_PIX_FMT_0BGR,
1157 AV_PIX_FMT_RGB0, AV_PIX_FMT_BGR0,
1158 AV_PIX_FMT_RGB48, AV_PIX_FMT_BGR48,
1159 AV_PIX_FMT_RGBA64, AV_PIX_FMT_BGRA64,
1160 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP,
1162 AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRAP10,
1163 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRAP12,
1165 AV_PIX_FMT_GBRP16, AV_PIX_FMT_GBRAP16,
1166 AV_PIX_FMT_GBRPF32, AV_PIX_FMT_GBRAPF32,
1169 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
1171 return AVERROR(ENOMEM);
1172 return ff_set_common_formats(ctx, fmts_list);
1175 static int config_input(AVFilterLink *inlink)
1177 int depth, is16bit, isfloat, planar;
1178 LUT3DContext *lut3d = inlink->dst->priv;
1179 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1181 depth = desc->comp[0].depth;
1182 is16bit = desc->comp[0].depth > 8;
1183 planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR;
1184 isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
1185 ff_fill_rgba_map(lut3d->rgba_map, inlink->format);
1186 lut3d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
1188 #define SET_FUNC(name) do { \
1189 if (planar && !isfloat) { \
1191 case 8: lut3d->interp = interp_8_##name##_p8; break; \
1192 case 9: lut3d->interp = interp_16_##name##_p9; break; \
1193 case 10: lut3d->interp = interp_16_##name##_p10; break; \
1194 case 12: lut3d->interp = interp_16_##name##_p12; break; \
1195 case 14: lut3d->interp = interp_16_##name##_p14; break; \
1196 case 16: lut3d->interp = interp_16_##name##_p16; break; \
1198 } else if (isfloat) { lut3d->interp = interp_##name##_pf32; \
1199 } else if (is16bit) { lut3d->interp = interp_16_##name; \
1200 } else { lut3d->interp = interp_8_##name; } \
1203 switch (lut3d->interpolation) {
1204 case INTERPOLATE_NEAREST: SET_FUNC(nearest); break;
1205 case INTERPOLATE_TRILINEAR: SET_FUNC(trilinear); break;
1206 case INTERPOLATE_TETRAHEDRAL: SET_FUNC(tetrahedral); break;
1207 case INTERPOLATE_PYRAMID: SET_FUNC(pyramid); break;
1208 case INTERPOLATE_PRISM: SET_FUNC(prism); break;
1216 static AVFrame *apply_lut(AVFilterLink *inlink, AVFrame *in)
1218 AVFilterContext *ctx = inlink->dst;
1219 LUT3DContext *lut3d = ctx->priv;
1220 AVFilterLink *outlink = inlink->dst->outputs[0];
1224 if (av_frame_is_writable(in)) {
1227 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
1232 av_frame_copy_props(out, in);
1237 ctx->internal->execute(ctx, lut3d->interp, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
1245 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
1247 AVFilterLink *outlink = inlink->dst->outputs[0];
1248 AVFrame *out = apply_lut(inlink, in);
1250 return AVERROR(ENOMEM);
1251 return ff_filter_frame(outlink, out);
1254 #if CONFIG_LUT3D_FILTER
1255 static const AVOption lut3d_options[] = {
1256 { "file", "set 3D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
1260 AVFILTER_DEFINE_CLASS(lut3d);
1262 static av_cold int lut3d_init(AVFilterContext *ctx)
1267 LUT3DContext *lut3d = ctx->priv;
1269 lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1272 return set_identity_matrix(ctx, 32);
1275 f = av_fopen_utf8(lut3d->file, "r");
1277 ret = AVERROR(errno);
1278 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret));
1282 ext = strrchr(lut3d->file, '.');
1284 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
1285 ret = AVERROR_INVALIDDATA;
1290 if (!av_strcasecmp(ext, "dat")) {
1291 ret = parse_dat(ctx, f);
1292 } else if (!av_strcasecmp(ext, "3dl")) {
1293 ret = parse_3dl(ctx, f);
1294 } else if (!av_strcasecmp(ext, "cube")) {
1295 ret = parse_cube(ctx, f);
1296 } else if (!av_strcasecmp(ext, "m3d")) {
1297 ret = parse_m3d(ctx, f);
1298 } else if (!av_strcasecmp(ext, "csp")) {
1299 ret = parse_cinespace(ctx, f);
1301 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
1302 ret = AVERROR(EINVAL);
1305 if (!ret && !lut3d->lutsize) {
1306 av_log(ctx, AV_LOG_ERROR, "3D LUT is empty\n");
1307 ret = AVERROR_INVALIDDATA;
1315 static av_cold void lut3d_uninit(AVFilterContext *ctx)
1317 LUT3DContext *lut3d = ctx->priv;
1319 av_freep(&lut3d->lut);
1321 for (i = 0; i < 3; i++) {
1322 av_freep(&lut3d->prelut.lut[i]);
1326 static const AVFilterPad lut3d_inputs[] = {
1329 .type = AVMEDIA_TYPE_VIDEO,
1330 .filter_frame = filter_frame,
1331 .config_props = config_input,
1336 static const AVFilterPad lut3d_outputs[] = {
1339 .type = AVMEDIA_TYPE_VIDEO,
1344 AVFilter ff_vf_lut3d = {
1346 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
1347 .priv_size = sizeof(LUT3DContext),
1349 .uninit = lut3d_uninit,
1350 .query_formats = query_formats,
1351 .inputs = lut3d_inputs,
1352 .outputs = lut3d_outputs,
1353 .priv_class = &lut3d_class,
1354 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
1358 #if CONFIG_HALDCLUT_FILTER
1360 static void update_clut_packed(LUT3DContext *lut3d, const AVFrame *frame)
1362 const uint8_t *data = frame->data[0];
1363 const int linesize = frame->linesize[0];
1364 const int w = lut3d->clut_width;
1365 const int step = lut3d->clut_step;
1366 const uint8_t *rgba_map = lut3d->clut_rgba_map;
1367 const int level = lut3d->lutsize;
1368 const int level2 = lut3d->lutsize2;
1370 #define LOAD_CLUT(nbits) do { \
1371 int i, j, k, x = 0, y = 0; \
1373 for (k = 0; k < level; k++) { \
1374 for (j = 0; j < level; j++) { \
1375 for (i = 0; i < level; i++) { \
1376 const uint##nbits##_t *src = (const uint##nbits##_t *) \
1377 (data + y*linesize + x*step); \
1378 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1379 vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \
1380 vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \
1381 vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \
1391 switch (lut3d->clut_bits) {
1392 case 8: LOAD_CLUT(8); break;
1393 case 16: LOAD_CLUT(16); break;
1397 static void update_clut_planar(LUT3DContext *lut3d, const AVFrame *frame)
1399 const uint8_t *datag = frame->data[0];
1400 const uint8_t *datab = frame->data[1];
1401 const uint8_t *datar = frame->data[2];
1402 const int glinesize = frame->linesize[0];
1403 const int blinesize = frame->linesize[1];
1404 const int rlinesize = frame->linesize[2];
1405 const int w = lut3d->clut_width;
1406 const int level = lut3d->lutsize;
1407 const int level2 = lut3d->lutsize2;
1409 #define LOAD_CLUT_PLANAR(nbits, depth) do { \
1410 int i, j, k, x = 0, y = 0; \
1412 for (k = 0; k < level; k++) { \
1413 for (j = 0; j < level; j++) { \
1414 for (i = 0; i < level; i++) { \
1415 const uint##nbits##_t *gsrc = (const uint##nbits##_t *) \
1416 (datag + y*glinesize); \
1417 const uint##nbits##_t *bsrc = (const uint##nbits##_t *) \
1418 (datab + y*blinesize); \
1419 const uint##nbits##_t *rsrc = (const uint##nbits##_t *) \
1420 (datar + y*rlinesize); \
1421 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1422 vec->r = gsrc[x] / (float)((1<<(depth)) - 1); \
1423 vec->g = bsrc[x] / (float)((1<<(depth)) - 1); \
1424 vec->b = rsrc[x] / (float)((1<<(depth)) - 1); \
1434 switch (lut3d->clut_bits) {
1435 case 8: LOAD_CLUT_PLANAR(8, 8); break;
1436 case 9: LOAD_CLUT_PLANAR(16, 9); break;
1437 case 10: LOAD_CLUT_PLANAR(16, 10); break;
1438 case 12: LOAD_CLUT_PLANAR(16, 12); break;
1439 case 14: LOAD_CLUT_PLANAR(16, 14); break;
1440 case 16: LOAD_CLUT_PLANAR(16, 16); break;
1444 static void update_clut_float(LUT3DContext *lut3d, const AVFrame *frame)
1446 const uint8_t *datag = frame->data[0];
1447 const uint8_t *datab = frame->data[1];
1448 const uint8_t *datar = frame->data[2];
1449 const int glinesize = frame->linesize[0];
1450 const int blinesize = frame->linesize[1];
1451 const int rlinesize = frame->linesize[2];
1452 const int w = lut3d->clut_width;
1453 const int level = lut3d->lutsize;
1454 const int level2 = lut3d->lutsize2;
1456 int i, j, k, x = 0, y = 0;
1458 for (k = 0; k < level; k++) {
1459 for (j = 0; j < level; j++) {
1460 for (i = 0; i < level; i++) {
1461 const float *gsrc = (const float *)(datag + y*glinesize);
1462 const float *bsrc = (const float *)(datab + y*blinesize);
1463 const float *rsrc = (const float *)(datar + y*rlinesize);
1464 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k];
1477 static int config_output(AVFilterLink *outlink)
1479 AVFilterContext *ctx = outlink->src;
1480 LUT3DContext *lut3d = ctx->priv;
1483 ret = ff_framesync_init_dualinput(&lut3d->fs, ctx);
1486 outlink->w = ctx->inputs[0]->w;
1487 outlink->h = ctx->inputs[0]->h;
1488 outlink->time_base = ctx->inputs[0]->time_base;
1489 if ((ret = ff_framesync_configure(&lut3d->fs)) < 0)
1494 static int activate(AVFilterContext *ctx)
1496 LUT3DContext *s = ctx->priv;
1497 return ff_framesync_activate(&s->fs);
1500 static int config_clut(AVFilterLink *inlink)
1502 int size, level, w, h;
1503 AVFilterContext *ctx = inlink->dst;
1504 LUT3DContext *lut3d = ctx->priv;
1505 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1509 lut3d->clut_bits = desc->comp[0].depth;
1510 lut3d->clut_planar = av_pix_fmt_count_planes(inlink->format) > 1;
1511 lut3d->clut_float = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
1513 lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3;
1514 ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format);
1516 if (inlink->w > inlink->h)
1517 av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the "
1518 "Hald CLUT will be ignored\n", inlink->w - inlink->h);
1519 else if (inlink->w < inlink->h)
1520 av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the "
1521 "Hald CLUT will be ignored\n", inlink->h - inlink->w);
1522 lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h);
1524 for (level = 1; level*level*level < w; level++);
1525 size = level*level*level;
1527 av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n");
1528 return AVERROR_INVALIDDATA;
1530 av_assert0(w == h && w == size);
1532 if (level > MAX_LEVEL) {
1533 const int max_clut_level = sqrt(MAX_LEVEL);
1534 const int max_clut_size = max_clut_level*max_clut_level*max_clut_level;
1535 av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT "
1536 "(maximum level is %d, or %dx%d CLUT)\n",
1537 max_clut_level, max_clut_size, max_clut_size);
1538 return AVERROR(EINVAL);
1541 return allocate_3dlut(ctx, level, 0);
1544 static int update_apply_clut(FFFrameSync *fs)
1546 AVFilterContext *ctx = fs->parent;
1547 LUT3DContext *lut3d = ctx->priv;
1548 AVFilterLink *inlink = ctx->inputs[0];
1549 AVFrame *master, *second, *out;
1552 ret = ff_framesync_dualinput_get(fs, &master, &second);
1556 return ff_filter_frame(ctx->outputs[0], master);
1557 if (lut3d->clut_float)
1558 update_clut_float(ctx->priv, second);
1559 else if (lut3d->clut_planar)
1560 update_clut_planar(ctx->priv, second);
1562 update_clut_packed(ctx->priv, second);
1563 out = apply_lut(inlink, master);
1564 return ff_filter_frame(ctx->outputs[0], out);
1567 static av_cold int haldclut_init(AVFilterContext *ctx)
1569 LUT3DContext *lut3d = ctx->priv;
1570 lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1571 lut3d->fs.on_event = update_apply_clut;
1575 static av_cold void haldclut_uninit(AVFilterContext *ctx)
1577 LUT3DContext *lut3d = ctx->priv;
1578 ff_framesync_uninit(&lut3d->fs);
1579 av_freep(&lut3d->lut);
1582 static const AVOption haldclut_options[] = {
1586 FRAMESYNC_DEFINE_CLASS(haldclut, LUT3DContext, fs);
1588 static const AVFilterPad haldclut_inputs[] = {
1591 .type = AVMEDIA_TYPE_VIDEO,
1592 .config_props = config_input,
1595 .type = AVMEDIA_TYPE_VIDEO,
1596 .config_props = config_clut,
1601 static const AVFilterPad haldclut_outputs[] = {
1604 .type = AVMEDIA_TYPE_VIDEO,
1605 .config_props = config_output,
1610 AVFilter ff_vf_haldclut = {
1612 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
1613 .priv_size = sizeof(LUT3DContext),
1614 .preinit = haldclut_framesync_preinit,
1615 .init = haldclut_init,
1616 .uninit = haldclut_uninit,
1617 .query_formats = query_formats,
1618 .activate = activate,
1619 .inputs = haldclut_inputs,
1620 .outputs = haldclut_outputs,
1621 .priv_class = &haldclut_class,
1622 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
1626 #if CONFIG_LUT1D_FILTER
1628 enum interp_1d_mode {
1629 INTERPOLATE_1D_NEAREST,
1630 INTERPOLATE_1D_LINEAR,
1631 INTERPOLATE_1D_CUBIC,
1632 INTERPOLATE_1D_COSINE,
1633 INTERPOLATE_1D_SPLINE,
1637 #define MAX_1D_LEVEL 65536
1639 typedef struct LUT1DContext {
1640 const AVClass *class;
1642 int interpolation; ///<interp_1d_mode
1643 struct rgbvec scale;
1644 uint8_t rgba_map[4];
1646 float lut[3][MAX_1D_LEVEL];
1648 avfilter_action_func *interp;
1652 #define OFFSET(x) offsetof(LUT1DContext, x)
1654 static void set_identity_matrix_1d(LUT1DContext *lut1d, int size)
1656 const float c = 1. / (size - 1);
1659 lut1d->lutsize = size;
1660 for (i = 0; i < size; i++) {
1661 lut1d->lut[0][i] = i * c;
1662 lut1d->lut[1][i] = i * c;
1663 lut1d->lut[2][i] = i * c;
1667 static int parse_cinespace_1d(AVFilterContext *ctx, FILE *f)
1669 LUT1DContext *lut1d = ctx->priv;
1670 char line[MAX_LINE_SIZE];
1671 float in_min[3] = {0.0, 0.0, 0.0};
1672 float in_max[3] = {1.0, 1.0, 1.0};
1673 float out_min[3] = {0.0, 0.0, 0.0};
1674 float out_max[3] = {1.0, 1.0, 1.0};
1675 int inside_metadata = 0, size;
1677 NEXT_LINE(skip_line(line));
1678 if (strncmp(line, "CSPLUTV100", 10)) {
1679 av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n");
1680 return AVERROR(EINVAL);
1683 NEXT_LINE(skip_line(line));
1684 if (strncmp(line, "1D", 2)) {
1685 av_log(ctx, AV_LOG_ERROR, "Not 1D LUT format\n");
1686 return AVERROR(EINVAL);
1690 NEXT_LINE(skip_line(line));
1692 if (!strncmp(line, "BEGIN METADATA", 14)) {
1693 inside_metadata = 1;
1696 if (!strncmp(line, "END METADATA", 12)) {
1697 inside_metadata = 0;
1700 if (inside_metadata == 0) {
1701 for (int i = 0; i < 3; i++) {
1702 int npoints = strtol(line, NULL, 0);
1705 av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n");
1706 return AVERROR_PATCHWELCOME;
1709 NEXT_LINE(skip_line(line));
1710 if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2)
1711 return AVERROR_INVALIDDATA;
1712 NEXT_LINE(skip_line(line));
1713 if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2)
1714 return AVERROR_INVALIDDATA;
1715 NEXT_LINE(skip_line(line));
1718 size = strtol(line, NULL, 0);
1720 if (size < 2 || size > MAX_1D_LEVEL) {
1721 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1722 return AVERROR(EINVAL);
1725 lut1d->lutsize = size;
1727 for (int i = 0; i < size; i++) {
1728 NEXT_LINE(skip_line(line));
1729 if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1730 return AVERROR_INVALIDDATA;
1731 lut1d->lut[0][i] *= out_max[0] - out_min[0];
1732 lut1d->lut[1][i] *= out_max[1] - out_min[1];
1733 lut1d->lut[2][i] *= out_max[2] - out_min[2];
1740 lut1d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f);
1741 lut1d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f);
1742 lut1d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f);
1747 static int parse_cube_1d(AVFilterContext *ctx, FILE *f)
1749 LUT1DContext *lut1d = ctx->priv;
1750 char line[MAX_LINE_SIZE];
1751 float min[3] = {0.0, 0.0, 0.0};
1752 float max[3] = {1.0, 1.0, 1.0};
1754 while (fgets(line, sizeof(line), f)) {
1755 if (!strncmp(line, "LUT_1D_SIZE", 11)) {
1756 const int size = strtol(line + 12, NULL, 0);
1759 if (size < 2 || size > MAX_1D_LEVEL) {
1760 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1761 return AVERROR(EINVAL);
1763 lut1d->lutsize = size;
1764 for (i = 0; i < size; i++) {
1768 if (!strncmp(line, "DOMAIN_", 7)) {
1770 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
1771 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
1773 return AVERROR_INVALIDDATA;
1774 av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
1775 av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
1776 min[0], min[1], min[2], max[0], max[1], max[2]);
1778 } else if (!strncmp(line, "LUT_1D_INPUT_RANGE ", 19)) {
1779 av_sscanf(line + 19, "%f %f", min, max);
1780 min[1] = min[2] = min[0];
1781 max[1] = max[2] = max[0];
1783 } else if (!strncmp(line, "TITLE", 5)) {
1786 } while (skip_line(line));
1787 if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1788 return AVERROR_INVALIDDATA;
1794 lut1d->scale.r = av_clipf(1. / (max[0] - min[0]), 0.f, 1.f);
1795 lut1d->scale.g = av_clipf(1. / (max[1] - min[1]), 0.f, 1.f);
1796 lut1d->scale.b = av_clipf(1. / (max[2] - min[2]), 0.f, 1.f);
1801 static const AVOption lut1d_options[] = {
1802 { "file", "set 1D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = TFLAGS },
1803 { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_1D_LINEAR}, 0, NB_INTERP_1D_MODE-1, TFLAGS, "interp_mode" },
1804 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_NEAREST}, 0, 0, TFLAGS, "interp_mode" },
1805 { "linear", "use values from the linear interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_LINEAR}, 0, 0, TFLAGS, "interp_mode" },
1806 { "cosine", "use values from the cosine interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_COSINE}, 0, 0, TFLAGS, "interp_mode" },
1807 { "cubic", "use values from the cubic interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_CUBIC}, 0, 0, TFLAGS, "interp_mode" },
1808 { "spline", "use values from the spline interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_SPLINE}, 0, 0, TFLAGS, "interp_mode" },
1812 AVFILTER_DEFINE_CLASS(lut1d);
1814 static inline float interp_1d_nearest(const LUT1DContext *lut1d,
1815 int idx, const float s)
1817 return lut1d->lut[idx][NEAR(s)];
1820 #define NEXT1D(x) (FFMIN((int)(x) + 1, lut1d->lutsize - 1))
1822 static inline float interp_1d_linear(const LUT1DContext *lut1d,
1823 int idx, const float s)
1825 const int prev = PREV(s);
1826 const int next = NEXT1D(s);
1827 const float d = s - prev;
1828 const float p = lut1d->lut[idx][prev];
1829 const float n = lut1d->lut[idx][next];
1831 return lerpf(p, n, d);
1834 static inline float interp_1d_cosine(const LUT1DContext *lut1d,
1835 int idx, const float s)
1837 const int prev = PREV(s);
1838 const int next = NEXT1D(s);
1839 const float d = s - prev;
1840 const float p = lut1d->lut[idx][prev];
1841 const float n = lut1d->lut[idx][next];
1842 const float m = (1.f - cosf(d * M_PI)) * .5f;
1844 return lerpf(p, n, m);
1847 static inline float interp_1d_cubic(const LUT1DContext *lut1d,
1848 int idx, const float s)
1850 const int prev = PREV(s);
1851 const int next = NEXT1D(s);
1852 const float mu = s - prev;
1853 float a0, a1, a2, a3, mu2;
1855 float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1856 float y1 = lut1d->lut[idx][prev];
1857 float y2 = lut1d->lut[idx][next];
1858 float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1862 a0 = y3 - y2 - y0 + y1;
1867 return a0 * mu * mu2 + a1 * mu2 + a2 * mu + a3;
1870 static inline float interp_1d_spline(const LUT1DContext *lut1d,
1871 int idx, const float s)
1873 const int prev = PREV(s);
1874 const int next = NEXT1D(s);
1875 const float x = s - prev;
1876 float c0, c1, c2, c3;
1878 float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1879 float y1 = lut1d->lut[idx][prev];
1880 float y2 = lut1d->lut[idx][next];
1881 float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1884 c1 = .5f * (y2 - y0);
1885 c2 = y0 - 2.5f * y1 + 2.f * y2 - .5f * y3;
1886 c3 = .5f * (y3 - y0) + 1.5f * (y1 - y2);
1888 return ((c3 * x + c2) * x + c1) * x + c0;
1891 #define DEFINE_INTERP_FUNC_PLANAR_1D(name, nbits, depth) \
1892 static int interp_1d_##nbits##_##name##_p##depth(AVFilterContext *ctx, \
1893 void *arg, int jobnr, \
1897 const LUT1DContext *lut1d = ctx->priv; \
1898 const ThreadData *td = arg; \
1899 const AVFrame *in = td->in; \
1900 const AVFrame *out = td->out; \
1901 const int direct = out == in; \
1902 const int slice_start = (in->height * jobnr ) / nb_jobs; \
1903 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
1904 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
1905 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
1906 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
1907 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
1908 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
1909 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
1910 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
1911 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
1912 const float factor = (1 << depth) - 1; \
1913 const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \
1914 const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \
1915 const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \
1917 for (y = slice_start; y < slice_end; y++) { \
1918 uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
1919 uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
1920 uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
1921 uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
1922 const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
1923 const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
1924 const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
1925 const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
1926 for (x = 0; x < in->width; x++) { \
1927 float r = srcr[x] * scale_r; \
1928 float g = srcg[x] * scale_g; \
1929 float b = srcb[x] * scale_b; \
1930 r = interp_1d_##name(lut1d, 0, r); \
1931 g = interp_1d_##name(lut1d, 1, g); \
1932 b = interp_1d_##name(lut1d, 2, b); \
1933 dstr[x] = av_clip_uintp2(r * factor, depth); \
1934 dstg[x] = av_clip_uintp2(g * factor, depth); \
1935 dstb[x] = av_clip_uintp2(b * factor, depth); \
1936 if (!direct && in->linesize[3]) \
1937 dsta[x] = srca[x]; \
1939 grow += out->linesize[0]; \
1940 brow += out->linesize[1]; \
1941 rrow += out->linesize[2]; \
1942 arow += out->linesize[3]; \
1943 srcgrow += in->linesize[0]; \
1944 srcbrow += in->linesize[1]; \
1945 srcrrow += in->linesize[2]; \
1946 srcarow += in->linesize[3]; \
1951 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 8, 8)
1952 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 8, 8)
1953 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 8, 8)
1954 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 8, 8)
1955 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 8, 8)
1957 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 9)
1958 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 9)
1959 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 9)
1960 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 9)
1961 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 9)
1963 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 10)
1964 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 10)
1965 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 10)
1966 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 10)
1967 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 10)
1969 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 12)
1970 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 12)
1971 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 12)
1972 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 12)
1973 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 12)
1975 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 14)
1976 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 14)
1977 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 14)
1978 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 14)
1979 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 14)
1981 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 16)
1982 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 16)
1983 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 16)
1984 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 16)
1985 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 16)
1987 #define DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(name, depth) \
1988 static int interp_1d_##name##_pf##depth(AVFilterContext *ctx, \
1989 void *arg, int jobnr, \
1993 const LUT1DContext *lut1d = ctx->priv; \
1994 const ThreadData *td = arg; \
1995 const AVFrame *in = td->in; \
1996 const AVFrame *out = td->out; \
1997 const int direct = out == in; \
1998 const int slice_start = (in->height * jobnr ) / nb_jobs; \
1999 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
2000 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
2001 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
2002 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
2003 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
2004 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
2005 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
2006 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
2007 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
2008 const float lutsize = lut1d->lutsize - 1; \
2009 const float scale_r = lut1d->scale.r * lutsize; \
2010 const float scale_g = lut1d->scale.g * lutsize; \
2011 const float scale_b = lut1d->scale.b * lutsize; \
2013 for (y = slice_start; y < slice_end; y++) { \
2014 float *dstg = (float *)grow; \
2015 float *dstb = (float *)brow; \
2016 float *dstr = (float *)rrow; \
2017 float *dsta = (float *)arow; \
2018 const float *srcg = (const float *)srcgrow; \
2019 const float *srcb = (const float *)srcbrow; \
2020 const float *srcr = (const float *)srcrrow; \
2021 const float *srca = (const float *)srcarow; \
2022 for (x = 0; x < in->width; x++) { \
2023 float r = av_clipf(sanitizef(srcr[x]) * scale_r, 0.0f, lutsize); \
2024 float g = av_clipf(sanitizef(srcg[x]) * scale_g, 0.0f, lutsize); \
2025 float b = av_clipf(sanitizef(srcb[x]) * scale_b, 0.0f, lutsize); \
2026 r = interp_1d_##name(lut1d, 0, r); \
2027 g = interp_1d_##name(lut1d, 1, g); \
2028 b = interp_1d_##name(lut1d, 2, b); \
2032 if (!direct && in->linesize[3]) \
2033 dsta[x] = srca[x]; \
2035 grow += out->linesize[0]; \
2036 brow += out->linesize[1]; \
2037 rrow += out->linesize[2]; \
2038 arow += out->linesize[3]; \
2039 srcgrow += in->linesize[0]; \
2040 srcbrow += in->linesize[1]; \
2041 srcrrow += in->linesize[2]; \
2042 srcarow += in->linesize[3]; \
2047 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(nearest, 32)
2048 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(linear, 32)
2049 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cosine, 32)
2050 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cubic, 32)
2051 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(spline, 32)
2053 #define DEFINE_INTERP_FUNC_1D(name, nbits) \
2054 static int interp_1d_##nbits##_##name(AVFilterContext *ctx, void *arg, \
2055 int jobnr, int nb_jobs) \
2058 const LUT1DContext *lut1d = ctx->priv; \
2059 const ThreadData *td = arg; \
2060 const AVFrame *in = td->in; \
2061 const AVFrame *out = td->out; \
2062 const int direct = out == in; \
2063 const int step = lut1d->step; \
2064 const uint8_t r = lut1d->rgba_map[R]; \
2065 const uint8_t g = lut1d->rgba_map[G]; \
2066 const uint8_t b = lut1d->rgba_map[B]; \
2067 const uint8_t a = lut1d->rgba_map[A]; \
2068 const int slice_start = (in->height * jobnr ) / nb_jobs; \
2069 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
2070 uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
2071 const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
2072 const float factor = (1 << nbits) - 1; \
2073 const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \
2074 const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \
2075 const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \
2077 for (y = slice_start; y < slice_end; y++) { \
2078 uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
2079 const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
2080 for (x = 0; x < in->width * step; x += step) { \
2081 float rr = src[x + r] * scale_r; \
2082 float gg = src[x + g] * scale_g; \
2083 float bb = src[x + b] * scale_b; \
2084 rr = interp_1d_##name(lut1d, 0, rr); \
2085 gg = interp_1d_##name(lut1d, 1, gg); \
2086 bb = interp_1d_##name(lut1d, 2, bb); \
2087 dst[x + r] = av_clip_uint##nbits(rr * factor); \
2088 dst[x + g] = av_clip_uint##nbits(gg * factor); \
2089 dst[x + b] = av_clip_uint##nbits(bb * factor); \
2090 if (!direct && step == 4) \
2091 dst[x + a] = src[x + a]; \
2093 dstrow += out->linesize[0]; \
2094 srcrow += in ->linesize[0]; \
2099 DEFINE_INTERP_FUNC_1D(nearest, 8)
2100 DEFINE_INTERP_FUNC_1D(linear, 8)
2101 DEFINE_INTERP_FUNC_1D(cosine, 8)
2102 DEFINE_INTERP_FUNC_1D(cubic, 8)
2103 DEFINE_INTERP_FUNC_1D(spline, 8)
2105 DEFINE_INTERP_FUNC_1D(nearest, 16)
2106 DEFINE_INTERP_FUNC_1D(linear, 16)
2107 DEFINE_INTERP_FUNC_1D(cosine, 16)
2108 DEFINE_INTERP_FUNC_1D(cubic, 16)
2109 DEFINE_INTERP_FUNC_1D(spline, 16)
2111 static int config_input_1d(AVFilterLink *inlink)
2113 int depth, is16bit, isfloat, planar;
2114 LUT1DContext *lut1d = inlink->dst->priv;
2115 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
2117 depth = desc->comp[0].depth;
2118 is16bit = desc->comp[0].depth > 8;
2119 planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR;
2120 isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
2121 ff_fill_rgba_map(lut1d->rgba_map, inlink->format);
2122 lut1d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
2124 #define SET_FUNC_1D(name) do { \
2125 if (planar && !isfloat) { \
2127 case 8: lut1d->interp = interp_1d_8_##name##_p8; break; \
2128 case 9: lut1d->interp = interp_1d_16_##name##_p9; break; \
2129 case 10: lut1d->interp = interp_1d_16_##name##_p10; break; \
2130 case 12: lut1d->interp = interp_1d_16_##name##_p12; break; \
2131 case 14: lut1d->interp = interp_1d_16_##name##_p14; break; \
2132 case 16: lut1d->interp = interp_1d_16_##name##_p16; break; \
2134 } else if (isfloat) { lut1d->interp = interp_1d_##name##_pf32; \
2135 } else if (is16bit) { lut1d->interp = interp_1d_16_##name; \
2136 } else { lut1d->interp = interp_1d_8_##name; } \
2139 switch (lut1d->interpolation) {
2140 case INTERPOLATE_1D_NEAREST: SET_FUNC_1D(nearest); break;
2141 case INTERPOLATE_1D_LINEAR: SET_FUNC_1D(linear); break;
2142 case INTERPOLATE_1D_COSINE: SET_FUNC_1D(cosine); break;
2143 case INTERPOLATE_1D_CUBIC: SET_FUNC_1D(cubic); break;
2144 case INTERPOLATE_1D_SPLINE: SET_FUNC_1D(spline); break;
2152 static av_cold int lut1d_init(AVFilterContext *ctx)
2157 LUT1DContext *lut1d = ctx->priv;
2159 lut1d->scale.r = lut1d->scale.g = lut1d->scale.b = 1.f;
2162 set_identity_matrix_1d(lut1d, 32);
2166 f = av_fopen_utf8(lut1d->file, "r");
2168 ret = AVERROR(errno);
2169 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut1d->file, av_err2str(ret));
2173 ext = strrchr(lut1d->file, '.');
2175 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
2176 ret = AVERROR_INVALIDDATA;
2181 if (!av_strcasecmp(ext, "cube") || !av_strcasecmp(ext, "1dlut")) {
2182 ret = parse_cube_1d(ctx, f);
2183 } else if (!av_strcasecmp(ext, "csp")) {
2184 ret = parse_cinespace_1d(ctx, f);
2186 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
2187 ret = AVERROR(EINVAL);
2190 if (!ret && !lut1d->lutsize) {
2191 av_log(ctx, AV_LOG_ERROR, "1D LUT is empty\n");
2192 ret = AVERROR_INVALIDDATA;
2200 static AVFrame *apply_1d_lut(AVFilterLink *inlink, AVFrame *in)
2202 AVFilterContext *ctx = inlink->dst;
2203 LUT1DContext *lut1d = ctx->priv;
2204 AVFilterLink *outlink = inlink->dst->outputs[0];
2208 if (av_frame_is_writable(in)) {
2211 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
2216 av_frame_copy_props(out, in);
2221 ctx->internal->execute(ctx, lut1d->interp, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
2229 static int filter_frame_1d(AVFilterLink *inlink, AVFrame *in)
2231 AVFilterLink *outlink = inlink->dst->outputs[0];
2232 AVFrame *out = apply_1d_lut(inlink, in);
2234 return AVERROR(ENOMEM);
2235 return ff_filter_frame(outlink, out);
2238 static int lut1d_process_command(AVFilterContext *ctx, const char *cmd, const char *args,
2239 char *res, int res_len, int flags)
2241 LUT1DContext *lut1d = ctx->priv;
2244 ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
2248 ret = lut1d_init(ctx);
2250 set_identity_matrix_1d(lut1d, 32);
2253 return config_input_1d(ctx->inputs[0]);
2256 static const AVFilterPad lut1d_inputs[] = {
2259 .type = AVMEDIA_TYPE_VIDEO,
2260 .filter_frame = filter_frame_1d,
2261 .config_props = config_input_1d,
2266 static const AVFilterPad lut1d_outputs[] = {
2269 .type = AVMEDIA_TYPE_VIDEO,
2274 AVFilter ff_vf_lut1d = {
2276 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 1D LUT."),
2277 .priv_size = sizeof(LUT1DContext),
2279 .query_formats = query_formats,
2280 .inputs = lut1d_inputs,
2281 .outputs = lut1d_outputs,
2282 .priv_class = &lut1d_class,
2283 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
2284 .process_command = lut1d_process_command,