1 /*
2 * Copyright (c) 2013 Clément Bœsch
3 * Copyright (c) 2018 Paul B Mahol
4 *
5 * This file is part of FFmpeg.
6 *
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.
11 *
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.
16 *
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
20 */
21
22 /**
23 * @file
24 * 3D Lookup table filter
25 */
26
27 #include "float.h"
28
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"
36 #include "avfilter.h"
37 #include "drawutils.h"
38 #include "formats.h"
39 #include "framesync.h"
40 #include "internal.h"
41 #include "video.h"
42
43 #define R 0
44 #define G 1
45 #define B 2
46 #define A 3
47
48 enum interp_mode {
49 INTERPOLATE_NEAREST,
50 INTERPOLATE_TRILINEAR,
51 INTERPOLATE_TETRAHEDRAL,
52 INTERPOLATE_PYRAMID,
53 INTERPOLATE_PRISM,
54 NB_INTERP_MODE
55 };
56
57 struct rgbvec {
58 float r, g, b;
59 };
60
61 /* 3D LUT don't often go up to level 32, but it is common to have a Hald CLUT
62 * of 512x512 (64x64x64) */
63 #define MAX_LEVEL 256
64 #define PRELUT_SIZE 65536
65
66 typedef struct Lut3DPreLut {
67 int size;
68 float min[3];
69 float max[3];
70 float scale[3];
71 float* lut[3];
72 } Lut3DPreLut;
73
74 typedef struct LUT3DContext {
75 const AVClass *class;
76 int interpolation; ///<interp_mode
77 char *file;
78 uint8_t rgba_map[4];
79 int step;
80 avfilter_action_func *interp;
81 struct rgbvec scale;
82 struct rgbvec *lut;
83 int lutsize;
84 int lutsize2;
85 Lut3DPreLut prelut;
86 #if CONFIG_HALDCLUT_FILTER
87 uint8_t clut_rgba_map[4];
88 int clut_step;
89 int clut_bits;
90 int clut_planar;
91 int clut_float;
92 int clut_width;
93 FFFrameSync fs;
94 #endif
95 } LUT3DContext;
96
97 typedef struct ThreadData {
98 AVFrame *in, *out;
99 } ThreadData;
100
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, TFLAGS, "interp_mode" }, \
106 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_NEAREST}, 0, 0, TFLAGS, "interp_mode" }, \
107 { "trilinear", "interpolate values using the 8 points defining a cube", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TRILINEAR}, 0, 0, TFLAGS, "interp_mode" }, \
108 { "tetrahedral", "interpolate values using a tetrahedron", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, 0, TFLAGS, "interp_mode" }, \
109 { "pyramid", "interpolate values using a pyramid", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PYRAMID}, 0, 0, TFLAGS, "interp_mode" }, \
110 { "prism", "interpolate values using a prism", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PRISM}, 0, 0, TFLAGS, "interp_mode" }, \
111 { NULL }
112
113 #define EXPONENT_MASK 0x7F800000
114 #define MANTISSA_MASK 0x007FFFFF
115 #define SIGN_MASK 0x80000000
116
sanitizef(float f)117 static inline float sanitizef(float f)
118 {
119 union av_intfloat32 t;
120 t.f = f;
121
122 if ((t.i & EXPONENT_MASK) == EXPONENT_MASK) {
123 if ((t.i & MANTISSA_MASK) != 0) {
124 // NAN
125 return 0.0f;
126 } else if (t.i & SIGN_MASK) {
127 // -INF
128 return -FLT_MAX;
129 } else {
130 // +INF
131 return FLT_MAX;
132 }
133 }
134 return f;
135 }
136
lerpf(float v0,float v1,float f)137 static inline float lerpf(float v0, float v1, float f)
138 {
139 return v0 + (v1 - v0) * f;
140 }
141
lerp(const struct rgbvec * v0,const struct rgbvec * v1,float f)142 static inline struct rgbvec lerp(const struct rgbvec *v0, const struct rgbvec *v1, float f)
143 {
144 struct rgbvec v = {
145 lerpf(v0->r, v1->r, f), lerpf(v0->g, v1->g, f), lerpf(v0->b, v1->b, f)
146 };
147 return v;
148 }
149
150 #define NEAR(x) ((int)((x) + .5))
151 #define PREV(x) ((int)(x))
152 #define NEXT(x) (FFMIN((int)(x) + 1, lut3d->lutsize - 1))
153
154 /**
155 * Get the nearest defined point
156 */
interp_nearest(const LUT3DContext * lut3d,const struct rgbvec * s)157 static inline struct rgbvec interp_nearest(const LUT3DContext *lut3d,
158 const struct rgbvec *s)
159 {
160 return lut3d->lut[NEAR(s->r) * lut3d->lutsize2 + NEAR(s->g) * lut3d->lutsize + NEAR(s->b)];
161 }
162
163 /**
164 * Interpolate using the 8 vertices of a cube
165 * @see https://en.wikipedia.org/wiki/Trilinear_interpolation
166 */
interp_trilinear(const LUT3DContext * lut3d,const struct rgbvec * s)167 static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d,
168 const struct rgbvec *s)
169 {
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);
190 return c;
191 }
192
interp_pyramid(const LUT3DContext * lut3d,const struct rgbvec * s)193 static inline struct rgbvec interp_pyramid(const LUT3DContext *lut3d,
194 const struct rgbvec *s)
195 {
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]];
203 struct rgbvec c;
204
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]];
209
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]];
220
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;
227 } else {
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]];
231
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;
238 }
239
240 return c;
241 }
242
interp_prism(const LUT3DContext * lut3d,const struct rgbvec * s)243 static inline struct rgbvec interp_prism(const LUT3DContext *lut3d,
244 const struct rgbvec *s)
245 {
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]];
255 struct rgbvec c;
256
257 if (d.b > d.r) {
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]];
260
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;
270 } else {
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]];
273
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;
283 }
284
285 return c;
286 }
287
288 /**
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
291 */
interp_tetrahedral(const LUT3DContext * lut3d,const struct rgbvec * s)292 static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d,
293 const struct rgbvec *s)
294 {
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]];
302 struct rgbvec c;
303 if (d.r > d.g) {
304 if (d.g > d.b) {
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;
316 } else {
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;
322 }
323 } else {
324 if (d.b > d.g) {
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;
336 } else {
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;
342 }
343 }
344 return c;
345 }
346
prelut_interp_1d_linear(const Lut3DPreLut * prelut,int idx,const float s)347 static inline float prelut_interp_1d_linear(const Lut3DPreLut *prelut,
348 int idx, const float s)
349 {
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);
359 }
360
apply_prelut(const Lut3DPreLut * prelut,const struct rgbvec * s)361 static inline struct rgbvec apply_prelut(const Lut3DPreLut *prelut,
362 const struct rgbvec *s)
363 {
364 struct rgbvec c;
365
366 if (prelut->size <= 0)
367 return *s;
368
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);
372 return c;
373 }
374
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) \
377 { \
378 int x, y; \
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; \
400 \
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, \
412 srcg[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]) \
423 dsta[x] = srca[x]; \
424 } \
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]; \
433 } \
434 return 0; \
435 }
436
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)
442
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)
448
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)
454
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)
460
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)
466
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)
472
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) \
475 { \
476 int x, y; \
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; \
497 \
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); \
516 dstr[x] = vec.r; \
517 dstg[x] = vec.g; \
518 dstb[x] = vec.b; \
519 if (!direct && in->linesize[3]) \
520 dsta[x] = srca[x]; \
521 } \
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]; \
530 } \
531 return 0; \
532 }
533
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)
539
540 #define DEFINE_INTERP_FUNC(name, nbits) \
541 static int interp_##nbits##_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
542 { \
543 int x, y; \
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; \
564 \
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]; \
582 } \
583 dstrow += out->linesize[0]; \
584 srcrow += in ->linesize[0]; \
585 } \
586 return 0; \
587 }
588
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)
594
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)
600
601 #define MAX_LINE_SIZE 512
602
skip_line(const char * p)603 static int skip_line(const char *p)
604 {
605 while (*p && av_isspace(*p))
606 p++;
607 return !*p || *p == '#';
608 }
609
fget_next_word(char * dst,int max,FILE * f)610 static char* fget_next_word(char* dst, int max, FILE* f)
611 {
612 int c;
613 char *p = dst;
614
615 /* for null */
616 max--;
617 /* skip until next non whitespace char */
618 while ((c = fgetc(f)) != EOF) {
619 if (av_isspace(c))
620 continue;
621
622 *p++ = c;
623 max--;
624 break;
625 }
626
627 /* get max bytes or up until next whitespace char */
628 for (; max > 0; max--) {
629 if ((c = fgetc(f)) == EOF)
630 break;
631
632 if (av_isspace(c))
633 break;
634
635 *p++ = c;
636 }
637
638 *p = 0;
639 if (p == dst)
640 return NULL;
641 return p;
642 }
643
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; \
648 } \
649 } while (loop_cond)
650
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; \
655 goto label; \
656 } \
657 } while (loop_cond)
658
allocate_3dlut(AVFilterContext * ctx,int lutsize,int prelut)659 static int allocate_3dlut(AVFilterContext *ctx, int lutsize, int prelut)
660 {
661 LUT3DContext *lut3d = ctx->priv;
662 int i;
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);
666 }
667
668 av_freep(&lut3d->lut);
669 lut3d->lut = av_malloc_array(lutsize * lutsize * lutsize, sizeof(*lut3d->lut));
670 if (!lut3d->lut)
671 return AVERROR(ENOMEM);
672
673 if (prelut) {
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);
680 }
681 } else {
682 lut3d->prelut.size = 0;
683 for (i = 0; i < 3; i++) {
684 av_freep(&lut3d->prelut.lut[i]);
685 }
686 }
687 lut3d->lutsize = lutsize;
688 lut3d->lutsize2 = lutsize * lutsize;
689 return 0;
690 }
691
692 /* Basically r g and b float values on each line, with a facultative 3DLUTSIZE
693 * directive; seems to be generated by Davinci */
parse_dat(AVFilterContext * ctx,FILE * f)694 static int parse_dat(AVFilterContext *ctx, FILE *f)
695 {
696 LUT3DContext *lut3d = ctx->priv;
697 char line[MAX_LINE_SIZE];
698 int ret, i, j, k, size, size2;
699
700 lut3d->lutsize = size = 33;
701 size2 = size * size;
702
703 NEXT_LINE(skip_line(line));
704 if (!strncmp(line, "3DLUTSIZE ", 10)) {
705 size = strtol(line + 10, NULL, 0);
706
707 NEXT_LINE(skip_line(line));
708 }
709
710 ret = allocate_3dlut(ctx, size, 0);
711 if (ret < 0)
712 return ret;
713
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;
722 }
723 }
724 }
725 return 0;
726 }
727
728 /* Iridas format */
parse_cube(AVFilterContext * ctx,FILE * f)729 static int parse_cube(AVFilterContext *ctx, FILE *f)
730 {
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};
735
736 while (fgets(line, sizeof(line), f)) {
737 if (!strncmp(line, "LUT_3D_SIZE", 11)) {
738 int ret, i, j, k;
739 const int size = strtol(line + 12, NULL, 0);
740 const int size2 = size * size;
741
742 ret = allocate_3dlut(ctx, size, 0);
743 if (ret < 0)
744 return ret;
745
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];
750
751 do {
752 try_again:
753 NEXT_LINE(0);
754 if (!strncmp(line, "DOMAIN_", 7)) {
755 float *vals = NULL;
756 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
757 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
758 if (!vals)
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]);
763 goto try_again;
764 } else if (!strncmp(line, "TITLE", 5)) {
765 goto try_again;
766 }
767 } while (skip_line(line));
768 if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
769 return AVERROR_INVALIDDATA;
770 }
771 }
772 }
773 break;
774 }
775 }
776
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);
780
781 return 0;
782 }
783
784 /* Assume 17x17x17 LUT with a 16-bit depth
785 * FIXME: it seems there are various 3dl formats */
parse_3dl(AVFilterContext * ctx,FILE * f)786 static int parse_3dl(AVFilterContext *ctx, FILE *f)
787 {
788 char line[MAX_LINE_SIZE];
789 LUT3DContext *lut3d = ctx->priv;
790 int ret, i, j, k;
791 const int size = 17;
792 const int size2 = 17 * 17;
793 const float scale = 16*16*16;
794
795 lut3d->lutsize = size;
796
797 ret = allocate_3dlut(ctx, size, 0);
798 if (ret < 0)
799 return ret;
800
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++) {
805 int r, g, b;
806 struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
807
808 NEXT_LINE(skip_line(line));
809 if (av_sscanf(line, "%d %d %d", &r, &g, &b) != 3)
810 return AVERROR_INVALIDDATA;
811 vec->r = r / scale;
812 vec->g = g / scale;
813 vec->b = b / scale;
814 }
815 }
816 }
817 return 0;
818 }
819
820 /* Pandora format */
parse_m3d(AVFilterContext * ctx,FILE * f)821 static int parse_m3d(AVFilterContext *ctx, FILE *f)
822 {
823 LUT3DContext *lut3d = ctx->priv;
824 float scale;
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};
828
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)) \
836 p++; \
837 switch (*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; \
841 } \
842 while (*p && !av_isspace(*p)) \
843 p++; \
844 } while (0)
845 SET_COLOR(0);
846 SET_COLOR(1);
847 SET_COLOR(2);
848 break;
849 }
850 }
851
852 if (in == -1 || out == -1) {
853 av_log(ctx, AV_LOG_ERROR, "in and out must be defined\n");
854 return AVERROR_INVALIDDATA;
855 }
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;
861 }
862 for (size = 1; size*size*size < in; size++);
863 lut3d->lutsize = size;
864 size2 = size * size;
865
866 ret = allocate_3dlut(ctx, size, 0);
867 if (ret < 0)
868 return ret;
869
870 scale = 1. / (out - 1);
871
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];
876 float val[3];
877
878 NEXT_LINE(0);
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;
884 }
885 }
886 }
887 return 0;
888 }
889
nearest_sample_index(float * data,float x,int low,int hi)890 static int nearest_sample_index(float *data, float x, int low, int hi)
891 {
892 int mid;
893 if (x < data[low])
894 return low;
895
896 if (x > data[hi])
897 return hi;
898
899 for (;;) {
900 av_assert0(x >= data[low]);
901 av_assert0(x <= data[hi]);
902 av_assert0((hi-low) > 0);
903
904 if (hi - low == 1)
905 return low;
906
907 mid = (low + hi) / 2;
908
909 if (x < data[mid])
910 hi = mid;
911 else
912 low = mid;
913 }
914
915 return 0;
916 }
917
918 #define NEXT_FLOAT_OR_GOTO(value, label) \
919 if (!fget_next_word(line, sizeof(line) ,f)) { \
920 ret = AVERROR_INVALIDDATA; \
921 goto label; \
922 } \
923 if (av_sscanf(line, "%f", &value) != 1) { \
924 ret = AVERROR_INVALIDDATA; \
925 goto label; \
926 }
927
parse_cinespace(AVFilterContext * ctx,FILE * f)928 static int parse_cinespace(AVFilterContext *ctx, FILE *f)
929 {
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;
937 int prelut = 0;
938 int ret = 0;
939
940 int prelut_sizes[3] = {0, 0, 0};
941 float *in_prelut[3] = {NULL, NULL, NULL};
942 float *out_prelut[3] = {NULL, NULL, NULL};
943
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);
948 goto end;
949 }
950
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);
955 goto end;
956 }
957
958 while (1) {
959 NEXT_LINE_OR_GOTO(skip_line(line), end);
960
961 if (!strncmp(line, "BEGIN METADATA", 14)) {
962 inside_metadata = 1;
963 continue;
964 }
965 if (!strncmp(line, "END METADATA", 12)) {
966 inside_metadata = 0;
967 continue;
968 }
969 if (inside_metadata == 0) {
970 int size_r, size_g, size_b;
971
972 for (int i = 0; i < 3; i++) {
973 int npoints = strtol(line, NULL, 0);
974
975 if (npoints > 2) {
976 float v,last;
977
978 if (npoints > PRELUT_SIZE) {
979 av_log(ctx, AV_LOG_ERROR, "Prelut size too large.\n");
980 ret = AVERROR_INVALIDDATA;
981 goto end;
982 }
983
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;
987 goto end;
988 }
989
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);
994 goto end;
995 }
996
997 prelut_sizes[i] = npoints;
998 in_min[i] = FLT_MAX;
999 in_max[i] = -FLT_MAX;
1000 out_min[i] = FLT_MAX;
1001 out_max[i] = -FLT_MAX;
1002
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);
1011 goto end;
1012 }
1013 last = v;
1014 }
1015
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;
1021 }
1022
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;
1027 goto end;
1028 }
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;
1032 goto end;
1033 }
1034
1035 } else {
1036 av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n");
1037 ret = AVERROR_PATCHWELCOME;
1038 goto end;
1039 }
1040
1041 NEXT_LINE_OR_GOTO(skip_line(line), end);
1042 }
1043
1044 if (av_sscanf(line, "%d %d %d", &size_r, &size_g, &size_b) != 3) {
1045 ret = AVERROR(EINVAL);
1046 goto end;
1047 }
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;
1051 goto end;
1052 }
1053
1054 size = size_r;
1055 size2 = size * size;
1056
1057 if (prelut_sizes[0] && prelut_sizes[1] && prelut_sizes[2])
1058 prelut = 1;
1059
1060 ret = allocate_3dlut(ctx, size, prelut);
1061 if (ret < 0)
1062 return ret;
1063
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];
1068
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;
1072 goto end;
1073 }
1074
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];
1078 }
1079 }
1080 }
1081
1082 break;
1083 }
1084 }
1085
1086 if (prelut) {
1087 for (int c = 0; c < 3; c++) {
1088
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);
1092
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;
1096
1097 int idx = nearest_sample_index(in_prelut[c], x, 0, prelut_sizes[c]-1);
1098 av_assert0(idx + 1 < prelut_sizes[c]);
1099
1100 a = out_prelut[c][idx + 0];
1101 b = out_prelut[c][idx + 1];
1102 mix = x - in_prelut[c][idx];
1103
1104 lut3d->prelut.lut[c][i] = sanitizef(lerpf(a, b, mix));
1105 }
1106 }
1107 lut3d->scale.r = 1.00f;
1108 lut3d->scale.g = 1.00f;
1109 lut3d->scale.b = 1.00f;
1110
1111 } else {
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);
1115 }
1116
1117 end:
1118 for (int c = 0; c < 3; c++) {
1119 av_freep(&in_prelut[c]);
1120 av_freep(&out_prelut[c]);
1121 }
1122 return ret;
1123 }
1124
set_identity_matrix(AVFilterContext * ctx,int size)1125 static int set_identity_matrix(AVFilterContext *ctx, int size)
1126 {
1127 LUT3DContext *lut3d = ctx->priv;
1128 int ret, i, j, k;
1129 const int size2 = size * size;
1130 const float c = 1. / (size - 1);
1131
1132 ret = allocate_3dlut(ctx, size, 0);
1133 if (ret < 0)
1134 return ret;
1135
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];
1140 vec->r = k * c;
1141 vec->g = j * c;
1142 vec->b = i * c;
1143 }
1144 }
1145 }
1146
1147 return 0;
1148 }
1149
query_formats(AVFilterContext * ctx)1150 static int query_formats(AVFilterContext *ctx)
1151 {
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,
1161 AV_PIX_FMT_GBRP9,
1162 AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRAP10,
1163 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRAP12,
1164 AV_PIX_FMT_GBRP14,
1165 AV_PIX_FMT_GBRP16, AV_PIX_FMT_GBRAP16,
1166 AV_PIX_FMT_GBRPF32, AV_PIX_FMT_GBRAPF32,
1167 AV_PIX_FMT_NONE
1168 };
1169 AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts);
1170 if (!fmts_list)
1171 return AVERROR(ENOMEM);
1172 return ff_set_common_formats(ctx, fmts_list);
1173 }
1174
config_input(AVFilterLink * inlink)1175 static int config_input(AVFilterLink *inlink)
1176 {
1177 int depth, is16bit, isfloat, planar;
1178 LUT3DContext *lut3d = inlink->dst->priv;
1179 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1180
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);
1187
1188 #define SET_FUNC(name) do { \
1189 if (planar && !isfloat) { \
1190 switch (depth) { \
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; \
1197 } \
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; } \
1201 } while (0)
1202
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;
1209 default:
1210 av_assert0(0);
1211 }
1212
1213 return 0;
1214 }
1215
apply_lut(AVFilterLink * inlink,AVFrame * in)1216 static AVFrame *apply_lut(AVFilterLink *inlink, AVFrame *in)
1217 {
1218 AVFilterContext *ctx = inlink->dst;
1219 LUT3DContext *lut3d = ctx->priv;
1220 AVFilterLink *outlink = inlink->dst->outputs[0];
1221 AVFrame *out;
1222 ThreadData td;
1223
1224 if (av_frame_is_writable(in)) {
1225 out = in;
1226 } else {
1227 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
1228 if (!out) {
1229 av_frame_free(&in);
1230 return NULL;
1231 }
1232 av_frame_copy_props(out, in);
1233 }
1234
1235 td.in = in;
1236 td.out = out;
1237 ctx->internal->execute(ctx, lut3d->interp, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
1238
1239 if (out != in)
1240 av_frame_free(&in);
1241
1242 return out;
1243 }
1244
filter_frame(AVFilterLink * inlink,AVFrame * in)1245 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
1246 {
1247 AVFilterLink *outlink = inlink->dst->outputs[0];
1248 AVFrame *out = apply_lut(inlink, in);
1249 if (!out)
1250 return AVERROR(ENOMEM);
1251 return ff_filter_frame(outlink, out);
1252 }
1253
process_command(AVFilterContext * ctx,const char * cmd,const char * args,char * res,int res_len,int flags)1254 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
1255 char *res, int res_len, int flags)
1256 {
1257 int ret;
1258
1259 ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
1260 if (ret < 0)
1261 return ret;
1262
1263 return config_input(ctx->inputs[0]);
1264 }
1265
1266 #if CONFIG_LUT3D_FILTER
1267 static const AVOption lut3d_options[] = {
1268 { "file", "set 3D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
1269 COMMON_OPTIONS
1270 };
1271
1272 AVFILTER_DEFINE_CLASS(lut3d);
1273
lut3d_init(AVFilterContext * ctx)1274 static av_cold int lut3d_init(AVFilterContext *ctx)
1275 {
1276 int ret;
1277 FILE *f;
1278 const char *ext;
1279 LUT3DContext *lut3d = ctx->priv;
1280
1281 lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1282
1283 if (!lut3d->file) {
1284 return set_identity_matrix(ctx, 32);
1285 }
1286
1287 f = av_fopen_utf8(lut3d->file, "r");
1288 if (!f) {
1289 ret = AVERROR(errno);
1290 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret));
1291 return ret;
1292 }
1293
1294 ext = strrchr(lut3d->file, '.');
1295 if (!ext) {
1296 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
1297 ret = AVERROR_INVALIDDATA;
1298 goto end;
1299 }
1300 ext++;
1301
1302 if (!av_strcasecmp(ext, "dat")) {
1303 ret = parse_dat(ctx, f);
1304 } else if (!av_strcasecmp(ext, "3dl")) {
1305 ret = parse_3dl(ctx, f);
1306 } else if (!av_strcasecmp(ext, "cube")) {
1307 ret = parse_cube(ctx, f);
1308 } else if (!av_strcasecmp(ext, "m3d")) {
1309 ret = parse_m3d(ctx, f);
1310 } else if (!av_strcasecmp(ext, "csp")) {
1311 ret = parse_cinespace(ctx, f);
1312 } else {
1313 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
1314 ret = AVERROR(EINVAL);
1315 }
1316
1317 if (!ret && !lut3d->lutsize) {
1318 av_log(ctx, AV_LOG_ERROR, "3D LUT is empty\n");
1319 ret = AVERROR_INVALIDDATA;
1320 }
1321
1322 end:
1323 fclose(f);
1324 return ret;
1325 }
1326
lut3d_uninit(AVFilterContext * ctx)1327 static av_cold void lut3d_uninit(AVFilterContext *ctx)
1328 {
1329 LUT3DContext *lut3d = ctx->priv;
1330 int i;
1331 av_freep(&lut3d->lut);
1332
1333 for (i = 0; i < 3; i++) {
1334 av_freep(&lut3d->prelut.lut[i]);
1335 }
1336 }
1337
1338 static const AVFilterPad lut3d_inputs[] = {
1339 {
1340 .name = "default",
1341 .type = AVMEDIA_TYPE_VIDEO,
1342 .filter_frame = filter_frame,
1343 .config_props = config_input,
1344 },
1345 { NULL }
1346 };
1347
1348 static const AVFilterPad lut3d_outputs[] = {
1349 {
1350 .name = "default",
1351 .type = AVMEDIA_TYPE_VIDEO,
1352 },
1353 { NULL }
1354 };
1355
1356 AVFilter ff_vf_lut3d = {
1357 .name = "lut3d",
1358 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
1359 .priv_size = sizeof(LUT3DContext),
1360 .init = lut3d_init,
1361 .uninit = lut3d_uninit,
1362 .query_formats = query_formats,
1363 .inputs = lut3d_inputs,
1364 .outputs = lut3d_outputs,
1365 .priv_class = &lut3d_class,
1366 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
1367 .process_command = process_command,
1368 };
1369 #endif
1370
1371 #if CONFIG_HALDCLUT_FILTER
1372
update_clut_packed(LUT3DContext * lut3d,const AVFrame * frame)1373 static void update_clut_packed(LUT3DContext *lut3d, const AVFrame *frame)
1374 {
1375 const uint8_t *data = frame->data[0];
1376 const int linesize = frame->linesize[0];
1377 const int w = lut3d->clut_width;
1378 const int step = lut3d->clut_step;
1379 const uint8_t *rgba_map = lut3d->clut_rgba_map;
1380 const int level = lut3d->lutsize;
1381 const int level2 = lut3d->lutsize2;
1382
1383 #define LOAD_CLUT(nbits) do { \
1384 int i, j, k, x = 0, y = 0; \
1385 \
1386 for (k = 0; k < level; k++) { \
1387 for (j = 0; j < level; j++) { \
1388 for (i = 0; i < level; i++) { \
1389 const uint##nbits##_t *src = (const uint##nbits##_t *) \
1390 (data + y*linesize + x*step); \
1391 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1392 vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \
1393 vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \
1394 vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \
1395 if (++x == w) { \
1396 x = 0; \
1397 y++; \
1398 } \
1399 } \
1400 } \
1401 } \
1402 } while (0)
1403
1404 switch (lut3d->clut_bits) {
1405 case 8: LOAD_CLUT(8); break;
1406 case 16: LOAD_CLUT(16); break;
1407 }
1408 }
1409
update_clut_planar(LUT3DContext * lut3d,const AVFrame * frame)1410 static void update_clut_planar(LUT3DContext *lut3d, const AVFrame *frame)
1411 {
1412 const uint8_t *datag = frame->data[0];
1413 const uint8_t *datab = frame->data[1];
1414 const uint8_t *datar = frame->data[2];
1415 const int glinesize = frame->linesize[0];
1416 const int blinesize = frame->linesize[1];
1417 const int rlinesize = frame->linesize[2];
1418 const int w = lut3d->clut_width;
1419 const int level = lut3d->lutsize;
1420 const int level2 = lut3d->lutsize2;
1421
1422 #define LOAD_CLUT_PLANAR(nbits, depth) do { \
1423 int i, j, k, x = 0, y = 0; \
1424 \
1425 for (k = 0; k < level; k++) { \
1426 for (j = 0; j < level; j++) { \
1427 for (i = 0; i < level; i++) { \
1428 const uint##nbits##_t *gsrc = (const uint##nbits##_t *) \
1429 (datag + y*glinesize); \
1430 const uint##nbits##_t *bsrc = (const uint##nbits##_t *) \
1431 (datab + y*blinesize); \
1432 const uint##nbits##_t *rsrc = (const uint##nbits##_t *) \
1433 (datar + y*rlinesize); \
1434 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1435 vec->r = gsrc[x] / (float)((1<<(depth)) - 1); \
1436 vec->g = bsrc[x] / (float)((1<<(depth)) - 1); \
1437 vec->b = rsrc[x] / (float)((1<<(depth)) - 1); \
1438 if (++x == w) { \
1439 x = 0; \
1440 y++; \
1441 } \
1442 } \
1443 } \
1444 } \
1445 } while (0)
1446
1447 switch (lut3d->clut_bits) {
1448 case 8: LOAD_CLUT_PLANAR(8, 8); break;
1449 case 9: LOAD_CLUT_PLANAR(16, 9); break;
1450 case 10: LOAD_CLUT_PLANAR(16, 10); break;
1451 case 12: LOAD_CLUT_PLANAR(16, 12); break;
1452 case 14: LOAD_CLUT_PLANAR(16, 14); break;
1453 case 16: LOAD_CLUT_PLANAR(16, 16); break;
1454 }
1455 }
1456
update_clut_float(LUT3DContext * lut3d,const AVFrame * frame)1457 static void update_clut_float(LUT3DContext *lut3d, const AVFrame *frame)
1458 {
1459 const uint8_t *datag = frame->data[0];
1460 const uint8_t *datab = frame->data[1];
1461 const uint8_t *datar = frame->data[2];
1462 const int glinesize = frame->linesize[0];
1463 const int blinesize = frame->linesize[1];
1464 const int rlinesize = frame->linesize[2];
1465 const int w = lut3d->clut_width;
1466 const int level = lut3d->lutsize;
1467 const int level2 = lut3d->lutsize2;
1468
1469 int i, j, k, x = 0, y = 0;
1470
1471 for (k = 0; k < level; k++) {
1472 for (j = 0; j < level; j++) {
1473 for (i = 0; i < level; i++) {
1474 const float *gsrc = (const float *)(datag + y*glinesize);
1475 const float *bsrc = (const float *)(datab + y*blinesize);
1476 const float *rsrc = (const float *)(datar + y*rlinesize);
1477 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k];
1478 vec->r = rsrc[x];
1479 vec->g = gsrc[x];
1480 vec->b = bsrc[x];
1481 if (++x == w) {
1482 x = 0;
1483 y++;
1484 }
1485 }
1486 }
1487 }
1488 }
1489
config_output(AVFilterLink * outlink)1490 static int config_output(AVFilterLink *outlink)
1491 {
1492 AVFilterContext *ctx = outlink->src;
1493 LUT3DContext *lut3d = ctx->priv;
1494 int ret;
1495
1496 ret = ff_framesync_init_dualinput(&lut3d->fs, ctx);
1497 if (ret < 0)
1498 return ret;
1499 outlink->w = ctx->inputs[0]->w;
1500 outlink->h = ctx->inputs[0]->h;
1501 outlink->time_base = ctx->inputs[0]->time_base;
1502 if ((ret = ff_framesync_configure(&lut3d->fs)) < 0)
1503 return ret;
1504 return 0;
1505 }
1506
activate(AVFilterContext * ctx)1507 static int activate(AVFilterContext *ctx)
1508 {
1509 LUT3DContext *s = ctx->priv;
1510 return ff_framesync_activate(&s->fs);
1511 }
1512
config_clut(AVFilterLink * inlink)1513 static int config_clut(AVFilterLink *inlink)
1514 {
1515 int size, level, w, h;
1516 AVFilterContext *ctx = inlink->dst;
1517 LUT3DContext *lut3d = ctx->priv;
1518 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1519
1520 av_assert0(desc);
1521
1522 lut3d->clut_bits = desc->comp[0].depth;
1523 lut3d->clut_planar = av_pix_fmt_count_planes(inlink->format) > 1;
1524 lut3d->clut_float = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
1525
1526 lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3;
1527 ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format);
1528
1529 if (inlink->w > inlink->h)
1530 av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the "
1531 "Hald CLUT will be ignored\n", inlink->w - inlink->h);
1532 else if (inlink->w < inlink->h)
1533 av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the "
1534 "Hald CLUT will be ignored\n", inlink->h - inlink->w);
1535 lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h);
1536
1537 for (level = 1; level*level*level < w; level++);
1538 size = level*level*level;
1539 if (size != w) {
1540 av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n");
1541 return AVERROR_INVALIDDATA;
1542 }
1543 av_assert0(w == h && w == size);
1544 level *= level;
1545 if (level > MAX_LEVEL) {
1546 const int max_clut_level = sqrt(MAX_LEVEL);
1547 const int max_clut_size = max_clut_level*max_clut_level*max_clut_level;
1548 av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT "
1549 "(maximum level is %d, or %dx%d CLUT)\n",
1550 max_clut_level, max_clut_size, max_clut_size);
1551 return AVERROR(EINVAL);
1552 }
1553
1554 return allocate_3dlut(ctx, level, 0);
1555 }
1556
update_apply_clut(FFFrameSync * fs)1557 static int update_apply_clut(FFFrameSync *fs)
1558 {
1559 AVFilterContext *ctx = fs->parent;
1560 LUT3DContext *lut3d = ctx->priv;
1561 AVFilterLink *inlink = ctx->inputs[0];
1562 AVFrame *master, *second, *out;
1563 int ret;
1564
1565 ret = ff_framesync_dualinput_get(fs, &master, &second);
1566 if (ret < 0)
1567 return ret;
1568 if (!second)
1569 return ff_filter_frame(ctx->outputs[0], master);
1570 if (lut3d->clut_float)
1571 update_clut_float(ctx->priv, second);
1572 else if (lut3d->clut_planar)
1573 update_clut_planar(ctx->priv, second);
1574 else
1575 update_clut_packed(ctx->priv, second);
1576 out = apply_lut(inlink, master);
1577 return ff_filter_frame(ctx->outputs[0], out);
1578 }
1579
haldclut_init(AVFilterContext * ctx)1580 static av_cold int haldclut_init(AVFilterContext *ctx)
1581 {
1582 LUT3DContext *lut3d = ctx->priv;
1583 lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1584 lut3d->fs.on_event = update_apply_clut;
1585 return 0;
1586 }
1587
haldclut_uninit(AVFilterContext * ctx)1588 static av_cold void haldclut_uninit(AVFilterContext *ctx)
1589 {
1590 LUT3DContext *lut3d = ctx->priv;
1591 ff_framesync_uninit(&lut3d->fs);
1592 av_freep(&lut3d->lut);
1593 }
1594
1595 static const AVOption haldclut_options[] = {
1596 COMMON_OPTIONS
1597 };
1598
1599 FRAMESYNC_DEFINE_CLASS(haldclut, LUT3DContext, fs);
1600
1601 static const AVFilterPad haldclut_inputs[] = {
1602 {
1603 .name = "main",
1604 .type = AVMEDIA_TYPE_VIDEO,
1605 .config_props = config_input,
1606 },{
1607 .name = "clut",
1608 .type = AVMEDIA_TYPE_VIDEO,
1609 .config_props = config_clut,
1610 },
1611 { NULL }
1612 };
1613
1614 static const AVFilterPad haldclut_outputs[] = {
1615 {
1616 .name = "default",
1617 .type = AVMEDIA_TYPE_VIDEO,
1618 .config_props = config_output,
1619 },
1620 { NULL }
1621 };
1622
1623 AVFilter ff_vf_haldclut = {
1624 .name = "haldclut",
1625 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
1626 .priv_size = sizeof(LUT3DContext),
1627 .preinit = haldclut_framesync_preinit,
1628 .init = haldclut_init,
1629 .uninit = haldclut_uninit,
1630 .query_formats = query_formats,
1631 .activate = activate,
1632 .inputs = haldclut_inputs,
1633 .outputs = haldclut_outputs,
1634 .priv_class = &haldclut_class,
1635 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
1636 .process_command = process_command,
1637 };
1638 #endif
1639
1640 #if CONFIG_LUT1D_FILTER
1641
1642 enum interp_1d_mode {
1643 INTERPOLATE_1D_NEAREST,
1644 INTERPOLATE_1D_LINEAR,
1645 INTERPOLATE_1D_CUBIC,
1646 INTERPOLATE_1D_COSINE,
1647 INTERPOLATE_1D_SPLINE,
1648 NB_INTERP_1D_MODE
1649 };
1650
1651 #define MAX_1D_LEVEL 65536
1652
1653 typedef struct LUT1DContext {
1654 const AVClass *class;
1655 char *file;
1656 int interpolation; ///<interp_1d_mode
1657 struct rgbvec scale;
1658 uint8_t rgba_map[4];
1659 int step;
1660 float lut[3][MAX_1D_LEVEL];
1661 int lutsize;
1662 avfilter_action_func *interp;
1663 } LUT1DContext;
1664
1665 #undef OFFSET
1666 #define OFFSET(x) offsetof(LUT1DContext, x)
1667
set_identity_matrix_1d(LUT1DContext * lut1d,int size)1668 static void set_identity_matrix_1d(LUT1DContext *lut1d, int size)
1669 {
1670 const float c = 1. / (size - 1);
1671 int i;
1672
1673 lut1d->lutsize = size;
1674 for (i = 0; i < size; i++) {
1675 lut1d->lut[0][i] = i * c;
1676 lut1d->lut[1][i] = i * c;
1677 lut1d->lut[2][i] = i * c;
1678 }
1679 }
1680
parse_cinespace_1d(AVFilterContext * ctx,FILE * f)1681 static int parse_cinespace_1d(AVFilterContext *ctx, FILE *f)
1682 {
1683 LUT1DContext *lut1d = ctx->priv;
1684 char line[MAX_LINE_SIZE];
1685 float in_min[3] = {0.0, 0.0, 0.0};
1686 float in_max[3] = {1.0, 1.0, 1.0};
1687 float out_min[3] = {0.0, 0.0, 0.0};
1688 float out_max[3] = {1.0, 1.0, 1.0};
1689 int inside_metadata = 0, size;
1690
1691 NEXT_LINE(skip_line(line));
1692 if (strncmp(line, "CSPLUTV100", 10)) {
1693 av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n");
1694 return AVERROR(EINVAL);
1695 }
1696
1697 NEXT_LINE(skip_line(line));
1698 if (strncmp(line, "1D", 2)) {
1699 av_log(ctx, AV_LOG_ERROR, "Not 1D LUT format\n");
1700 return AVERROR(EINVAL);
1701 }
1702
1703 while (1) {
1704 NEXT_LINE(skip_line(line));
1705
1706 if (!strncmp(line, "BEGIN METADATA", 14)) {
1707 inside_metadata = 1;
1708 continue;
1709 }
1710 if (!strncmp(line, "END METADATA", 12)) {
1711 inside_metadata = 0;
1712 continue;
1713 }
1714 if (inside_metadata == 0) {
1715 for (int i = 0; i < 3; i++) {
1716 int npoints = strtol(line, NULL, 0);
1717
1718 if (npoints != 2) {
1719 av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n");
1720 return AVERROR_PATCHWELCOME;
1721 }
1722
1723 NEXT_LINE(skip_line(line));
1724 if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2)
1725 return AVERROR_INVALIDDATA;
1726 NEXT_LINE(skip_line(line));
1727 if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2)
1728 return AVERROR_INVALIDDATA;
1729 NEXT_LINE(skip_line(line));
1730 }
1731
1732 size = strtol(line, NULL, 0);
1733
1734 if (size < 2 || size > MAX_1D_LEVEL) {
1735 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1736 return AVERROR(EINVAL);
1737 }
1738
1739 lut1d->lutsize = size;
1740
1741 for (int i = 0; i < size; i++) {
1742 NEXT_LINE(skip_line(line));
1743 if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1744 return AVERROR_INVALIDDATA;
1745 lut1d->lut[0][i] *= out_max[0] - out_min[0];
1746 lut1d->lut[1][i] *= out_max[1] - out_min[1];
1747 lut1d->lut[2][i] *= out_max[2] - out_min[2];
1748 }
1749
1750 break;
1751 }
1752 }
1753
1754 lut1d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f);
1755 lut1d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f);
1756 lut1d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f);
1757
1758 return 0;
1759 }
1760
parse_cube_1d(AVFilterContext * ctx,FILE * f)1761 static int parse_cube_1d(AVFilterContext *ctx, FILE *f)
1762 {
1763 LUT1DContext *lut1d = ctx->priv;
1764 char line[MAX_LINE_SIZE];
1765 float min[3] = {0.0, 0.0, 0.0};
1766 float max[3] = {1.0, 1.0, 1.0};
1767
1768 while (fgets(line, sizeof(line), f)) {
1769 if (!strncmp(line, "LUT_1D_SIZE", 11)) {
1770 const int size = strtol(line + 12, NULL, 0);
1771 int i;
1772
1773 if (size < 2 || size > MAX_1D_LEVEL) {
1774 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1775 return AVERROR(EINVAL);
1776 }
1777 lut1d->lutsize = size;
1778 for (i = 0; i < size; i++) {
1779 do {
1780 try_again:
1781 NEXT_LINE(0);
1782 if (!strncmp(line, "DOMAIN_", 7)) {
1783 float *vals = NULL;
1784 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
1785 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
1786 if (!vals)
1787 return AVERROR_INVALIDDATA;
1788 av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
1789 av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
1790 min[0], min[1], min[2], max[0], max[1], max[2]);
1791 goto try_again;
1792 } else if (!strncmp(line, "LUT_1D_INPUT_RANGE ", 19)) {
1793 av_sscanf(line + 19, "%f %f", min, max);
1794 min[1] = min[2] = min[0];
1795 max[1] = max[2] = max[0];
1796 goto try_again;
1797 } else if (!strncmp(line, "TITLE", 5)) {
1798 goto try_again;
1799 }
1800 } while (skip_line(line));
1801 if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1802 return AVERROR_INVALIDDATA;
1803 }
1804 break;
1805 }
1806 }
1807
1808 lut1d->scale.r = av_clipf(1. / (max[0] - min[0]), 0.f, 1.f);
1809 lut1d->scale.g = av_clipf(1. / (max[1] - min[1]), 0.f, 1.f);
1810 lut1d->scale.b = av_clipf(1. / (max[2] - min[2]), 0.f, 1.f);
1811
1812 return 0;
1813 }
1814
1815 static const AVOption lut1d_options[] = {
1816 { "file", "set 1D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = TFLAGS },
1817 { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_1D_LINEAR}, 0, NB_INTERP_1D_MODE-1, TFLAGS, "interp_mode" },
1818 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_NEAREST}, 0, 0, TFLAGS, "interp_mode" },
1819 { "linear", "use values from the linear interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_LINEAR}, 0, 0, TFLAGS, "interp_mode" },
1820 { "cosine", "use values from the cosine interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_COSINE}, 0, 0, TFLAGS, "interp_mode" },
1821 { "cubic", "use values from the cubic interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_CUBIC}, 0, 0, TFLAGS, "interp_mode" },
1822 { "spline", "use values from the spline interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_SPLINE}, 0, 0, TFLAGS, "interp_mode" },
1823 { NULL }
1824 };
1825
1826 AVFILTER_DEFINE_CLASS(lut1d);
1827
interp_1d_nearest(const LUT1DContext * lut1d,int idx,const float s)1828 static inline float interp_1d_nearest(const LUT1DContext *lut1d,
1829 int idx, const float s)
1830 {
1831 return lut1d->lut[idx][NEAR(s)];
1832 }
1833
1834 #define NEXT1D(x) (FFMIN((int)(x) + 1, lut1d->lutsize - 1))
1835
interp_1d_linear(const LUT1DContext * lut1d,int idx,const float s)1836 static inline float interp_1d_linear(const LUT1DContext *lut1d,
1837 int idx, const float s)
1838 {
1839 const int prev = PREV(s);
1840 const int next = NEXT1D(s);
1841 const float d = s - prev;
1842 const float p = lut1d->lut[idx][prev];
1843 const float n = lut1d->lut[idx][next];
1844
1845 return lerpf(p, n, d);
1846 }
1847
interp_1d_cosine(const LUT1DContext * lut1d,int idx,const float s)1848 static inline float interp_1d_cosine(const LUT1DContext *lut1d,
1849 int idx, const float s)
1850 {
1851 const int prev = PREV(s);
1852 const int next = NEXT1D(s);
1853 const float d = s - prev;
1854 const float p = lut1d->lut[idx][prev];
1855 const float n = lut1d->lut[idx][next];
1856 const float m = (1.f - cosf(d * M_PI)) * .5f;
1857
1858 return lerpf(p, n, m);
1859 }
1860
interp_1d_cubic(const LUT1DContext * lut1d,int idx,const float s)1861 static inline float interp_1d_cubic(const LUT1DContext *lut1d,
1862 int idx, const float s)
1863 {
1864 const int prev = PREV(s);
1865 const int next = NEXT1D(s);
1866 const float mu = s - prev;
1867 float a0, a1, a2, a3, mu2;
1868
1869 float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1870 float y1 = lut1d->lut[idx][prev];
1871 float y2 = lut1d->lut[idx][next];
1872 float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1873
1874
1875 mu2 = mu * mu;
1876 a0 = y3 - y2 - y0 + y1;
1877 a1 = y0 - y1 - a0;
1878 a2 = y2 - y0;
1879 a3 = y1;
1880
1881 return a0 * mu * mu2 + a1 * mu2 + a2 * mu + a3;
1882 }
1883
interp_1d_spline(const LUT1DContext * lut1d,int idx,const float s)1884 static inline float interp_1d_spline(const LUT1DContext *lut1d,
1885 int idx, const float s)
1886 {
1887 const int prev = PREV(s);
1888 const int next = NEXT1D(s);
1889 const float x = s - prev;
1890 float c0, c1, c2, c3;
1891
1892 float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1893 float y1 = lut1d->lut[idx][prev];
1894 float y2 = lut1d->lut[idx][next];
1895 float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1896
1897 c0 = y1;
1898 c1 = .5f * (y2 - y0);
1899 c2 = y0 - 2.5f * y1 + 2.f * y2 - .5f * y3;
1900 c3 = .5f * (y3 - y0) + 1.5f * (y1 - y2);
1901
1902 return ((c3 * x + c2) * x + c1) * x + c0;
1903 }
1904
1905 #define DEFINE_INTERP_FUNC_PLANAR_1D(name, nbits, depth) \
1906 static int interp_1d_##nbits##_##name##_p##depth(AVFilterContext *ctx, \
1907 void *arg, int jobnr, \
1908 int nb_jobs) \
1909 { \
1910 int x, y; \
1911 const LUT1DContext *lut1d = ctx->priv; \
1912 const ThreadData *td = arg; \
1913 const AVFrame *in = td->in; \
1914 const AVFrame *out = td->out; \
1915 const int direct = out == in; \
1916 const int slice_start = (in->height * jobnr ) / nb_jobs; \
1917 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
1918 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
1919 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
1920 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
1921 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
1922 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
1923 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
1924 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
1925 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
1926 const float factor = (1 << depth) - 1; \
1927 const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \
1928 const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \
1929 const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \
1930 \
1931 for (y = slice_start; y < slice_end; y++) { \
1932 uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
1933 uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
1934 uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
1935 uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
1936 const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
1937 const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
1938 const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
1939 const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
1940 for (x = 0; x < in->width; x++) { \
1941 float r = srcr[x] * scale_r; \
1942 float g = srcg[x] * scale_g; \
1943 float b = srcb[x] * scale_b; \
1944 r = interp_1d_##name(lut1d, 0, r); \
1945 g = interp_1d_##name(lut1d, 1, g); \
1946 b = interp_1d_##name(lut1d, 2, b); \
1947 dstr[x] = av_clip_uintp2(r * factor, depth); \
1948 dstg[x] = av_clip_uintp2(g * factor, depth); \
1949 dstb[x] = av_clip_uintp2(b * factor, depth); \
1950 if (!direct && in->linesize[3]) \
1951 dsta[x] = srca[x]; \
1952 } \
1953 grow += out->linesize[0]; \
1954 brow += out->linesize[1]; \
1955 rrow += out->linesize[2]; \
1956 arow += out->linesize[3]; \
1957 srcgrow += in->linesize[0]; \
1958 srcbrow += in->linesize[1]; \
1959 srcrrow += in->linesize[2]; \
1960 srcarow += in->linesize[3]; \
1961 } \
1962 return 0; \
1963 }
1964
1965 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 8, 8)
1966 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 8, 8)
1967 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 8, 8)
1968 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 8, 8)
1969 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 8, 8)
1970
1971 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 9)
1972 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 9)
1973 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 9)
1974 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 9)
1975 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 9)
1976
1977 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 10)
1978 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 10)
1979 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 10)
1980 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 10)
1981 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 10)
1982
1983 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 12)
1984 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 12)
1985 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 12)
1986 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 12)
1987 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 12)
1988
1989 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 14)
1990 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 14)
1991 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 14)
1992 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 14)
1993 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 14)
1994
1995 DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 16)
1996 DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 16)
1997 DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 16)
1998 DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 16)
1999 DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 16)
2000
2001 #define DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(name, depth) \
2002 static int interp_1d_##name##_pf##depth(AVFilterContext *ctx, \
2003 void *arg, int jobnr, \
2004 int nb_jobs) \
2005 { \
2006 int x, y; \
2007 const LUT1DContext *lut1d = ctx->priv; \
2008 const ThreadData *td = arg; \
2009 const AVFrame *in = td->in; \
2010 const AVFrame *out = td->out; \
2011 const int direct = out == in; \
2012 const int slice_start = (in->height * jobnr ) / nb_jobs; \
2013 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
2014 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
2015 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
2016 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
2017 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
2018 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
2019 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
2020 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
2021 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
2022 const float lutsize = lut1d->lutsize - 1; \
2023 const float scale_r = lut1d->scale.r * lutsize; \
2024 const float scale_g = lut1d->scale.g * lutsize; \
2025 const float scale_b = lut1d->scale.b * lutsize; \
2026 \
2027 for (y = slice_start; y < slice_end; y++) { \
2028 float *dstg = (float *)grow; \
2029 float *dstb = (float *)brow; \
2030 float *dstr = (float *)rrow; \
2031 float *dsta = (float *)arow; \
2032 const float *srcg = (const float *)srcgrow; \
2033 const float *srcb = (const float *)srcbrow; \
2034 const float *srcr = (const float *)srcrrow; \
2035 const float *srca = (const float *)srcarow; \
2036 for (x = 0; x < in->width; x++) { \
2037 float r = av_clipf(sanitizef(srcr[x]) * scale_r, 0.0f, lutsize); \
2038 float g = av_clipf(sanitizef(srcg[x]) * scale_g, 0.0f, lutsize); \
2039 float b = av_clipf(sanitizef(srcb[x]) * scale_b, 0.0f, lutsize); \
2040 r = interp_1d_##name(lut1d, 0, r); \
2041 g = interp_1d_##name(lut1d, 1, g); \
2042 b = interp_1d_##name(lut1d, 2, b); \
2043 dstr[x] = r; \
2044 dstg[x] = g; \
2045 dstb[x] = b; \
2046 if (!direct && in->linesize[3]) \
2047 dsta[x] = srca[x]; \
2048 } \
2049 grow += out->linesize[0]; \
2050 brow += out->linesize[1]; \
2051 rrow += out->linesize[2]; \
2052 arow += out->linesize[3]; \
2053 srcgrow += in->linesize[0]; \
2054 srcbrow += in->linesize[1]; \
2055 srcrrow += in->linesize[2]; \
2056 srcarow += in->linesize[3]; \
2057 } \
2058 return 0; \
2059 }
2060
2061 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(nearest, 32)
2062 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(linear, 32)
2063 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cosine, 32)
2064 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cubic, 32)
2065 DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(spline, 32)
2066
2067 #define DEFINE_INTERP_FUNC_1D(name, nbits) \
2068 static int interp_1d_##nbits##_##name(AVFilterContext *ctx, void *arg, \
2069 int jobnr, int nb_jobs) \
2070 { \
2071 int x, y; \
2072 const LUT1DContext *lut1d = ctx->priv; \
2073 const ThreadData *td = arg; \
2074 const AVFrame *in = td->in; \
2075 const AVFrame *out = td->out; \
2076 const int direct = out == in; \
2077 const int step = lut1d->step; \
2078 const uint8_t r = lut1d->rgba_map[R]; \
2079 const uint8_t g = lut1d->rgba_map[G]; \
2080 const uint8_t b = lut1d->rgba_map[B]; \
2081 const uint8_t a = lut1d->rgba_map[A]; \
2082 const int slice_start = (in->height * jobnr ) / nb_jobs; \
2083 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
2084 uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
2085 const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
2086 const float factor = (1 << nbits) - 1; \
2087 const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \
2088 const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \
2089 const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \
2090 \
2091 for (y = slice_start; y < slice_end; y++) { \
2092 uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
2093 const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
2094 for (x = 0; x < in->width * step; x += step) { \
2095 float rr = src[x + r] * scale_r; \
2096 float gg = src[x + g] * scale_g; \
2097 float bb = src[x + b] * scale_b; \
2098 rr = interp_1d_##name(lut1d, 0, rr); \
2099 gg = interp_1d_##name(lut1d, 1, gg); \
2100 bb = interp_1d_##name(lut1d, 2, bb); \
2101 dst[x + r] = av_clip_uint##nbits(rr * factor); \
2102 dst[x + g] = av_clip_uint##nbits(gg * factor); \
2103 dst[x + b] = av_clip_uint##nbits(bb * factor); \
2104 if (!direct && step == 4) \
2105 dst[x + a] = src[x + a]; \
2106 } \
2107 dstrow += out->linesize[0]; \
2108 srcrow += in ->linesize[0]; \
2109 } \
2110 return 0; \
2111 }
2112
2113 DEFINE_INTERP_FUNC_1D(nearest, 8)
2114 DEFINE_INTERP_FUNC_1D(linear, 8)
2115 DEFINE_INTERP_FUNC_1D(cosine, 8)
2116 DEFINE_INTERP_FUNC_1D(cubic, 8)
2117 DEFINE_INTERP_FUNC_1D(spline, 8)
2118
2119 DEFINE_INTERP_FUNC_1D(nearest, 16)
2120 DEFINE_INTERP_FUNC_1D(linear, 16)
2121 DEFINE_INTERP_FUNC_1D(cosine, 16)
2122 DEFINE_INTERP_FUNC_1D(cubic, 16)
2123 DEFINE_INTERP_FUNC_1D(spline, 16)
2124
config_input_1d(AVFilterLink * inlink)2125 static int config_input_1d(AVFilterLink *inlink)
2126 {
2127 int depth, is16bit, isfloat, planar;
2128 LUT1DContext *lut1d = inlink->dst->priv;
2129 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
2130
2131 depth = desc->comp[0].depth;
2132 is16bit = desc->comp[0].depth > 8;
2133 planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR;
2134 isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
2135 ff_fill_rgba_map(lut1d->rgba_map, inlink->format);
2136 lut1d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
2137
2138 #define SET_FUNC_1D(name) do { \
2139 if (planar && !isfloat) { \
2140 switch (depth) { \
2141 case 8: lut1d->interp = interp_1d_8_##name##_p8; break; \
2142 case 9: lut1d->interp = interp_1d_16_##name##_p9; break; \
2143 case 10: lut1d->interp = interp_1d_16_##name##_p10; break; \
2144 case 12: lut1d->interp = interp_1d_16_##name##_p12; break; \
2145 case 14: lut1d->interp = interp_1d_16_##name##_p14; break; \
2146 case 16: lut1d->interp = interp_1d_16_##name##_p16; break; \
2147 } \
2148 } else if (isfloat) { lut1d->interp = interp_1d_##name##_pf32; \
2149 } else if (is16bit) { lut1d->interp = interp_1d_16_##name; \
2150 } else { lut1d->interp = interp_1d_8_##name; } \
2151 } while (0)
2152
2153 switch (lut1d->interpolation) {
2154 case INTERPOLATE_1D_NEAREST: SET_FUNC_1D(nearest); break;
2155 case INTERPOLATE_1D_LINEAR: SET_FUNC_1D(linear); break;
2156 case INTERPOLATE_1D_COSINE: SET_FUNC_1D(cosine); break;
2157 case INTERPOLATE_1D_CUBIC: SET_FUNC_1D(cubic); break;
2158 case INTERPOLATE_1D_SPLINE: SET_FUNC_1D(spline); break;
2159 default:
2160 av_assert0(0);
2161 }
2162
2163 return 0;
2164 }
2165
lut1d_init(AVFilterContext * ctx)2166 static av_cold int lut1d_init(AVFilterContext *ctx)
2167 {
2168 int ret;
2169 FILE *f;
2170 const char *ext;
2171 LUT1DContext *lut1d = ctx->priv;
2172
2173 lut1d->scale.r = lut1d->scale.g = lut1d->scale.b = 1.f;
2174
2175 if (!lut1d->file) {
2176 set_identity_matrix_1d(lut1d, 32);
2177 return 0;
2178 }
2179
2180 f = av_fopen_utf8(lut1d->file, "r");
2181 if (!f) {
2182 ret = AVERROR(errno);
2183 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut1d->file, av_err2str(ret));
2184 return ret;
2185 }
2186
2187 ext = strrchr(lut1d->file, '.');
2188 if (!ext) {
2189 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
2190 ret = AVERROR_INVALIDDATA;
2191 goto end;
2192 }
2193 ext++;
2194
2195 if (!av_strcasecmp(ext, "cube") || !av_strcasecmp(ext, "1dlut")) {
2196 ret = parse_cube_1d(ctx, f);
2197 } else if (!av_strcasecmp(ext, "csp")) {
2198 ret = parse_cinespace_1d(ctx, f);
2199 } else {
2200 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
2201 ret = AVERROR(EINVAL);
2202 }
2203
2204 if (!ret && !lut1d->lutsize) {
2205 av_log(ctx, AV_LOG_ERROR, "1D LUT is empty\n");
2206 ret = AVERROR_INVALIDDATA;
2207 }
2208
2209 end:
2210 fclose(f);
2211 return ret;
2212 }
2213
apply_1d_lut(AVFilterLink * inlink,AVFrame * in)2214 static AVFrame *apply_1d_lut(AVFilterLink *inlink, AVFrame *in)
2215 {
2216 AVFilterContext *ctx = inlink->dst;
2217 LUT1DContext *lut1d = ctx->priv;
2218 AVFilterLink *outlink = inlink->dst->outputs[0];
2219 AVFrame *out;
2220 ThreadData td;
2221
2222 if (av_frame_is_writable(in)) {
2223 out = in;
2224 } else {
2225 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
2226 if (!out) {
2227 av_frame_free(&in);
2228 return NULL;
2229 }
2230 av_frame_copy_props(out, in);
2231 }
2232
2233 td.in = in;
2234 td.out = out;
2235 ctx->internal->execute(ctx, lut1d->interp, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
2236
2237 if (out != in)
2238 av_frame_free(&in);
2239
2240 return out;
2241 }
2242
filter_frame_1d(AVFilterLink * inlink,AVFrame * in)2243 static int filter_frame_1d(AVFilterLink *inlink, AVFrame *in)
2244 {
2245 AVFilterLink *outlink = inlink->dst->outputs[0];
2246 AVFrame *out = apply_1d_lut(inlink, in);
2247 if (!out)
2248 return AVERROR(ENOMEM);
2249 return ff_filter_frame(outlink, out);
2250 }
2251
lut1d_process_command(AVFilterContext * ctx,const char * cmd,const char * args,char * res,int res_len,int flags)2252 static int lut1d_process_command(AVFilterContext *ctx, const char *cmd, const char *args,
2253 char *res, int res_len, int flags)
2254 {
2255 LUT1DContext *lut1d = ctx->priv;
2256 int ret;
2257
2258 ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
2259 if (ret < 0)
2260 return ret;
2261
2262 ret = lut1d_init(ctx);
2263 if (ret < 0) {
2264 set_identity_matrix_1d(lut1d, 32);
2265 return ret;
2266 }
2267 return config_input_1d(ctx->inputs[0]);
2268 }
2269
2270 static const AVFilterPad lut1d_inputs[] = {
2271 {
2272 .name = "default",
2273 .type = AVMEDIA_TYPE_VIDEO,
2274 .filter_frame = filter_frame_1d,
2275 .config_props = config_input_1d,
2276 },
2277 { NULL }
2278 };
2279
2280 static const AVFilterPad lut1d_outputs[] = {
2281 {
2282 .name = "default",
2283 .type = AVMEDIA_TYPE_VIDEO,
2284 },
2285 { NULL }
2286 };
2287
2288 AVFilter ff_vf_lut1d = {
2289 .name = "lut1d",
2290 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 1D LUT."),
2291 .priv_size = sizeof(LUT1DContext),
2292 .init = lut1d_init,
2293 .query_formats = query_formats,
2294 .inputs = lut1d_inputs,
2295 .outputs = lut1d_outputs,
2296 .priv_class = &lut1d_class,
2297 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
2298 .process_command = lut1d_process_command,
2299 };
2300 #endif
2301