FFmpeg coverage

Directory: ../../../ffmpeg/
File: src/libavfilter/vf_lut3d.c
Date: 2025-01-20 09:27:23
Exec Total Coverage
Lines: 0 1029 0.0%
Functions: 0 140 0.0%
Branches: 0 1528 0.0%
Line Branch Exec Source
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 "config_components.h"
30
31 #include "libavutil/mem.h"
32 #include "libavutil/opt.h"
33 #include "libavutil/file_open.h"
34 #include "libavutil/intfloat.h"
35 #include "libavutil/avassert.h"
36 #include "libavutil/avstring.h"
37 #include "drawutils.h"
38 #include "filters.h"
39 #include "video.h"
40 #include "lut3d.h"
41
42 #define R 0
43 #define G 1
44 #define B 2
45 #define A 3
46
47 #define OFFSET(x) offsetof(LUT3DContext, x)
48 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
49 #define TFLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
50 #define COMMON_OPTIONS \
51 { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, NB_INTERP_MODE-1, TFLAGS, .unit = "interp_mode" }, \
52 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_NEAREST}, 0, 0, TFLAGS, .unit = "interp_mode" }, \
53 { "trilinear", "interpolate values using the 8 points defining a cube", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TRILINEAR}, 0, 0, TFLAGS, .unit = "interp_mode" }, \
54 { "tetrahedral", "interpolate values using a tetrahedron", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, 0, TFLAGS, .unit = "interp_mode" }, \
55 { "pyramid", "interpolate values using a pyramid", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PYRAMID}, 0, 0, TFLAGS, .unit = "interp_mode" }, \
56 { "prism", "interpolate values using a prism", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PRISM}, 0, 0, TFLAGS, .unit = "interp_mode" }, \
57 { NULL }
58
59 #define EXPONENT_MASK 0x7F800000
60 #define MANTISSA_MASK 0x007FFFFF
61 #define SIGN_MASK 0x80000000
62
63 ✗ static inline float sanitizef(float f)
64 {
65 union av_intfloat32 t;
66 ✗ t.f = f;
67
68 ✗ if ((t.i & EXPONENT_MASK) == EXPONENT_MASK) {
69 ✗ if ((t.i & MANTISSA_MASK) != 0) {
70 // NAN
71 ✗ return 0.0f;
72 ✗ } else if (t.i & SIGN_MASK) {
73 // -INF
74 ✗ return -FLT_MAX;
75 } else {
76 // +INF
77 ✗ return FLT_MAX;
78 }
79 }
80 ✗ return f;
81 }
82
83 ✗ static inline float lerpf(float v0, float v1, float f)
84 {
85 ✗ return v0 + (v1 - v0) * f;
86 }
87
88 ✗ static inline struct rgbvec lerp(const struct rgbvec *v0, const struct rgbvec *v1, float f)
89 {
90 ✗ struct rgbvec v = {
91 ✗ lerpf(v0->r, v1->r, f), lerpf(v0->g, v1->g, f), lerpf(v0->b, v1->b, f)
92 };
93 ✗ return v;
94 }
95
96 #define NEAR(x) ((int)((x) + .5))
97 #define PREV(x) ((int)(x))
98 #define NEXT(x) (FFMIN((int)(x) + 1, lut3d->lutsize - 1))
99
100 /**
101 * Get the nearest defined point
102 */
103 ✗ static inline struct rgbvec interp_nearest(const LUT3DContext *lut3d,
104 const struct rgbvec *s)
105 {
106 ✗ return lut3d->lut[NEAR(s->r) * lut3d->lutsize2 + NEAR(s->g) * lut3d->lutsize + NEAR(s->b)];
107 }
108
109 /**
110 * Interpolate using the 8 vertices of a cube
111 * @see https://en.wikipedia.org/wiki/Trilinear_interpolation
112 */
113 ✗ static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d,
114 const struct rgbvec *s)
115 {
116 ✗ const int lutsize2 = lut3d->lutsize2;
117 ✗ const int lutsize = lut3d->lutsize;
118 ✗ const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
119 ✗ const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
120 ✗ const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
121 ✗ const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
122 ✗ const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
123 ✗ const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
124 ✗ const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
125 ✗ const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
126 ✗ const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
127 ✗ const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
128 ✗ const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
129 ✗ const struct rgbvec c00 = lerp(&c000, &c100, d.r);
130 ✗ const struct rgbvec c10 = lerp(&c010, &c110, d.r);
131 ✗ const struct rgbvec c01 = lerp(&c001, &c101, d.r);
132 ✗ const struct rgbvec c11 = lerp(&c011, &c111, d.r);
133 ✗ const struct rgbvec c0 = lerp(&c00, &c10, d.g);
134 ✗ const struct rgbvec c1 = lerp(&c01, &c11, d.g);
135 ✗ const struct rgbvec c = lerp(&c0, &c1, d.b);
136 ✗ return c;
137 }
138
139 ✗ static inline struct rgbvec interp_pyramid(const LUT3DContext *lut3d,
140 const struct rgbvec *s)
141 {
142 ✗ const int lutsize2 = lut3d->lutsize2;
143 ✗ const int lutsize = lut3d->lutsize;
144 ✗ const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
145 ✗ const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
146 ✗ const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
147 ✗ const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
148 ✗ const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
149 struct rgbvec c;
150
151 ✗ if (d.g > d.r && d.b > d.r) {
152 ✗ const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
153 ✗ const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
154 ✗ const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
155
156 ✗ c.r = c000.r + (c111.r - c011.r) * d.r + (c010.r - c000.r) * d.g + (c001.r - c000.r) * d.b +
157 ✗ (c011.r - c001.r - c010.r + c000.r) * d.g * d.b;
158 ✗ c.g = c000.g + (c111.g - c011.g) * d.r + (c010.g - c000.g) * d.g + (c001.g - c000.g) * d.b +
159 ✗ (c011.g - c001.g - c010.g + c000.g) * d.g * d.b;
160 ✗ c.b = c000.b + (c111.b - c011.b) * d.r + (c010.b - c000.b) * d.g + (c001.b - c000.b) * d.b +
161 ✗ (c011.b - c001.b - c010.b + c000.b) * d.g * d.b;
162 ✗ } else if (d.r > d.g && d.b > d.g) {
163 ✗ const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
164 ✗ const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
165 ✗ const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
166
167 ✗ c.r = c000.r + (c100.r - c000.r) * d.r + (c111.r - c101.r) * d.g + (c001.r - c000.r) * d.b +
168 ✗ (c101.r - c001.r - c100.r + c000.r) * d.r * d.b;
169 ✗ c.g = c000.g + (c100.g - c000.g) * d.r + (c111.g - c101.g) * d.g + (c001.g - c000.g) * d.b +
170 ✗ (c101.g - c001.g - c100.g + c000.g) * d.r * d.b;
171 ✗ c.b = c000.b + (c100.b - c000.b) * d.r + (c111.b - c101.b) * d.g + (c001.b - c000.b) * d.b +
172 ✗ (c101.b - c001.b - c100.b + c000.b) * d.r * d.b;
173 } else {
174 ✗ const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
175 ✗ const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
176 ✗ const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
177
178 ✗ c.r = c000.r + (c100.r - c000.r) * d.r + (c010.r - c000.r) * d.g + (c111.r - c110.r) * d.b +
179 ✗ (c110.r - c100.r - c010.r + c000.r) * d.r * d.g;
180 ✗ c.g = c000.g + (c100.g - c000.g) * d.r + (c010.g - c000.g) * d.g + (c111.g - c110.g) * d.b +
181 ✗ (c110.g - c100.g - c010.g + c000.g) * d.r * d.g;
182 ✗ c.b = c000.b + (c100.b - c000.b) * d.r + (c010.b - c000.b) * d.g + (c111.b - c110.b) * d.b +
183 ✗ (c110.b - c100.b - c010.b + c000.b) * d.r * d.g;
184 }
185
186 ✗ return c;
187 }
188
189 ✗ static inline struct rgbvec interp_prism(const LUT3DContext *lut3d,
190 const struct rgbvec *s)
191 {
192 ✗ const int lutsize2 = lut3d->lutsize2;
193 ✗ const int lutsize = lut3d->lutsize;
194 ✗ const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
195 ✗ const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
196 ✗ const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
197 ✗ const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
198 ✗ const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
199 ✗ const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
200 ✗ const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
201 struct rgbvec c;
202
203 ✗ if (d.b > d.r) {
204 ✗ const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
205 ✗ const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
206
207 ✗ c.r = c000.r + (c001.r - c000.r) * d.b + (c101.r - c001.r) * d.r + (c010.r - c000.r) * d.g +
208 ✗ (c000.r - c010.r - c001.r + c011.r) * d.b * d.g +
209 ✗ (c001.r - c011.r - c101.r + c111.r) * d.r * d.g;
210 ✗ c.g = c000.g + (c001.g - c000.g) * d.b + (c101.g - c001.g) * d.r + (c010.g - c000.g) * d.g +
211 ✗ (c000.g - c010.g - c001.g + c011.g) * d.b * d.g +
212 ✗ (c001.g - c011.g - c101.g + c111.g) * d.r * d.g;
213 ✗ c.b = c000.b + (c001.b - c000.b) * d.b + (c101.b - c001.b) * d.r + (c010.b - c000.b) * d.g +
214 ✗ (c000.b - c010.b - c001.b + c011.b) * d.b * d.g +
215 ✗ (c001.b - c011.b - c101.b + c111.b) * d.r * d.g;
216 } else {
217 ✗ const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
218 ✗ const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
219
220 ✗ c.r = c000.r + (c101.r - c100.r) * d.b + (c100.r - c000.r) * d.r + (c010.r - c000.r) * d.g +
221 ✗ (c100.r - c110.r - c101.r + c111.r) * d.b * d.g +
222 ✗ (c000.r - c010.r - c100.r + c110.r) * d.r * d.g;
223 ✗ c.g = c000.g + (c101.g - c100.g) * d.b + (c100.g - c000.g) * d.r + (c010.g - c000.g) * d.g +
224 ✗ (c100.g - c110.g - c101.g + c111.g) * d.b * d.g +
225 ✗ (c000.g - c010.g - c100.g + c110.g) * d.r * d.g;
226 ✗ c.b = c000.b + (c101.b - c100.b) * d.b + (c100.b - c000.b) * d.r + (c010.b - c000.b) * d.g +
227 ✗ (c100.b - c110.b - c101.b + c111.b) * d.b * d.g +
228 ✗ (c000.b - c010.b - c100.b + c110.b) * d.r * d.g;
229 }
230
231 ✗ return c;
232 }
233
234 /**
235 * Tetrahedral interpolation. Based on code found in Truelight Software Library paper.
236 * @see http://www.filmlight.ltd.uk/pdf/whitepapers/FL-TL-TN-0057-SoftwareLib.pdf
237 */
238 ✗ static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d,
239 const struct rgbvec *s)
240 {
241 ✗ const int lutsize2 = lut3d->lutsize2;
242 ✗ const int lutsize = lut3d->lutsize;
243 ✗ const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
244 ✗ const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
245 ✗ const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
246 ✗ const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
247 ✗ const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
248 struct rgbvec c;
249 ✗ if (d.r > d.g) {
250 ✗ if (d.g > d.b) {
251 ✗ const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
252 ✗ const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
253 ✗ c.r = (1-d.r) * c000.r + (d.r-d.g) * c100.r + (d.g-d.b) * c110.r + (d.b) * c111.r;
254 ✗ c.g = (1-d.r) * c000.g + (d.r-d.g) * c100.g + (d.g-d.b) * c110.g + (d.b) * c111.g;
255 ✗ c.b = (1-d.r) * c000.b + (d.r-d.g) * c100.b + (d.g-d.b) * c110.b + (d.b) * c111.b;
256 ✗ } else if (d.r > d.b) {
257 ✗ const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
258 ✗ const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
259 ✗ c.r = (1-d.r) * c000.r + (d.r-d.b) * c100.r + (d.b-d.g) * c101.r + (d.g) * c111.r;
260 ✗ c.g = (1-d.r) * c000.g + (d.r-d.b) * c100.g + (d.b-d.g) * c101.g + (d.g) * c111.g;
261 ✗ c.b = (1-d.r) * c000.b + (d.r-d.b) * c100.b + (d.b-d.g) * c101.b + (d.g) * c111.b;
262 } else {
263 ✗ const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
264 ✗ const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
265 ✗ c.r = (1-d.b) * c000.r + (d.b-d.r) * c001.r + (d.r-d.g) * c101.r + (d.g) * c111.r;
266 ✗ c.g = (1-d.b) * c000.g + (d.b-d.r) * c001.g + (d.r-d.g) * c101.g + (d.g) * c111.g;
267 ✗ c.b = (1-d.b) * c000.b + (d.b-d.r) * c001.b + (d.r-d.g) * c101.b + (d.g) * c111.b;
268 }
269 } else {
270 ✗ if (d.b > d.g) {
271 ✗ const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
272 ✗ const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
273 ✗ c.r = (1-d.b) * c000.r + (d.b-d.g) * c001.r + (d.g-d.r) * c011.r + (d.r) * c111.r;
274 ✗ c.g = (1-d.b) * c000.g + (d.b-d.g) * c001.g + (d.g-d.r) * c011.g + (d.r) * c111.g;
275 ✗ c.b = (1-d.b) * c000.b + (d.b-d.g) * c001.b + (d.g-d.r) * c011.b + (d.r) * c111.b;
276 ✗ } else if (d.b > d.r) {
277 ✗ const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
278 ✗ const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
279 ✗ c.r = (1-d.g) * c000.r + (d.g-d.b) * c010.r + (d.b-d.r) * c011.r + (d.r) * c111.r;
280 ✗ c.g = (1-d.g) * c000.g + (d.g-d.b) * c010.g + (d.b-d.r) * c011.g + (d.r) * c111.g;
281 ✗ c.b = (1-d.g) * c000.b + (d.g-d.b) * c010.b + (d.b-d.r) * c011.b + (d.r) * c111.b;
282 } else {
283 ✗ const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
284 ✗ const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
285 ✗ c.r = (1-d.g) * c000.r + (d.g-d.r) * c010.r + (d.r-d.b) * c110.r + (d.b) * c111.r;
286 ✗ c.g = (1-d.g) * c000.g + (d.g-d.r) * c010.g + (d.r-d.b) * c110.g + (d.b) * c111.g;
287 ✗ c.b = (1-d.g) * c000.b + (d.g-d.r) * c010.b + (d.r-d.b) * c110.b + (d.b) * c111.b;
288 }
289 }
290 ✗ return c;
291 }
292
293 ✗ static inline float prelut_interp_1d_linear(const Lut3DPreLut *prelut,
294 int idx, const float s)
295 {
296 ✗ const int lut_max = prelut->size - 1;
297 ✗ const float scaled = (s - prelut->min[idx]) * prelut->scale[idx];
298 ✗ const float x = av_clipf(scaled, 0.0f, lut_max);
299 ✗ const int prev = PREV(x);
300 ✗ const int next = FFMIN((int)(x) + 1, lut_max);
301 ✗ const float p = prelut->lut[idx][prev];
302 ✗ const float n = prelut->lut[idx][next];
303 ✗ const float d = x - (float)prev;
304 ✗ return lerpf(p, n, d);
305 }
306
307 ✗ static inline struct rgbvec apply_prelut(const Lut3DPreLut *prelut,
308 const struct rgbvec *s)
309 {
310 struct rgbvec c;
311
312 ✗ if (prelut->size <= 0)
313 ✗ return *s;
314
315 ✗ c.r = prelut_interp_1d_linear(prelut, 0, s->r);
316 ✗ c.g = prelut_interp_1d_linear(prelut, 1, s->g);
317 ✗ c.b = prelut_interp_1d_linear(prelut, 2, s->b);
318 ✗ return c;
319 }
320
321 #define DEFINE_INTERP_FUNC_PLANAR(name, nbits, depth) \
322 static int interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
323 { \
324 int x, y; \
325 const LUT3DContext *lut3d = ctx->priv; \
326 const Lut3DPreLut *prelut = &lut3d->prelut; \
327 const ThreadData *td = arg; \
328 const AVFrame *in = td->in; \
329 const AVFrame *out = td->out; \
330 const int direct = out == in; \
331 const int slice_start = (in->height * jobnr ) / nb_jobs; \
332 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
333 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
334 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
335 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
336 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
337 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
338 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
339 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
340 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
341 const float lut_max = lut3d->lutsize - 1; \
342 const float scale_f = 1.0f / ((1<<depth) - 1); \
343 const float scale_r = lut3d->scale.r * lut_max; \
344 const float scale_g = lut3d->scale.g * lut_max; \
345 const float scale_b = lut3d->scale.b * lut_max; \
346 \
347 for (y = slice_start; y < slice_end; y++) { \
348 uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
349 uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
350 uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
351 uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
352 const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
353 const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
354 const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
355 const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
356 for (x = 0; x < in->width; x++) { \
357 const struct rgbvec rgb = {srcr[x] * scale_f, \
358 srcg[x] * scale_f, \
359 srcb[x] * scale_f}; \
360 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
361 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
362 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
363 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
364 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
365 dstr[x] = av_clip_uintp2(vec.r * (float)((1<<depth) - 1), depth); \
366 dstg[x] = av_clip_uintp2(vec.g * (float)((1<<depth) - 1), depth); \
367 dstb[x] = av_clip_uintp2(vec.b * (float)((1<<depth) - 1), depth); \
368 if (!direct && in->linesize[3]) \
369 dsta[x] = srca[x]; \
370 } \
371 grow += out->linesize[0]; \
372 brow += out->linesize[1]; \
373 rrow += out->linesize[2]; \
374 arow += out->linesize[3]; \
375 srcgrow += in->linesize[0]; \
376 srcbrow += in->linesize[1]; \
377 srcrrow += in->linesize[2]; \
378 srcarow += in->linesize[3]; \
379 } \
380 return 0; \
381 }
382
383 ✗ DEFINE_INTERP_FUNC_PLANAR(nearest, 8, 8)
384 ✗ DEFINE_INTERP_FUNC_PLANAR(trilinear, 8, 8)
385 ✗ DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 8, 8)
386 ✗ DEFINE_INTERP_FUNC_PLANAR(pyramid, 8, 8)
387 ✗ DEFINE_INTERP_FUNC_PLANAR(prism, 8, 8)
388
389 ✗ DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 9)
390 ✗ DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 9)
391 ✗ DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 9)
392 ✗ DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 9)
393 ✗ DEFINE_INTERP_FUNC_PLANAR(prism, 16, 9)
394
395 ✗ DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 10)
396 ✗ DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 10)
397 ✗ DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 10)
398 ✗ DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 10)
399 ✗ DEFINE_INTERP_FUNC_PLANAR(prism, 16, 10)
400
401 ✗ DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 12)
402 ✗ DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 12)
403 ✗ DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 12)
404 ✗ DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 12)
405 ✗ DEFINE_INTERP_FUNC_PLANAR(prism, 16, 12)
406
407 ✗ DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 14)
408 ✗ DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 14)
409 ✗ DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 14)
410 ✗ DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 14)
411 ✗ DEFINE_INTERP_FUNC_PLANAR(prism, 16, 14)
412
413 ✗ DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 16)
414 ✗ DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 16)
415 ✗ DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 16)
416 ✗ DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 16)
417 ✗ DEFINE_INTERP_FUNC_PLANAR(prism, 16, 16)
418
419 #define DEFINE_INTERP_FUNC_PLANAR_FLOAT(name, depth) \
420 static int interp_##name##_pf##depth(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
421 { \
422 int x, y; \
423 const LUT3DContext *lut3d = ctx->priv; \
424 const Lut3DPreLut *prelut = &lut3d->prelut; \
425 const ThreadData *td = arg; \
426 const AVFrame *in = td->in; \
427 const AVFrame *out = td->out; \
428 const int direct = out == in; \
429 const int slice_start = (in->height * jobnr ) / nb_jobs; \
430 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
431 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
432 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
433 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
434 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
435 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
436 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
437 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
438 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
439 const float lut_max = lut3d->lutsize - 1; \
440 const float scale_r = lut3d->scale.r * lut_max; \
441 const float scale_g = lut3d->scale.g * lut_max; \
442 const float scale_b = lut3d->scale.b * lut_max; \
443 \
444 for (y = slice_start; y < slice_end; y++) { \
445 float *dstg = (float *)grow; \
446 float *dstb = (float *)brow; \
447 float *dstr = (float *)rrow; \
448 float *dsta = (float *)arow; \
449 const float *srcg = (const float *)srcgrow; \
450 const float *srcb = (const float *)srcbrow; \
451 const float *srcr = (const float *)srcrrow; \
452 const float *srca = (const float *)srcarow; \
453 for (x = 0; x < in->width; x++) { \
454 const struct rgbvec rgb = {sanitizef(srcr[x]), \
455 sanitizef(srcg[x]), \
456 sanitizef(srcb[x])}; \
457 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
458 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
459 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
460 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
461 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
462 dstr[x] = vec.r; \
463 dstg[x] = vec.g; \
464 dstb[x] = vec.b; \
465 if (!direct && in->linesize[3]) \
466 dsta[x] = srca[x]; \
467 } \
468 grow += out->linesize[0]; \
469 brow += out->linesize[1]; \
470 rrow += out->linesize[2]; \
471 arow += out->linesize[3]; \
472 srcgrow += in->linesize[0]; \
473 srcbrow += in->linesize[1]; \
474 srcrrow += in->linesize[2]; \
475 srcarow += in->linesize[3]; \
476 } \
477 return 0; \
478 }
479
480 ✗ DEFINE_INTERP_FUNC_PLANAR_FLOAT(nearest, 32)
481 ✗ DEFINE_INTERP_FUNC_PLANAR_FLOAT(trilinear, 32)
482 ✗ DEFINE_INTERP_FUNC_PLANAR_FLOAT(tetrahedral, 32)
483 ✗ DEFINE_INTERP_FUNC_PLANAR_FLOAT(pyramid, 32)
484 ✗ DEFINE_INTERP_FUNC_PLANAR_FLOAT(prism, 32)
485
486 #define DEFINE_INTERP_FUNC(name, nbits) \
487 static int interp_##nbits##_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
488 { \
489 int x, y; \
490 const LUT3DContext *lut3d = ctx->priv; \
491 const Lut3DPreLut *prelut = &lut3d->prelut; \
492 const ThreadData *td = arg; \
493 const AVFrame *in = td->in; \
494 const AVFrame *out = td->out; \
495 const int direct = out == in; \
496 const int step = lut3d->step; \
497 const uint8_t r = lut3d->rgba_map[R]; \
498 const uint8_t g = lut3d->rgba_map[G]; \
499 const uint8_t b = lut3d->rgba_map[B]; \
500 const uint8_t a = lut3d->rgba_map[A]; \
501 const int slice_start = (in->height * jobnr ) / nb_jobs; \
502 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
503 uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
504 const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
505 const float lut_max = lut3d->lutsize - 1; \
506 const float scale_f = 1.0f / ((1<<nbits) - 1); \
507 const float scale_r = lut3d->scale.r * lut_max; \
508 const float scale_g = lut3d->scale.g * lut_max; \
509 const float scale_b = lut3d->scale.b * lut_max; \
510 \
511 for (y = slice_start; y < slice_end; y++) { \
512 uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
513 const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
514 for (x = 0; x < in->width * step; x += step) { \
515 const struct rgbvec rgb = {src[x + r] * scale_f, \
516 src[x + g] * scale_f, \
517 src[x + b] * scale_f}; \
518 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
519 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
520 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
521 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
522 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
523 dst[x + r] = av_clip_uint##nbits(vec.r * (float)((1<<nbits) - 1)); \
524 dst[x + g] = av_clip_uint##nbits(vec.g * (float)((1<<nbits) - 1)); \
525 dst[x + b] = av_clip_uint##nbits(vec.b * (float)((1<<nbits) - 1)); \
526 if (!direct && step == 4) \
527 dst[x + a] = src[x + a]; \
528 } \
529 dstrow += out->linesize[0]; \
530 srcrow += in ->linesize[0]; \
531 } \
532 return 0; \
533 }
534
535 ✗ DEFINE_INTERP_FUNC(nearest, 8)
536 ✗ DEFINE_INTERP_FUNC(trilinear, 8)
537 ✗ DEFINE_INTERP_FUNC(tetrahedral, 8)
538 ✗ DEFINE_INTERP_FUNC(pyramid, 8)
539 ✗ DEFINE_INTERP_FUNC(prism, 8)
540
541 ✗ DEFINE_INTERP_FUNC(nearest, 16)
542 ✗ DEFINE_INTERP_FUNC(trilinear, 16)
543 ✗ DEFINE_INTERP_FUNC(tetrahedral, 16)
544 ✗ DEFINE_INTERP_FUNC(pyramid, 16)
545 ✗ DEFINE_INTERP_FUNC(prism, 16)
546
547 #define MAX_LINE_SIZE 512
548
549 ✗ static int skip_line(const char *p)
550 {
551 ✗ while (*p && av_isspace(*p))
552 ✗ p++;
553 ✗ return !*p || *p == '#';
554 }
555
556 ✗ static char* fget_next_word(char* dst, int max, FILE* f)
557 {
558 int c;
559 ✗ char *p = dst;
560
561 /* for null */
562 ✗ max--;
563 /* skip until next non whitespace char */
564 ✗ while ((c = fgetc(f)) != EOF) {
565 ✗ if (av_isspace(c))
566 ✗ continue;
567
568 ✗ *p++ = c;
569 ✗ max--;
570 ✗ break;
571 }
572
573 /* get max bytes or up until next whitespace char */
574 ✗ for (; max > 0; max--) {
575 ✗ if ((c = fgetc(f)) == EOF)
576 ✗ break;
577
578 ✗ if (av_isspace(c))
579 ✗ break;
580
581 ✗ *p++ = c;
582 }
583
584 ✗ *p = 0;
585 ✗ if (p == dst)
586 ✗ return NULL;
587 ✗ return p;
588 }
589
590 #define NEXT_LINE(loop_cond) do { \
591 if (!fgets(line, sizeof(line), f)) { \
592 av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \
593 return AVERROR_INVALIDDATA; \
594 } \
595 } while (loop_cond)
596
597 #define NEXT_LINE_OR_GOTO(loop_cond, label) do { \
598 if (!fgets(line, sizeof(line), f)) { \
599 av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \
600 ret = AVERROR_INVALIDDATA; \
601 goto label; \
602 } \
603 } while (loop_cond)
604
605 ✗ static int allocate_3dlut(AVFilterContext *ctx, int lutsize, int prelut)
606 {
607 ✗ LUT3DContext *lut3d = ctx->priv;
608 int i;
609 ✗ if (lutsize < 2 || lutsize > MAX_LEVEL) {
610 ✗ av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n");
611 ✗ return AVERROR(EINVAL);
612 }
613
614 ✗ av_freep(&lut3d->lut);
615 ✗ lut3d->lut = av_malloc_array(lutsize * lutsize * lutsize, sizeof(*lut3d->lut));
616 ✗ if (!lut3d->lut)
617 ✗ return AVERROR(ENOMEM);
618
619 ✗ if (prelut) {
620 ✗ lut3d->prelut.size = PRELUT_SIZE;
621 ✗ for (i = 0; i < 3; i++) {
622 ✗ av_freep(&lut3d->prelut.lut[i]);
623 ✗ lut3d->prelut.lut[i] = av_malloc_array(PRELUT_SIZE, sizeof(*lut3d->prelut.lut[0]));
624 ✗ if (!lut3d->prelut.lut[i])
625 ✗ return AVERROR(ENOMEM);
626 }
627 } else {
628 ✗ lut3d->prelut.size = 0;
629 ✗ for (i = 0; i < 3; i++) {
630 ✗ av_freep(&lut3d->prelut.lut[i]);
631 }
632 }
633 ✗ lut3d->lutsize = lutsize;
634 ✗ lut3d->lutsize2 = lutsize * lutsize;
635 ✗ return 0;
636 }
637
638 /* Basically r g and b float values on each line, with a facultative 3DLUTSIZE
639 * directive; seems to be generated by Davinci */
640 ✗ static int parse_dat(AVFilterContext *ctx, FILE *f)
641 {
642 ✗ LUT3DContext *lut3d = ctx->priv;
643 char line[MAX_LINE_SIZE];
644 int ret, i, j, k, size, size2;
645
646 ✗ lut3d->lutsize = size = 33;
647 ✗ size2 = size * size;
648
649 ✗ NEXT_LINE(skip_line(line));
650 ✗ if (!strncmp(line, "3DLUTSIZE ", 10)) {
651 ✗ size = strtol(line + 10, NULL, 0);
652
653 ✗ NEXT_LINE(skip_line(line));
654 }
655
656 ✗ ret = allocate_3dlut(ctx, size, 0);
657 ✗ if (ret < 0)
658 ✗ return ret;
659
660 ✗ for (k = 0; k < size; k++) {
661 ✗ for (j = 0; j < size; j++) {
662 ✗ for (i = 0; i < size; i++) {
663 ✗ struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
664 ✗ if (k != 0 || j != 0 || i != 0)
665 ✗ NEXT_LINE(skip_line(line));
666 ✗ if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
667 ✗ return AVERROR_INVALIDDATA;
668 }
669 }
670 }
671 ✗ return 0;
672 }
673
674 /* Iridas format */
675 ✗ static int parse_cube(AVFilterContext *ctx, FILE *f)
676 {
677 ✗ LUT3DContext *lut3d = ctx->priv;
678 char line[MAX_LINE_SIZE];
679 ✗ float min[3] = {0.0, 0.0, 0.0};
680 ✗ float max[3] = {1.0, 1.0, 1.0};
681
682 ✗ while (fgets(line, sizeof(line), f)) {
683 ✗ if (!strncmp(line, "LUT_3D_SIZE", 11)) {
684 int ret, i, j, k;
685 ✗ const int size = strtol(line + 12, NULL, 0);
686 ✗ const int size2 = size * size;
687
688 ✗ ret = allocate_3dlut(ctx, size, 0);
689 ✗ if (ret < 0)
690 ✗ return ret;
691
692 ✗ for (k = 0; k < size; k++) {
693 ✗ for (j = 0; j < size; j++) {
694 ✗ for (i = 0; i < size; i++) {
695 ✗ struct rgbvec *vec = &lut3d->lut[i * size2 + j * size + k];
696
697 do {
698 ✗ try_again:
699 ✗ NEXT_LINE(0);
700 ✗ if (!strncmp(line, "DOMAIN_", 7)) {
701 ✗ float *vals = NULL;
702 ✗ if (!strncmp(line + 7, "MIN ", 4)) vals = min;
703 ✗ else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
704 ✗ if (!vals)
705 ✗ return AVERROR_INVALIDDATA;
706 ✗ if (av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2) != 3)
707 ✗ return AVERROR_INVALIDDATA;
708 ✗ av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
709 ✗ min[0], min[1], min[2], max[0], max[1], max[2]);
710 ✗ goto try_again;
711 ✗ } else if (!strncmp(line, "TITLE", 5)) {
712 ✗ goto try_again;
713 }
714 ✗ } while (skip_line(line));
715 ✗ if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
716 ✗ return AVERROR_INVALIDDATA;
717 }
718 }
719 }
720 ✗ break;
721 }
722 }
723
724 ✗ lut3d->scale.r = av_clipf(1. / (max[0] - min[0]), 0.f, 1.f);
725 ✗ lut3d->scale.g = av_clipf(1. / (max[1] - min[1]), 0.f, 1.f);
726 ✗ lut3d->scale.b = av_clipf(1. / (max[2] - min[2]), 0.f, 1.f);
727
728 ✗ return 0;
729 }
730
731 /* Assume 17x17x17 LUT with a 16-bit depth
732 * FIXME: it seems there are various 3dl formats */
733 ✗ static int parse_3dl(AVFilterContext *ctx, FILE *f)
734 {
735 char line[MAX_LINE_SIZE];
736 ✗ LUT3DContext *lut3d = ctx->priv;
737 int ret, i, j, k;
738 ✗ const int size = 17;
739 ✗ const int size2 = 17 * 17;
740 ✗ const float scale = 16*16*16;
741
742 ✗ lut3d->lutsize = size;
743
744 ✗ ret = allocate_3dlut(ctx, size, 0);
745 ✗ if (ret < 0)
746 ✗ return ret;
747
748 ✗ NEXT_LINE(skip_line(line));
749 ✗ for (k = 0; k < size; k++) {
750 ✗ for (j = 0; j < size; j++) {
751 ✗ for (i = 0; i < size; i++) {
752 int r, g, b;
753 ✗ struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
754
755 ✗ NEXT_LINE(skip_line(line));
756 ✗ if (av_sscanf(line, "%d %d %d", &r, &g, &b) != 3)
757 ✗ return AVERROR_INVALIDDATA;
758 ✗ vec->r = r / scale;
759 ✗ vec->g = g / scale;
760 ✗ vec->b = b / scale;
761 }
762 }
763 }
764 ✗ return 0;
765 }
766
767 /* Pandora format */
768 ✗ static int parse_m3d(AVFilterContext *ctx, FILE *f)
769 {
770 ✗ LUT3DContext *lut3d = ctx->priv;
771 float scale;
772 ✗ int ret, i, j, k, size, size2, in = -1, out = -1;
773 char line[MAX_LINE_SIZE];
774 ✗ uint8_t rgb_map[3] = {0, 1, 2};
775
776 ✗ while (fgets(line, sizeof(line), f)) {
777 ✗ if (!strncmp(line, "in", 2)) in = strtol(line + 2, NULL, 0);
778 ✗ else if (!strncmp(line, "out", 3)) out = strtol(line + 3, NULL, 0);
779 ✗ else if (!strncmp(line, "values", 6)) {
780 ✗ const char *p = line + 6;
781 #define SET_COLOR(id) do { \
782 while (av_isspace(*p)) \
783 p++; \
784 switch (*p) { \
785 case 'r': rgb_map[id] = 0; break; \
786 case 'g': rgb_map[id] = 1; break; \
787 case 'b': rgb_map[id] = 2; break; \
788 } \
789 while (*p && !av_isspace(*p)) \
790 p++; \
791 } while (0)
792 ✗ SET_COLOR(0);
793 ✗ SET_COLOR(1);
794 ✗ SET_COLOR(2);
795 ✗ break;
796 }
797 }
798
799 ✗ if (in == -1 || out == -1) {
800 ✗ av_log(ctx, AV_LOG_ERROR, "in and out must be defined\n");
801 ✗ return AVERROR_INVALIDDATA;
802 }
803 ✗ if (in < 2 || out < 2 ||
804 ✗ in > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL ||
805 out > MAX_LEVEL*MAX_LEVEL*MAX_LEVEL) {
806 ✗ av_log(ctx, AV_LOG_ERROR, "invalid in (%d) or out (%d)\n", in, out);
807 ✗ return AVERROR_INVALIDDATA;
808 }
809 ✗ for (size = 1; size*size*size < in; size++);
810 ✗ lut3d->lutsize = size;
811 ✗ size2 = size * size;
812
813 ✗ ret = allocate_3dlut(ctx, size, 0);
814 ✗ if (ret < 0)
815 ✗ return ret;
816
817 ✗ scale = 1. / (out - 1);
818
819 ✗ for (k = 0; k < size; k++) {
820 ✗ for (j = 0; j < size; j++) {
821 ✗ for (i = 0; i < size; i++) {
822 ✗ struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
823 float val[3];
824
825 ✗ NEXT_LINE(0);
826 ✗ if (av_sscanf(line, "%f %f %f", val, val + 1, val + 2) != 3)
827 ✗ return AVERROR_INVALIDDATA;
828 ✗ vec->r = val[rgb_map[0]] * scale;
829 ✗ vec->g = val[rgb_map[1]] * scale;
830 ✗ vec->b = val[rgb_map[2]] * scale;
831 }
832 }
833 }
834 ✗ return 0;
835 }
836
837 ✗ static int nearest_sample_index(float *data, float x, int low, int hi)
838 {
839 int mid;
840 ✗ if (x < data[low])
841 ✗ return low;
842
843 ✗ if (x > data[hi])
844 ✗ return hi;
845
846 for (;;) {
847 ✗ av_assert0(x >= data[low]);
848 ✗ av_assert0(x <= data[hi]);
849 ✗ av_assert0((hi-low) > 0);
850
851 ✗ if (hi - low == 1)
852 ✗ return low;
853
854 ✗ mid = (low + hi) / 2;
855
856 ✗ if (x < data[mid])
857 ✗ hi = mid;
858 else
859 ✗ low = mid;
860 }
861
862 return 0;
863 }
864
865 #define NEXT_FLOAT_OR_GOTO(value, label) \
866 if (!fget_next_word(line, sizeof(line) ,f)) { \
867 ret = AVERROR_INVALIDDATA; \
868 goto label; \
869 } \
870 if (av_sscanf(line, "%f", &value) != 1) { \
871 ret = AVERROR_INVALIDDATA; \
872 goto label; \
873 }
874
875 ✗ static int parse_cinespace(AVFilterContext *ctx, FILE *f)
876 {
877 ✗ LUT3DContext *lut3d = ctx->priv;
878 char line[MAX_LINE_SIZE];
879 ✗ float in_min[3] = {0.0, 0.0, 0.0};
880 ✗ float in_max[3] = {1.0, 1.0, 1.0};
881 ✗ float out_min[3] = {0.0, 0.0, 0.0};
882 ✗ float out_max[3] = {1.0, 1.0, 1.0};
883 ✗ int inside_metadata = 0, size, size2;
884 ✗ int prelut = 0;
885 ✗ int ret = 0;
886
887 ✗ int prelut_sizes[3] = {0, 0, 0};
888 ✗ float *in_prelut[3] = {NULL, NULL, NULL};
889 ✗ float *out_prelut[3] = {NULL, NULL, NULL};
890
891 ✗ NEXT_LINE_OR_GOTO(skip_line(line), end);
892 ✗ if (strncmp(line, "CSPLUTV100", 10)) {
893 ✗ av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n");
894 ✗ ret = AVERROR(EINVAL);
895 ✗ goto end;
896 }
897
898 ✗ NEXT_LINE_OR_GOTO(skip_line(line), end);
899 ✗ if (strncmp(line, "3D", 2)) {
900 ✗ av_log(ctx, AV_LOG_ERROR, "Not 3D LUT format\n");
901 ✗ ret = AVERROR(EINVAL);
902 ✗ goto end;
903 }
904
905 ✗ while (1) {
906 ✗ NEXT_LINE_OR_GOTO(skip_line(line), end);
907
908 ✗ if (!strncmp(line, "BEGIN METADATA", 14)) {
909 ✗ inside_metadata = 1;
910 ✗ continue;
911 }
912 ✗ if (!strncmp(line, "END METADATA", 12)) {
913 ✗ inside_metadata = 0;
914 ✗ continue;
915 }
916 ✗ if (inside_metadata == 0) {
917 int size_r, size_g, size_b;
918
919 ✗ for (int i = 0; i < 3; i++) {
920 ✗ int npoints = strtol(line, NULL, 0);
921
922 ✗ if (npoints > 2) {
923 float v,last;
924
925 ✗ if (npoints > PRELUT_SIZE) {
926 ✗ av_log(ctx, AV_LOG_ERROR, "Prelut size too large.\n");
927 ✗ ret = AVERROR_INVALIDDATA;
928 ✗ goto end;
929 }
930
931 ✗ if (in_prelut[i] || out_prelut[i]) {
932 ✗ av_log(ctx, AV_LOG_ERROR, "Invalid file has multiple preluts.\n");
933 ✗ ret = AVERROR_INVALIDDATA;
934 ✗ goto end;
935 }
936
937 ✗ in_prelut[i] = (float*)av_malloc(npoints * sizeof(float));
938 ✗ out_prelut[i] = (float*)av_malloc(npoints * sizeof(float));
939 ✗ if (!in_prelut[i] || !out_prelut[i]) {
940 ✗ ret = AVERROR(ENOMEM);
941 ✗ goto end;
942 }
943
944 ✗ prelut_sizes[i] = npoints;
945 ✗ in_min[i] = FLT_MAX;
946 ✗ in_max[i] = -FLT_MAX;
947 ✗ out_min[i] = FLT_MAX;
948 ✗ out_max[i] = -FLT_MAX;
949
950 ✗ for (int j = 0; j < npoints; j++) {
951 ✗ NEXT_FLOAT_OR_GOTO(v, end)
952 ✗ in_min[i] = FFMIN(in_min[i], v);
953 ✗ in_max[i] = FFMAX(in_max[i], v);
954 ✗ in_prelut[i][j] = v;
955 ✗ if (j > 0 && v < last) {
956 ✗ av_log(ctx, AV_LOG_ERROR, "Invalid file, non increasing prelut.\n");
957 ✗ ret = AVERROR(ENOMEM);
958 ✗ goto end;
959 }
960 ✗ last = v;
961 }
962
963 ✗ for (int j = 0; j < npoints; j++) {
964 ✗ NEXT_FLOAT_OR_GOTO(v, end)
965 ✗ out_min[i] = FFMIN(out_min[i], v);
966 ✗ out_max[i] = FFMAX(out_max[i], v);
967 ✗ out_prelut[i][j] = v;
968 }
969
970 ✗ } else if (npoints == 2) {
971 ✗ NEXT_LINE_OR_GOTO(skip_line(line), end);
972 ✗ if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2) {
973 ✗ ret = AVERROR_INVALIDDATA;
974 ✗ goto end;
975 }
976 ✗ NEXT_LINE_OR_GOTO(skip_line(line), end);
977 ✗ if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2) {
978 ✗ ret = AVERROR_INVALIDDATA;
979 ✗ goto end;
980 }
981
982 } else {
983 ✗ av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n");
984 ✗ ret = AVERROR_PATCHWELCOME;
985 ✗ goto end;
986 }
987
988 ✗ NEXT_LINE_OR_GOTO(skip_line(line), end);
989 }
990
991 ✗ if (av_sscanf(line, "%d %d %d", &size_r, &size_g, &size_b) != 3) {
992 ✗ ret = AVERROR(EINVAL);
993 ✗ goto end;
994 }
995 ✗ if (size_r != size_g || size_r != size_b) {
996 ✗ av_log(ctx, AV_LOG_ERROR, "Unsupported size combination: %dx%dx%d.\n", size_r, size_g, size_b);
997 ✗ ret = AVERROR_PATCHWELCOME;
998 ✗ goto end;
999 }
1000
1001 ✗ size = size_r;
1002 ✗ size2 = size * size;
1003
1004 ✗ if (prelut_sizes[0] && prelut_sizes[1] && prelut_sizes[2])
1005 ✗ prelut = 1;
1006
1007 ✗ ret = allocate_3dlut(ctx, size, prelut);
1008 ✗ if (ret < 0)
1009 ✗ return ret;
1010
1011 ✗ for (int k = 0; k < size; k++) {
1012 ✗ for (int j = 0; j < size; j++) {
1013 ✗ for (int i = 0; i < size; i++) {
1014 ✗ struct rgbvec *vec = &lut3d->lut[i * size2 + j * size + k];
1015
1016 ✗ NEXT_LINE_OR_GOTO(skip_line(line), end);
1017 ✗ if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) {
1018 ✗ ret = AVERROR_INVALIDDATA;
1019 ✗ goto end;
1020 }
1021
1022 ✗ vec->r *= out_max[0] - out_min[0];
1023 ✗ vec->g *= out_max[1] - out_min[1];
1024 ✗ vec->b *= out_max[2] - out_min[2];
1025 }
1026 }
1027 }
1028
1029 ✗ break;
1030 }
1031 }
1032
1033 ✗ if (prelut) {
1034 ✗ for (int c = 0; c < 3; c++) {
1035
1036 ✗ lut3d->prelut.min[c] = in_min[c];
1037 ✗ lut3d->prelut.max[c] = in_max[c];
1038 ✗ lut3d->prelut.scale[c] = (1.0f / (float)(in_max[c] - in_min[c])) * (lut3d->prelut.size - 1);
1039
1040 ✗ for (int i = 0; i < lut3d->prelut.size; ++i) {
1041 ✗ float mix = (float) i / (float)(lut3d->prelut.size - 1);
1042 ✗ float x = lerpf(in_min[c], in_max[c], mix), a, b;
1043
1044 ✗ int idx = nearest_sample_index(in_prelut[c], x, 0, prelut_sizes[c]-1);
1045 ✗ av_assert0(idx + 1 < prelut_sizes[c]);
1046
1047 ✗ a = out_prelut[c][idx + 0];
1048 ✗ b = out_prelut[c][idx + 1];
1049 ✗ mix = x - in_prelut[c][idx];
1050
1051 ✗ lut3d->prelut.lut[c][i] = sanitizef(lerpf(a, b, mix));
1052 }
1053 }
1054 ✗ lut3d->scale.r = 1.00f;
1055 ✗ lut3d->scale.g = 1.00f;
1056 ✗ lut3d->scale.b = 1.00f;
1057
1058 } else {
1059 ✗ lut3d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f);
1060 ✗ lut3d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f);
1061 ✗ lut3d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f);
1062 }
1063
1064 ✗ end:
1065 ✗ for (int c = 0; c < 3; c++) {
1066 ✗ av_freep(&in_prelut[c]);
1067 ✗ av_freep(&out_prelut[c]);
1068 }
1069 ✗ return ret;
1070 }
1071
1072 ✗ static int set_identity_matrix(AVFilterContext *ctx, int size)
1073 {
1074 ✗ LUT3DContext *lut3d = ctx->priv;
1075 int ret, i, j, k;
1076 ✗ const int size2 = size * size;
1077 ✗ const float c = 1. / (size - 1);
1078
1079 ✗ ret = allocate_3dlut(ctx, size, 0);
1080 ✗ if (ret < 0)
1081 ✗ return ret;
1082
1083 ✗ for (k = 0; k < size; k++) {
1084 ✗ for (j = 0; j < size; j++) {
1085 ✗ for (i = 0; i < size; i++) {
1086 ✗ struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
1087 ✗ vec->r = k * c;
1088 ✗ vec->g = j * c;
1089 ✗ vec->b = i * c;
1090 }
1091 }
1092 }
1093
1094 ✗ return 0;
1095 }
1096
1097 static const enum AVPixelFormat pix_fmts[] = {
1098 AV_PIX_FMT_RGB24, AV_PIX_FMT_BGR24,
1099 AV_PIX_FMT_RGBA, AV_PIX_FMT_BGRA,
1100 AV_PIX_FMT_ARGB, AV_PIX_FMT_ABGR,
1101 AV_PIX_FMT_0RGB, AV_PIX_FMT_0BGR,
1102 AV_PIX_FMT_RGB0, AV_PIX_FMT_BGR0,
1103 AV_PIX_FMT_RGB48, AV_PIX_FMT_BGR48,
1104 AV_PIX_FMT_RGBA64, AV_PIX_FMT_BGRA64,
1105 AV_PIX_FMT_GBRP, AV_PIX_FMT_GBRAP,
1106 AV_PIX_FMT_GBRP9,
1107 AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRAP10,
1108 AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRAP12,
1109 AV_PIX_FMT_GBRP14,
1110 AV_PIX_FMT_GBRP16, AV_PIX_FMT_GBRAP16,
1111 AV_PIX_FMT_GBRPF32, AV_PIX_FMT_GBRAPF32,
1112 AV_PIX_FMT_NONE
1113 };
1114
1115 #if CONFIG_LUT3D_FILTER || CONFIG_HALDCLUT_FILTER
1116
1117 ✗ static int config_input(AVFilterLink *inlink)
1118 {
1119 int depth, is16bit, isfloat, planar;
1120 ✗ LUT3DContext *lut3d = inlink->dst->priv;
1121 ✗ const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1122
1123 ✗ depth = desc->comp[0].depth;
1124 ✗ is16bit = desc->comp[0].depth > 8;
1125 ✗ planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR;
1126 ✗ isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
1127 ✗ ff_fill_rgba_map(lut3d->rgba_map, inlink->format);
1128 ✗ lut3d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
1129
1130 #define SET_FUNC(name) do { \
1131 if (planar && !isfloat) { \
1132 switch (depth) { \
1133 case 8: lut3d->interp = interp_8_##name##_p8; break; \
1134 case 9: lut3d->interp = interp_16_##name##_p9; break; \
1135 case 10: lut3d->interp = interp_16_##name##_p10; break; \
1136 case 12: lut3d->interp = interp_16_##name##_p12; break; \
1137 case 14: lut3d->interp = interp_16_##name##_p14; break; \
1138 case 16: lut3d->interp = interp_16_##name##_p16; break; \
1139 } \
1140 } else if (isfloat) { lut3d->interp = interp_##name##_pf32; \
1141 } else if (is16bit) { lut3d->interp = interp_16_##name; \
1142 } else { lut3d->interp = interp_8_##name; } \
1143 } while (0)
1144
1145 ✗ switch (lut3d->interpolation) {
1146 ✗ case INTERPOLATE_NEAREST: SET_FUNC(nearest); break;
1147 ✗ case INTERPOLATE_TRILINEAR: SET_FUNC(trilinear); break;
1148 ✗ case INTERPOLATE_TETRAHEDRAL: SET_FUNC(tetrahedral); break;
1149 ✗ case INTERPOLATE_PYRAMID: SET_FUNC(pyramid); break;
1150 ✗ case INTERPOLATE_PRISM: SET_FUNC(prism); break;
1151 ✗ default:
1152 ✗ av_assert0(0);
1153 }
1154
1155 #if ARCH_X86
1156 ✗ ff_lut3d_init_x86(lut3d, desc);
1157 #endif
1158
1159 ✗ return 0;
1160 }
1161
1162 ✗ static AVFrame *apply_lut(AVFilterLink *inlink, AVFrame *in)
1163 {
1164 ✗ AVFilterContext *ctx = inlink->dst;
1165 ✗ LUT3DContext *lut3d = ctx->priv;
1166 ✗ AVFilterLink *outlink = inlink->dst->outputs[0];
1167 AVFrame *out;
1168 ThreadData td;
1169
1170 ✗ if (av_frame_is_writable(in)) {
1171 ✗ out = in;
1172 } else {
1173 ✗ out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
1174 ✗ if (!out) {
1175 ✗ av_frame_free(&in);
1176 ✗ return NULL;
1177 }
1178 ✗ av_frame_copy_props(out, in);
1179 }
1180
1181 ✗ av_frame_side_data_remove_by_props(&out->side_data, &out->nb_side_data,
1182 AV_SIDE_DATA_PROP_COLOR_DEPENDENT);
1183
1184 ✗ td.in = in;
1185 ✗ td.out = out;
1186 ✗ ff_filter_execute(ctx, lut3d->interp, &td, NULL,
1187 ✗ FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
1188
1189 ✗ if (out != in)
1190 ✗ av_frame_free(&in);
1191
1192 ✗ return out;
1193 }
1194
1195 ✗ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
1196 {
1197 ✗ AVFilterLink *outlink = inlink->dst->outputs[0];
1198 ✗ AVFrame *out = apply_lut(inlink, in);
1199 ✗ if (!out)
1200 ✗ return AVERROR(ENOMEM);
1201 ✗ return ff_filter_frame(outlink, out);
1202 }
1203
1204 ✗ static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
1205 char *res, int res_len, int flags)
1206 {
1207 int ret;
1208
1209 ✗ ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
1210 ✗ if (ret < 0)
1211 ✗ return ret;
1212
1213 ✗ return config_input(ctx->inputs[0]);
1214 }
1215
1216 /* These options are shared between several filters;
1217 * &lut3d_haldclut_options[COMMON_OPTIONS_OFFSET] must always
1218 * point to the first of the COMMON_OPTIONS. */
1219 #define COMMON_OPTIONS_OFFSET CONFIG_LUT3D_FILTER
1220 static const AVOption lut3d_haldclut_options[] = {
1221 #if CONFIG_LUT3D_FILTER
1222 { "file", "set 3D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
1223 #endif
1224 #if CONFIG_HALDCLUT_FILTER
1225 { "clut", "when to process CLUT", OFFSET(clut), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, .flags = TFLAGS, .unit = "clut" },
1226 { "first", "process only first CLUT, ignore rest", 0, AV_OPT_TYPE_CONST, {.i64=0}, .flags = TFLAGS, .unit = "clut" },
1227 { "all", "process all CLUTs", 0, AV_OPT_TYPE_CONST, {.i64=1}, .flags = TFLAGS, .unit = "clut" },
1228 #endif
1229 COMMON_OPTIONS
1230 };
1231
1232 #if CONFIG_LUT3D_FILTER
1233
1234 AVFILTER_DEFINE_CLASS_EXT(lut3d, "lut3d", lut3d_haldclut_options);
1235
1236 ✗ static av_cold int lut3d_init(AVFilterContext *ctx)
1237 {
1238 int ret;
1239 FILE *f;
1240 const char *ext;
1241 ✗ LUT3DContext *lut3d = ctx->priv;
1242
1243 ✗ lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1244
1245 ✗ if (!lut3d->file) {
1246 ✗ return set_identity_matrix(ctx, 32);
1247 }
1248
1249 ✗ f = avpriv_fopen_utf8(lut3d->file, "r");
1250 ✗ if (!f) {
1251 ✗ ret = AVERROR(errno);
1252 ✗ av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret));
1253 ✗ return ret;
1254 }
1255
1256 ✗ ext = strrchr(lut3d->file, '.');
1257 ✗ if (!ext) {
1258 ✗ av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
1259 ✗ ret = AVERROR_INVALIDDATA;
1260 ✗ goto end;
1261 }
1262 ✗ ext++;
1263
1264 ✗ if (!av_strcasecmp(ext, "dat")) {
1265 ✗ ret = parse_dat(ctx, f);
1266 ✗ } else if (!av_strcasecmp(ext, "3dl")) {
1267 ✗ ret = parse_3dl(ctx, f);
1268 ✗ } else if (!av_strcasecmp(ext, "cube")) {
1269 ✗ ret = parse_cube(ctx, f);
1270 ✗ } else if (!av_strcasecmp(ext, "m3d")) {
1271 ✗ ret = parse_m3d(ctx, f);
1272 ✗ } else if (!av_strcasecmp(ext, "csp")) {
1273 ✗ ret = parse_cinespace(ctx, f);
1274 } else {
1275 ✗ av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
1276 ✗ ret = AVERROR(EINVAL);
1277 }
1278
1279 ✗ if (!ret && !lut3d->lutsize) {
1280 ✗ av_log(ctx, AV_LOG_ERROR, "3D LUT is empty\n");
1281 ✗ ret = AVERROR_INVALIDDATA;
1282 }
1283
1284 ✗ end:
1285 ✗ fclose(f);
1286 ✗ return ret;
1287 }
1288
1289 ✗ static av_cold void lut3d_uninit(AVFilterContext *ctx)
1290 {
1291 ✗ LUT3DContext *lut3d = ctx->priv;
1292 int i;
1293 ✗ av_freep(&lut3d->lut);
1294
1295 ✗ for (i = 0; i < 3; i++) {
1296 ✗ av_freep(&lut3d->prelut.lut[i]);
1297 }
1298 ✗ }
1299
1300 static const AVFilterPad lut3d_inputs[] = {
1301 {
1302 .name = "default",
1303 .type = AVMEDIA_TYPE_VIDEO,
1304 .filter_frame = filter_frame,
1305 .config_props = config_input,
1306 },
1307 };
1308
1309 const FFFilter ff_vf_lut3d = {
1310 .p.name = "lut3d",
1311 .p.description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
1312 .p.priv_class = &lut3d_class,
1313 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC |
1314 AVFILTER_FLAG_SLICE_THREADS,
1315 .priv_size = sizeof(LUT3DContext),
1316 .init = lut3d_init,
1317 .uninit = lut3d_uninit,
1318 FILTER_INPUTS(lut3d_inputs),
1319 FILTER_OUTPUTS(ff_video_default_filterpad),
1320 FILTER_PIXFMTS_ARRAY(pix_fmts),
1321 .process_command = process_command,
1322 };
1323 #endif
1324
1325 #if CONFIG_HALDCLUT_FILTER
1326
1327 ✗ static void update_clut_packed(LUT3DContext *lut3d, const AVFrame *frame)
1328 {
1329 ✗ const uint8_t *data = frame->data[0];
1330 ✗ const ptrdiff_t linesize = frame->linesize[0];
1331 ✗ const int w = lut3d->clut_width;
1332 ✗ const int step = lut3d->clut_step;
1333 ✗ const uint8_t *rgba_map = lut3d->clut_rgba_map;
1334 ✗ const int level = lut3d->lutsize;
1335 ✗ const int level2 = lut3d->lutsize2;
1336
1337 #define LOAD_CLUT(nbits) do { \
1338 int i, j, k, x = 0, y = 0; \
1339 \
1340 for (k = 0; k < level; k++) { \
1341 for (j = 0; j < level; j++) { \
1342 for (i = 0; i < level; i++) { \
1343 const uint##nbits##_t *src = (const uint##nbits##_t *) \
1344 (data + y*linesize + x*step); \
1345 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1346 vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \
1347 vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \
1348 vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \
1349 if (++x == w) { \
1350 x = 0; \
1351 y++; \
1352 } \
1353 } \
1354 } \
1355 } \
1356 } while (0)
1357
1358 ✗ switch (lut3d->clut_bits) {
1359 ✗ case 8: LOAD_CLUT(8); break;
1360 ✗ case 16: LOAD_CLUT(16); break;
1361 }
1362 ✗ }
1363
1364 ✗ static void update_clut_planar(LUT3DContext *lut3d, const AVFrame *frame)
1365 {
1366 ✗ const uint8_t *datag = frame->data[0];
1367 ✗ const uint8_t *datab = frame->data[1];
1368 ✗ const uint8_t *datar = frame->data[2];
1369 ✗ const ptrdiff_t glinesize = frame->linesize[0];
1370 ✗ const ptrdiff_t blinesize = frame->linesize[1];
1371 ✗ const ptrdiff_t rlinesize = frame->linesize[2];
1372 ✗ const int w = lut3d->clut_width;
1373 ✗ const int level = lut3d->lutsize;
1374 ✗ const int level2 = lut3d->lutsize2;
1375
1376 #define LOAD_CLUT_PLANAR(nbits, depth) do { \
1377 int i, j, k, x = 0, y = 0; \
1378 \
1379 for (k = 0; k < level; k++) { \
1380 for (j = 0; j < level; j++) { \
1381 for (i = 0; i < level; i++) { \
1382 const uint##nbits##_t *gsrc = (const uint##nbits##_t *) \
1383 (datag + y*glinesize); \
1384 const uint##nbits##_t *bsrc = (const uint##nbits##_t *) \
1385 (datab + y*blinesize); \
1386 const uint##nbits##_t *rsrc = (const uint##nbits##_t *) \
1387 (datar + y*rlinesize); \
1388 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1389 vec->r = gsrc[x] / (float)((1<<(depth)) - 1); \
1390 vec->g = bsrc[x] / (float)((1<<(depth)) - 1); \
1391 vec->b = rsrc[x] / (float)((1<<(depth)) - 1); \
1392 if (++x == w) { \
1393 x = 0; \
1394 y++; \
1395 } \
1396 } \
1397 } \
1398 } \
1399 } while (0)
1400
1401 ✗ switch (lut3d->clut_bits) {
1402 ✗ case 8: LOAD_CLUT_PLANAR(8, 8); break;
1403 ✗ case 9: LOAD_CLUT_PLANAR(16, 9); break;
1404 ✗ case 10: LOAD_CLUT_PLANAR(16, 10); break;
1405 ✗ case 12: LOAD_CLUT_PLANAR(16, 12); break;
1406 ✗ case 14: LOAD_CLUT_PLANAR(16, 14); break;
1407 ✗ case 16: LOAD_CLUT_PLANAR(16, 16); break;
1408 }
1409 ✗ }
1410
1411 ✗ static void update_clut_float(LUT3DContext *lut3d, const AVFrame *frame)
1412 {
1413 ✗ const uint8_t *datag = frame->data[0];
1414 ✗ const uint8_t *datab = frame->data[1];
1415 ✗ const uint8_t *datar = frame->data[2];
1416 ✗ const ptrdiff_t glinesize = frame->linesize[0];
1417 ✗ const ptrdiff_t blinesize = frame->linesize[1];
1418 ✗ const ptrdiff_t rlinesize = frame->linesize[2];
1419 ✗ const int w = lut3d->clut_width;
1420 ✗ const int level = lut3d->lutsize;
1421 ✗ const int level2 = lut3d->lutsize2;
1422
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 float *gsrc = (const float *)(datag + y*glinesize);
1429 ✗ const float *bsrc = (const float *)(datab + y*blinesize);
1430 ✗ const float *rsrc = (const float *)(datar + y*rlinesize);
1431 ✗ struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k];
1432 ✗ vec->r = rsrc[x];
1433 ✗ vec->g = gsrc[x];
1434 ✗ vec->b = bsrc[x];
1435 ✗ if (++x == w) {
1436 ✗ x = 0;
1437 ✗ y++;
1438 }
1439 }
1440 }
1441 }
1442 ✗ }
1443
1444 ✗ static int config_output(AVFilterLink *outlink)
1445 {
1446 ✗ AVFilterContext *ctx = outlink->src;
1447 ✗ LUT3DContext *lut3d = ctx->priv;
1448 int ret;
1449
1450 ✗ ret = ff_framesync_init_dualinput(&lut3d->fs, ctx);
1451 ✗ if (ret < 0)
1452 ✗ return ret;
1453 ✗ outlink->w = ctx->inputs[0]->w;
1454 ✗ outlink->h = ctx->inputs[0]->h;
1455 ✗ outlink->time_base = ctx->inputs[0]->time_base;
1456 ✗ if ((ret = ff_framesync_configure(&lut3d->fs)) < 0)
1457 ✗ return ret;
1458 ✗ return 0;
1459 }
1460
1461 ✗ static int activate(AVFilterContext *ctx)
1462 {
1463 ✗ LUT3DContext *s = ctx->priv;
1464 ✗ return ff_framesync_activate(&s->fs);
1465 }
1466
1467 ✗ static int config_clut(AVFilterLink *inlink)
1468 {
1469 int size, level, w, h;
1470 ✗ AVFilterContext *ctx = inlink->dst;
1471 ✗ LUT3DContext *lut3d = ctx->priv;
1472 ✗ const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1473
1474 ✗ av_assert0(desc);
1475
1476 ✗ lut3d->clut_bits = desc->comp[0].depth;
1477 ✗ lut3d->clut_planar = av_pix_fmt_count_planes(inlink->format) > 1;
1478 ✗ lut3d->clut_float = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
1479
1480 ✗ lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3;
1481 ✗ ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format);
1482
1483 ✗ if (inlink->w > inlink->h)
1484 ✗ av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the "
1485 ✗ "Hald CLUT will be ignored\n", inlink->w - inlink->h);
1486 ✗ else if (inlink->w < inlink->h)
1487 ✗ av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the "
1488 ✗ "Hald CLUT will be ignored\n", inlink->h - inlink->w);
1489 ✗ lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h);
1490
1491 ✗ for (level = 1; level*level*level < w; level++);
1492 ✗ size = level*level*level;
1493 ✗ if (size != w) {
1494 ✗ av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n");
1495 ✗ return AVERROR_INVALIDDATA;
1496 }
1497 ✗ av_assert0(w == h && w == size);
1498 ✗ level *= level;
1499 ✗ if (level > MAX_LEVEL) {
1500 ✗ const int max_clut_level = sqrt(MAX_LEVEL);
1501 ✗ const int max_clut_size = max_clut_level*max_clut_level*max_clut_level;
1502 ✗ av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT "
1503 "(maximum level is %d, or %dx%d CLUT)\n",
1504 max_clut_level, max_clut_size, max_clut_size);
1505 ✗ return AVERROR(EINVAL);
1506 }
1507
1508 ✗ return allocate_3dlut(ctx, level, 0);
1509 }
1510
1511 ✗ static int update_apply_clut(FFFrameSync *fs)
1512 {
1513 ✗ AVFilterContext *ctx = fs->parent;
1514 ✗ LUT3DContext *lut3d = ctx->priv;
1515 ✗ AVFilterLink *inlink = ctx->inputs[0];
1516 AVFrame *master, *second, *out;
1517 int ret;
1518
1519 ✗ ret = ff_framesync_dualinput_get(fs, &master, &second);
1520 ✗ if (ret < 0)
1521 ✗ return ret;
1522 ✗ if (!second)
1523 ✗ return ff_filter_frame(ctx->outputs[0], master);
1524 ✗ if (lut3d->clut || !lut3d->got_clut) {
1525 ✗ if (lut3d->clut_float)
1526 ✗ update_clut_float(ctx->priv, second);
1527 ✗ else if (lut3d->clut_planar)
1528 ✗ update_clut_planar(ctx->priv, second);
1529 else
1530 ✗ update_clut_packed(ctx->priv, second);
1531 ✗ lut3d->got_clut = 1;
1532 }
1533 ✗ out = apply_lut(inlink, master);
1534 ✗ return ff_filter_frame(ctx->outputs[0], out);
1535 }
1536
1537 ✗ static av_cold int haldclut_init(AVFilterContext *ctx)
1538 {
1539 ✗ LUT3DContext *lut3d = ctx->priv;
1540 ✗ lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1541 ✗ lut3d->fs.on_event = update_apply_clut;
1542 ✗ return 0;
1543 }
1544
1545 ✗ static av_cold void haldclut_uninit(AVFilterContext *ctx)
1546 {
1547 ✗ LUT3DContext *lut3d = ctx->priv;
1548 ✗ ff_framesync_uninit(&lut3d->fs);
1549 ✗ av_freep(&lut3d->lut);
1550 ✗ }
1551
1552 ✗ FRAMESYNC_DEFINE_CLASS_EXT(haldclut, LUT3DContext, fs,
1553 &lut3d_haldclut_options[COMMON_OPTIONS_OFFSET]);
1554
1555 static const AVFilterPad haldclut_inputs[] = {
1556 {
1557 .name = "main",
1558 .type = AVMEDIA_TYPE_VIDEO,
1559 .config_props = config_input,
1560 },{
1561 .name = "clut",
1562 .type = AVMEDIA_TYPE_VIDEO,
1563 .config_props = config_clut,
1564 },
1565 };
1566
1567 static const AVFilterPad haldclut_outputs[] = {
1568 {
1569 .name = "default",
1570 .type = AVMEDIA_TYPE_VIDEO,
1571 .config_props = config_output,
1572 },
1573 };
1574
1575 const FFFilter ff_vf_haldclut = {
1576 .p.name = "haldclut",
1577 .p.description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
1578 .p.priv_class = &haldclut_class,
1579 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL |
1580 AVFILTER_FLAG_SLICE_THREADS,
1581 .priv_size = sizeof(LUT3DContext),
1582 .preinit = haldclut_framesync_preinit,
1583 .init = haldclut_init,
1584 .uninit = haldclut_uninit,
1585 .activate = activate,
1586 FILTER_INPUTS(haldclut_inputs),
1587 FILTER_OUTPUTS(haldclut_outputs),
1588 FILTER_PIXFMTS_ARRAY(pix_fmts),
1589 .process_command = process_command,
1590 };
1591 #endif
1592
1593 #endif /* CONFIG_LUT3D_FILTER || CONFIG_HALDCLUT_FILTER */
1594
1595 #if CONFIG_LUT1D_FILTER
1596
1597 enum interp_1d_mode {
1598 INTERPOLATE_1D_NEAREST,
1599 INTERPOLATE_1D_LINEAR,
1600 INTERPOLATE_1D_CUBIC,
1601 INTERPOLATE_1D_COSINE,
1602 INTERPOLATE_1D_SPLINE,
1603 NB_INTERP_1D_MODE
1604 };
1605
1606 #define MAX_1D_LEVEL 65536
1607
1608 typedef struct LUT1DContext {
1609 const AVClass *class;
1610 char *file;
1611 int interpolation; ///<interp_1d_mode
1612 struct rgbvec scale;
1613 uint8_t rgba_map[4];
1614 int step;
1615 float lut[3][MAX_1D_LEVEL];
1616 int lutsize;
1617 avfilter_action_func *interp;
1618 } LUT1DContext;
1619
1620 #undef OFFSET
1621 #define OFFSET(x) offsetof(LUT1DContext, x)
1622
1623 ✗ static void set_identity_matrix_1d(LUT1DContext *lut1d, int size)
1624 {
1625 ✗ const float c = 1. / (size - 1);
1626 int i;
1627
1628 ✗ lut1d->lutsize = size;
1629 ✗ for (i = 0; i < size; i++) {
1630 ✗ lut1d->lut[0][i] = i * c;
1631 ✗ lut1d->lut[1][i] = i * c;
1632 ✗ lut1d->lut[2][i] = i * c;
1633 }
1634 ✗ }
1635
1636 ✗ static int parse_cinespace_1d(AVFilterContext *ctx, FILE *f)
1637 {
1638 ✗ LUT1DContext *lut1d = ctx->priv;
1639 char line[MAX_LINE_SIZE];
1640 ✗ float in_min[3] = {0.0, 0.0, 0.0};
1641 ✗ float in_max[3] = {1.0, 1.0, 1.0};
1642 ✗ float out_min[3] = {0.0, 0.0, 0.0};
1643 ✗ float out_max[3] = {1.0, 1.0, 1.0};
1644 ✗ int inside_metadata = 0, size;
1645
1646 ✗ NEXT_LINE(skip_line(line));
1647 ✗ if (strncmp(line, "CSPLUTV100", 10)) {
1648 ✗ av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n");
1649 ✗ return AVERROR(EINVAL);
1650 }
1651
1652 ✗ NEXT_LINE(skip_line(line));
1653 ✗ if (strncmp(line, "1D", 2)) {
1654 ✗ av_log(ctx, AV_LOG_ERROR, "Not 1D LUT format\n");
1655 ✗ return AVERROR(EINVAL);
1656 }
1657
1658 ✗ while (1) {
1659 ✗ NEXT_LINE(skip_line(line));
1660
1661 ✗ if (!strncmp(line, "BEGIN METADATA", 14)) {
1662 ✗ inside_metadata = 1;
1663 ✗ continue;
1664 }
1665 ✗ if (!strncmp(line, "END METADATA", 12)) {
1666 ✗ inside_metadata = 0;
1667 ✗ continue;
1668 }
1669 ✗ if (inside_metadata == 0) {
1670 ✗ for (int i = 0; i < 3; i++) {
1671 ✗ int npoints = strtol(line, NULL, 0);
1672
1673 ✗ if (npoints != 2) {
1674 ✗ av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n");
1675 ✗ return AVERROR_PATCHWELCOME;
1676 }
1677
1678 ✗ NEXT_LINE(skip_line(line));
1679 ✗ if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2)
1680 ✗ return AVERROR_INVALIDDATA;
1681 ✗ NEXT_LINE(skip_line(line));
1682 ✗ if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2)
1683 ✗ return AVERROR_INVALIDDATA;
1684 ✗ NEXT_LINE(skip_line(line));
1685 }
1686
1687 ✗ size = strtol(line, NULL, 0);
1688
1689 ✗ if (size < 2 || size > MAX_1D_LEVEL) {
1690 ✗ av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1691 ✗ return AVERROR(EINVAL);
1692 }
1693
1694 ✗ lut1d->lutsize = size;
1695
1696 ✗ for (int i = 0; i < size; i++) {
1697 ✗ NEXT_LINE(skip_line(line));
1698 ✗ if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1699 ✗ return AVERROR_INVALIDDATA;
1700 ✗ lut1d->lut[0][i] *= out_max[0] - out_min[0];
1701 ✗ lut1d->lut[1][i] *= out_max[1] - out_min[1];
1702 ✗ lut1d->lut[2][i] *= out_max[2] - out_min[2];
1703 }
1704
1705 ✗ break;
1706 }
1707 }
1708
1709 ✗ lut1d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f);
1710 ✗ lut1d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f);
1711 ✗ lut1d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f);
1712
1713 ✗ return 0;
1714 }
1715
1716 ✗ static int parse_cube_1d(AVFilterContext *ctx, FILE *f)
1717 {
1718 ✗ LUT1DContext *lut1d = ctx->priv;
1719 char line[MAX_LINE_SIZE];
1720 ✗ float min[3] = {0.0, 0.0, 0.0};
1721 ✗ float max[3] = {1.0, 1.0, 1.0};
1722
1723 ✗ while (fgets(line, sizeof(line), f)) {
1724 ✗ if (!strncmp(line, "LUT_1D_SIZE", 11)) {
1725 ✗ const int size = strtol(line + 12, NULL, 0);
1726 int i;
1727
1728 ✗ if (size < 2 || size > MAX_1D_LEVEL) {
1729 ✗ av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1730 ✗ return AVERROR(EINVAL);
1731 }
1732 ✗ lut1d->lutsize = size;
1733 ✗ for (i = 0; i < size; i++) {
1734 do {
1735 ✗ try_again:
1736 ✗ NEXT_LINE(0);
1737 ✗ if (!strncmp(line, "DOMAIN_", 7)) {
1738 ✗ float *vals = NULL;
1739 ✗ if (!strncmp(line + 7, "MIN ", 4)) vals = min;
1740 ✗ else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
1741 ✗ if (!vals)
1742 ✗ return AVERROR_INVALIDDATA;
1743 ✗ if (av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2) != 3)
1744 ✗ return AVERROR_INVALIDDATA;
1745 ✗ av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
1746 ✗ min[0], min[1], min[2], max[0], max[1], max[2]);
1747 ✗ goto try_again;
1748 ✗ } else if (!strncmp(line, "LUT_1D_INPUT_RANGE ", 19)) {
1749 ✗ if (av_sscanf(line + 19, "%f %f", min, max) != 2)
1750 ✗ return AVERROR_INVALIDDATA;
1751 ✗ min[1] = min[2] = min[0];
1752 ✗ max[1] = max[2] = max[0];
1753 ✗ goto try_again;
1754 ✗ } else if (!strncmp(line, "TITLE", 5)) {
1755 ✗ goto try_again;
1756 }
1757 ✗ } while (skip_line(line));
1758 ✗ if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1759 ✗ return AVERROR_INVALIDDATA;
1760 }
1761 ✗ break;
1762 }
1763 }
1764
1765 ✗ lut1d->scale.r = av_clipf(1. / (max[0] - min[0]), 0.f, 1.f);
1766 ✗ lut1d->scale.g = av_clipf(1. / (max[1] - min[1]), 0.f, 1.f);
1767 ✗ lut1d->scale.b = av_clipf(1. / (max[2] - min[2]), 0.f, 1.f);
1768
1769 ✗ return 0;
1770 }
1771
1772 static const AVOption lut1d_options[] = {
1773 { "file", "set 1D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = TFLAGS },
1774 { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_1D_LINEAR}, 0, NB_INTERP_1D_MODE-1, TFLAGS, .unit = "interp_mode" },
1775 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_NEAREST}, 0, 0, TFLAGS, .unit = "interp_mode" },
1776 { "linear", "use values from the linear interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_LINEAR}, 0, 0, TFLAGS, .unit = "interp_mode" },
1777 { "cosine", "use values from the cosine interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_COSINE}, 0, 0, TFLAGS, .unit = "interp_mode" },
1778 { "cubic", "use values from the cubic interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_CUBIC}, 0, 0, TFLAGS, .unit = "interp_mode" },
1779 { "spline", "use values from the spline interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_SPLINE}, 0, 0, TFLAGS, .unit = "interp_mode" },
1780 { NULL }
1781 };
1782
1783 AVFILTER_DEFINE_CLASS(lut1d);
1784
1785 ✗ static inline float interp_1d_nearest(const LUT1DContext *lut1d,
1786 int idx, const float s)
1787 {
1788 ✗ return lut1d->lut[idx][NEAR(s)];
1789 }
1790
1791 #define NEXT1D(x) (FFMIN((int)(x) + 1, lut1d->lutsize - 1))
1792
1793 ✗ static inline float interp_1d_linear(const LUT1DContext *lut1d,
1794 int idx, const float s)
1795 {
1796 ✗ const int prev = PREV(s);
1797 ✗ const int next = NEXT1D(s);
1798 ✗ const float d = s - prev;
1799 ✗ const float p = lut1d->lut[idx][prev];
1800 ✗ const float n = lut1d->lut[idx][next];
1801
1802 ✗ return lerpf(p, n, d);
1803 }
1804
1805 ✗ static inline float interp_1d_cosine(const LUT1DContext *lut1d,
1806 int idx, const float s)
1807 {
1808 ✗ const int prev = PREV(s);
1809 ✗ const int next = NEXT1D(s);
1810 ✗ const float d = s - prev;
1811 ✗ const float p = lut1d->lut[idx][prev];
1812 ✗ const float n = lut1d->lut[idx][next];
1813 ✗ const float m = (1.f - cosf(d * M_PI)) * .5f;
1814
1815 ✗ return lerpf(p, n, m);
1816 }
1817
1818 ✗ static inline float interp_1d_cubic(const LUT1DContext *lut1d,
1819 int idx, const float s)
1820 {
1821 ✗ const int prev = PREV(s);
1822 ✗ const int next = NEXT1D(s);
1823 ✗ const float mu = s - prev;
1824 float a0, a1, a2, a3, mu2;
1825
1826 ✗ float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1827 ✗ float y1 = lut1d->lut[idx][prev];
1828 ✗ float y2 = lut1d->lut[idx][next];
1829 ✗ float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1830
1831
1832 ✗ mu2 = mu * mu;
1833 ✗ a0 = y3 - y2 - y0 + y1;
1834 ✗ a1 = y0 - y1 - a0;
1835 ✗ a2 = y2 - y0;
1836 ✗ a3 = y1;
1837
1838 ✗ return a0 * mu * mu2 + a1 * mu2 + a2 * mu + a3;
1839 }
1840
1841 ✗ static inline float interp_1d_spline(const LUT1DContext *lut1d,
1842 int idx, const float s)
1843 {
1844 ✗ const int prev = PREV(s);
1845 ✗ const int next = NEXT1D(s);
1846 ✗ const float x = s - prev;
1847 float c0, c1, c2, c3;
1848
1849 ✗ float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1850 ✗ float y1 = lut1d->lut[idx][prev];
1851 ✗ float y2 = lut1d->lut[idx][next];
1852 ✗ float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1853
1854 ✗ c0 = y1;
1855 ✗ c1 = .5f * (y2 - y0);
1856 ✗ c2 = y0 - 2.5f * y1 + 2.f * y2 - .5f * y3;
1857 ✗ c3 = .5f * (y3 - y0) + 1.5f * (y1 - y2);
1858
1859 ✗ return ((c3 * x + c2) * x + c1) * x + c0;
1860 }
1861
1862 #define DEFINE_INTERP_FUNC_PLANAR_1D(name, nbits, depth) \
1863 static int interp_1d_##nbits##_##name##_p##depth(AVFilterContext *ctx, \
1864 void *arg, int jobnr, \
1865 int nb_jobs) \
1866 { \
1867 int x, y; \
1868 const LUT1DContext *lut1d = ctx->priv; \
1869 const ThreadData *td = arg; \
1870 const AVFrame *in = td->in; \
1871 const AVFrame *out = td->out; \
1872 const int direct = out == in; \
1873 const int slice_start = (in->height * jobnr ) / nb_jobs; \
1874 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
1875 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
1876 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
1877 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
1878 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
1879 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
1880 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
1881 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
1882 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
1883 const float factor = (1 << depth) - 1; \
1884 const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \
1885 const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \
1886 const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \
1887 \
1888 for (y = slice_start; y < slice_end; y++) { \
1889 uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
1890 uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
1891 uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
1892 uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
1893 const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
1894 const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
1895 const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
1896 const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
1897 for (x = 0; x < in->width; x++) { \
1898 float r = srcr[x] * scale_r; \
1899 float g = srcg[x] * scale_g; \
1900 float b = srcb[x] * scale_b; \
1901 r = interp_1d_##name(lut1d, 0, r); \
1902 g = interp_1d_##name(lut1d, 1, g); \
1903 b = interp_1d_##name(lut1d, 2, b); \
1904 dstr[x] = av_clip_uintp2(r * factor, depth); \
1905 dstg[x] = av_clip_uintp2(g * factor, depth); \
1906 dstb[x] = av_clip_uintp2(b * factor, depth); \
1907 if (!direct && in->linesize[3]) \
1908 dsta[x] = srca[x]; \
1909 } \
1910 grow += out->linesize[0]; \
1911 brow += out->linesize[1]; \
1912 rrow += out->linesize[2]; \
1913 arow += out->linesize[3]; \
1914 srcgrow += in->linesize[0]; \
1915 srcbrow += in->linesize[1]; \
1916 srcrrow += in->linesize[2]; \
1917 srcarow += in->linesize[3]; \
1918 } \
1919 return 0; \
1920 }
1921
1922 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 8, 8)
1923 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(linear, 8, 8)
1924 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 8, 8)
1925 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 8, 8)
1926 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(spline, 8, 8)
1927
1928 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 9)
1929 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 9)
1930 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 9)
1931 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 9)
1932 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 9)
1933
1934 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 10)
1935 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 10)
1936 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 10)
1937 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 10)
1938 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 10)
1939
1940 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 12)
1941 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 12)
1942 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 12)
1943 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 12)
1944 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 12)
1945
1946 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 14)
1947 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 14)
1948 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 14)
1949 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 14)
1950 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 14)
1951
1952 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 16)
1953 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 16)
1954 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 16)
1955 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 16)
1956 ✗ DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 16)
1957
1958 #define DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(name, depth) \
1959 static int interp_1d_##name##_pf##depth(AVFilterContext *ctx, \
1960 void *arg, int jobnr, \
1961 int nb_jobs) \
1962 { \
1963 int x, y; \
1964 const LUT1DContext *lut1d = ctx->priv; \
1965 const ThreadData *td = arg; \
1966 const AVFrame *in = td->in; \
1967 const AVFrame *out = td->out; \
1968 const int direct = out == in; \
1969 const int slice_start = (in->height * jobnr ) / nb_jobs; \
1970 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
1971 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
1972 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
1973 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
1974 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
1975 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
1976 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
1977 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
1978 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
1979 const float lutsize = lut1d->lutsize - 1; \
1980 const float scale_r = lut1d->scale.r * lutsize; \
1981 const float scale_g = lut1d->scale.g * lutsize; \
1982 const float scale_b = lut1d->scale.b * lutsize; \
1983 \
1984 for (y = slice_start; y < slice_end; y++) { \
1985 float *dstg = (float *)grow; \
1986 float *dstb = (float *)brow; \
1987 float *dstr = (float *)rrow; \
1988 float *dsta = (float *)arow; \
1989 const float *srcg = (const float *)srcgrow; \
1990 const float *srcb = (const float *)srcbrow; \
1991 const float *srcr = (const float *)srcrrow; \
1992 const float *srca = (const float *)srcarow; \
1993 for (x = 0; x < in->width; x++) { \
1994 float r = av_clipf(sanitizef(srcr[x]) * scale_r, 0.0f, lutsize); \
1995 float g = av_clipf(sanitizef(srcg[x]) * scale_g, 0.0f, lutsize); \
1996 float b = av_clipf(sanitizef(srcb[x]) * scale_b, 0.0f, lutsize); \
1997 r = interp_1d_##name(lut1d, 0, r); \
1998 g = interp_1d_##name(lut1d, 1, g); \
1999 b = interp_1d_##name(lut1d, 2, b); \
2000 dstr[x] = r; \
2001 dstg[x] = g; \
2002 dstb[x] = b; \
2003 if (!direct && in->linesize[3]) \
2004 dsta[x] = srca[x]; \
2005 } \
2006 grow += out->linesize[0]; \
2007 brow += out->linesize[1]; \
2008 rrow += out->linesize[2]; \
2009 arow += out->linesize[3]; \
2010 srcgrow += in->linesize[0]; \
2011 srcbrow += in->linesize[1]; \
2012 srcrrow += in->linesize[2]; \
2013 srcarow += in->linesize[3]; \
2014 } \
2015 return 0; \
2016 }
2017
2018 ✗ DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(nearest, 32)
2019 ✗ DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(linear, 32)
2020 ✗ DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cosine, 32)
2021 ✗ DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cubic, 32)
2022 ✗ DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(spline, 32)
2023
2024 #define DEFINE_INTERP_FUNC_1D(name, nbits) \
2025 static int interp_1d_##nbits##_##name(AVFilterContext *ctx, void *arg, \
2026 int jobnr, int nb_jobs) \
2027 { \
2028 int x, y; \
2029 const LUT1DContext *lut1d = ctx->priv; \
2030 const ThreadData *td = arg; \
2031 const AVFrame *in = td->in; \
2032 const AVFrame *out = td->out; \
2033 const int direct = out == in; \
2034 const int step = lut1d->step; \
2035 const uint8_t r = lut1d->rgba_map[R]; \
2036 const uint8_t g = lut1d->rgba_map[G]; \
2037 const uint8_t b = lut1d->rgba_map[B]; \
2038 const uint8_t a = lut1d->rgba_map[A]; \
2039 const int slice_start = (in->height * jobnr ) / nb_jobs; \
2040 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
2041 uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
2042 const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
2043 const float factor = (1 << nbits) - 1; \
2044 const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \
2045 const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \
2046 const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \
2047 \
2048 for (y = slice_start; y < slice_end; y++) { \
2049 uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
2050 const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
2051 for (x = 0; x < in->width * step; x += step) { \
2052 float rr = src[x + r] * scale_r; \
2053 float gg = src[x + g] * scale_g; \
2054 float bb = src[x + b] * scale_b; \
2055 rr = interp_1d_##name(lut1d, 0, rr); \
2056 gg = interp_1d_##name(lut1d, 1, gg); \
2057 bb = interp_1d_##name(lut1d, 2, bb); \
2058 dst[x + r] = av_clip_uint##nbits(rr * factor); \
2059 dst[x + g] = av_clip_uint##nbits(gg * factor); \
2060 dst[x + b] = av_clip_uint##nbits(bb * factor); \
2061 if (!direct && step == 4) \
2062 dst[x + a] = src[x + a]; \
2063 } \
2064 dstrow += out->linesize[0]; \
2065 srcrow += in ->linesize[0]; \
2066 } \
2067 return 0; \
2068 }
2069
2070 ✗ DEFINE_INTERP_FUNC_1D(nearest, 8)
2071 ✗ DEFINE_INTERP_FUNC_1D(linear, 8)
2072 ✗ DEFINE_INTERP_FUNC_1D(cosine, 8)
2073 ✗ DEFINE_INTERP_FUNC_1D(cubic, 8)
2074 ✗ DEFINE_INTERP_FUNC_1D(spline, 8)
2075
2076 ✗ DEFINE_INTERP_FUNC_1D(nearest, 16)
2077 ✗ DEFINE_INTERP_FUNC_1D(linear, 16)
2078 ✗ DEFINE_INTERP_FUNC_1D(cosine, 16)
2079 ✗ DEFINE_INTERP_FUNC_1D(cubic, 16)
2080 ✗ DEFINE_INTERP_FUNC_1D(spline, 16)
2081
2082 ✗ static int config_input_1d(AVFilterLink *inlink)
2083 {
2084 int depth, is16bit, isfloat, planar;
2085 ✗ LUT1DContext *lut1d = inlink->dst->priv;
2086 ✗ const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
2087
2088 ✗ depth = desc->comp[0].depth;
2089 ✗ is16bit = desc->comp[0].depth > 8;
2090 ✗ planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR;
2091 ✗ isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
2092 ✗ ff_fill_rgba_map(lut1d->rgba_map, inlink->format);
2093 ✗ lut1d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
2094
2095 #define SET_FUNC_1D(name) do { \
2096 if (planar && !isfloat) { \
2097 switch (depth) { \
2098 case 8: lut1d->interp = interp_1d_8_##name##_p8; break; \
2099 case 9: lut1d->interp = interp_1d_16_##name##_p9; break; \
2100 case 10: lut1d->interp = interp_1d_16_##name##_p10; break; \
2101 case 12: lut1d->interp = interp_1d_16_##name##_p12; break; \
2102 case 14: lut1d->interp = interp_1d_16_##name##_p14; break; \
2103 case 16: lut1d->interp = interp_1d_16_##name##_p16; break; \
2104 } \
2105 } else if (isfloat) { lut1d->interp = interp_1d_##name##_pf32; \
2106 } else if (is16bit) { lut1d->interp = interp_1d_16_##name; \
2107 } else { lut1d->interp = interp_1d_8_##name; } \
2108 } while (0)
2109
2110 ✗ switch (lut1d->interpolation) {
2111 ✗ case INTERPOLATE_1D_NEAREST: SET_FUNC_1D(nearest); break;
2112 ✗ case INTERPOLATE_1D_LINEAR: SET_FUNC_1D(linear); break;
2113 ✗ case INTERPOLATE_1D_COSINE: SET_FUNC_1D(cosine); break;
2114 ✗ case INTERPOLATE_1D_CUBIC: SET_FUNC_1D(cubic); break;
2115 ✗ case INTERPOLATE_1D_SPLINE: SET_FUNC_1D(spline); break;
2116 ✗ default:
2117 ✗ av_assert0(0);
2118 }
2119
2120 ✗ return 0;
2121 }
2122
2123 ✗ static av_cold int lut1d_init(AVFilterContext *ctx)
2124 {
2125 int ret;
2126 FILE *f;
2127 const char *ext;
2128 ✗ LUT1DContext *lut1d = ctx->priv;
2129
2130 ✗ lut1d->scale.r = lut1d->scale.g = lut1d->scale.b = 1.f;
2131
2132 ✗ if (!lut1d->file) {
2133 ✗ set_identity_matrix_1d(lut1d, 32);
2134 ✗ return 0;
2135 }
2136
2137 ✗ f = avpriv_fopen_utf8(lut1d->file, "r");
2138 ✗ if (!f) {
2139 ✗ ret = AVERROR(errno);
2140 ✗ av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut1d->file, av_err2str(ret));
2141 ✗ return ret;
2142 }
2143
2144 ✗ ext = strrchr(lut1d->file, '.');
2145 ✗ if (!ext) {
2146 ✗ av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
2147 ✗ ret = AVERROR_INVALIDDATA;
2148 ✗ goto end;
2149 }
2150 ✗ ext++;
2151
2152 ✗ if (!av_strcasecmp(ext, "cube") || !av_strcasecmp(ext, "1dlut")) {
2153 ✗ ret = parse_cube_1d(ctx, f);
2154 ✗ } else if (!av_strcasecmp(ext, "csp")) {
2155 ✗ ret = parse_cinespace_1d(ctx, f);
2156 } else {
2157 ✗ av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
2158 ✗ ret = AVERROR(EINVAL);
2159 }
2160
2161 ✗ if (!ret && !lut1d->lutsize) {
2162 ✗ av_log(ctx, AV_LOG_ERROR, "1D LUT is empty\n");
2163 ✗ ret = AVERROR_INVALIDDATA;
2164 }
2165
2166 ✗ end:
2167 ✗ fclose(f);
2168 ✗ return ret;
2169 }
2170
2171 ✗ static AVFrame *apply_1d_lut(AVFilterLink *inlink, AVFrame *in)
2172 {
2173 ✗ AVFilterContext *ctx = inlink->dst;
2174 ✗ LUT1DContext *lut1d = ctx->priv;
2175 ✗ AVFilterLink *outlink = inlink->dst->outputs[0];
2176 AVFrame *out;
2177 ThreadData td;
2178
2179 ✗ if (av_frame_is_writable(in)) {
2180 ✗ out = in;
2181 } else {
2182 ✗ out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
2183 ✗ if (!out) {
2184 ✗ av_frame_free(&in);
2185 ✗ return NULL;
2186 }
2187 ✗ av_frame_copy_props(out, in);
2188 }
2189
2190 ✗ td.in = in;
2191 ✗ td.out = out;
2192 ✗ ff_filter_execute(ctx, lut1d->interp, &td, NULL,
2193 ✗ FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
2194
2195 ✗ if (out != in)
2196 ✗ av_frame_free(&in);
2197
2198 ✗ return out;
2199 }
2200
2201 ✗ static int filter_frame_1d(AVFilterLink *inlink, AVFrame *in)
2202 {
2203 ✗ AVFilterLink *outlink = inlink->dst->outputs[0];
2204 ✗ AVFrame *out = apply_1d_lut(inlink, in);
2205 ✗ if (!out)
2206 ✗ return AVERROR(ENOMEM);
2207 ✗ return ff_filter_frame(outlink, out);
2208 }
2209
2210 ✗ static int lut1d_process_command(AVFilterContext *ctx, const char *cmd, const char *args,
2211 char *res, int res_len, int flags)
2212 {
2213 ✗ LUT1DContext *lut1d = ctx->priv;
2214 int ret;
2215
2216 ✗ ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
2217 ✗ if (ret < 0)
2218 ✗ return ret;
2219
2220 ✗ ret = lut1d_init(ctx);
2221 ✗ if (ret < 0) {
2222 ✗ set_identity_matrix_1d(lut1d, 32);
2223 ✗ return ret;
2224 }
2225 ✗ return config_input_1d(ctx->inputs[0]);
2226 }
2227
2228 static const AVFilterPad lut1d_inputs[] = {
2229 {
2230 .name = "default",
2231 .type = AVMEDIA_TYPE_VIDEO,
2232 .filter_frame = filter_frame_1d,
2233 .config_props = config_input_1d,
2234 },
2235 };
2236
2237 const FFFilter ff_vf_lut1d = {
2238 .p.name = "lut1d",
2239 .p.description = NULL_IF_CONFIG_SMALL("Adjust colors using a 1D LUT."),
2240 .p.priv_class = &lut1d_class,
2241 .p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC |
2242 AVFILTER_FLAG_SLICE_THREADS,
2243 .priv_size = sizeof(LUT1DContext),
2244 .init = lut1d_init,
2245 FILTER_INPUTS(lut1d_inputs),
2246 FILTER_OUTPUTS(ff_video_default_filterpad),
2247 FILTER_PIXFMTS_ARRAY(pix_fmts),
2248 .process_command = lut1d_process_command,
2249 };
2250 #endif
2251