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 "config_components.h"
28
29 #include "float.h"
30
31 #include "libavutil/opt.h"
32 #include "libavutil/file.h"
33 #include "libavutil/intreadwrite.h"
34 #include "libavutil/intfloat.h"
35 #include "libavutil/avassert.h"
36 #include "libavutil/avstring.h"
37 #include "drawutils.h"
38 #include "formats.h"
39 #include "internal.h"
40 #include "video.h"
41 #include "lut3d.h"
42
43 #define R 0
44 #define G 1
45 #define B 2
46 #define A 3
47
48 #define OFFSET(x) offsetof(LUT3DContext, x)
49 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
50 #define TFLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM
51 #define COMMON_OPTIONS \
52 { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, NB_INTERP_MODE-1, TFLAGS, "interp_mode" }, \
53 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_NEAREST}, 0, 0, TFLAGS, "interp_mode" }, \
54 { "trilinear", "interpolate values using the 8 points defining a cube", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TRILINEAR}, 0, 0, TFLAGS, "interp_mode" }, \
55 { "tetrahedral", "interpolate values using a tetrahedron", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, 0, TFLAGS, "interp_mode" }, \
56 { "pyramid", "interpolate values using a pyramid", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PYRAMID}, 0, 0, TFLAGS, "interp_mode" }, \
57 { "prism", "interpolate values using a prism", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PRISM}, 0, 0, TFLAGS, "interp_mode" }, \
58 { NULL }
59
60 #define EXPONENT_MASK 0x7F800000
61 #define MANTISSA_MASK 0x007FFFFF
62 #define SIGN_MASK 0x80000000
63
sanitizef(float f)64 static inline float sanitizef(float f)
65 {
66 union av_intfloat32 t;
67 t.f = f;
68
69 if ((t.i & EXPONENT_MASK) == EXPONENT_MASK) {
70 if ((t.i & MANTISSA_MASK) != 0) {
71 // NAN
72 return 0.0f;
73 } else if (t.i & SIGN_MASK) {
74 // -INF
75 return -FLT_MAX;
76 } else {
77 // +INF
78 return FLT_MAX;
79 }
80 }
81 return f;
82 }
83
lerpf(float v0,float v1,float f)84 static inline float lerpf(float v0, float v1, float f)
85 {
86 return v0 + (v1 - v0) * f;
87 }
88
lerp(const struct rgbvec * v0,const struct rgbvec * v1,float f)89 static inline struct rgbvec lerp(const struct rgbvec *v0, const struct rgbvec *v1, float f)
90 {
91 struct rgbvec v = {
92 lerpf(v0->r, v1->r, f), lerpf(v0->g, v1->g, f), lerpf(v0->b, v1->b, f)
93 };
94 return v;
95 }
96
97 #define NEAR(x) ((int)((x) + .5))
98 #define PREV(x) ((int)(x))
99 #define NEXT(x) (FFMIN((int)(x) + 1, lut3d->lutsize - 1))
100
101 /**
102 * Get the nearest defined point
103 */
interp_nearest(const LUT3DContext * lut3d,const struct rgbvec * s)104 static inline struct rgbvec interp_nearest(const LUT3DContext *lut3d,
105 const struct rgbvec *s)
106 {
107 return lut3d->lut[NEAR(s->r) * lut3d->lutsize2 + NEAR(s->g) * lut3d->lutsize + NEAR(s->b)];
108 }
109
110 /**
111 * Interpolate using the 8 vertices of a cube
112 * @see https://en.wikipedia.org/wiki/Trilinear_interpolation
113 */
interp_trilinear(const LUT3DContext * lut3d,const struct rgbvec * s)114 static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d,
115 const struct rgbvec *s)
116 {
117 const int lutsize2 = lut3d->lutsize2;
118 const int lutsize = lut3d->lutsize;
119 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
120 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
121 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
122 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
123 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
124 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
125 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
126 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
127 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
128 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
129 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
130 const struct rgbvec c00 = lerp(&c000, &c100, d.r);
131 const struct rgbvec c10 = lerp(&c010, &c110, d.r);
132 const struct rgbvec c01 = lerp(&c001, &c101, d.r);
133 const struct rgbvec c11 = lerp(&c011, &c111, d.r);
134 const struct rgbvec c0 = lerp(&c00, &c10, d.g);
135 const struct rgbvec c1 = lerp(&c01, &c11, d.g);
136 const struct rgbvec c = lerp(&c0, &c1, d.b);
137 return c;
138 }
139
interp_pyramid(const LUT3DContext * lut3d,const struct rgbvec * s)140 static inline struct rgbvec interp_pyramid(const LUT3DContext *lut3d,
141 const struct rgbvec *s)
142 {
143 const int lutsize2 = lut3d->lutsize2;
144 const int lutsize = lut3d->lutsize;
145 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
146 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
147 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
148 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
149 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
150 struct rgbvec c;
151
152 if (d.g > d.r && d.b > d.r) {
153 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
154 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
155 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
156
157 c.r = c000.r + (c111.r - c011.r) * d.r + (c010.r - c000.r) * d.g + (c001.r - c000.r) * d.b +
158 (c011.r - c001.r - c010.r + c000.r) * d.g * d.b;
159 c.g = c000.g + (c111.g - c011.g) * d.r + (c010.g - c000.g) * d.g + (c001.g - c000.g) * d.b +
160 (c011.g - c001.g - c010.g + c000.g) * d.g * d.b;
161 c.b = c000.b + (c111.b - c011.b) * d.r + (c010.b - c000.b) * d.g + (c001.b - c000.b) * d.b +
162 (c011.b - c001.b - c010.b + c000.b) * d.g * d.b;
163 } else if (d.r > d.g && d.b > d.g) {
164 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
165 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
166 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
167
168 c.r = c000.r + (c100.r - c000.r) * d.r + (c111.r - c101.r) * d.g + (c001.r - c000.r) * d.b +
169 (c101.r - c001.r - c100.r + c000.r) * d.r * d.b;
170 c.g = c000.g + (c100.g - c000.g) * d.r + (c111.g - c101.g) * d.g + (c001.g - c000.g) * d.b +
171 (c101.g - c001.g - c100.g + c000.g) * d.r * d.b;
172 c.b = c000.b + (c100.b - c000.b) * d.r + (c111.b - c101.b) * d.g + (c001.b - c000.b) * d.b +
173 (c101.b - c001.b - c100.b + c000.b) * d.r * d.b;
174 } else {
175 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
176 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
177 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
178
179 c.r = c000.r + (c100.r - c000.r) * d.r + (c010.r - c000.r) * d.g + (c111.r - c110.r) * d.b +
180 (c110.r - c100.r - c010.r + c000.r) * d.r * d.g;
181 c.g = c000.g + (c100.g - c000.g) * d.r + (c010.g - c000.g) * d.g + (c111.g - c110.g) * d.b +
182 (c110.g - c100.g - c010.g + c000.g) * d.r * d.g;
183 c.b = c000.b + (c100.b - c000.b) * d.r + (c010.b - c000.b) * d.g + (c111.b - c110.b) * d.b +
184 (c110.b - c100.b - c010.b + c000.b) * d.r * d.g;
185 }
186
187 return c;
188 }
189
interp_prism(const LUT3DContext * lut3d,const struct rgbvec * s)190 static inline struct rgbvec interp_prism(const LUT3DContext *lut3d,
191 const struct rgbvec *s)
192 {
193 const int lutsize2 = lut3d->lutsize2;
194 const int lutsize = lut3d->lutsize;
195 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
196 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
197 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
198 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
199 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
200 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
201 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
202 struct rgbvec c;
203
204 if (d.b > d.r) {
205 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
206 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
207
208 c.r = c000.r + (c001.r - c000.r) * d.b + (c101.r - c001.r) * d.r + (c010.r - c000.r) * d.g +
209 (c000.r - c010.r - c001.r + c011.r) * d.b * d.g +
210 (c001.r - c011.r - c101.r + c111.r) * d.r * d.g;
211 c.g = c000.g + (c001.g - c000.g) * d.b + (c101.g - c001.g) * d.r + (c010.g - c000.g) * d.g +
212 (c000.g - c010.g - c001.g + c011.g) * d.b * d.g +
213 (c001.g - c011.g - c101.g + c111.g) * d.r * d.g;
214 c.b = c000.b + (c001.b - c000.b) * d.b + (c101.b - c001.b) * d.r + (c010.b - c000.b) * d.g +
215 (c000.b - c010.b - c001.b + c011.b) * d.b * d.g +
216 (c001.b - c011.b - c101.b + c111.b) * d.r * d.g;
217 } else {
218 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
219 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
220
221 c.r = c000.r + (c101.r - c100.r) * d.b + (c100.r - c000.r) * d.r + (c010.r - c000.r) * d.g +
222 (c100.r - c110.r - c101.r + c111.r) * d.b * d.g +
223 (c000.r - c010.r - c100.r + c110.r) * d.r * d.g;
224 c.g = c000.g + (c101.g - c100.g) * d.b + (c100.g - c000.g) * d.r + (c010.g - c000.g) * d.g +
225 (c100.g - c110.g - c101.g + c111.g) * d.b * d.g +
226 (c000.g - c010.g - c100.g + c110.g) * d.r * d.g;
227 c.b = c000.b + (c101.b - c100.b) * d.b + (c100.b - c000.b) * d.r + (c010.b - c000.b) * d.g +
228 (c100.b - c110.b - c101.b + c111.b) * d.b * d.g +
229 (c000.b - c010.b - c100.b + c110.b) * d.r * d.g;
230 }
231
232 return c;
233 }
234
235 /**
236 * Tetrahedral interpolation. Based on code found in Truelight Software Library paper.
237 * @see http://www.filmlight.ltd.uk/pdf/whitepapers/FL-TL-TN-0057-SoftwareLib.pdf
238 */
interp_tetrahedral(const LUT3DContext * lut3d,const struct rgbvec * s)239 static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d,
240 const struct rgbvec *s)
241 {
242 const int lutsize2 = lut3d->lutsize2;
243 const int lutsize = lut3d->lutsize;
244 const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)};
245 const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)};
246 const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]};
247 const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]];
248 const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]];
249 struct rgbvec c;
250 if (d.r > d.g) {
251 if (d.g > d.b) {
252 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
253 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
254 c.r = (1-d.r) * c000.r + (d.r-d.g) * c100.r + (d.g-d.b) * c110.r + (d.b) * c111.r;
255 c.g = (1-d.r) * c000.g + (d.r-d.g) * c100.g + (d.g-d.b) * c110.g + (d.b) * c111.g;
256 c.b = (1-d.r) * c000.b + (d.r-d.g) * c100.b + (d.g-d.b) * c110.b + (d.b) * c111.b;
257 } else if (d.r > d.b) {
258 const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]];
259 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
260 c.r = (1-d.r) * c000.r + (d.r-d.b) * c100.r + (d.b-d.g) * c101.r + (d.g) * c111.r;
261 c.g = (1-d.r) * c000.g + (d.r-d.b) * c100.g + (d.b-d.g) * c101.g + (d.g) * c111.g;
262 c.b = (1-d.r) * c000.b + (d.r-d.b) * c100.b + (d.b-d.g) * c101.b + (d.g) * c111.b;
263 } else {
264 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
265 const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]];
266 c.r = (1-d.b) * c000.r + (d.b-d.r) * c001.r + (d.r-d.g) * c101.r + (d.g) * c111.r;
267 c.g = (1-d.b) * c000.g + (d.b-d.r) * c001.g + (d.r-d.g) * c101.g + (d.g) * c111.g;
268 c.b = (1-d.b) * c000.b + (d.b-d.r) * c001.b + (d.r-d.g) * c101.b + (d.g) * c111.b;
269 }
270 } else {
271 if (d.b > d.g) {
272 const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]];
273 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
274 c.r = (1-d.b) * c000.r + (d.b-d.g) * c001.r + (d.g-d.r) * c011.r + (d.r) * c111.r;
275 c.g = (1-d.b) * c000.g + (d.b-d.g) * c001.g + (d.g-d.r) * c011.g + (d.r) * c111.g;
276 c.b = (1-d.b) * c000.b + (d.b-d.g) * c001.b + (d.g-d.r) * c011.b + (d.r) * c111.b;
277 } else if (d.b > d.r) {
278 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
279 const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]];
280 c.r = (1-d.g) * c000.r + (d.g-d.b) * c010.r + (d.b-d.r) * c011.r + (d.r) * c111.r;
281 c.g = (1-d.g) * c000.g + (d.g-d.b) * c010.g + (d.b-d.r) * c011.g + (d.r) * c111.g;
282 c.b = (1-d.g) * c000.b + (d.g-d.b) * c010.b + (d.b-d.r) * c011.b + (d.r) * c111.b;
283 } else {
284 const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]];
285 const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]];
286 c.r = (1-d.g) * c000.r + (d.g-d.r) * c010.r + (d.r-d.b) * c110.r + (d.b) * c111.r;
287 c.g = (1-d.g) * c000.g + (d.g-d.r) * c010.g + (d.r-d.b) * c110.g + (d.b) * c111.g;
288 c.b = (1-d.g) * c000.b + (d.g-d.r) * c010.b + (d.r-d.b) * c110.b + (d.b) * c111.b;
289 }
290 }
291 return c;
292 }
293
prelut_interp_1d_linear(const Lut3DPreLut * prelut,int idx,const float s)294 static inline float prelut_interp_1d_linear(const Lut3DPreLut *prelut,
295 int idx, const float s)
296 {
297 const int lut_max = prelut->size - 1;
298 const float scaled = (s - prelut->min[idx]) * prelut->scale[idx];
299 const float x = av_clipf(scaled, 0.0f, lut_max);
300 const int prev = PREV(x);
301 const int next = FFMIN((int)(x) + 1, lut_max);
302 const float p = prelut->lut[idx][prev];
303 const float n = prelut->lut[idx][next];
304 const float d = x - (float)prev;
305 return lerpf(p, n, d);
306 }
307
apply_prelut(const Lut3DPreLut * prelut,const struct rgbvec * s)308 static inline struct rgbvec apply_prelut(const Lut3DPreLut *prelut,
309 const struct rgbvec *s)
310 {
311 struct rgbvec c;
312
313 if (prelut->size <= 0)
314 return *s;
315
316 c.r = prelut_interp_1d_linear(prelut, 0, s->r);
317 c.g = prelut_interp_1d_linear(prelut, 1, s->g);
318 c.b = prelut_interp_1d_linear(prelut, 2, s->b);
319 return c;
320 }
321
322 #define DEFINE_INTERP_FUNC_PLANAR(name, nbits, depth) \
323 static int interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
324 { \
325 int x, y; \
326 const LUT3DContext *lut3d = ctx->priv; \
327 const Lut3DPreLut *prelut = &lut3d->prelut; \
328 const ThreadData *td = arg; \
329 const AVFrame *in = td->in; \
330 const AVFrame *out = td->out; \
331 const int direct = out == in; \
332 const int slice_start = (in->height * jobnr ) / nb_jobs; \
333 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
334 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
335 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
336 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
337 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
338 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
339 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
340 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
341 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
342 const float lut_max = lut3d->lutsize - 1; \
343 const float scale_f = 1.0f / ((1<<depth) - 1); \
344 const float scale_r = lut3d->scale.r * lut_max; \
345 const float scale_g = lut3d->scale.g * lut_max; \
346 const float scale_b = lut3d->scale.b * lut_max; \
347 \
348 for (y = slice_start; y < slice_end; y++) { \
349 uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
350 uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
351 uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
352 uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
353 const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
354 const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
355 const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
356 const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
357 for (x = 0; x < in->width; x++) { \
358 const struct rgbvec rgb = {srcr[x] * scale_f, \
359 srcg[x] * scale_f, \
360 srcb[x] * scale_f}; \
361 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
362 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
363 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
364 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
365 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
366 dstr[x] = av_clip_uintp2(vec.r * (float)((1<<depth) - 1), depth); \
367 dstg[x] = av_clip_uintp2(vec.g * (float)((1<<depth) - 1), depth); \
368 dstb[x] = av_clip_uintp2(vec.b * (float)((1<<depth) - 1), depth); \
369 if (!direct && in->linesize[3]) \
370 dsta[x] = srca[x]; \
371 } \
372 grow += out->linesize[0]; \
373 brow += out->linesize[1]; \
374 rrow += out->linesize[2]; \
375 arow += out->linesize[3]; \
376 srcgrow += in->linesize[0]; \
377 srcbrow += in->linesize[1]; \
378 srcrrow += in->linesize[2]; \
379 srcarow += in->linesize[3]; \
380 } \
381 return 0; \
382 }
383
384 DEFINE_INTERP_FUNC_PLANAR(nearest, 8, 8)
385 DEFINE_INTERP_FUNC_PLANAR(trilinear, 8, 8)
386 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 8, 8)
387 DEFINE_INTERP_FUNC_PLANAR(pyramid, 8, 8)
388 DEFINE_INTERP_FUNC_PLANAR(prism, 8, 8)
389
390 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 9)
391 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 9)
392 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 9)
393 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 9)
394 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 9)
395
396 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 10)
397 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 10)
398 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 10)
399 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 10)
400 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 10)
401
402 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 12)
403 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 12)
404 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 12)
405 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 12)
406 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 12)
407
408 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 14)
409 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 14)
410 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 14)
411 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 14)
412 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 14)
413
414 DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 16)
415 DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 16)
416 DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 16)
417 DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 16)
418 DEFINE_INTERP_FUNC_PLANAR(prism, 16, 16)
419
420 #define DEFINE_INTERP_FUNC_PLANAR_FLOAT(name, depth) \
421 static int interp_##name##_pf##depth(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
422 { \
423 int x, y; \
424 const LUT3DContext *lut3d = ctx->priv; \
425 const Lut3DPreLut *prelut = &lut3d->prelut; \
426 const ThreadData *td = arg; \
427 const AVFrame *in = td->in; \
428 const AVFrame *out = td->out; \
429 const int direct = out == in; \
430 const int slice_start = (in->height * jobnr ) / nb_jobs; \
431 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
432 uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
433 uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
434 uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
435 uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
436 const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
437 const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
438 const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
439 const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
440 const float lut_max = lut3d->lutsize - 1; \
441 const float scale_r = lut3d->scale.r * lut_max; \
442 const float scale_g = lut3d->scale.g * lut_max; \
443 const float scale_b = lut3d->scale.b * lut_max; \
444 \
445 for (y = slice_start; y < slice_end; y++) { \
446 float *dstg = (float *)grow; \
447 float *dstb = (float *)brow; \
448 float *dstr = (float *)rrow; \
449 float *dsta = (float *)arow; \
450 const float *srcg = (const float *)srcgrow; \
451 const float *srcb = (const float *)srcbrow; \
452 const float *srcr = (const float *)srcrrow; \
453 const float *srca = (const float *)srcarow; \
454 for (x = 0; x < in->width; x++) { \
455 const struct rgbvec rgb = {sanitizef(srcr[x]), \
456 sanitizef(srcg[x]), \
457 sanitizef(srcb[x])}; \
458 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
459 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
460 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
461 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
462 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
463 dstr[x] = vec.r; \
464 dstg[x] = vec.g; \
465 dstb[x] = vec.b; \
466 if (!direct && in->linesize[3]) \
467 dsta[x] = srca[x]; \
468 } \
469 grow += out->linesize[0]; \
470 brow += out->linesize[1]; \
471 rrow += out->linesize[2]; \
472 arow += out->linesize[3]; \
473 srcgrow += in->linesize[0]; \
474 srcbrow += in->linesize[1]; \
475 srcrrow += in->linesize[2]; \
476 srcarow += in->linesize[3]; \
477 } \
478 return 0; \
479 }
480
481 DEFINE_INTERP_FUNC_PLANAR_FLOAT(nearest, 32)
482 DEFINE_INTERP_FUNC_PLANAR_FLOAT(trilinear, 32)
483 DEFINE_INTERP_FUNC_PLANAR_FLOAT(tetrahedral, 32)
484 DEFINE_INTERP_FUNC_PLANAR_FLOAT(pyramid, 32)
485 DEFINE_INTERP_FUNC_PLANAR_FLOAT(prism, 32)
486
487 #define DEFINE_INTERP_FUNC(name, nbits) \
488 static int interp_##nbits##_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
489 { \
490 int x, y; \
491 const LUT3DContext *lut3d = ctx->priv; \
492 const Lut3DPreLut *prelut = &lut3d->prelut; \
493 const ThreadData *td = arg; \
494 const AVFrame *in = td->in; \
495 const AVFrame *out = td->out; \
496 const int direct = out == in; \
497 const int step = lut3d->step; \
498 const uint8_t r = lut3d->rgba_map[R]; \
499 const uint8_t g = lut3d->rgba_map[G]; \
500 const uint8_t b = lut3d->rgba_map[B]; \
501 const uint8_t a = lut3d->rgba_map[A]; \
502 const int slice_start = (in->height * jobnr ) / nb_jobs; \
503 const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
504 uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
505 const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
506 const float lut_max = lut3d->lutsize - 1; \
507 const float scale_f = 1.0f / ((1<<nbits) - 1); \
508 const float scale_r = lut3d->scale.r * lut_max; \
509 const float scale_g = lut3d->scale.g * lut_max; \
510 const float scale_b = lut3d->scale.b * lut_max; \
511 \
512 for (y = slice_start; y < slice_end; y++) { \
513 uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
514 const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
515 for (x = 0; x < in->width * step; x += step) { \
516 const struct rgbvec rgb = {src[x + r] * scale_f, \
517 src[x + g] * scale_f, \
518 src[x + b] * scale_f}; \
519 const struct rgbvec prelut_rgb = apply_prelut(prelut, &rgb); \
520 const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \
521 av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \
522 av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \
523 struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \
524 dst[x + r] = av_clip_uint##nbits(vec.r * (float)((1<<nbits) - 1)); \
525 dst[x + g] = av_clip_uint##nbits(vec.g * (float)((1<<nbits) - 1)); \
526 dst[x + b] = av_clip_uint##nbits(vec.b * (float)((1<<nbits) - 1)); \
527 if (!direct && step == 4) \
528 dst[x + a] = src[x + a]; \
529 } \
530 dstrow += out->linesize[0]; \
531 srcrow += in ->linesize[0]; \
532 } \
533 return 0; \
534 }
535
536 DEFINE_INTERP_FUNC(nearest, 8)
537 DEFINE_INTERP_FUNC(trilinear, 8)
538 DEFINE_INTERP_FUNC(tetrahedral, 8)
539 DEFINE_INTERP_FUNC(pyramid, 8)
540 DEFINE_INTERP_FUNC(prism, 8)
541
542 DEFINE_INTERP_FUNC(nearest, 16)
543 DEFINE_INTERP_FUNC(trilinear, 16)
544 DEFINE_INTERP_FUNC(tetrahedral, 16)
545 DEFINE_INTERP_FUNC(pyramid, 16)
546 DEFINE_INTERP_FUNC(prism, 16)
547
548 #define MAX_LINE_SIZE 512
549
skip_line(const char * p)550 static int skip_line(const char *p)
551 {
552 while (*p && av_isspace(*p))
553 p++;
554 return !*p || *p == '#';
555 }
556
fget_next_word(char * dst,int max,FILE * f)557 static char* fget_next_word(char* dst, int max, FILE* f)
558 {
559 int c;
560 char *p = dst;
561
562 /* for null */
563 max--;
564 /* skip until next non whitespace char */
565 while ((c = fgetc(f)) != EOF) {
566 if (av_isspace(c))
567 continue;
568
569 *p++ = c;
570 max--;
571 break;
572 }
573
574 /* get max bytes or up until next whitespace char */
575 for (; max > 0; max--) {
576 if ((c = fgetc(f)) == EOF)
577 break;
578
579 if (av_isspace(c))
580 break;
581
582 *p++ = c;
583 }
584
585 *p = 0;
586 if (p == dst)
587 return NULL;
588 return p;
589 }
590
591 #define NEXT_LINE(loop_cond) do { \
592 if (!fgets(line, sizeof(line), f)) { \
593 av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \
594 return AVERROR_INVALIDDATA; \
595 } \
596 } while (loop_cond)
597
598 #define NEXT_LINE_OR_GOTO(loop_cond, label) do { \
599 if (!fgets(line, sizeof(line), f)) { \
600 av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \
601 ret = AVERROR_INVALIDDATA; \
602 goto label; \
603 } \
604 } while (loop_cond)
605
allocate_3dlut(AVFilterContext * ctx,int lutsize,int prelut)606 static int allocate_3dlut(AVFilterContext *ctx, int lutsize, int prelut)
607 {
608 LUT3DContext *lut3d = ctx->priv;
609 int i;
610 if (lutsize < 2 || lutsize > MAX_LEVEL) {
611 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n");
612 return AVERROR(EINVAL);
613 }
614
615 av_freep(&lut3d->lut);
616 lut3d->lut = av_malloc_array(lutsize * lutsize * lutsize, sizeof(*lut3d->lut));
617 if (!lut3d->lut)
618 return AVERROR(ENOMEM);
619
620 if (prelut) {
621 lut3d->prelut.size = PRELUT_SIZE;
622 for (i = 0; i < 3; i++) {
623 av_freep(&lut3d->prelut.lut[i]);
624 lut3d->prelut.lut[i] = av_malloc_array(PRELUT_SIZE, sizeof(*lut3d->prelut.lut[0]));
625 if (!lut3d->prelut.lut[i])
626 return AVERROR(ENOMEM);
627 }
628 } else {
629 lut3d->prelut.size = 0;
630 for (i = 0; i < 3; i++) {
631 av_freep(&lut3d->prelut.lut[i]);
632 }
633 }
634 lut3d->lutsize = lutsize;
635 lut3d->lutsize2 = lutsize * lutsize;
636 return 0;
637 }
638
639 /* Basically r g and b float values on each line, with a facultative 3DLUTSIZE
640 * directive; seems to be generated by Davinci */
parse_dat(AVFilterContext * ctx,FILE * f)641 static int parse_dat(AVFilterContext *ctx, FILE *f)
642 {
643 LUT3DContext *lut3d = ctx->priv;
644 char line[MAX_LINE_SIZE];
645 int ret, i, j, k, size, size2;
646
647 lut3d->lutsize = size = 33;
648 size2 = size * size;
649
650 NEXT_LINE(skip_line(line));
651 if (!strncmp(line, "3DLUTSIZE ", 10)) {
652 size = strtol(line + 10, NULL, 0);
653
654 NEXT_LINE(skip_line(line));
655 }
656
657 ret = allocate_3dlut(ctx, size, 0);
658 if (ret < 0)
659 return ret;
660
661 for (k = 0; k < size; k++) {
662 for (j = 0; j < size; j++) {
663 for (i = 0; i < size; i++) {
664 struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i];
665 if (k != 0 || j != 0 || i != 0)
666 NEXT_LINE(skip_line(line));
667 if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3)
668 return AVERROR_INVALIDDATA;
669 }
670 }
671 }
672 return 0;
673 }
674
675 /* Iridas format */
parse_cube(AVFilterContext * ctx,FILE * f)676 static int parse_cube(AVFilterContext *ctx, FILE *f)
677 {
678 LUT3DContext *lut3d = ctx->priv;
679 char line[MAX_LINE_SIZE];
680 float min[3] = {0.0, 0.0, 0.0};
681 float max[3] = {1.0, 1.0, 1.0};
682
683 while (fgets(line, sizeof(line), f)) {
684 if (!strncmp(line, "LUT_3D_SIZE", 11)) {
685 int ret, i, j, k;
686 const int size = strtol(line + 12, NULL, 0);
687 const int size2 = size * size;
688
689 ret = allocate_3dlut(ctx, size, 0);
690 if (ret < 0)
691 return ret;
692
693 for (k = 0; k < size; k++) {
694 for (j = 0; j < size; j++) {
695 for (i = 0; i < size; i++) {
696 struct rgbvec *vec = &lut3d->lut[i * size2 + j * size + k];
697
698 do {
699 try_again:
700 NEXT_LINE(0);
701 if (!strncmp(line, "DOMAIN_", 7)) {
702 float *vals = NULL;
703 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
704 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
705 if (!vals)
706 return AVERROR_INVALIDDATA;
707 av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
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 */
parse_3dl(AVFilterContext * ctx,FILE * f)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 */
parse_m3d(AVFilterContext * ctx,FILE * f)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
nearest_sample_index(float * data,float x,int low,int hi)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
parse_cinespace(AVFilterContext * ctx,FILE * f)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
set_identity_matrix(AVFilterContext * ctx,int size)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
config_input(AVFilterLink * inlink)1115 static int config_input(AVFilterLink *inlink)
1116 {
1117 int depth, is16bit, isfloat, planar;
1118 LUT3DContext *lut3d = inlink->dst->priv;
1119 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1120
1121 depth = desc->comp[0].depth;
1122 is16bit = desc->comp[0].depth > 8;
1123 planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR;
1124 isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
1125 ff_fill_rgba_map(lut3d->rgba_map, inlink->format);
1126 lut3d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
1127
1128 #define SET_FUNC(name) do { \
1129 if (planar && !isfloat) { \
1130 switch (depth) { \
1131 case 8: lut3d->interp = interp_8_##name##_p8; break; \
1132 case 9: lut3d->interp = interp_16_##name##_p9; break; \
1133 case 10: lut3d->interp = interp_16_##name##_p10; break; \
1134 case 12: lut3d->interp = interp_16_##name##_p12; break; \
1135 case 14: lut3d->interp = interp_16_##name##_p14; break; \
1136 case 16: lut3d->interp = interp_16_##name##_p16; break; \
1137 } \
1138 } else if (isfloat) { lut3d->interp = interp_##name##_pf32; \
1139 } else if (is16bit) { lut3d->interp = interp_16_##name; \
1140 } else { lut3d->interp = interp_8_##name; } \
1141 } while (0)
1142
1143 switch (lut3d->interpolation) {
1144 case INTERPOLATE_NEAREST: SET_FUNC(nearest); break;
1145 case INTERPOLATE_TRILINEAR: SET_FUNC(trilinear); break;
1146 case INTERPOLATE_TETRAHEDRAL: SET_FUNC(tetrahedral); break;
1147 case INTERPOLATE_PYRAMID: SET_FUNC(pyramid); break;
1148 case INTERPOLATE_PRISM: SET_FUNC(prism); break;
1149 default:
1150 av_assert0(0);
1151 }
1152
1153 #if ARCH_X86
1154 ff_lut3d_init_x86(lut3d, desc);
1155 #endif
1156
1157 return 0;
1158 }
1159
apply_lut(AVFilterLink * inlink,AVFrame * in)1160 static AVFrame *apply_lut(AVFilterLink *inlink, AVFrame *in)
1161 {
1162 AVFilterContext *ctx = inlink->dst;
1163 LUT3DContext *lut3d = ctx->priv;
1164 AVFilterLink *outlink = inlink->dst->outputs[0];
1165 AVFrame *out;
1166 ThreadData td;
1167
1168 if (av_frame_is_writable(in)) {
1169 out = in;
1170 } else {
1171 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
1172 if (!out) {
1173 av_frame_free(&in);
1174 return NULL;
1175 }
1176 av_frame_copy_props(out, in);
1177 }
1178
1179 td.in = in;
1180 td.out = out;
1181 ff_filter_execute(ctx, lut3d->interp, &td, NULL,
1182 FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
1183
1184 if (out != in)
1185 av_frame_free(&in);
1186
1187 return out;
1188 }
1189
filter_frame(AVFilterLink * inlink,AVFrame * in)1190 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
1191 {
1192 AVFilterLink *outlink = inlink->dst->outputs[0];
1193 AVFrame *out = apply_lut(inlink, in);
1194 if (!out)
1195 return AVERROR(ENOMEM);
1196 return ff_filter_frame(outlink, out);
1197 }
1198
process_command(AVFilterContext * ctx,const char * cmd,const char * args,char * res,int res_len,int flags)1199 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
1200 char *res, int res_len, int flags)
1201 {
1202 int ret;
1203
1204 ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
1205 if (ret < 0)
1206 return ret;
1207
1208 return config_input(ctx->inputs[0]);
1209 }
1210
1211 #if CONFIG_LUT3D_FILTER || CONFIG_HALDCLUT_FILTER
1212
1213 /* These options are shared between several filters;
1214 * &lut3d_haldclut_options[COMMON_OPTIONS_OFFSET] must always
1215 * point to the first of the COMMON_OPTIONS. */
1216 #define COMMON_OPTIONS_OFFSET CONFIG_LUT3D_FILTER
1217 static const AVOption lut3d_haldclut_options[] = {
1218 #if CONFIG_LUT3D_FILTER
1219 { "file", "set 3D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
1220 #endif
1221 #if CONFIG_HALDCLUT_FILTER
1222 { "clut", "when to process CLUT", OFFSET(clut), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, .flags = TFLAGS, "clut" },
1223 { "first", "process only first CLUT, ignore rest", 0, AV_OPT_TYPE_CONST, {.i64=0}, .flags = TFLAGS, "clut" },
1224 { "all", "process all CLUTs", 0, AV_OPT_TYPE_CONST, {.i64=1}, .flags = TFLAGS, "clut" },
1225 #endif
1226 COMMON_OPTIONS
1227 };
1228
1229 #if CONFIG_LUT3D_FILTER
1230
1231 AVFILTER_DEFINE_CLASS_EXT(lut3d, "lut3d", lut3d_haldclut_options);
1232
lut3d_init(AVFilterContext * ctx)1233 static av_cold int lut3d_init(AVFilterContext *ctx)
1234 {
1235 int ret;
1236 FILE *f;
1237 const char *ext;
1238 LUT3DContext *lut3d = ctx->priv;
1239
1240 lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1241
1242 if (!lut3d->file) {
1243 return set_identity_matrix(ctx, 32);
1244 }
1245
1246 f = avpriv_fopen_utf8(lut3d->file, "r");
1247 if (!f) {
1248 ret = AVERROR(errno);
1249 av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret));
1250 return ret;
1251 }
1252
1253 ext = strrchr(lut3d->file, '.');
1254 if (!ext) {
1255 av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
1256 ret = AVERROR_INVALIDDATA;
1257 goto end;
1258 }
1259 ext++;
1260
1261 if (!av_strcasecmp(ext, "dat")) {
1262 ret = parse_dat(ctx, f);
1263 } else if (!av_strcasecmp(ext, "3dl")) {
1264 ret = parse_3dl(ctx, f);
1265 } else if (!av_strcasecmp(ext, "cube")) {
1266 ret = parse_cube(ctx, f);
1267 } else if (!av_strcasecmp(ext, "m3d")) {
1268 ret = parse_m3d(ctx, f);
1269 } else if (!av_strcasecmp(ext, "csp")) {
1270 ret = parse_cinespace(ctx, f);
1271 } else {
1272 av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
1273 ret = AVERROR(EINVAL);
1274 }
1275
1276 if (!ret && !lut3d->lutsize) {
1277 av_log(ctx, AV_LOG_ERROR, "3D LUT is empty\n");
1278 ret = AVERROR_INVALIDDATA;
1279 }
1280
1281 end:
1282 fclose(f);
1283 return ret;
1284 }
1285
lut3d_uninit(AVFilterContext * ctx)1286 static av_cold void lut3d_uninit(AVFilterContext *ctx)
1287 {
1288 LUT3DContext *lut3d = ctx->priv;
1289 int i;
1290 av_freep(&lut3d->lut);
1291
1292 for (i = 0; i < 3; i++) {
1293 av_freep(&lut3d->prelut.lut[i]);
1294 }
1295 }
1296
1297 static const AVFilterPad lut3d_inputs[] = {
1298 {
1299 .name = "default",
1300 .type = AVMEDIA_TYPE_VIDEO,
1301 .filter_frame = filter_frame,
1302 .config_props = config_input,
1303 },
1304 };
1305
1306 static const AVFilterPad lut3d_outputs[] = {
1307 {
1308 .name = "default",
1309 .type = AVMEDIA_TYPE_VIDEO,
1310 },
1311 };
1312
1313 const AVFilter ff_vf_lut3d = {
1314 .name = "lut3d",
1315 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
1316 .priv_size = sizeof(LUT3DContext),
1317 .init = lut3d_init,
1318 .uninit = lut3d_uninit,
1319 FILTER_INPUTS(lut3d_inputs),
1320 FILTER_OUTPUTS(lut3d_outputs),
1321 FILTER_PIXFMTS_ARRAY(pix_fmts),
1322 .priv_class = &lut3d_class,
1323 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
1324 .process_command = process_command,
1325 };
1326 #endif
1327
1328 #if CONFIG_HALDCLUT_FILTER
1329
update_clut_packed(LUT3DContext * lut3d,const AVFrame * frame)1330 static void update_clut_packed(LUT3DContext *lut3d, const AVFrame *frame)
1331 {
1332 const uint8_t *data = frame->data[0];
1333 const int linesize = frame->linesize[0];
1334 const int w = lut3d->clut_width;
1335 const int step = lut3d->clut_step;
1336 const uint8_t *rgba_map = lut3d->clut_rgba_map;
1337 const int level = lut3d->lutsize;
1338 const int level2 = lut3d->lutsize2;
1339
1340 #define LOAD_CLUT(nbits) do { \
1341 int i, j, k, x = 0, y = 0; \
1342 \
1343 for (k = 0; k < level; k++) { \
1344 for (j = 0; j < level; j++) { \
1345 for (i = 0; i < level; i++) { \
1346 const uint##nbits##_t *src = (const uint##nbits##_t *) \
1347 (data + y*linesize + x*step); \
1348 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1349 vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \
1350 vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \
1351 vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \
1352 if (++x == w) { \
1353 x = 0; \
1354 y++; \
1355 } \
1356 } \
1357 } \
1358 } \
1359 } while (0)
1360
1361 switch (lut3d->clut_bits) {
1362 case 8: LOAD_CLUT(8); break;
1363 case 16: LOAD_CLUT(16); break;
1364 }
1365 }
1366
update_clut_planar(LUT3DContext * lut3d,const AVFrame * frame)1367 static void update_clut_planar(LUT3DContext *lut3d, const AVFrame *frame)
1368 {
1369 const uint8_t *datag = frame->data[0];
1370 const uint8_t *datab = frame->data[1];
1371 const uint8_t *datar = frame->data[2];
1372 const int glinesize = frame->linesize[0];
1373 const int blinesize = frame->linesize[1];
1374 const int rlinesize = frame->linesize[2];
1375 const int w = lut3d->clut_width;
1376 const int level = lut3d->lutsize;
1377 const int level2 = lut3d->lutsize2;
1378
1379 #define LOAD_CLUT_PLANAR(nbits, depth) do { \
1380 int i, j, k, x = 0, y = 0; \
1381 \
1382 for (k = 0; k < level; k++) { \
1383 for (j = 0; j < level; j++) { \
1384 for (i = 0; i < level; i++) { \
1385 const uint##nbits##_t *gsrc = (const uint##nbits##_t *) \
1386 (datag + y*glinesize); \
1387 const uint##nbits##_t *bsrc = (const uint##nbits##_t *) \
1388 (datab + y*blinesize); \
1389 const uint##nbits##_t *rsrc = (const uint##nbits##_t *) \
1390 (datar + y*rlinesize); \
1391 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1392 vec->r = gsrc[x] / (float)((1<<(depth)) - 1); \
1393 vec->g = bsrc[x] / (float)((1<<(depth)) - 1); \
1394 vec->b = rsrc[x] / (float)((1<<(depth)) - 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_PLANAR(8, 8); break;
1406 case 9: LOAD_CLUT_PLANAR(16, 9); break;
1407 case 10: LOAD_CLUT_PLANAR(16, 10); break;
1408 case 12: LOAD_CLUT_PLANAR(16, 12); break;
1409 case 14: LOAD_CLUT_PLANAR(16, 14); break;
1410 case 16: LOAD_CLUT_PLANAR(16, 16); break;
1411 }
1412 }
1413
update_clut_float(LUT3DContext * lut3d,const AVFrame * frame)1414 static void update_clut_float(LUT3DContext *lut3d, const AVFrame *frame)
1415 {
1416 const uint8_t *datag = frame->data[0];
1417 const uint8_t *datab = frame->data[1];
1418 const uint8_t *datar = frame->data[2];
1419 const int glinesize = frame->linesize[0];
1420 const int blinesize = frame->linesize[1];
1421 const int rlinesize = frame->linesize[2];
1422 const int w = lut3d->clut_width;
1423 const int level = lut3d->lutsize;
1424 const int level2 = lut3d->lutsize2;
1425
1426 int i, j, k, x = 0, y = 0;
1427
1428 for (k = 0; k < level; k++) {
1429 for (j = 0; j < level; j++) {
1430 for (i = 0; i < level; i++) {
1431 const float *gsrc = (const float *)(datag + y*glinesize);
1432 const float *bsrc = (const float *)(datab + y*blinesize);
1433 const float *rsrc = (const float *)(datar + y*rlinesize);
1434 struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k];
1435 vec->r = rsrc[x];
1436 vec->g = gsrc[x];
1437 vec->b = bsrc[x];
1438 if (++x == w) {
1439 x = 0;
1440 y++;
1441 }
1442 }
1443 }
1444 }
1445 }
1446
config_output(AVFilterLink * outlink)1447 static int config_output(AVFilterLink *outlink)
1448 {
1449 AVFilterContext *ctx = outlink->src;
1450 LUT3DContext *lut3d = ctx->priv;
1451 int ret;
1452
1453 ret = ff_framesync_init_dualinput(&lut3d->fs, ctx);
1454 if (ret < 0)
1455 return ret;
1456 outlink->w = ctx->inputs[0]->w;
1457 outlink->h = ctx->inputs[0]->h;
1458 outlink->time_base = ctx->inputs[0]->time_base;
1459 if ((ret = ff_framesync_configure(&lut3d->fs)) < 0)
1460 return ret;
1461 return 0;
1462 }
1463
activate(AVFilterContext * ctx)1464 static int activate(AVFilterContext *ctx)
1465 {
1466 LUT3DContext *s = ctx->priv;
1467 return ff_framesync_activate(&s->fs);
1468 }
1469
config_clut(AVFilterLink * inlink)1470 static int config_clut(AVFilterLink *inlink)
1471 {
1472 int size, level, w, h;
1473 AVFilterContext *ctx = inlink->dst;
1474 LUT3DContext *lut3d = ctx->priv;
1475 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1476
1477 av_assert0(desc);
1478
1479 lut3d->clut_bits = desc->comp[0].depth;
1480 lut3d->clut_planar = av_pix_fmt_count_planes(inlink->format) > 1;
1481 lut3d->clut_float = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
1482
1483 lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3;
1484 ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format);
1485
1486 if (inlink->w > inlink->h)
1487 av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the "
1488 "Hald CLUT will be ignored\n", inlink->w - inlink->h);
1489 else if (inlink->w < inlink->h)
1490 av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the "
1491 "Hald CLUT will be ignored\n", inlink->h - inlink->w);
1492 lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h);
1493
1494 for (level = 1; level*level*level < w; level++);
1495 size = level*level*level;
1496 if (size != w) {
1497 av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n");
1498 return AVERROR_INVALIDDATA;
1499 }
1500 av_assert0(w == h && w == size);
1501 level *= level;
1502 if (level > MAX_LEVEL) {
1503 const int max_clut_level = sqrt(MAX_LEVEL);
1504 const int max_clut_size = max_clut_level*max_clut_level*max_clut_level;
1505 av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT "
1506 "(maximum level is %d, or %dx%d CLUT)\n",
1507 max_clut_level, max_clut_size, max_clut_size);
1508 return AVERROR(EINVAL);
1509 }
1510
1511 return allocate_3dlut(ctx, level, 0);
1512 }
1513
update_apply_clut(FFFrameSync * fs)1514 static int update_apply_clut(FFFrameSync *fs)
1515 {
1516 AVFilterContext *ctx = fs->parent;
1517 LUT3DContext *lut3d = ctx->priv;
1518 AVFilterLink *inlink = ctx->inputs[0];
1519 AVFrame *master, *second, *out;
1520 int ret;
1521
1522 ret = ff_framesync_dualinput_get(fs, &master, &second);
1523 if (ret < 0)
1524 return ret;
1525 if (!second)
1526 return ff_filter_frame(ctx->outputs[0], master);
1527 if (lut3d->clut || !lut3d->got_clut) {
1528 if (lut3d->clut_float)
1529 update_clut_float(ctx->priv, second);
1530 else if (lut3d->clut_planar)
1531 update_clut_planar(ctx->priv, second);
1532 else
1533 update_clut_packed(ctx->priv, second);
1534 lut3d->got_clut = 1;
1535 }
1536 out = apply_lut(inlink, master);
1537 return ff_filter_frame(ctx->outputs[0], out);
1538 }
1539
haldclut_init(AVFilterContext * ctx)1540 static av_cold int haldclut_init(AVFilterContext *ctx)
1541 {
1542 LUT3DContext *lut3d = ctx->priv;
1543 lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1544 lut3d->fs.on_event = update_apply_clut;
1545 return 0;
1546 }
1547
haldclut_uninit(AVFilterContext * ctx)1548 static av_cold void haldclut_uninit(AVFilterContext *ctx)
1549 {
1550 LUT3DContext *lut3d = ctx->priv;
1551 ff_framesync_uninit(&lut3d->fs);
1552 av_freep(&lut3d->lut);
1553 }
1554
1555 FRAMESYNC_DEFINE_CLASS_EXT(haldclut, LUT3DContext, fs,
1556 &lut3d_haldclut_options[COMMON_OPTIONS_OFFSET]);
1557
1558 static const AVFilterPad haldclut_inputs[] = {
1559 {
1560 .name = "main",
1561 .type = AVMEDIA_TYPE_VIDEO,
1562 .config_props = config_input,
1563 },{
1564 .name = "clut",
1565 .type = AVMEDIA_TYPE_VIDEO,
1566 .config_props = config_clut,
1567 },
1568 };
1569
1570 static const AVFilterPad haldclut_outputs[] = {
1571 {
1572 .name = "default",
1573 .type = AVMEDIA_TYPE_VIDEO,
1574 .config_props = config_output,
1575 },
1576 };
1577
1578 const AVFilter ff_vf_haldclut = {
1579 .name = "haldclut",
1580 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
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 .priv_class = &haldclut_class,
1590 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
1591 .process_command = process_command,
1592 };
1593 #endif
1594
1595 #endif /* CONFIG_LUT3D_FILTER || CONFIG_HALDCLUT_FILTER */
1596
1597 #if CONFIG_LUT1D_FILTER
1598
1599 enum interp_1d_mode {
1600 INTERPOLATE_1D_NEAREST,
1601 INTERPOLATE_1D_LINEAR,
1602 INTERPOLATE_1D_CUBIC,
1603 INTERPOLATE_1D_COSINE,
1604 INTERPOLATE_1D_SPLINE,
1605 NB_INTERP_1D_MODE
1606 };
1607
1608 #define MAX_1D_LEVEL 65536
1609
1610 typedef struct LUT1DContext {
1611 const AVClass *class;
1612 char *file;
1613 int interpolation; ///<interp_1d_mode
1614 struct rgbvec scale;
1615 uint8_t rgba_map[4];
1616 int step;
1617 float lut[3][MAX_1D_LEVEL];
1618 int lutsize;
1619 avfilter_action_func *interp;
1620 } LUT1DContext;
1621
1622 #undef OFFSET
1623 #define OFFSET(x) offsetof(LUT1DContext, x)
1624
set_identity_matrix_1d(LUT1DContext * lut1d,int size)1625 static void set_identity_matrix_1d(LUT1DContext *lut1d, int size)
1626 {
1627 const float c = 1. / (size - 1);
1628 int i;
1629
1630 lut1d->lutsize = size;
1631 for (i = 0; i < size; i++) {
1632 lut1d->lut[0][i] = i * c;
1633 lut1d->lut[1][i] = i * c;
1634 lut1d->lut[2][i] = i * c;
1635 }
1636 }
1637
parse_cinespace_1d(AVFilterContext * ctx,FILE * f)1638 static int parse_cinespace_1d(AVFilterContext *ctx, FILE *f)
1639 {
1640 LUT1DContext *lut1d = ctx->priv;
1641 char line[MAX_LINE_SIZE];
1642 float in_min[3] = {0.0, 0.0, 0.0};
1643 float in_max[3] = {1.0, 1.0, 1.0};
1644 float out_min[3] = {0.0, 0.0, 0.0};
1645 float out_max[3] = {1.0, 1.0, 1.0};
1646 int inside_metadata = 0, size;
1647
1648 NEXT_LINE(skip_line(line));
1649 if (strncmp(line, "CSPLUTV100", 10)) {
1650 av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n");
1651 return AVERROR(EINVAL);
1652 }
1653
1654 NEXT_LINE(skip_line(line));
1655 if (strncmp(line, "1D", 2)) {
1656 av_log(ctx, AV_LOG_ERROR, "Not 1D LUT format\n");
1657 return AVERROR(EINVAL);
1658 }
1659
1660 while (1) {
1661 NEXT_LINE(skip_line(line));
1662
1663 if (!strncmp(line, "BEGIN METADATA", 14)) {
1664 inside_metadata = 1;
1665 continue;
1666 }
1667 if (!strncmp(line, "END METADATA", 12)) {
1668 inside_metadata = 0;
1669 continue;
1670 }
1671 if (inside_metadata == 0) {
1672 for (int i = 0; i < 3; i++) {
1673 int npoints = strtol(line, NULL, 0);
1674
1675 if (npoints != 2) {
1676 av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n");
1677 return AVERROR_PATCHWELCOME;
1678 }
1679
1680 NEXT_LINE(skip_line(line));
1681 if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2)
1682 return AVERROR_INVALIDDATA;
1683 NEXT_LINE(skip_line(line));
1684 if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2)
1685 return AVERROR_INVALIDDATA;
1686 NEXT_LINE(skip_line(line));
1687 }
1688
1689 size = strtol(line, NULL, 0);
1690
1691 if (size < 2 || size > MAX_1D_LEVEL) {
1692 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1693 return AVERROR(EINVAL);
1694 }
1695
1696 lut1d->lutsize = size;
1697
1698 for (int i = 0; i < size; i++) {
1699 NEXT_LINE(skip_line(line));
1700 if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1701 return AVERROR_INVALIDDATA;
1702 lut1d->lut[0][i] *= out_max[0] - out_min[0];
1703 lut1d->lut[1][i] *= out_max[1] - out_min[1];
1704 lut1d->lut[2][i] *= out_max[2] - out_min[2];
1705 }
1706
1707 break;
1708 }
1709 }
1710
1711 lut1d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f);
1712 lut1d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f);
1713 lut1d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f);
1714
1715 return 0;
1716 }
1717
parse_cube_1d(AVFilterContext * ctx,FILE * f)1718 static int parse_cube_1d(AVFilterContext *ctx, FILE *f)
1719 {
1720 LUT1DContext *lut1d = ctx->priv;
1721 char line[MAX_LINE_SIZE];
1722 float min[3] = {0.0, 0.0, 0.0};
1723 float max[3] = {1.0, 1.0, 1.0};
1724
1725 while (fgets(line, sizeof(line), f)) {
1726 if (!strncmp(line, "LUT_1D_SIZE", 11)) {
1727 const int size = strtol(line + 12, NULL, 0);
1728 int i;
1729
1730 if (size < 2 || size > MAX_1D_LEVEL) {
1731 av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1732 return AVERROR(EINVAL);
1733 }
1734 lut1d->lutsize = size;
1735 for (i = 0; i < size; i++) {
1736 do {
1737 try_again:
1738 NEXT_LINE(0);
1739 if (!strncmp(line, "DOMAIN_", 7)) {
1740 float *vals = NULL;
1741 if (!strncmp(line + 7, "MIN ", 4)) vals = min;
1742 else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
1743 if (!vals)
1744 return AVERROR_INVALIDDATA;
1745 av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2);
1746 av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
1747 min[0], min[1], min[2], max[0], max[1], max[2]);
1748 goto try_again;
1749 } else if (!strncmp(line, "LUT_1D_INPUT_RANGE ", 19)) {
1750 av_sscanf(line + 19, "%f %f", min, max);
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, "interp_mode" },
1775 { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_NEAREST}, 0, 0, TFLAGS, "interp_mode" },
1776 { "linear", "use values from the linear interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_LINEAR}, 0, 0, TFLAGS, "interp_mode" },
1777 { "cosine", "use values from the cosine interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_COSINE}, 0, 0, TFLAGS, "interp_mode" },
1778 { "cubic", "use values from the cubic interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_CUBIC}, 0, 0, TFLAGS, "interp_mode" },
1779 { "spline", "use values from the spline interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_SPLINE}, 0, 0, TFLAGS, "interp_mode" },
1780 { NULL }
1781 };
1782
1783 AVFILTER_DEFINE_CLASS(lut1d);
1784
interp_1d_nearest(const LUT1DContext * lut1d,int idx,const float s)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
interp_1d_linear(const LUT1DContext * lut1d,int idx,const float s)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
interp_1d_cosine(const LUT1DContext * lut1d,int idx,const float s)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
interp_1d_cubic(const LUT1DContext * lut1d,int idx,const float s)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
interp_1d_spline(const LUT1DContext * lut1d,int idx,const float s)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
config_input_1d(AVFilterLink * inlink)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
lut1d_init(AVFilterContext * ctx)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
apply_1d_lut(AVFilterLink * inlink,AVFrame * in)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
filter_frame_1d(AVFilterLink * inlink,AVFrame * in)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
lut1d_process_command(AVFilterContext * ctx,const char * cmd,const char * args,char * res,int res_len,int flags)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 static const AVFilterPad lut1d_outputs[] = {
2238 {
2239 .name = "default",
2240 .type = AVMEDIA_TYPE_VIDEO,
2241 },
2242 };
2243
2244 const AVFilter ff_vf_lut1d = {
2245 .name = "lut1d",
2246 .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 1D LUT."),
2247 .priv_size = sizeof(LUT1DContext),
2248 .init = lut1d_init,
2249 FILTER_INPUTS(lut1d_inputs),
2250 FILTER_OUTPUTS(lut1d_outputs),
2251 FILTER_PIXFMTS_ARRAY(pix_fmts),
2252 .priv_class = &lut1d_class,
2253 .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
2254 .process_command = lut1d_process_command,
2255 };
2256 #endif
2257