FFmpeg coverage

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 "internal.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		✗	td.in = in;
1182		✗	td.out = out;
1183		✗	ff_filter_execute(ctx, lut3d->interp, &td, NULL,
1184		✗	FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
1185
1186		✗	if (out != in)
1187		✗	av_frame_free(&in);
1188
1189		✗	return out;
1190			}
1191
1192		✗	static int filter_frame(AVFilterLink *inlink, AVFrame *in)
1193			{
1194		✗	AVFilterLink *outlink = inlink->dst->outputs[0];
1195		✗	AVFrame *out = apply_lut(inlink, in);
1196		✗	if (!out)
1197		✗	return AVERROR(ENOMEM);
1198		✗	return ff_filter_frame(outlink, out);
1199			}
1200
1201		✗	static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
1202			char *res, int res_len, int flags)
1203			{
1204			int ret;
1205
1206		✗	ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
1207		✗	if (ret < 0)
1208		✗	return ret;
1209
1210		✗	return config_input(ctx->inputs[0]);
1211			}
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, .unit = "clut" },
1223			{ "first", "process only first CLUT, ignore rest", 0, AV_OPT_TYPE_CONST, {.i64=0}, .flags = TFLAGS, .unit = "clut" },
1224			{ "all", "process all CLUTs", 0, AV_OPT_TYPE_CONST, {.i64=1}, .flags = TFLAGS, .unit = "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
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
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			const AVFilter ff_vf_lut3d = {
1307			.name = "lut3d",
1308			.description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
1309			.priv_size = sizeof(LUT3DContext),
1310			.init = lut3d_init,
1311			.uninit = lut3d_uninit,
1312			FILTER_INPUTS(lut3d_inputs),
1313			FILTER_OUTPUTS(ff_video_default_filterpad),
1314			FILTER_PIXFMTS_ARRAY(pix_fmts),
1315			.priv_class = &lut3d_class,
1316			.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
1317			.process_command = process_command,
1318			};
1319			#endif
1320
1321			#if CONFIG_HALDCLUT_FILTER
1322
1323		✗	static void update_clut_packed(LUT3DContext *lut3d, const AVFrame *frame)
1324			{
1325		✗	const uint8_t *data = frame->data[0];
1326		✗	const ptrdiff_t linesize = frame->linesize[0];
1327		✗	const int w = lut3d->clut_width;
1328		✗	const int step = lut3d->clut_step;
1329		✗	const uint8_t *rgba_map = lut3d->clut_rgba_map;
1330		✗	const int level = lut3d->lutsize;
1331		✗	const int level2 = lut3d->lutsize2;
1332
1333			#define LOAD_CLUT(nbits) do { \
1334			int i, j, k, x = 0, y = 0; \
1335			\
1336			for (k = 0; k < level; k++) { \
1337			for (j = 0; j < level; j++) { \
1338			for (i = 0; i < level; i++) { \
1339			const uint##nbits##_t *src = (const uint##nbits##_t *) \
1340			(data + y*linesize + x*step); \
1341			struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1342			vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \
1343			vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \
1344			vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \
1345			if (++x == w) { \
1346			x = 0; \
1347			y++; \
1348			} \
1349			} \
1350			} \
1351			} \
1352			} while (0)
1353
1354		✗	switch (lut3d->clut_bits) {
1355		✗	case 8: LOAD_CLUT(8); break;
1356		✗	case 16: LOAD_CLUT(16); break;
1357			}
1358		✗	}
1359
1360		✗	static void update_clut_planar(LUT3DContext *lut3d, const AVFrame *frame)
1361			{
1362		✗	const uint8_t *datag = frame->data[0];
1363		✗	const uint8_t *datab = frame->data[1];
1364		✗	const uint8_t *datar = frame->data[2];
1365		✗	const ptrdiff_t glinesize = frame->linesize[0];
1366		✗	const ptrdiff_t blinesize = frame->linesize[1];
1367		✗	const ptrdiff_t rlinesize = frame->linesize[2];
1368		✗	const int w = lut3d->clut_width;
1369		✗	const int level = lut3d->lutsize;
1370		✗	const int level2 = lut3d->lutsize2;
1371
1372			#define LOAD_CLUT_PLANAR(nbits, depth) do { \
1373			int i, j, k, x = 0, y = 0; \
1374			\
1375			for (k = 0; k < level; k++) { \
1376			for (j = 0; j < level; j++) { \
1377			for (i = 0; i < level; i++) { \
1378			const uint##nbits##_t *gsrc = (const uint##nbits##_t *) \
1379			(datag + y*glinesize); \
1380			const uint##nbits##_t *bsrc = (const uint##nbits##_t *) \
1381			(datab + y*blinesize); \
1382			const uint##nbits##_t *rsrc = (const uint##nbits##_t *) \
1383			(datar + y*rlinesize); \
1384			struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \
1385			vec->r = gsrc[x] / (float)((1<<(depth)) - 1); \
1386			vec->g = bsrc[x] / (float)((1<<(depth)) - 1); \
1387			vec->b = rsrc[x] / (float)((1<<(depth)) - 1); \
1388			if (++x == w) { \
1389			x = 0; \
1390			y++; \
1391			} \
1392			} \
1393			} \
1394			} \
1395			} while (0)
1396
1397		✗	switch (lut3d->clut_bits) {
1398		✗	case 8: LOAD_CLUT_PLANAR(8, 8); break;
1399		✗	case 9: LOAD_CLUT_PLANAR(16, 9); break;
1400		✗	case 10: LOAD_CLUT_PLANAR(16, 10); break;
1401		✗	case 12: LOAD_CLUT_PLANAR(16, 12); break;
1402		✗	case 14: LOAD_CLUT_PLANAR(16, 14); break;
1403		✗	case 16: LOAD_CLUT_PLANAR(16, 16); break;
1404			}
1405		✗	}
1406
1407		✗	static void update_clut_float(LUT3DContext *lut3d, const AVFrame *frame)
1408			{
1409		✗	const uint8_t *datag = frame->data[0];
1410		✗	const uint8_t *datab = frame->data[1];
1411		✗	const uint8_t *datar = frame->data[2];
1412		✗	const ptrdiff_t glinesize = frame->linesize[0];
1413		✗	const ptrdiff_t blinesize = frame->linesize[1];
1414		✗	const ptrdiff_t rlinesize = frame->linesize[2];
1415		✗	const int w = lut3d->clut_width;
1416		✗	const int level = lut3d->lutsize;
1417		✗	const int level2 = lut3d->lutsize2;
1418
1419		✗	int i, j, k, x = 0, y = 0;
1420
1421		✗	for (k = 0; k < level; k++) {
1422		✗	for (j = 0; j < level; j++) {
1423		✗	for (i = 0; i < level; i++) {
1424		✗	const float *gsrc = (const float *)(datag + y*glinesize);
1425		✗	const float *bsrc = (const float *)(datab + y*blinesize);
1426		✗	const float *rsrc = (const float *)(datar + y*rlinesize);
1427		✗	struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k];
1428		✗	vec->r = rsrc[x];
1429		✗	vec->g = gsrc[x];
1430		✗	vec->b = bsrc[x];
1431		✗	if (++x == w) {
1432		✗	x = 0;
1433		✗	y++;
1434			}
1435			}
1436			}
1437			}
1438		✗	}
1439
1440		✗	static int config_output(AVFilterLink *outlink)
1441			{
1442		✗	AVFilterContext *ctx = outlink->src;
1443		✗	LUT3DContext *lut3d = ctx->priv;
1444			int ret;
1445
1446		✗	ret = ff_framesync_init_dualinput(&lut3d->fs, ctx);
1447		✗	if (ret < 0)
1448		✗	return ret;
1449		✗	outlink->w = ctx->inputs[0]->w;
1450		✗	outlink->h = ctx->inputs[0]->h;
1451		✗	outlink->time_base = ctx->inputs[0]->time_base;
1452		✗	if ((ret = ff_framesync_configure(&lut3d->fs)) < 0)
1453		✗	return ret;
1454		✗	return 0;
1455			}
1456
1457		✗	static int activate(AVFilterContext *ctx)
1458			{
1459		✗	LUT3DContext *s = ctx->priv;
1460		✗	return ff_framesync_activate(&s->fs);
1461			}
1462
1463		✗	static int config_clut(AVFilterLink *inlink)
1464			{
1465			int size, level, w, h;
1466		✗	AVFilterContext *ctx = inlink->dst;
1467		✗	LUT3DContext *lut3d = ctx->priv;
1468		✗	const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
1469
1470		✗	av_assert0(desc);
1471
1472		✗	lut3d->clut_bits = desc->comp[0].depth;
1473		✗	lut3d->clut_planar = av_pix_fmt_count_planes(inlink->format) > 1;
1474		✗	lut3d->clut_float = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
1475
1476		✗	lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3;
1477		✗	ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format);
1478
1479		✗	if (inlink->w > inlink->h)
1480		✗	av_log(ctx, AV_LOG_INFO, "Padding on the right (%dpx) of the "
1481		✗	"Hald CLUT will be ignored\n", inlink->w - inlink->h);
1482		✗	else if (inlink->w < inlink->h)
1483		✗	av_log(ctx, AV_LOG_INFO, "Padding at the bottom (%dpx) of the "
1484		✗	"Hald CLUT will be ignored\n", inlink->h - inlink->w);
1485		✗	lut3d->clut_width = w = h = FFMIN(inlink->w, inlink->h);
1486
1487		✗	for (level = 1; level*level*level < w; level++);
1488		✗	size = level*level*level;
1489		✗	if (size != w) {
1490		✗	av_log(ctx, AV_LOG_WARNING, "The Hald CLUT width does not match the level\n");
1491		✗	return AVERROR_INVALIDDATA;
1492			}
1493		✗	av_assert0(w == h && w == size);
1494		✗	level *= level;
1495		✗	if (level > MAX_LEVEL) {
1496		✗	const int max_clut_level = sqrt(MAX_LEVEL);
1497		✗	const int max_clut_size = max_clut_level*max_clut_level*max_clut_level;
1498		✗	av_log(ctx, AV_LOG_ERROR, "Too large Hald CLUT "
1499			"(maximum level is %d, or %dx%d CLUT)\n",
1500			max_clut_level, max_clut_size, max_clut_size);
1501		✗	return AVERROR(EINVAL);
1502			}
1503
1504		✗	return allocate_3dlut(ctx, level, 0);
1505			}
1506
1507		✗	static int update_apply_clut(FFFrameSync *fs)
1508			{
1509		✗	AVFilterContext *ctx = fs->parent;
1510		✗	LUT3DContext *lut3d = ctx->priv;
1511		✗	AVFilterLink *inlink = ctx->inputs[0];
1512			AVFrame *master, *second, *out;
1513			int ret;
1514
1515		✗	ret = ff_framesync_dualinput_get(fs, &master, &second);
1516		✗	if (ret < 0)
1517		✗	return ret;
1518		✗	if (!second)
1519		✗	return ff_filter_frame(ctx->outputs[0], master);
1520		✗	if (lut3d->clut || !lut3d->got_clut) {
1521		✗	if (lut3d->clut_float)
1522		✗	update_clut_float(ctx->priv, second);
1523		✗	else if (lut3d->clut_planar)
1524		✗	update_clut_planar(ctx->priv, second);
1525			else
1526		✗	update_clut_packed(ctx->priv, second);
1527		✗	lut3d->got_clut = 1;
1528			}
1529		✗	out = apply_lut(inlink, master);
1530		✗	return ff_filter_frame(ctx->outputs[0], out);
1531			}
1532
1533		✗	static av_cold int haldclut_init(AVFilterContext *ctx)
1534			{
1535		✗	LUT3DContext *lut3d = ctx->priv;
1536		✗	lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f;
1537		✗	lut3d->fs.on_event = update_apply_clut;
1538		✗	return 0;
1539			}
1540
1541		✗	static av_cold void haldclut_uninit(AVFilterContext *ctx)
1542			{
1543		✗	LUT3DContext *lut3d = ctx->priv;
1544		✗	ff_framesync_uninit(&lut3d->fs);
1545		✗	av_freep(&lut3d->lut);
1546		✗	}
1547
1548		✗	FRAMESYNC_DEFINE_CLASS_EXT(haldclut, LUT3DContext, fs,
1549			&lut3d_haldclut_options[COMMON_OPTIONS_OFFSET]);
1550
1551			static const AVFilterPad haldclut_inputs[] = {
1552			{
1553			.name = "main",
1554			.type = AVMEDIA_TYPE_VIDEO,
1555			.config_props = config_input,
1556			},{
1557			.name = "clut",
1558			.type = AVMEDIA_TYPE_VIDEO,
1559			.config_props = config_clut,
1560			},
1561			};
1562
1563			static const AVFilterPad haldclut_outputs[] = {
1564			{
1565			.name = "default",
1566			.type = AVMEDIA_TYPE_VIDEO,
1567			.config_props = config_output,
1568			},
1569			};
1570
1571			const AVFilter ff_vf_haldclut = {
1572			.name = "haldclut",
1573			.description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
1574			.priv_size = sizeof(LUT3DContext),
1575			.preinit = haldclut_framesync_preinit,
1576			.init = haldclut_init,
1577			.uninit = haldclut_uninit,
1578			.activate = activate,
1579			FILTER_INPUTS(haldclut_inputs),
1580			FILTER_OUTPUTS(haldclut_outputs),
1581			FILTER_PIXFMTS_ARRAY(pix_fmts),
1582			.priv_class = &haldclut_class,
1583			.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
1584			.process_command = process_command,
1585			};
1586			#endif
1587
1588			#endif /* CONFIG_LUT3D_FILTER || CONFIG_HALDCLUT_FILTER */
1589
1590			#if CONFIG_LUT1D_FILTER
1591
1592			enum interp_1d_mode {
1593			INTERPOLATE_1D_NEAREST,
1594			INTERPOLATE_1D_LINEAR,
1595			INTERPOLATE_1D_CUBIC,
1596			INTERPOLATE_1D_COSINE,
1597			INTERPOLATE_1D_SPLINE,
1598			NB_INTERP_1D_MODE
1599			};
1600
1601			#define MAX_1D_LEVEL 65536
1602
1603			typedef struct LUT1DContext {
1604			const AVClass *class;
1605			char *file;
1606			int interpolation; ///<interp_1d_mode
1607			struct rgbvec scale;
1608			uint8_t rgba_map[4];
1609			int step;
1610			float lut[3][MAX_1D_LEVEL];
1611			int lutsize;
1612			avfilter_action_func *interp;
1613			} LUT1DContext;
1614
1615			#undef OFFSET
1616			#define OFFSET(x) offsetof(LUT1DContext, x)
1617
1618		✗	static void set_identity_matrix_1d(LUT1DContext *lut1d, int size)
1619			{
1620		✗	const float c = 1. / (size - 1);
1621			int i;
1622
1623		✗	lut1d->lutsize = size;
1624		✗	for (i = 0; i < size; i++) {
1625		✗	lut1d->lut[0][i] = i * c;
1626		✗	lut1d->lut[1][i] = i * c;
1627		✗	lut1d->lut[2][i] = i * c;
1628			}
1629		✗	}
1630
1631		✗	static int parse_cinespace_1d(AVFilterContext *ctx, FILE *f)
1632			{
1633		✗	LUT1DContext *lut1d = ctx->priv;
1634			char line[MAX_LINE_SIZE];
1635		✗	float in_min[3] = {0.0, 0.0, 0.0};
1636		✗	float in_max[3] = {1.0, 1.0, 1.0};
1637		✗	float out_min[3] = {0.0, 0.0, 0.0};
1638		✗	float out_max[3] = {1.0, 1.0, 1.0};
1639		✗	int inside_metadata = 0, size;
1640
1641		✗	NEXT_LINE(skip_line(line));
1642		✗	if (strncmp(line, "CSPLUTV100", 10)) {
1643		✗	av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n");
1644		✗	return AVERROR(EINVAL);
1645			}
1646
1647		✗	NEXT_LINE(skip_line(line));
1648		✗	if (strncmp(line, "1D", 2)) {
1649		✗	av_log(ctx, AV_LOG_ERROR, "Not 1D LUT format\n");
1650		✗	return AVERROR(EINVAL);
1651			}
1652
1653		✗	while (1) {
1654		✗	NEXT_LINE(skip_line(line));
1655
1656		✗	if (!strncmp(line, "BEGIN METADATA", 14)) {
1657		✗	inside_metadata = 1;
1658		✗	continue;
1659			}
1660		✗	if (!strncmp(line, "END METADATA", 12)) {
1661		✗	inside_metadata = 0;
1662		✗	continue;
1663			}
1664		✗	if (inside_metadata == 0) {
1665		✗	for (int i = 0; i < 3; i++) {
1666		✗	int npoints = strtol(line, NULL, 0);
1667
1668		✗	if (npoints != 2) {
1669		✗	av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n");
1670		✗	return AVERROR_PATCHWELCOME;
1671			}
1672
1673		✗	NEXT_LINE(skip_line(line));
1674		✗	if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2)
1675		✗	return AVERROR_INVALIDDATA;
1676		✗	NEXT_LINE(skip_line(line));
1677		✗	if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2)
1678		✗	return AVERROR_INVALIDDATA;
1679		✗	NEXT_LINE(skip_line(line));
1680			}
1681
1682		✗	size = strtol(line, NULL, 0);
1683
1684		✗	if (size < 2 || size > MAX_1D_LEVEL) {
1685		✗	av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1686		✗	return AVERROR(EINVAL);
1687			}
1688
1689		✗	lut1d->lutsize = size;
1690
1691		✗	for (int i = 0; i < size; i++) {
1692		✗	NEXT_LINE(skip_line(line));
1693		✗	if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1694		✗	return AVERROR_INVALIDDATA;
1695		✗	lut1d->lut[0][i] *= out_max[0] - out_min[0];
1696		✗	lut1d->lut[1][i] *= out_max[1] - out_min[1];
1697		✗	lut1d->lut[2][i] *= out_max[2] - out_min[2];
1698			}
1699
1700		✗	break;
1701			}
1702			}
1703
1704		✗	lut1d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f);
1705		✗	lut1d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f);
1706		✗	lut1d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f);
1707
1708		✗	return 0;
1709			}
1710
1711		✗	static int parse_cube_1d(AVFilterContext *ctx, FILE *f)
1712			{
1713		✗	LUT1DContext *lut1d = ctx->priv;
1714			char line[MAX_LINE_SIZE];
1715		✗	float min[3] = {0.0, 0.0, 0.0};
1716		✗	float max[3] = {1.0, 1.0, 1.0};
1717
1718		✗	while (fgets(line, sizeof(line), f)) {
1719		✗	if (!strncmp(line, "LUT_1D_SIZE", 11)) {
1720		✗	const int size = strtol(line + 12, NULL, 0);
1721			int i;
1722
1723		✗	if (size < 2 || size > MAX_1D_LEVEL) {
1724		✗	av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n");
1725		✗	return AVERROR(EINVAL);
1726			}
1727		✗	lut1d->lutsize = size;
1728		✗	for (i = 0; i < size; i++) {
1729			do {
1730		✗	try_again:
1731		✗	NEXT_LINE(0);
1732		✗	if (!strncmp(line, "DOMAIN_", 7)) {
1733		✗	float *vals = NULL;
1734		✗	if (!strncmp(line + 7, "MIN ", 4)) vals = min;
1735		✗	else if (!strncmp(line + 7, "MAX ", 4)) vals = max;
1736		✗	if (!vals)
1737		✗	return AVERROR_INVALIDDATA;
1738		✗	if (av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2) != 3)
1739		✗	return AVERROR_INVALIDDATA;
1740		✗	av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n",
1741		✗	min[0], min[1], min[2], max[0], max[1], max[2]);
1742		✗	goto try_again;
1743		✗	} else if (!strncmp(line, "LUT_1D_INPUT_RANGE ", 19)) {
1744		✗	if (av_sscanf(line + 19, "%f %f", min, max) != 2)
1745		✗	return AVERROR_INVALIDDATA;
1746		✗	min[1] = min[2] = min[0];
1747		✗	max[1] = max[2] = max[0];
1748		✗	goto try_again;
1749		✗	} else if (!strncmp(line, "TITLE", 5)) {
1750		✗	goto try_again;
1751			}
1752		✗	} while (skip_line(line));
1753		✗	if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3)
1754		✗	return AVERROR_INVALIDDATA;
1755			}
1756		✗	break;
1757			}
1758			}
1759
1760		✗	lut1d->scale.r = av_clipf(1. / (max[0] - min[0]), 0.f, 1.f);
1761		✗	lut1d->scale.g = av_clipf(1. / (max[1] - min[1]), 0.f, 1.f);
1762		✗	lut1d->scale.b = av_clipf(1. / (max[2] - min[2]), 0.f, 1.f);
1763
1764		✗	return 0;
1765			}
1766
1767			static const AVOption lut1d_options[] = {
1768			{ "file", "set 1D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = TFLAGS },
1769			{ "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_1D_LINEAR}, 0, NB_INTERP_1D_MODE-1, TFLAGS, .unit = "interp_mode" },
1770			{ "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_NEAREST}, 0, 0, TFLAGS, .unit = "interp_mode" },
1771			{ "linear", "use values from the linear interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_LINEAR}, 0, 0, TFLAGS, .unit = "interp_mode" },
1772			{ "cosine", "use values from the cosine interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_COSINE}, 0, 0, TFLAGS, .unit = "interp_mode" },
1773			{ "cubic", "use values from the cubic interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_CUBIC}, 0, 0, TFLAGS, .unit = "interp_mode" },
1774			{ "spline", "use values from the spline interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_SPLINE}, 0, 0, TFLAGS, .unit = "interp_mode" },
1775			{ NULL }
1776			};
1777
1778			AVFILTER_DEFINE_CLASS(lut1d);
1779
1780		✗	static inline float interp_1d_nearest(const LUT1DContext *lut1d,
1781			int idx, const float s)
1782			{
1783		✗	return lut1d->lut[idx][NEAR(s)];
1784			}
1785
1786			#define NEXT1D(x) (FFMIN((int)(x) + 1, lut1d->lutsize - 1))
1787
1788		✗	static inline float interp_1d_linear(const LUT1DContext *lut1d,
1789			int idx, const float s)
1790			{
1791		✗	const int prev = PREV(s);
1792		✗	const int next = NEXT1D(s);
1793		✗	const float d = s - prev;
1794		✗	const float p = lut1d->lut[idx][prev];
1795		✗	const float n = lut1d->lut[idx][next];
1796
1797		✗	return lerpf(p, n, d);
1798			}
1799
1800		✗	static inline float interp_1d_cosine(const LUT1DContext *lut1d,
1801			int idx, const float s)
1802			{
1803		✗	const int prev = PREV(s);
1804		✗	const int next = NEXT1D(s);
1805		✗	const float d = s - prev;
1806		✗	const float p = lut1d->lut[idx][prev];
1807		✗	const float n = lut1d->lut[idx][next];
1808		✗	const float m = (1.f - cosf(d * M_PI)) * .5f;
1809
1810		✗	return lerpf(p, n, m);
1811			}
1812
1813		✗	static inline float interp_1d_cubic(const LUT1DContext *lut1d,
1814			int idx, const float s)
1815			{
1816		✗	const int prev = PREV(s);
1817		✗	const int next = NEXT1D(s);
1818		✗	const float mu = s - prev;
1819			float a0, a1, a2, a3, mu2;
1820
1821		✗	float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1822		✗	float y1 = lut1d->lut[idx][prev];
1823		✗	float y2 = lut1d->lut[idx][next];
1824		✗	float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1825
1826
1827		✗	mu2 = mu * mu;
1828		✗	a0 = y3 - y2 - y0 + y1;
1829		✗	a1 = y0 - y1 - a0;
1830		✗	a2 = y2 - y0;
1831		✗	a3 = y1;
1832
1833		✗	return a0 * mu * mu2 + a1 * mu2 + a2 * mu + a3;
1834			}
1835
1836		✗	static inline float interp_1d_spline(const LUT1DContext *lut1d,
1837			int idx, const float s)
1838			{
1839		✗	const int prev = PREV(s);
1840		✗	const int next = NEXT1D(s);
1841		✗	const float x = s - prev;
1842			float c0, c1, c2, c3;
1843
1844		✗	float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)];
1845		✗	float y1 = lut1d->lut[idx][prev];
1846		✗	float y2 = lut1d->lut[idx][next];
1847		✗	float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)];
1848
1849		✗	c0 = y1;
1850		✗	c1 = .5f * (y2 - y0);
1851		✗	c2 = y0 - 2.5f * y1 + 2.f * y2 - .5f * y3;
1852		✗	c3 = .5f * (y3 - y0) + 1.5f * (y1 - y2);
1853
1854		✗	return ((c3 * x + c2) * x + c1) * x + c0;
1855			}
1856
1857			#define DEFINE_INTERP_FUNC_PLANAR_1D(name, nbits, depth) \
1858			static int interp_1d_##nbits##_##name##_p##depth(AVFilterContext *ctx, \
1859			void *arg, int jobnr, \
1860			int nb_jobs) \
1861			{ \
1862			int x, y; \
1863			const LUT1DContext *lut1d = ctx->priv; \
1864			const ThreadData *td = arg; \
1865			const AVFrame *in = td->in; \
1866			const AVFrame *out = td->out; \
1867			const int direct = out == in; \
1868			const int slice_start = (in->height * jobnr ) / nb_jobs; \
1869			const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
1870			uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
1871			uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
1872			uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
1873			uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
1874			const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
1875			const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
1876			const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
1877			const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
1878			const float factor = (1 << depth) - 1; \
1879			const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \
1880			const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \
1881			const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \
1882			\
1883			for (y = slice_start; y < slice_end; y++) { \
1884			uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \
1885			uint##nbits##_t *dstb = (uint##nbits##_t *)brow; \
1886			uint##nbits##_t *dstr = (uint##nbits##_t *)rrow; \
1887			uint##nbits##_t *dsta = (uint##nbits##_t *)arow; \
1888			const uint##nbits##_t *srcg = (const uint##nbits##_t *)srcgrow; \
1889			const uint##nbits##_t *srcb = (const uint##nbits##_t *)srcbrow; \
1890			const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \
1891			const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \
1892			for (x = 0; x < in->width; x++) { \
1893			float r = srcr[x] * scale_r; \
1894			float g = srcg[x] * scale_g; \
1895			float b = srcb[x] * scale_b; \
1896			r = interp_1d_##name(lut1d, 0, r); \
1897			g = interp_1d_##name(lut1d, 1, g); \
1898			b = interp_1d_##name(lut1d, 2, b); \
1899			dstr[x] = av_clip_uintp2(r * factor, depth); \
1900			dstg[x] = av_clip_uintp2(g * factor, depth); \
1901			dstb[x] = av_clip_uintp2(b * factor, depth); \
1902			if (!direct && in->linesize[3]) \
1903			dsta[x] = srca[x]; \
1904			} \
1905			grow += out->linesize[0]; \
1906			brow += out->linesize[1]; \
1907			rrow += out->linesize[2]; \
1908			arow += out->linesize[3]; \
1909			srcgrow += in->linesize[0]; \
1910			srcbrow += in->linesize[1]; \
1911			srcrrow += in->linesize[2]; \
1912			srcarow += in->linesize[3]; \
1913			} \
1914			return 0; \
1915			}
1916
1917		✗	DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 8, 8)
1918		✗	DEFINE_INTERP_FUNC_PLANAR_1D(linear, 8, 8)
1919		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 8, 8)
1920		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 8, 8)
1921		✗	DEFINE_INTERP_FUNC_PLANAR_1D(spline, 8, 8)
1922
1923		✗	DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 9)
1924		✗	DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 9)
1925		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 9)
1926		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 9)
1927		✗	DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 9)
1928
1929		✗	DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 10)
1930		✗	DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 10)
1931		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 10)
1932		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 10)
1933		✗	DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 10)
1934
1935		✗	DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 12)
1936		✗	DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 12)
1937		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 12)
1938		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 12)
1939		✗	DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 12)
1940
1941		✗	DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 14)
1942		✗	DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 14)
1943		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 14)
1944		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 14)
1945		✗	DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 14)
1946
1947		✗	DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 16)
1948		✗	DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 16)
1949		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 16)
1950		✗	DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 16)
1951		✗	DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 16)
1952
1953			#define DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(name, depth) \
1954			static int interp_1d_##name##_pf##depth(AVFilterContext *ctx, \
1955			void *arg, int jobnr, \
1956			int nb_jobs) \
1957			{ \
1958			int x, y; \
1959			const LUT1DContext *lut1d = ctx->priv; \
1960			const ThreadData *td = arg; \
1961			const AVFrame *in = td->in; \
1962			const AVFrame *out = td->out; \
1963			const int direct = out == in; \
1964			const int slice_start = (in->height * jobnr ) / nb_jobs; \
1965			const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
1966			uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \
1967			uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \
1968			uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \
1969			uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \
1970			const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \
1971			const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \
1972			const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \
1973			const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \
1974			const float lutsize = lut1d->lutsize - 1; \
1975			const float scale_r = lut1d->scale.r * lutsize; \
1976			const float scale_g = lut1d->scale.g * lutsize; \
1977			const float scale_b = lut1d->scale.b * lutsize; \
1978			\
1979			for (y = slice_start; y < slice_end; y++) { \
1980			float *dstg = (float *)grow; \
1981			float *dstb = (float *)brow; \
1982			float *dstr = (float *)rrow; \
1983			float *dsta = (float *)arow; \
1984			const float *srcg = (const float *)srcgrow; \
1985			const float *srcb = (const float *)srcbrow; \
1986			const float *srcr = (const float *)srcrrow; \
1987			const float *srca = (const float *)srcarow; \
1988			for (x = 0; x < in->width; x++) { \
1989			float r = av_clipf(sanitizef(srcr[x]) * scale_r, 0.0f, lutsize); \
1990			float g = av_clipf(sanitizef(srcg[x]) * scale_g, 0.0f, lutsize); \
1991			float b = av_clipf(sanitizef(srcb[x]) * scale_b, 0.0f, lutsize); \
1992			r = interp_1d_##name(lut1d, 0, r); \
1993			g = interp_1d_##name(lut1d, 1, g); \
1994			b = interp_1d_##name(lut1d, 2, b); \
1995			dstr[x] = r; \
1996			dstg[x] = g; \
1997			dstb[x] = b; \
1998			if (!direct && in->linesize[3]) \
1999			dsta[x] = srca[x]; \
2000			} \
2001			grow += out->linesize[0]; \
2002			brow += out->linesize[1]; \
2003			rrow += out->linesize[2]; \
2004			arow += out->linesize[3]; \
2005			srcgrow += in->linesize[0]; \
2006			srcbrow += in->linesize[1]; \
2007			srcrrow += in->linesize[2]; \
2008			srcarow += in->linesize[3]; \
2009			} \
2010			return 0; \
2011			}
2012
2013		✗	DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(nearest, 32)
2014		✗	DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(linear, 32)
2015		✗	DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cosine, 32)
2016		✗	DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cubic, 32)
2017		✗	DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(spline, 32)
2018
2019			#define DEFINE_INTERP_FUNC_1D(name, nbits) \
2020			static int interp_1d_##nbits##_##name(AVFilterContext *ctx, void *arg, \
2021			int jobnr, int nb_jobs) \
2022			{ \
2023			int x, y; \
2024			const LUT1DContext *lut1d = ctx->priv; \
2025			const ThreadData *td = arg; \
2026			const AVFrame *in = td->in; \
2027			const AVFrame *out = td->out; \
2028			const int direct = out == in; \
2029			const int step = lut1d->step; \
2030			const uint8_t r = lut1d->rgba_map[R]; \
2031			const uint8_t g = lut1d->rgba_map[G]; \
2032			const uint8_t b = lut1d->rgba_map[B]; \
2033			const uint8_t a = lut1d->rgba_map[A]; \
2034			const int slice_start = (in->height * jobnr ) / nb_jobs; \
2035			const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \
2036			uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \
2037			const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \
2038			const float factor = (1 << nbits) - 1; \
2039			const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \
2040			const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \
2041			const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \
2042			\
2043			for (y = slice_start; y < slice_end; y++) { \
2044			uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \
2045			const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \
2046			for (x = 0; x < in->width * step; x += step) { \
2047			float rr = src[x + r] * scale_r; \
2048			float gg = src[x + g] * scale_g; \
2049			float bb = src[x + b] * scale_b; \
2050			rr = interp_1d_##name(lut1d, 0, rr); \
2051			gg = interp_1d_##name(lut1d, 1, gg); \
2052			bb = interp_1d_##name(lut1d, 2, bb); \
2053			dst[x + r] = av_clip_uint##nbits(rr * factor); \
2054			dst[x + g] = av_clip_uint##nbits(gg * factor); \
2055			dst[x + b] = av_clip_uint##nbits(bb * factor); \
2056			if (!direct && step == 4) \
2057			dst[x + a] = src[x + a]; \
2058			} \
2059			dstrow += out->linesize[0]; \
2060			srcrow += in ->linesize[0]; \
2061			} \
2062			return 0; \
2063			}
2064
2065		✗	DEFINE_INTERP_FUNC_1D(nearest, 8)
2066		✗	DEFINE_INTERP_FUNC_1D(linear, 8)
2067		✗	DEFINE_INTERP_FUNC_1D(cosine, 8)
2068		✗	DEFINE_INTERP_FUNC_1D(cubic, 8)
2069		✗	DEFINE_INTERP_FUNC_1D(spline, 8)
2070
2071		✗	DEFINE_INTERP_FUNC_1D(nearest, 16)
2072		✗	DEFINE_INTERP_FUNC_1D(linear, 16)
2073		✗	DEFINE_INTERP_FUNC_1D(cosine, 16)
2074		✗	DEFINE_INTERP_FUNC_1D(cubic, 16)
2075		✗	DEFINE_INTERP_FUNC_1D(spline, 16)
2076
2077		✗	static int config_input_1d(AVFilterLink *inlink)
2078			{
2079			int depth, is16bit, isfloat, planar;
2080		✗	LUT1DContext *lut1d = inlink->dst->priv;
2081		✗	const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
2082
2083		✗	depth = desc->comp[0].depth;
2084		✗	is16bit = desc->comp[0].depth > 8;
2085		✗	planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR;
2086		✗	isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT;
2087		✗	ff_fill_rgba_map(lut1d->rgba_map, inlink->format);
2088		✗	lut1d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit);
2089
2090			#define SET_FUNC_1D(name) do { \
2091			if (planar && !isfloat) { \
2092			switch (depth) { \
2093			case 8: lut1d->interp = interp_1d_8_##name##_p8; break; \
2094			case 9: lut1d->interp = interp_1d_16_##name##_p9; break; \
2095			case 10: lut1d->interp = interp_1d_16_##name##_p10; break; \
2096			case 12: lut1d->interp = interp_1d_16_##name##_p12; break; \
2097			case 14: lut1d->interp = interp_1d_16_##name##_p14; break; \
2098			case 16: lut1d->interp = interp_1d_16_##name##_p16; break; \
2099			} \
2100			} else if (isfloat) { lut1d->interp = interp_1d_##name##_pf32; \
2101			} else if (is16bit) { lut1d->interp = interp_1d_16_##name; \
2102			} else { lut1d->interp = interp_1d_8_##name; } \
2103			} while (0)
2104
2105		✗	switch (lut1d->interpolation) {
2106		✗	case INTERPOLATE_1D_NEAREST: SET_FUNC_1D(nearest); break;
2107		✗	case INTERPOLATE_1D_LINEAR: SET_FUNC_1D(linear); break;
2108		✗	case INTERPOLATE_1D_COSINE: SET_FUNC_1D(cosine); break;
2109		✗	case INTERPOLATE_1D_CUBIC: SET_FUNC_1D(cubic); break;
2110		✗	case INTERPOLATE_1D_SPLINE: SET_FUNC_1D(spline); break;
2111		✗	default:
2112		✗	av_assert0(0);
2113			}
2114
2115		✗	return 0;
2116			}
2117
2118		✗	static av_cold int lut1d_init(AVFilterContext *ctx)
2119			{
2120			int ret;
2121			FILE *f;
2122			const char *ext;
2123		✗	LUT1DContext *lut1d = ctx->priv;
2124
2125		✗	lut1d->scale.r = lut1d->scale.g = lut1d->scale.b = 1.f;
2126
2127		✗	if (!lut1d->file) {
2128		✗	set_identity_matrix_1d(lut1d, 32);
2129		✗	return 0;
2130			}
2131
2132		✗	f = avpriv_fopen_utf8(lut1d->file, "r");
2133		✗	if (!f) {
2134		✗	ret = AVERROR(errno);
2135		✗	av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut1d->file, av_err2str(ret));
2136		✗	return ret;
2137			}
2138
2139		✗	ext = strrchr(lut1d->file, '.');
2140		✗	if (!ext) {
2141		✗	av_log(ctx, AV_LOG_ERROR, "Unable to guess the format from the extension\n");
2142		✗	ret = AVERROR_INVALIDDATA;
2143		✗	goto end;
2144			}
2145		✗	ext++;
2146
2147		✗	if (!av_strcasecmp(ext, "cube") || !av_strcasecmp(ext, "1dlut")) {
2148		✗	ret = parse_cube_1d(ctx, f);
2149		✗	} else if (!av_strcasecmp(ext, "csp")) {
2150		✗	ret = parse_cinespace_1d(ctx, f);
2151			} else {
2152		✗	av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext);
2153		✗	ret = AVERROR(EINVAL);
2154			}
2155
2156		✗	if (!ret && !lut1d->lutsize) {
2157		✗	av_log(ctx, AV_LOG_ERROR, "1D LUT is empty\n");
2158		✗	ret = AVERROR_INVALIDDATA;
2159			}
2160
2161		✗	end:
2162		✗	fclose(f);
2163		✗	return ret;
2164			}
2165
2166		✗	static AVFrame *apply_1d_lut(AVFilterLink *inlink, AVFrame *in)
2167			{
2168		✗	AVFilterContext *ctx = inlink->dst;
2169		✗	LUT1DContext *lut1d = ctx->priv;
2170		✗	AVFilterLink *outlink = inlink->dst->outputs[0];
2171			AVFrame *out;
2172			ThreadData td;
2173
2174		✗	if (av_frame_is_writable(in)) {
2175		✗	out = in;
2176			} else {
2177		✗	out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
2178		✗	if (!out) {
2179		✗	av_frame_free(&in);
2180		✗	return NULL;
2181			}
2182		✗	av_frame_copy_props(out, in);
2183			}
2184
2185		✗	td.in = in;
2186		✗	td.out = out;
2187		✗	ff_filter_execute(ctx, lut1d->interp, &td, NULL,
2188		✗	FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
2189
2190		✗	if (out != in)
2191		✗	av_frame_free(&in);
2192
2193		✗	return out;
2194			}
2195
2196		✗	static int filter_frame_1d(AVFilterLink *inlink, AVFrame *in)
2197			{
2198		✗	AVFilterLink *outlink = inlink->dst->outputs[0];
2199		✗	AVFrame *out = apply_1d_lut(inlink, in);
2200		✗	if (!out)
2201		✗	return AVERROR(ENOMEM);
2202		✗	return ff_filter_frame(outlink, out);
2203			}
2204
2205		✗	static int lut1d_process_command(AVFilterContext *ctx, const char *cmd, const char *args,
2206			char *res, int res_len, int flags)
2207			{
2208		✗	LUT1DContext *lut1d = ctx->priv;
2209			int ret;
2210
2211		✗	ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
2212		✗	if (ret < 0)
2213		✗	return ret;
2214
2215		✗	ret = lut1d_init(ctx);
2216		✗	if (ret < 0) {
2217		✗	set_identity_matrix_1d(lut1d, 32);
2218		✗	return ret;
2219			}
2220		✗	return config_input_1d(ctx->inputs[0]);
2221			}
2222
2223			static const AVFilterPad lut1d_inputs[] = {
2224			{
2225			.name = "default",
2226			.type = AVMEDIA_TYPE_VIDEO,
2227			.filter_frame = filter_frame_1d,
2228			.config_props = config_input_1d,
2229			},
2230			};
2231
2232			const AVFilter ff_vf_lut1d = {
2233			.name = "lut1d",
2234			.description = NULL_IF_CONFIG_SMALL("Adjust colors using a 1D LUT."),
2235			.priv_size = sizeof(LUT1DContext),
2236			.init = lut1d_init,
2237			FILTER_INPUTS(lut1d_inputs),
2238			FILTER_OUTPUTS(ff_video_default_filterpad),
2239			FILTER_PIXFMTS_ARRAY(pix_fmts),
2240			.priv_class = &lut1d_class,
2241			.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
2242			.process_command = lut1d_process_command,
2243			};
2244			#endif
2245