1 /*
2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version 2
5 * of the License, or (at your option) any later version.
6 *
7 * This program is distributed in the hope that it will be useful,
8 * but WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10 * GNU General Public License for more details.
11 *
12 * You should have received a copy of the GNU General Public License
13 * along with this program; if not, write to the Free Software Foundation,
14 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
15 *
16 * The Original Code is Copyright (C) 2001-2002 by NaN Holding BV.
17 * All rights reserved.
18 *
19 * The Original Code is: some of this file.
20 *
21 * */
22
23 /** \file
24 * \ingroup bli
25 */
26
27 #include "BLI_math_base.h"
28 #include "BLI_math_color.h"
29 #include "BLI_utildefines.h"
30
31 #include "math.h"
32
33 #ifndef __MATH_COLOR_INLINE_C__
34 # define __MATH_COLOR_INLINE_C__
35
36 /******************************** Color Space ********************************/
37
38 # ifdef __SSE2__
39
srgb_to_linearrgb_v4_simd(const __m128 c)40 MALWAYS_INLINE __m128 srgb_to_linearrgb_v4_simd(const __m128 c)
41 {
42 __m128 cmp = _mm_cmplt_ps(c, _mm_set1_ps(0.04045f));
43 __m128 lt = _mm_max_ps(_mm_mul_ps(c, _mm_set1_ps(1.0f / 12.92f)), _mm_set1_ps(0.0f));
44 __m128 gtebase = _mm_mul_ps(_mm_add_ps(c, _mm_set1_ps(0.055f)),
45 _mm_set1_ps(1.0f / 1.055f)); /* fma */
46 __m128 gte = _bli_math_fastpow24(gtebase);
47 return _bli_math_blend_sse(cmp, lt, gte);
48 }
49
linearrgb_to_srgb_v4_simd(const __m128 c)50 MALWAYS_INLINE __m128 linearrgb_to_srgb_v4_simd(const __m128 c)
51 {
52 __m128 cmp = _mm_cmplt_ps(c, _mm_set1_ps(0.0031308f));
53 __m128 lt = _mm_max_ps(_mm_mul_ps(c, _mm_set1_ps(12.92f)), _mm_set1_ps(0.0f));
54 __m128 gte = _mm_add_ps(_mm_mul_ps(_mm_set1_ps(1.055f), _bli_math_fastpow512(c)),
55 _mm_set1_ps(-0.055f));
56 return _bli_math_blend_sse(cmp, lt, gte);
57 }
58
srgb_to_linearrgb_v3_v3(float linear[3],const float srgb[3])59 MINLINE void srgb_to_linearrgb_v3_v3(float linear[3], const float srgb[3])
60 {
61 float r[4] = {srgb[0], srgb[1], srgb[2], 1.0f};
62 __m128 *rv = (__m128 *)&r;
63 *rv = srgb_to_linearrgb_v4_simd(*rv);
64 linear[0] = r[0];
65 linear[1] = r[1];
66 linear[2] = r[2];
67 }
68
linearrgb_to_srgb_v3_v3(float srgb[3],const float linear[3])69 MINLINE void linearrgb_to_srgb_v3_v3(float srgb[3], const float linear[3])
70 {
71 float r[4] = {linear[0], linear[1], linear[2], 1.0f};
72 __m128 *rv = (__m128 *)&r;
73 *rv = linearrgb_to_srgb_v4_simd(*rv);
74 srgb[0] = r[0];
75 srgb[1] = r[1];
76 srgb[2] = r[2];
77 }
78
79 # else /* __SSE2__ */
80
srgb_to_linearrgb_v3_v3(float linear[3],const float srgb[3])81 MINLINE void srgb_to_linearrgb_v3_v3(float linear[3], const float srgb[3])
82 {
83 linear[0] = srgb_to_linearrgb(srgb[0]);
84 linear[1] = srgb_to_linearrgb(srgb[1]);
85 linear[2] = srgb_to_linearrgb(srgb[2]);
86 }
87
linearrgb_to_srgb_v3_v3(float srgb[3],const float linear[3])88 MINLINE void linearrgb_to_srgb_v3_v3(float srgb[3], const float linear[3])
89 {
90 srgb[0] = linearrgb_to_srgb(linear[0]);
91 srgb[1] = linearrgb_to_srgb(linear[1]);
92 srgb[2] = linearrgb_to_srgb(linear[2]);
93 }
94 # endif /* __SSE2__ */
95
srgb_to_linearrgb_v4(float linear[4],const float srgb[4])96 MINLINE void srgb_to_linearrgb_v4(float linear[4], const float srgb[4])
97 {
98 srgb_to_linearrgb_v3_v3(linear, srgb);
99 linear[3] = srgb[3];
100 }
101
linearrgb_to_srgb_v4(float srgb[4],const float linear[4])102 MINLINE void linearrgb_to_srgb_v4(float srgb[4], const float linear[4])
103 {
104 linearrgb_to_srgb_v3_v3(srgb, linear);
105 srgb[3] = linear[3];
106 }
107
linearrgb_to_srgb_uchar3(unsigned char srgb[3],const float linear[3])108 MINLINE void linearrgb_to_srgb_uchar3(unsigned char srgb[3], const float linear[3])
109 {
110 float srgb_f[3];
111
112 linearrgb_to_srgb_v3_v3(srgb_f, linear);
113 unit_float_to_uchar_clamp_v3(srgb, srgb_f);
114 }
115
linearrgb_to_srgb_uchar4(unsigned char srgb[4],const float linear[4])116 MINLINE void linearrgb_to_srgb_uchar4(unsigned char srgb[4], const float linear[4])
117 {
118 float srgb_f[4];
119
120 linearrgb_to_srgb_v4(srgb_f, linear);
121 unit_float_to_uchar_clamp_v4(srgb, srgb_f);
122 }
123
124 /* predivide versions to work on associated/pre-multiplied alpha. if this should
125 * be done or not depends on the background the image will be composited over,
126 * ideally you would never do color space conversion on an image with alpha
127 * because it is ill defined */
128
srgb_to_linearrgb_predivide_v4(float linear[4],const float srgb[4])129 MINLINE void srgb_to_linearrgb_predivide_v4(float linear[4], const float srgb[4])
130 {
131 float alpha, inv_alpha;
132
133 if (srgb[3] == 1.0f || srgb[3] == 0.0f) {
134 alpha = 1.0f;
135 inv_alpha = 1.0f;
136 }
137 else {
138 alpha = srgb[3];
139 inv_alpha = 1.0f / alpha;
140 }
141
142 linear[0] = srgb[0] * inv_alpha;
143 linear[1] = srgb[1] * inv_alpha;
144 linear[2] = srgb[2] * inv_alpha;
145 linear[3] = srgb[3];
146 srgb_to_linearrgb_v3_v3(linear, linear);
147 linear[0] *= alpha;
148 linear[1] *= alpha;
149 linear[2] *= alpha;
150 }
151
linearrgb_to_srgb_predivide_v4(float srgb[4],const float linear[4])152 MINLINE void linearrgb_to_srgb_predivide_v4(float srgb[4], const float linear[4])
153 {
154 float alpha, inv_alpha;
155
156 if (linear[3] == 1.0f || linear[3] == 0.0f) {
157 alpha = 1.0f;
158 inv_alpha = 1.0f;
159 }
160 else {
161 alpha = linear[3];
162 inv_alpha = 1.0f / alpha;
163 }
164
165 srgb[0] = linear[0] * inv_alpha;
166 srgb[1] = linear[1] * inv_alpha;
167 srgb[2] = linear[2] * inv_alpha;
168 srgb[3] = linear[3];
169 linearrgb_to_srgb_v3_v3(srgb, srgb);
170 srgb[0] *= alpha;
171 srgb[1] *= alpha;
172 srgb[2] *= alpha;
173 }
174
175 /* LUT accelerated conversions */
176
177 extern float BLI_color_from_srgb_table[256];
178 extern unsigned short BLI_color_to_srgb_table[0x10000];
179
to_srgb_table_lookup(const float f)180 MINLINE unsigned short to_srgb_table_lookup(const float f)
181 {
182
183 union {
184 float f;
185 unsigned short us[2];
186 } tmp;
187 tmp.f = f;
188 # ifdef __BIG_ENDIAN__
189 return BLI_color_to_srgb_table[tmp.us[0]];
190 # else
191 return BLI_color_to_srgb_table[tmp.us[1]];
192 # endif
193 }
194
linearrgb_to_srgb_ushort4(unsigned short srgb[4],const float linear[4])195 MINLINE void linearrgb_to_srgb_ushort4(unsigned short srgb[4], const float linear[4])
196 {
197 srgb[0] = to_srgb_table_lookup(linear[0]);
198 srgb[1] = to_srgb_table_lookup(linear[1]);
199 srgb[2] = to_srgb_table_lookup(linear[2]);
200 srgb[3] = unit_float_to_ushort_clamp(linear[3]);
201 }
202
srgb_to_linearrgb_uchar4(float linear[4],const unsigned char srgb[4])203 MINLINE void srgb_to_linearrgb_uchar4(float linear[4], const unsigned char srgb[4])
204 {
205 linear[0] = BLI_color_from_srgb_table[srgb[0]];
206 linear[1] = BLI_color_from_srgb_table[srgb[1]];
207 linear[2] = BLI_color_from_srgb_table[srgb[2]];
208 linear[3] = srgb[3] * (1.0f / 255.0f);
209 }
210
srgb_to_linearrgb_uchar4_predivide(float linear[4],const unsigned char srgb[4])211 MINLINE void srgb_to_linearrgb_uchar4_predivide(float linear[4], const unsigned char srgb[4])
212 {
213 float fsrgb[4];
214 int i;
215
216 if (srgb[3] == 255 || srgb[3] == 0) {
217 srgb_to_linearrgb_uchar4(linear, srgb);
218 return;
219 }
220
221 for (i = 0; i < 4; i++) {
222 fsrgb[i] = srgb[i] * (1.0f / 255.0f);
223 }
224
225 srgb_to_linearrgb_predivide_v4(linear, fsrgb);
226 }
227
rgba_uchar_args_set(uchar col[4],const uchar r,const uchar g,const uchar b,const uchar a)228 MINLINE void rgba_uchar_args_set(
229 uchar col[4], const uchar r, const uchar g, const uchar b, const uchar a)
230 {
231 col[0] = r;
232 col[1] = g;
233 col[2] = b;
234 col[3] = a;
235 }
236
rgba_float_args_set(float col[4],const float r,const float g,const float b,const float a)237 MINLINE void rgba_float_args_set(
238 float col[4], const float r, const float g, const float b, const float a)
239 {
240 col[0] = r;
241 col[1] = g;
242 col[2] = b;
243 col[3] = a;
244 }
245
rgba_uchar_args_test_set(uchar col[4],const uchar r,const uchar g,const uchar b,const uchar a)246 MINLINE void rgba_uchar_args_test_set(
247 uchar col[4], const uchar r, const uchar g, const uchar b, const uchar a)
248 {
249 if (col[3] == 0) {
250 col[0] = r;
251 col[1] = g;
252 col[2] = b;
253 col[3] = a;
254 }
255 }
256
cpack_cpy_3ub(unsigned char r_col[3],const unsigned int pack)257 MINLINE void cpack_cpy_3ub(unsigned char r_col[3], const unsigned int pack)
258 {
259 r_col[0] = ((pack) >> 0) & 0xFF;
260 r_col[1] = ((pack) >> 8) & 0xFF;
261 r_col[2] = ((pack) >> 16) & 0xFF;
262 }
263
264 /** \name RGB/Grayscale Functions
265 *
266 * \warning
267 * These are only an approximation,
268 * in almost _all_ cases, #IMB_colormanagement_get_luminance should be used instead. However for
269 * screen-only colors which don't depend on the currently loaded profile - this is preferred.
270 * Checking theme colors for contrast, etc. Basically anything outside the render pipeline.
271 *
272 * \{ */
273
274 /**
275 * ITU-R BT.709 primaries
276 * https://en.wikipedia.org/wiki/Relative_luminance
277 *
278 * Real values are:
279 * ``Y = 0.2126390059(R) + 0.7151686788(G) + 0.0721923154(B)``
280 * according to: "Derivation of Basic Television Color Equations", RP 177-1993
281 *
282 * As this sums slightly above 1.0, the document recommends to use:
283 * ``0.2126(R) + 0.7152(G) + 0.0722(B)``, as used here.
284 *
285 * The high precision values are used to calculate the rounded byte weights so they add up to 255:
286 * ``54(R) + 182(G) + 19(B)``
287 */
rgb_to_grayscale(const float rgb[3])288 MINLINE float rgb_to_grayscale(const float rgb[3])
289 {
290 return (0.2126f * rgb[0]) + (0.7152f * rgb[1]) + (0.0722f * rgb[2]);
291 }
292
rgb_to_grayscale_byte(const unsigned char rgb[3])293 MINLINE unsigned char rgb_to_grayscale_byte(const unsigned char rgb[3])
294 {
295 return (unsigned char)(((54 * (unsigned short)rgb[0]) + (182 * (unsigned short)rgb[1]) +
296 (19 * (unsigned short)rgb[2])) /
297 255);
298 }
299
300 /** \} */
301
compare_rgb_uchar(const unsigned char col_a[3],const unsigned char col_b[3],const int limit)302 MINLINE int compare_rgb_uchar(const unsigned char col_a[3],
303 const unsigned char col_b[3],
304 const int limit)
305 {
306 const int r = (int)col_a[0] - (int)col_b[0];
307 if (abs(r) < limit) {
308 const int g = (int)col_a[1] - (int)col_b[1];
309 if (abs(g) < limit) {
310 const int b = (int)col_a[2] - (int)col_b[2];
311 if (abs(b) < limit) {
312 return 1;
313 }
314 }
315 }
316
317 return 0;
318 }
319
320 /* Using a triangle distribution which gives a more final uniform noise.
321 * See Banding in Games:A Noisy Rant(revision 5) Mikkel Gjøl, Playdead (slide 27) */
322 /* Return triangle noise in [-0.5..1.5[ range */
dither_random_value(float s,float t)323 MINLINE float dither_random_value(float s, float t)
324 {
325 /* Original code from https://www.shadertoy.com/view/4t2SDh */
326 /* The noise reshaping technique does not work on CPU.
327 * We generate and merge two distribution instead */
328 /* Uniform noise in [0..1[ range */
329 float nrnd0 = sinf(s * 12.9898f + t * 78.233f) * 43758.5453f;
330 float nrnd1 = sinf(s * 19.9898f + t * 119.233f) * 43798.5453f;
331 nrnd0 -= floorf(nrnd0);
332 nrnd1 -= floorf(nrnd1);
333 /* Convert uniform distribution into triangle-shaped distribution. */
334 return nrnd0 + nrnd1 - 0.5f;
335 }
336
float_to_byte_dither_v3(unsigned char b[3],const float f[3],float dither,float s,float t)337 MINLINE void float_to_byte_dither_v3(
338 unsigned char b[3], const float f[3], float dither, float s, float t)
339 {
340 float dither_value = dither_random_value(s, t) * 0.0033f * dither;
341
342 b[0] = unit_float_to_uchar_clamp(dither_value + f[0]);
343 b[1] = unit_float_to_uchar_clamp(dither_value + f[1]);
344 b[2] = unit_float_to_uchar_clamp(dither_value + f[2]);
345 }
346
347 /**************** Alpha Transformations *****************/
348
premul_to_straight_v4_v4(float straight[4],const float premul[4])349 MINLINE void premul_to_straight_v4_v4(float straight[4], const float premul[4])
350 {
351 if (premul[3] == 0.0f || premul[3] == 1.0f) {
352 straight[0] = premul[0];
353 straight[1] = premul[1];
354 straight[2] = premul[2];
355 straight[3] = premul[3];
356 }
357 else {
358 const float alpha_inv = 1.0f / premul[3];
359 straight[0] = premul[0] * alpha_inv;
360 straight[1] = premul[1] * alpha_inv;
361 straight[2] = premul[2] * alpha_inv;
362 straight[3] = premul[3];
363 }
364 }
365
premul_to_straight_v4(float color[4])366 MINLINE void premul_to_straight_v4(float color[4])
367 {
368 premul_to_straight_v4_v4(color, color);
369 }
370
straight_to_premul_v4_v4(float premul[4],const float straight[4])371 MINLINE void straight_to_premul_v4_v4(float premul[4], const float straight[4])
372 {
373 const float alpha = straight[3];
374 premul[0] = straight[0] * alpha;
375 premul[1] = straight[1] * alpha;
376 premul[2] = straight[2] * alpha;
377 premul[3] = straight[3];
378 }
379
straight_to_premul_v4(float color[4])380 MINLINE void straight_to_premul_v4(float color[4])
381 {
382 straight_to_premul_v4_v4(color, color);
383 }
384
straight_uchar_to_premul_float(float result[4],const unsigned char color[4])385 MINLINE void straight_uchar_to_premul_float(float result[4], const unsigned char color[4])
386 {
387 const float alpha = color[3] * (1.0f / 255.0f);
388 const float fac = alpha * (1.0f / 255.0f);
389
390 result[0] = color[0] * fac;
391 result[1] = color[1] * fac;
392 result[2] = color[2] * fac;
393 result[3] = alpha;
394 }
395
premul_float_to_straight_uchar(unsigned char * result,const float color[4])396 MINLINE void premul_float_to_straight_uchar(unsigned char *result, const float color[4])
397 {
398 if (color[3] == 0.0f || color[3] == 1.0f) {
399 result[0] = unit_float_to_uchar_clamp(color[0]);
400 result[1] = unit_float_to_uchar_clamp(color[1]);
401 result[2] = unit_float_to_uchar_clamp(color[2]);
402 result[3] = unit_float_to_uchar_clamp(color[3]);
403 }
404 else {
405 const float alpha_inv = 1.0f / color[3];
406
407 /* hopefully this would be optimized */
408 result[0] = unit_float_to_uchar_clamp(color[0] * alpha_inv);
409 result[1] = unit_float_to_uchar_clamp(color[1] * alpha_inv);
410 result[2] = unit_float_to_uchar_clamp(color[2] * alpha_inv);
411 result[3] = unit_float_to_uchar_clamp(color[3]);
412 }
413 }
414
415 #endif /* __MATH_COLOR_INLINE_C__ */
416