1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/gfx/qcms/transform.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,1395 @@
1.4 +/* vim: set ts=8 sw=8 noexpandtab: */
1.5 +// qcms
1.6 +// Copyright (C) 2009 Mozilla Corporation
1.7 +// Copyright (C) 1998-2007 Marti Maria
1.8 +//
1.9 +// Permission is hereby granted, free of charge, to any person obtaining
1.10 +// a copy of this software and associated documentation files (the "Software"),
1.11 +// to deal in the Software without restriction, including without limitation
1.12 +// the rights to use, copy, modify, merge, publish, distribute, sublicense,
1.13 +// and/or sell copies of the Software, and to permit persons to whom the Software
1.14 +// is furnished to do so, subject to the following conditions:
1.15 +//
1.16 +// The above copyright notice and this permission notice shall be included in
1.17 +// all copies or substantial portions of the Software.
1.18 +//
1.19 +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
1.20 +// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
1.21 +// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
1.22 +// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
1.23 +// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
1.24 +// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
1.25 +// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
1.26 +
1.27 +#include <stdlib.h>
1.28 +#include <math.h>
1.29 +#include <assert.h>
1.30 +#include <string.h> //memcpy
1.31 +#include "qcmsint.h"
1.32 +#include "chain.h"
1.33 +#include "matrix.h"
1.34 +#include "transform_util.h"
1.35 +
1.36 +/* for MSVC, GCC, Intel, and Sun compilers */
1.37 +#if defined(_M_IX86) || defined(__i386__) || defined(__i386) || defined(_M_AMD64) || defined(__x86_64__) || defined(__x86_64)
1.38 +#define X86
1.39 +#endif /* _M_IX86 || __i386__ || __i386 || _M_AMD64 || __x86_64__ || __x86_64 */
1.40 +
1.41 +/**
1.42 + * AltiVec detection for PowerPC CPUs
1.43 + * In case we have a method of detecting do the runtime detection.
1.44 + * Otherwise statically choose the AltiVec path in case the compiler
1.45 + * was told to build with AltiVec support.
1.46 + */
1.47 +#if (defined(__POWERPC__) || defined(__powerpc__))
1.48 +#if defined(__linux__)
1.49 +#include <unistd.h>
1.50 +#include <fcntl.h>
1.51 +#include <stdio.h>
1.52 +#include <elf.h>
1.53 +#include <linux/auxvec.h>
1.54 +#include <asm/cputable.h>
1.55 +#include <link.h>
1.56 +
1.57 +static inline qcms_bool have_altivec() {
1.58 + static int available = -1;
1.59 + int new_avail = 0;
1.60 + ElfW(auxv_t) auxv;
1.61 + ssize_t count;
1.62 + int fd, i;
1.63 +
1.64 + if (available != -1)
1.65 + return (available != 0 ? true : false);
1.66 +
1.67 + fd = open("/proc/self/auxv", O_RDONLY);
1.68 + if (fd < 0)
1.69 + goto out;
1.70 + do {
1.71 + count = read(fd, &auxv, sizeof(auxv));
1.72 + if (count < 0)
1.73 + goto out_close;
1.74 +
1.75 + if (auxv.a_type == AT_HWCAP) {
1.76 + new_avail = !!(auxv.a_un.a_val & PPC_FEATURE_HAS_ALTIVEC);
1.77 + goto out_close;
1.78 + }
1.79 + } while (auxv.a_type != AT_NULL);
1.80 +
1.81 +out_close:
1.82 + close(fd);
1.83 +out:
1.84 + available = new_avail;
1.85 + return (available != 0 ? true : false);
1.86 +}
1.87 +#elif defined(__APPLE__) && defined(__MACH__)
1.88 +#include <sys/sysctl.h>
1.89 +
1.90 +/**
1.91 + * rip-off from ffmpeg AltiVec detection code.
1.92 + * this code also appears on Apple's AltiVec pages.
1.93 + */
1.94 +static inline qcms_bool have_altivec() {
1.95 + int sels[2] = {CTL_HW, HW_VECTORUNIT};
1.96 + static int available = -1;
1.97 + size_t len = sizeof(available);
1.98 + int err;
1.99 +
1.100 + if (available != -1)
1.101 + return (available != 0 ? true : false);
1.102 +
1.103 + err = sysctl(sels, 2, &available, &len, NULL, 0);
1.104 +
1.105 + if (err == 0)
1.106 + if (available != 0)
1.107 + return true;
1.108 +
1.109 + return false;
1.110 +}
1.111 +#elif defined(__ALTIVEC__) || defined(__APPLE_ALTIVEC__)
1.112 +#define have_altivec() true
1.113 +#else
1.114 +#define have_altivec() false
1.115 +#endif
1.116 +#endif // (defined(__POWERPC__) || defined(__powerpc__))
1.117 +
1.118 +// Build a White point, primary chromas transfer matrix from RGB to CIE XYZ
1.119 +// This is just an approximation, I am not handling all the non-linear
1.120 +// aspects of the RGB to XYZ process, and assumming that the gamma correction
1.121 +// has transitive property in the tranformation chain.
1.122 +//
1.123 +// the alghoritm:
1.124 +//
1.125 +// - First I build the absolute conversion matrix using
1.126 +// primaries in XYZ. This matrix is next inverted
1.127 +// - Then I eval the source white point across this matrix
1.128 +// obtaining the coeficients of the transformation
1.129 +// - Then, I apply these coeficients to the original matrix
1.130 +static struct matrix build_RGB_to_XYZ_transfer_matrix(qcms_CIE_xyY white, qcms_CIE_xyYTRIPLE primrs)
1.131 +{
1.132 + struct matrix primaries;
1.133 + struct matrix primaries_invert;
1.134 + struct matrix result;
1.135 + struct vector white_point;
1.136 + struct vector coefs;
1.137 +
1.138 + double xn, yn;
1.139 + double xr, yr;
1.140 + double xg, yg;
1.141 + double xb, yb;
1.142 +
1.143 + xn = white.x;
1.144 + yn = white.y;
1.145 +
1.146 + if (yn == 0.0)
1.147 + return matrix_invalid();
1.148 +
1.149 + xr = primrs.red.x;
1.150 + yr = primrs.red.y;
1.151 + xg = primrs.green.x;
1.152 + yg = primrs.green.y;
1.153 + xb = primrs.blue.x;
1.154 + yb = primrs.blue.y;
1.155 +
1.156 + primaries.m[0][0] = xr;
1.157 + primaries.m[0][1] = xg;
1.158 + primaries.m[0][2] = xb;
1.159 +
1.160 + primaries.m[1][0] = yr;
1.161 + primaries.m[1][1] = yg;
1.162 + primaries.m[1][2] = yb;
1.163 +
1.164 + primaries.m[2][0] = 1 - xr - yr;
1.165 + primaries.m[2][1] = 1 - xg - yg;
1.166 + primaries.m[2][2] = 1 - xb - yb;
1.167 + primaries.invalid = false;
1.168 +
1.169 + white_point.v[0] = xn/yn;
1.170 + white_point.v[1] = 1.;
1.171 + white_point.v[2] = (1.0-xn-yn)/yn;
1.172 +
1.173 + primaries_invert = matrix_invert(primaries);
1.174 +
1.175 + coefs = matrix_eval(primaries_invert, white_point);
1.176 +
1.177 + result.m[0][0] = coefs.v[0]*xr;
1.178 + result.m[0][1] = coefs.v[1]*xg;
1.179 + result.m[0][2] = coefs.v[2]*xb;
1.180 +
1.181 + result.m[1][0] = coefs.v[0]*yr;
1.182 + result.m[1][1] = coefs.v[1]*yg;
1.183 + result.m[1][2] = coefs.v[2]*yb;
1.184 +
1.185 + result.m[2][0] = coefs.v[0]*(1.-xr-yr);
1.186 + result.m[2][1] = coefs.v[1]*(1.-xg-yg);
1.187 + result.m[2][2] = coefs.v[2]*(1.-xb-yb);
1.188 + result.invalid = primaries_invert.invalid;
1.189 +
1.190 + return result;
1.191 +}
1.192 +
1.193 +struct CIE_XYZ {
1.194 + double X;
1.195 + double Y;
1.196 + double Z;
1.197 +};
1.198 +
1.199 +/* CIE Illuminant D50 */
1.200 +static const struct CIE_XYZ D50_XYZ = {
1.201 + 0.9642,
1.202 + 1.0000,
1.203 + 0.8249
1.204 +};
1.205 +
1.206 +/* from lcms: xyY2XYZ()
1.207 + * corresponds to argyll: icmYxy2XYZ() */
1.208 +static struct CIE_XYZ xyY2XYZ(qcms_CIE_xyY source)
1.209 +{
1.210 + struct CIE_XYZ dest;
1.211 + dest.X = (source.x / source.y) * source.Y;
1.212 + dest.Y = source.Y;
1.213 + dest.Z = ((1 - source.x - source.y) / source.y) * source.Y;
1.214 + return dest;
1.215 +}
1.216 +
1.217 +/* from lcms: ComputeChromaticAdaption */
1.218 +// Compute chromatic adaption matrix using chad as cone matrix
1.219 +static struct matrix
1.220 +compute_chromatic_adaption(struct CIE_XYZ source_white_point,
1.221 + struct CIE_XYZ dest_white_point,
1.222 + struct matrix chad)
1.223 +{
1.224 + struct matrix chad_inv;
1.225 + struct vector cone_source_XYZ, cone_source_rgb;
1.226 + struct vector cone_dest_XYZ, cone_dest_rgb;
1.227 + struct matrix cone, tmp;
1.228 +
1.229 + tmp = chad;
1.230 + chad_inv = matrix_invert(tmp);
1.231 +
1.232 + cone_source_XYZ.v[0] = source_white_point.X;
1.233 + cone_source_XYZ.v[1] = source_white_point.Y;
1.234 + cone_source_XYZ.v[2] = source_white_point.Z;
1.235 +
1.236 + cone_dest_XYZ.v[0] = dest_white_point.X;
1.237 + cone_dest_XYZ.v[1] = dest_white_point.Y;
1.238 + cone_dest_XYZ.v[2] = dest_white_point.Z;
1.239 +
1.240 + cone_source_rgb = matrix_eval(chad, cone_source_XYZ);
1.241 + cone_dest_rgb = matrix_eval(chad, cone_dest_XYZ);
1.242 +
1.243 + cone.m[0][0] = cone_dest_rgb.v[0]/cone_source_rgb.v[0];
1.244 + cone.m[0][1] = 0;
1.245 + cone.m[0][2] = 0;
1.246 + cone.m[1][0] = 0;
1.247 + cone.m[1][1] = cone_dest_rgb.v[1]/cone_source_rgb.v[1];
1.248 + cone.m[1][2] = 0;
1.249 + cone.m[2][0] = 0;
1.250 + cone.m[2][1] = 0;
1.251 + cone.m[2][2] = cone_dest_rgb.v[2]/cone_source_rgb.v[2];
1.252 + cone.invalid = false;
1.253 +
1.254 + // Normalize
1.255 + return matrix_multiply(chad_inv, matrix_multiply(cone, chad));
1.256 +}
1.257 +
1.258 +/* from lcms: cmsAdaptionMatrix */
1.259 +// Returns the final chrmatic adaptation from illuminant FromIll to Illuminant ToIll
1.260 +// Bradford is assumed
1.261 +static struct matrix
1.262 +adaption_matrix(struct CIE_XYZ source_illumination, struct CIE_XYZ target_illumination)
1.263 +{
1.264 + struct matrix lam_rigg = {{ // Bradford matrix
1.265 + { 0.8951, 0.2664, -0.1614 },
1.266 + { -0.7502, 1.7135, 0.0367 },
1.267 + { 0.0389, -0.0685, 1.0296 }
1.268 + }};
1.269 + return compute_chromatic_adaption(source_illumination, target_illumination, lam_rigg);
1.270 +}
1.271 +
1.272 +/* from lcms: cmsAdaptMatrixToD50 */
1.273 +static struct matrix adapt_matrix_to_D50(struct matrix r, qcms_CIE_xyY source_white_pt)
1.274 +{
1.275 + struct CIE_XYZ Dn;
1.276 + struct matrix Bradford;
1.277 +
1.278 + if (source_white_pt.y == 0.0)
1.279 + return matrix_invalid();
1.280 +
1.281 + Dn = xyY2XYZ(source_white_pt);
1.282 +
1.283 + Bradford = adaption_matrix(Dn, D50_XYZ);
1.284 + return matrix_multiply(Bradford, r);
1.285 +}
1.286 +
1.287 +qcms_bool set_rgb_colorants(qcms_profile *profile, qcms_CIE_xyY white_point, qcms_CIE_xyYTRIPLE primaries)
1.288 +{
1.289 + struct matrix colorants;
1.290 + colorants = build_RGB_to_XYZ_transfer_matrix(white_point, primaries);
1.291 + colorants = adapt_matrix_to_D50(colorants, white_point);
1.292 +
1.293 + if (colorants.invalid)
1.294 + return false;
1.295 +
1.296 + /* note: there's a transpose type of operation going on here */
1.297 + profile->redColorant.X = double_to_s15Fixed16Number(colorants.m[0][0]);
1.298 + profile->redColorant.Y = double_to_s15Fixed16Number(colorants.m[1][0]);
1.299 + profile->redColorant.Z = double_to_s15Fixed16Number(colorants.m[2][0]);
1.300 +
1.301 + profile->greenColorant.X = double_to_s15Fixed16Number(colorants.m[0][1]);
1.302 + profile->greenColorant.Y = double_to_s15Fixed16Number(colorants.m[1][1]);
1.303 + profile->greenColorant.Z = double_to_s15Fixed16Number(colorants.m[2][1]);
1.304 +
1.305 + profile->blueColorant.X = double_to_s15Fixed16Number(colorants.m[0][2]);
1.306 + profile->blueColorant.Y = double_to_s15Fixed16Number(colorants.m[1][2]);
1.307 + profile->blueColorant.Z = double_to_s15Fixed16Number(colorants.m[2][2]);
1.308 +
1.309 + return true;
1.310 +}
1.311 +
1.312 +qcms_bool get_rgb_colorants(struct matrix *colorants, qcms_CIE_xyY white_point, qcms_CIE_xyYTRIPLE primaries)
1.313 +{
1.314 + *colorants = build_RGB_to_XYZ_transfer_matrix(white_point, primaries);
1.315 + *colorants = adapt_matrix_to_D50(*colorants, white_point);
1.316 +
1.317 + return (colorants->invalid ? true : false);
1.318 +}
1.319 +
1.320 +#if 0
1.321 +static void qcms_transform_data_rgb_out_pow(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length)
1.322 +{
1.323 + int i;
1.324 + float (*mat)[4] = transform->matrix;
1.325 + for (i=0; i<length; i++) {
1.326 + unsigned char device_r = *src++;
1.327 + unsigned char device_g = *src++;
1.328 + unsigned char device_b = *src++;
1.329 +
1.330 + float linear_r = transform->input_gamma_table_r[device_r];
1.331 + float linear_g = transform->input_gamma_table_g[device_g];
1.332 + float linear_b = transform->input_gamma_table_b[device_b];
1.333 +
1.334 + float out_linear_r = mat[0][0]*linear_r + mat[1][0]*linear_g + mat[2][0]*linear_b;
1.335 + float out_linear_g = mat[0][1]*linear_r + mat[1][1]*linear_g + mat[2][1]*linear_b;
1.336 + float out_linear_b = mat[0][2]*linear_r + mat[1][2]*linear_g + mat[2][2]*linear_b;
1.337 +
1.338 + float out_device_r = pow(out_linear_r, transform->out_gamma_r);
1.339 + float out_device_g = pow(out_linear_g, transform->out_gamma_g);
1.340 + float out_device_b = pow(out_linear_b, transform->out_gamma_b);
1.341 +
1.342 + dest[OUTPUT_R_INDEX] = clamp_u8(255*out_device_r);
1.343 + dest[OUTPUT_G_INDEX] = clamp_u8(255*out_device_g);
1.344 + dest[OUTPUT_B_INDEX] = clamp_u8(255*out_device_b);
1.345 + dest += RGB_OUTPUT_COMPONENTS;
1.346 + }
1.347 +}
1.348 +#endif
1.349 +
1.350 +static void qcms_transform_data_gray_out_lut(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length)
1.351 +{
1.352 + unsigned int i;
1.353 + for (i = 0; i < length; i++) {
1.354 + float out_device_r, out_device_g, out_device_b;
1.355 + unsigned char device = *src++;
1.356 +
1.357 + float linear = transform->input_gamma_table_gray[device];
1.358 +
1.359 + out_device_r = lut_interp_linear(linear, transform->output_gamma_lut_r, transform->output_gamma_lut_r_length);
1.360 + out_device_g = lut_interp_linear(linear, transform->output_gamma_lut_g, transform->output_gamma_lut_g_length);
1.361 + out_device_b = lut_interp_linear(linear, transform->output_gamma_lut_b, transform->output_gamma_lut_b_length);
1.362 +
1.363 + dest[OUTPUT_R_INDEX] = clamp_u8(out_device_r*255);
1.364 + dest[OUTPUT_G_INDEX] = clamp_u8(out_device_g*255);
1.365 + dest[OUTPUT_B_INDEX] = clamp_u8(out_device_b*255);
1.366 + dest += RGB_OUTPUT_COMPONENTS;
1.367 + }
1.368 +}
1.369 +
1.370 +/* Alpha is not corrected.
1.371 + A rationale for this is found in Alvy Ray's "Should Alpha Be Nonlinear If
1.372 + RGB Is?" Tech Memo 17 (December 14, 1998).
1.373 + See: ftp://ftp.alvyray.com/Acrobat/17_Nonln.pdf
1.374 +*/
1.375 +
1.376 +static void qcms_transform_data_graya_out_lut(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length)
1.377 +{
1.378 + unsigned int i;
1.379 + for (i = 0; i < length; i++) {
1.380 + float out_device_r, out_device_g, out_device_b;
1.381 + unsigned char device = *src++;
1.382 + unsigned char alpha = *src++;
1.383 +
1.384 + float linear = transform->input_gamma_table_gray[device];
1.385 +
1.386 + out_device_r = lut_interp_linear(linear, transform->output_gamma_lut_r, transform->output_gamma_lut_r_length);
1.387 + out_device_g = lut_interp_linear(linear, transform->output_gamma_lut_g, transform->output_gamma_lut_g_length);
1.388 + out_device_b = lut_interp_linear(linear, transform->output_gamma_lut_b, transform->output_gamma_lut_b_length);
1.389 +
1.390 + dest[OUTPUT_R_INDEX] = clamp_u8(out_device_r*255);
1.391 + dest[OUTPUT_G_INDEX] = clamp_u8(out_device_g*255);
1.392 + dest[OUTPUT_B_INDEX] = clamp_u8(out_device_b*255);
1.393 + dest[OUTPUT_A_INDEX] = alpha;
1.394 + dest += RGBA_OUTPUT_COMPONENTS;
1.395 + }
1.396 +}
1.397 +
1.398 +
1.399 +static void qcms_transform_data_gray_out_precache(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length)
1.400 +{
1.401 + unsigned int i;
1.402 + for (i = 0; i < length; i++) {
1.403 + unsigned char device = *src++;
1.404 + uint16_t gray;
1.405 +
1.406 + float linear = transform->input_gamma_table_gray[device];
1.407 +
1.408 + /* we could round here... */
1.409 + gray = linear * PRECACHE_OUTPUT_MAX;
1.410 +
1.411 + dest[OUTPUT_R_INDEX] = transform->output_table_r->data[gray];
1.412 + dest[OUTPUT_G_INDEX] = transform->output_table_g->data[gray];
1.413 + dest[OUTPUT_B_INDEX] = transform->output_table_b->data[gray];
1.414 + dest += RGB_OUTPUT_COMPONENTS;
1.415 + }
1.416 +}
1.417 +
1.418 +static void qcms_transform_data_graya_out_precache(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length)
1.419 +{
1.420 + unsigned int i;
1.421 + for (i = 0; i < length; i++) {
1.422 + unsigned char device = *src++;
1.423 + unsigned char alpha = *src++;
1.424 + uint16_t gray;
1.425 +
1.426 + float linear = transform->input_gamma_table_gray[device];
1.427 +
1.428 + /* we could round here... */
1.429 + gray = linear * PRECACHE_OUTPUT_MAX;
1.430 +
1.431 + dest[OUTPUT_R_INDEX] = transform->output_table_r->data[gray];
1.432 + dest[OUTPUT_G_INDEX] = transform->output_table_g->data[gray];
1.433 + dest[OUTPUT_B_INDEX] = transform->output_table_b->data[gray];
1.434 + dest[OUTPUT_A_INDEX] = alpha;
1.435 + dest += RGBA_OUTPUT_COMPONENTS;
1.436 + }
1.437 +}
1.438 +
1.439 +static void qcms_transform_data_rgb_out_lut_precache(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length)
1.440 +{
1.441 + unsigned int i;
1.442 + float (*mat)[4] = transform->matrix;
1.443 + for (i = 0; i < length; i++) {
1.444 + unsigned char device_r = *src++;
1.445 + unsigned char device_g = *src++;
1.446 + unsigned char device_b = *src++;
1.447 + uint16_t r, g, b;
1.448 +
1.449 + float linear_r = transform->input_gamma_table_r[device_r];
1.450 + float linear_g = transform->input_gamma_table_g[device_g];
1.451 + float linear_b = transform->input_gamma_table_b[device_b];
1.452 +
1.453 + float out_linear_r = mat[0][0]*linear_r + mat[1][0]*linear_g + mat[2][0]*linear_b;
1.454 + float out_linear_g = mat[0][1]*linear_r + mat[1][1]*linear_g + mat[2][1]*linear_b;
1.455 + float out_linear_b = mat[0][2]*linear_r + mat[1][2]*linear_g + mat[2][2]*linear_b;
1.456 +
1.457 + out_linear_r = clamp_float(out_linear_r);
1.458 + out_linear_g = clamp_float(out_linear_g);
1.459 + out_linear_b = clamp_float(out_linear_b);
1.460 +
1.461 + /* we could round here... */
1.462 + r = out_linear_r * PRECACHE_OUTPUT_MAX;
1.463 + g = out_linear_g * PRECACHE_OUTPUT_MAX;
1.464 + b = out_linear_b * PRECACHE_OUTPUT_MAX;
1.465 +
1.466 + dest[OUTPUT_R_INDEX] = transform->output_table_r->data[r];
1.467 + dest[OUTPUT_G_INDEX] = transform->output_table_g->data[g];
1.468 + dest[OUTPUT_B_INDEX] = transform->output_table_b->data[b];
1.469 + dest += RGB_OUTPUT_COMPONENTS;
1.470 + }
1.471 +}
1.472 +
1.473 +static void qcms_transform_data_rgba_out_lut_precache(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length)
1.474 +{
1.475 + unsigned int i;
1.476 + float (*mat)[4] = transform->matrix;
1.477 + for (i = 0; i < length; i++) {
1.478 + unsigned char device_r = *src++;
1.479 + unsigned char device_g = *src++;
1.480 + unsigned char device_b = *src++;
1.481 + unsigned char alpha = *src++;
1.482 + uint16_t r, g, b;
1.483 +
1.484 + float linear_r = transform->input_gamma_table_r[device_r];
1.485 + float linear_g = transform->input_gamma_table_g[device_g];
1.486 + float linear_b = transform->input_gamma_table_b[device_b];
1.487 +
1.488 + float out_linear_r = mat[0][0]*linear_r + mat[1][0]*linear_g + mat[2][0]*linear_b;
1.489 + float out_linear_g = mat[0][1]*linear_r + mat[1][1]*linear_g + mat[2][1]*linear_b;
1.490 + float out_linear_b = mat[0][2]*linear_r + mat[1][2]*linear_g + mat[2][2]*linear_b;
1.491 +
1.492 + out_linear_r = clamp_float(out_linear_r);
1.493 + out_linear_g = clamp_float(out_linear_g);
1.494 + out_linear_b = clamp_float(out_linear_b);
1.495 +
1.496 + /* we could round here... */
1.497 + r = out_linear_r * PRECACHE_OUTPUT_MAX;
1.498 + g = out_linear_g * PRECACHE_OUTPUT_MAX;
1.499 + b = out_linear_b * PRECACHE_OUTPUT_MAX;
1.500 +
1.501 + dest[OUTPUT_R_INDEX] = transform->output_table_r->data[r];
1.502 + dest[OUTPUT_G_INDEX] = transform->output_table_g->data[g];
1.503 + dest[OUTPUT_B_INDEX] = transform->output_table_b->data[b];
1.504 + dest[OUTPUT_A_INDEX] = alpha;
1.505 + dest += RGBA_OUTPUT_COMPONENTS;
1.506 + }
1.507 +}
1.508 +
1.509 +// Not used
1.510 +/*
1.511 +static void qcms_transform_data_clut(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length) {
1.512 + unsigned int i;
1.513 + int xy_len = 1;
1.514 + int x_len = transform->grid_size;
1.515 + int len = x_len * x_len;
1.516 + float* r_table = transform->r_clut;
1.517 + float* g_table = transform->g_clut;
1.518 + float* b_table = transform->b_clut;
1.519 +
1.520 + for (i = 0; i < length; i++) {
1.521 + unsigned char in_r = *src++;
1.522 + unsigned char in_g = *src++;
1.523 + unsigned char in_b = *src++;
1.524 + float linear_r = in_r/255.0f, linear_g=in_g/255.0f, linear_b = in_b/255.0f;
1.525 +
1.526 + int x = floorf(linear_r * (transform->grid_size-1));
1.527 + int y = floorf(linear_g * (transform->grid_size-1));
1.528 + int z = floorf(linear_b * (transform->grid_size-1));
1.529 + int x_n = ceilf(linear_r * (transform->grid_size-1));
1.530 + int y_n = ceilf(linear_g * (transform->grid_size-1));
1.531 + int z_n = ceilf(linear_b * (transform->grid_size-1));
1.532 + float x_d = linear_r * (transform->grid_size-1) - x;
1.533 + float y_d = linear_g * (transform->grid_size-1) - y;
1.534 + float z_d = linear_b * (transform->grid_size-1) - z;
1.535 +
1.536 + float r_x1 = lerp(CLU(r_table,x,y,z), CLU(r_table,x_n,y,z), x_d);
1.537 + float r_x2 = lerp(CLU(r_table,x,y_n,z), CLU(r_table,x_n,y_n,z), x_d);
1.538 + float r_y1 = lerp(r_x1, r_x2, y_d);
1.539 + float r_x3 = lerp(CLU(r_table,x,y,z_n), CLU(r_table,x_n,y,z_n), x_d);
1.540 + float r_x4 = lerp(CLU(r_table,x,y_n,z_n), CLU(r_table,x_n,y_n,z_n), x_d);
1.541 + float r_y2 = lerp(r_x3, r_x4, y_d);
1.542 + float clut_r = lerp(r_y1, r_y2, z_d);
1.543 +
1.544 + float g_x1 = lerp(CLU(g_table,x,y,z), CLU(g_table,x_n,y,z), x_d);
1.545 + float g_x2 = lerp(CLU(g_table,x,y_n,z), CLU(g_table,x_n,y_n,z), x_d);
1.546 + float g_y1 = lerp(g_x1, g_x2, y_d);
1.547 + float g_x3 = lerp(CLU(g_table,x,y,z_n), CLU(g_table,x_n,y,z_n), x_d);
1.548 + float g_x4 = lerp(CLU(g_table,x,y_n,z_n), CLU(g_table,x_n,y_n,z_n), x_d);
1.549 + float g_y2 = lerp(g_x3, g_x4, y_d);
1.550 + float clut_g = lerp(g_y1, g_y2, z_d);
1.551 +
1.552 + float b_x1 = lerp(CLU(b_table,x,y,z), CLU(b_table,x_n,y,z), x_d);
1.553 + float b_x2 = lerp(CLU(b_table,x,y_n,z), CLU(b_table,x_n,y_n,z), x_d);
1.554 + float b_y1 = lerp(b_x1, b_x2, y_d);
1.555 + float b_x3 = lerp(CLU(b_table,x,y,z_n), CLU(b_table,x_n,y,z_n), x_d);
1.556 + float b_x4 = lerp(CLU(b_table,x,y_n,z_n), CLU(b_table,x_n,y_n,z_n), x_d);
1.557 + float b_y2 = lerp(b_x3, b_x4, y_d);
1.558 + float clut_b = lerp(b_y1, b_y2, z_d);
1.559 +
1.560 + *dest++ = clamp_u8(clut_r*255.0f);
1.561 + *dest++ = clamp_u8(clut_g*255.0f);
1.562 + *dest++ = clamp_u8(clut_b*255.0f);
1.563 + }
1.564 +}
1.565 +*/
1.566 +
1.567 +static int int_div_ceil(int value, int div) {
1.568 + return ((value + div - 1) / div);
1.569 +}
1.570 +
1.571 +// Using lcms' tetra interpolation algorithm.
1.572 +static void qcms_transform_data_tetra_clut_rgba(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length) {
1.573 + unsigned int i;
1.574 + int xy_len = 1;
1.575 + int x_len = transform->grid_size;
1.576 + int len = x_len * x_len;
1.577 + float* r_table = transform->r_clut;
1.578 + float* g_table = transform->g_clut;
1.579 + float* b_table = transform->b_clut;
1.580 + float c0_r, c1_r, c2_r, c3_r;
1.581 + float c0_g, c1_g, c2_g, c3_g;
1.582 + float c0_b, c1_b, c2_b, c3_b;
1.583 + float clut_r, clut_g, clut_b;
1.584 + for (i = 0; i < length; i++) {
1.585 + unsigned char in_r = *src++;
1.586 + unsigned char in_g = *src++;
1.587 + unsigned char in_b = *src++;
1.588 + unsigned char in_a = *src++;
1.589 + float linear_r = in_r/255.0f, linear_g=in_g/255.0f, linear_b = in_b/255.0f;
1.590 +
1.591 + int x = in_r * (transform->grid_size-1) / 255;
1.592 + int y = in_g * (transform->grid_size-1) / 255;
1.593 + int z = in_b * (transform->grid_size-1) / 255;
1.594 + int x_n = int_div_ceil(in_r * (transform->grid_size-1), 255);
1.595 + int y_n = int_div_ceil(in_g * (transform->grid_size-1), 255);
1.596 + int z_n = int_div_ceil(in_b * (transform->grid_size-1), 255);
1.597 + float rx = linear_r * (transform->grid_size-1) - x;
1.598 + float ry = linear_g * (transform->grid_size-1) - y;
1.599 + float rz = linear_b * (transform->grid_size-1) - z;
1.600 +
1.601 + c0_r = CLU(r_table, x, y, z);
1.602 + c0_g = CLU(g_table, x, y, z);
1.603 + c0_b = CLU(b_table, x, y, z);
1.604 +
1.605 + if( rx >= ry ) {
1.606 + if (ry >= rz) { //rx >= ry && ry >= rz
1.607 + c1_r = CLU(r_table, x_n, y, z) - c0_r;
1.608 + c2_r = CLU(r_table, x_n, y_n, z) - CLU(r_table, x_n, y, z);
1.609 + c3_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y_n, z);
1.610 + c1_g = CLU(g_table, x_n, y, z) - c0_g;
1.611 + c2_g = CLU(g_table, x_n, y_n, z) - CLU(g_table, x_n, y, z);
1.612 + c3_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y_n, z);
1.613 + c1_b = CLU(b_table, x_n, y, z) - c0_b;
1.614 + c2_b = CLU(b_table, x_n, y_n, z) - CLU(b_table, x_n, y, z);
1.615 + c3_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y_n, z);
1.616 + } else {
1.617 + if (rx >= rz) { //rx >= rz && rz >= ry
1.618 + c1_r = CLU(r_table, x_n, y, z) - c0_r;
1.619 + c2_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y, z_n);
1.620 + c3_r = CLU(r_table, x_n, y, z_n) - CLU(r_table, x_n, y, z);
1.621 + c1_g = CLU(g_table, x_n, y, z) - c0_g;
1.622 + c2_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y, z_n);
1.623 + c3_g = CLU(g_table, x_n, y, z_n) - CLU(g_table, x_n, y, z);
1.624 + c1_b = CLU(b_table, x_n, y, z) - c0_b;
1.625 + c2_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y, z_n);
1.626 + c3_b = CLU(b_table, x_n, y, z_n) - CLU(b_table, x_n, y, z);
1.627 + } else { //rz > rx && rx >= ry
1.628 + c1_r = CLU(r_table, x_n, y, z_n) - CLU(r_table, x, y, z_n);
1.629 + c2_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y, z_n);
1.630 + c3_r = CLU(r_table, x, y, z_n) - c0_r;
1.631 + c1_g = CLU(g_table, x_n, y, z_n) - CLU(g_table, x, y, z_n);
1.632 + c2_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y, z_n);
1.633 + c3_g = CLU(g_table, x, y, z_n) - c0_g;
1.634 + c1_b = CLU(b_table, x_n, y, z_n) - CLU(b_table, x, y, z_n);
1.635 + c2_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y, z_n);
1.636 + c3_b = CLU(b_table, x, y, z_n) - c0_b;
1.637 + }
1.638 + }
1.639 + } else {
1.640 + if (rx >= rz) { //ry > rx && rx >= rz
1.641 + c1_r = CLU(r_table, x_n, y_n, z) - CLU(r_table, x, y_n, z);
1.642 + c2_r = CLU(r_table, x, y_n, z) - c0_r;
1.643 + c3_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y_n, z);
1.644 + c1_g = CLU(g_table, x_n, y_n, z) - CLU(g_table, x, y_n, z);
1.645 + c2_g = CLU(g_table, x, y_n, z) - c0_g;
1.646 + c3_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y_n, z);
1.647 + c1_b = CLU(b_table, x_n, y_n, z) - CLU(b_table, x, y_n, z);
1.648 + c2_b = CLU(b_table, x, y_n, z) - c0_b;
1.649 + c3_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y_n, z);
1.650 + } else {
1.651 + if (ry >= rz) { //ry >= rz && rz > rx
1.652 + c1_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x, y_n, z_n);
1.653 + c2_r = CLU(r_table, x, y_n, z) - c0_r;
1.654 + c3_r = CLU(r_table, x, y_n, z_n) - CLU(r_table, x, y_n, z);
1.655 + c1_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x, y_n, z_n);
1.656 + c2_g = CLU(g_table, x, y_n, z) - c0_g;
1.657 + c3_g = CLU(g_table, x, y_n, z_n) - CLU(g_table, x, y_n, z);
1.658 + c1_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x, y_n, z_n);
1.659 + c2_b = CLU(b_table, x, y_n, z) - c0_b;
1.660 + c3_b = CLU(b_table, x, y_n, z_n) - CLU(b_table, x, y_n, z);
1.661 + } else { //rz > ry && ry > rx
1.662 + c1_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x, y_n, z_n);
1.663 + c2_r = CLU(r_table, x, y_n, z_n) - CLU(r_table, x, y, z_n);
1.664 + c3_r = CLU(r_table, x, y, z_n) - c0_r;
1.665 + c1_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x, y_n, z_n);
1.666 + c2_g = CLU(g_table, x, y_n, z_n) - CLU(g_table, x, y, z_n);
1.667 + c3_g = CLU(g_table, x, y, z_n) - c0_g;
1.668 + c1_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x, y_n, z_n);
1.669 + c2_b = CLU(b_table, x, y_n, z_n) - CLU(b_table, x, y, z_n);
1.670 + c3_b = CLU(b_table, x, y, z_n) - c0_b;
1.671 + }
1.672 + }
1.673 + }
1.674 +
1.675 + clut_r = c0_r + c1_r*rx + c2_r*ry + c3_r*rz;
1.676 + clut_g = c0_g + c1_g*rx + c2_g*ry + c3_g*rz;
1.677 + clut_b = c0_b + c1_b*rx + c2_b*ry + c3_b*rz;
1.678 +
1.679 + dest[OUTPUT_R_INDEX] = clamp_u8(clut_r*255.0f);
1.680 + dest[OUTPUT_G_INDEX] = clamp_u8(clut_g*255.0f);
1.681 + dest[OUTPUT_B_INDEX] = clamp_u8(clut_b*255.0f);
1.682 + dest[OUTPUT_A_INDEX] = in_a;
1.683 + dest += RGBA_OUTPUT_COMPONENTS;
1.684 + }
1.685 +}
1.686 +
1.687 +// Using lcms' tetra interpolation code.
1.688 +static void qcms_transform_data_tetra_clut(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length) {
1.689 + unsigned int i;
1.690 + int xy_len = 1;
1.691 + int x_len = transform->grid_size;
1.692 + int len = x_len * x_len;
1.693 + float* r_table = transform->r_clut;
1.694 + float* g_table = transform->g_clut;
1.695 + float* b_table = transform->b_clut;
1.696 + float c0_r, c1_r, c2_r, c3_r;
1.697 + float c0_g, c1_g, c2_g, c3_g;
1.698 + float c0_b, c1_b, c2_b, c3_b;
1.699 + float clut_r, clut_g, clut_b;
1.700 + for (i = 0; i < length; i++) {
1.701 + unsigned char in_r = *src++;
1.702 + unsigned char in_g = *src++;
1.703 + unsigned char in_b = *src++;
1.704 + float linear_r = in_r/255.0f, linear_g=in_g/255.0f, linear_b = in_b/255.0f;
1.705 +
1.706 + int x = in_r * (transform->grid_size-1) / 255;
1.707 + int y = in_g * (transform->grid_size-1) / 255;
1.708 + int z = in_b * (transform->grid_size-1) / 255;
1.709 + int x_n = int_div_ceil(in_r * (transform->grid_size-1), 255);
1.710 + int y_n = int_div_ceil(in_g * (transform->grid_size-1), 255);
1.711 + int z_n = int_div_ceil(in_b * (transform->grid_size-1), 255);
1.712 + float rx = linear_r * (transform->grid_size-1) - x;
1.713 + float ry = linear_g * (transform->grid_size-1) - y;
1.714 + float rz = linear_b * (transform->grid_size-1) - z;
1.715 +
1.716 + c0_r = CLU(r_table, x, y, z);
1.717 + c0_g = CLU(g_table, x, y, z);
1.718 + c0_b = CLU(b_table, x, y, z);
1.719 +
1.720 + if( rx >= ry ) {
1.721 + if (ry >= rz) { //rx >= ry && ry >= rz
1.722 + c1_r = CLU(r_table, x_n, y, z) - c0_r;
1.723 + c2_r = CLU(r_table, x_n, y_n, z) - CLU(r_table, x_n, y, z);
1.724 + c3_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y_n, z);
1.725 + c1_g = CLU(g_table, x_n, y, z) - c0_g;
1.726 + c2_g = CLU(g_table, x_n, y_n, z) - CLU(g_table, x_n, y, z);
1.727 + c3_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y_n, z);
1.728 + c1_b = CLU(b_table, x_n, y, z) - c0_b;
1.729 + c2_b = CLU(b_table, x_n, y_n, z) - CLU(b_table, x_n, y, z);
1.730 + c3_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y_n, z);
1.731 + } else {
1.732 + if (rx >= rz) { //rx >= rz && rz >= ry
1.733 + c1_r = CLU(r_table, x_n, y, z) - c0_r;
1.734 + c2_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y, z_n);
1.735 + c3_r = CLU(r_table, x_n, y, z_n) - CLU(r_table, x_n, y, z);
1.736 + c1_g = CLU(g_table, x_n, y, z) - c0_g;
1.737 + c2_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y, z_n);
1.738 + c3_g = CLU(g_table, x_n, y, z_n) - CLU(g_table, x_n, y, z);
1.739 + c1_b = CLU(b_table, x_n, y, z) - c0_b;
1.740 + c2_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y, z_n);
1.741 + c3_b = CLU(b_table, x_n, y, z_n) - CLU(b_table, x_n, y, z);
1.742 + } else { //rz > rx && rx >= ry
1.743 + c1_r = CLU(r_table, x_n, y, z_n) - CLU(r_table, x, y, z_n);
1.744 + c2_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y, z_n);
1.745 + c3_r = CLU(r_table, x, y, z_n) - c0_r;
1.746 + c1_g = CLU(g_table, x_n, y, z_n) - CLU(g_table, x, y, z_n);
1.747 + c2_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y, z_n);
1.748 + c3_g = CLU(g_table, x, y, z_n) - c0_g;
1.749 + c1_b = CLU(b_table, x_n, y, z_n) - CLU(b_table, x, y, z_n);
1.750 + c2_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y, z_n);
1.751 + c3_b = CLU(b_table, x, y, z_n) - c0_b;
1.752 + }
1.753 + }
1.754 + } else {
1.755 + if (rx >= rz) { //ry > rx && rx >= rz
1.756 + c1_r = CLU(r_table, x_n, y_n, z) - CLU(r_table, x, y_n, z);
1.757 + c2_r = CLU(r_table, x, y_n, z) - c0_r;
1.758 + c3_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x_n, y_n, z);
1.759 + c1_g = CLU(g_table, x_n, y_n, z) - CLU(g_table, x, y_n, z);
1.760 + c2_g = CLU(g_table, x, y_n, z) - c0_g;
1.761 + c3_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x_n, y_n, z);
1.762 + c1_b = CLU(b_table, x_n, y_n, z) - CLU(b_table, x, y_n, z);
1.763 + c2_b = CLU(b_table, x, y_n, z) - c0_b;
1.764 + c3_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x_n, y_n, z);
1.765 + } else {
1.766 + if (ry >= rz) { //ry >= rz && rz > rx
1.767 + c1_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x, y_n, z_n);
1.768 + c2_r = CLU(r_table, x, y_n, z) - c0_r;
1.769 + c3_r = CLU(r_table, x, y_n, z_n) - CLU(r_table, x, y_n, z);
1.770 + c1_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x, y_n, z_n);
1.771 + c2_g = CLU(g_table, x, y_n, z) - c0_g;
1.772 + c3_g = CLU(g_table, x, y_n, z_n) - CLU(g_table, x, y_n, z);
1.773 + c1_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x, y_n, z_n);
1.774 + c2_b = CLU(b_table, x, y_n, z) - c0_b;
1.775 + c3_b = CLU(b_table, x, y_n, z_n) - CLU(b_table, x, y_n, z);
1.776 + } else { //rz > ry && ry > rx
1.777 + c1_r = CLU(r_table, x_n, y_n, z_n) - CLU(r_table, x, y_n, z_n);
1.778 + c2_r = CLU(r_table, x, y_n, z_n) - CLU(r_table, x, y, z_n);
1.779 + c3_r = CLU(r_table, x, y, z_n) - c0_r;
1.780 + c1_g = CLU(g_table, x_n, y_n, z_n) - CLU(g_table, x, y_n, z_n);
1.781 + c2_g = CLU(g_table, x, y_n, z_n) - CLU(g_table, x, y, z_n);
1.782 + c3_g = CLU(g_table, x, y, z_n) - c0_g;
1.783 + c1_b = CLU(b_table, x_n, y_n, z_n) - CLU(b_table, x, y_n, z_n);
1.784 + c2_b = CLU(b_table, x, y_n, z_n) - CLU(b_table, x, y, z_n);
1.785 + c3_b = CLU(b_table, x, y, z_n) - c0_b;
1.786 + }
1.787 + }
1.788 + }
1.789 +
1.790 + clut_r = c0_r + c1_r*rx + c2_r*ry + c3_r*rz;
1.791 + clut_g = c0_g + c1_g*rx + c2_g*ry + c3_g*rz;
1.792 + clut_b = c0_b + c1_b*rx + c2_b*ry + c3_b*rz;
1.793 +
1.794 + dest[OUTPUT_R_INDEX] = clamp_u8(clut_r*255.0f);
1.795 + dest[OUTPUT_G_INDEX] = clamp_u8(clut_g*255.0f);
1.796 + dest[OUTPUT_B_INDEX] = clamp_u8(clut_b*255.0f);
1.797 + dest += RGB_OUTPUT_COMPONENTS;
1.798 + }
1.799 +}
1.800 +
1.801 +static void qcms_transform_data_rgb_out_lut(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length)
1.802 +{
1.803 + unsigned int i;
1.804 + float (*mat)[4] = transform->matrix;
1.805 + for (i = 0; i < length; i++) {
1.806 + unsigned char device_r = *src++;
1.807 + unsigned char device_g = *src++;
1.808 + unsigned char device_b = *src++;
1.809 + float out_device_r, out_device_g, out_device_b;
1.810 +
1.811 + float linear_r = transform->input_gamma_table_r[device_r];
1.812 + float linear_g = transform->input_gamma_table_g[device_g];
1.813 + float linear_b = transform->input_gamma_table_b[device_b];
1.814 +
1.815 + float out_linear_r = mat[0][0]*linear_r + mat[1][0]*linear_g + mat[2][0]*linear_b;
1.816 + float out_linear_g = mat[0][1]*linear_r + mat[1][1]*linear_g + mat[2][1]*linear_b;
1.817 + float out_linear_b = mat[0][2]*linear_r + mat[1][2]*linear_g + mat[2][2]*linear_b;
1.818 +
1.819 + out_linear_r = clamp_float(out_linear_r);
1.820 + out_linear_g = clamp_float(out_linear_g);
1.821 + out_linear_b = clamp_float(out_linear_b);
1.822 +
1.823 + out_device_r = lut_interp_linear(out_linear_r,
1.824 + transform->output_gamma_lut_r, transform->output_gamma_lut_r_length);
1.825 + out_device_g = lut_interp_linear(out_linear_g,
1.826 + transform->output_gamma_lut_g, transform->output_gamma_lut_g_length);
1.827 + out_device_b = lut_interp_linear(out_linear_b,
1.828 + transform->output_gamma_lut_b, transform->output_gamma_lut_b_length);
1.829 +
1.830 + dest[OUTPUT_R_INDEX] = clamp_u8(out_device_r*255);
1.831 + dest[OUTPUT_G_INDEX] = clamp_u8(out_device_g*255);
1.832 + dest[OUTPUT_B_INDEX] = clamp_u8(out_device_b*255);
1.833 + dest += RGB_OUTPUT_COMPONENTS;
1.834 + }
1.835 +}
1.836 +
1.837 +static void qcms_transform_data_rgba_out_lut(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length)
1.838 +{
1.839 + unsigned int i;
1.840 + float (*mat)[4] = transform->matrix;
1.841 + for (i = 0; i < length; i++) {
1.842 + unsigned char device_r = *src++;
1.843 + unsigned char device_g = *src++;
1.844 + unsigned char device_b = *src++;
1.845 + unsigned char alpha = *src++;
1.846 + float out_device_r, out_device_g, out_device_b;
1.847 +
1.848 + float linear_r = transform->input_gamma_table_r[device_r];
1.849 + float linear_g = transform->input_gamma_table_g[device_g];
1.850 + float linear_b = transform->input_gamma_table_b[device_b];
1.851 +
1.852 + float out_linear_r = mat[0][0]*linear_r + mat[1][0]*linear_g + mat[2][0]*linear_b;
1.853 + float out_linear_g = mat[0][1]*linear_r + mat[1][1]*linear_g + mat[2][1]*linear_b;
1.854 + float out_linear_b = mat[0][2]*linear_r + mat[1][2]*linear_g + mat[2][2]*linear_b;
1.855 +
1.856 + out_linear_r = clamp_float(out_linear_r);
1.857 + out_linear_g = clamp_float(out_linear_g);
1.858 + out_linear_b = clamp_float(out_linear_b);
1.859 +
1.860 + out_device_r = lut_interp_linear(out_linear_r,
1.861 + transform->output_gamma_lut_r, transform->output_gamma_lut_r_length);
1.862 + out_device_g = lut_interp_linear(out_linear_g,
1.863 + transform->output_gamma_lut_g, transform->output_gamma_lut_g_length);
1.864 + out_device_b = lut_interp_linear(out_linear_b,
1.865 + transform->output_gamma_lut_b, transform->output_gamma_lut_b_length);
1.866 +
1.867 + dest[OUTPUT_R_INDEX] = clamp_u8(out_device_r*255);
1.868 + dest[OUTPUT_G_INDEX] = clamp_u8(out_device_g*255);
1.869 + dest[OUTPUT_B_INDEX] = clamp_u8(out_device_b*255);
1.870 + dest[OUTPUT_A_INDEX] = alpha;
1.871 + dest += RGBA_OUTPUT_COMPONENTS;
1.872 + }
1.873 +}
1.874 +
1.875 +#if 0
1.876 +static void qcms_transform_data_rgb_out_linear(qcms_transform *transform, unsigned char *src, unsigned char *dest, size_t length)
1.877 +{
1.878 + int i;
1.879 + float (*mat)[4] = transform->matrix;
1.880 + for (i = 0; i < length; i++) {
1.881 + unsigned char device_r = *src++;
1.882 + unsigned char device_g = *src++;
1.883 + unsigned char device_b = *src++;
1.884 +
1.885 + float linear_r = transform->input_gamma_table_r[device_r];
1.886 + float linear_g = transform->input_gamma_table_g[device_g];
1.887 + float linear_b = transform->input_gamma_table_b[device_b];
1.888 +
1.889 + float out_linear_r = mat[0][0]*linear_r + mat[1][0]*linear_g + mat[2][0]*linear_b;
1.890 + float out_linear_g = mat[0][1]*linear_r + mat[1][1]*linear_g + mat[2][1]*linear_b;
1.891 + float out_linear_b = mat[0][2]*linear_r + mat[1][2]*linear_g + mat[2][2]*linear_b;
1.892 +
1.893 + *dest++ = clamp_u8(out_linear_r*255);
1.894 + *dest++ = clamp_u8(out_linear_g*255);
1.895 + *dest++ = clamp_u8(out_linear_b*255);
1.896 + }
1.897 +}
1.898 +#endif
1.899 +
1.900 +/*
1.901 + * If users create and destroy objects on different threads, even if the same
1.902 + * objects aren't used on different threads at the same time, we can still run
1.903 + * in to trouble with refcounts if they aren't atomic.
1.904 + *
1.905 + * This can lead to us prematurely deleting the precache if threads get unlucky
1.906 + * and write the wrong value to the ref count.
1.907 + */
1.908 +static struct precache_output *precache_reference(struct precache_output *p)
1.909 +{
1.910 + qcms_atomic_increment(p->ref_count);
1.911 + return p;
1.912 +}
1.913 +
1.914 +static struct precache_output *precache_create()
1.915 +{
1.916 + struct precache_output *p = malloc(sizeof(struct precache_output));
1.917 + if (p)
1.918 + p->ref_count = 1;
1.919 + return p;
1.920 +}
1.921 +
1.922 +void precache_release(struct precache_output *p)
1.923 +{
1.924 + if (qcms_atomic_decrement(p->ref_count) == 0) {
1.925 + free(p);
1.926 + }
1.927 +}
1.928 +
1.929 +#ifdef HAS_POSIX_MEMALIGN
1.930 +static qcms_transform *transform_alloc(void)
1.931 +{
1.932 + qcms_transform *t;
1.933 + if (!posix_memalign(&t, 16, sizeof(*t))) {
1.934 + return t;
1.935 + } else {
1.936 + return NULL;
1.937 + }
1.938 +}
1.939 +static void transform_free(qcms_transform *t)
1.940 +{
1.941 + free(t);
1.942 +}
1.943 +#else
1.944 +static qcms_transform *transform_alloc(void)
1.945 +{
1.946 + /* transform needs to be aligned on a 16byte boundrary */
1.947 + char *original_block = calloc(sizeof(qcms_transform) + sizeof(void*) + 16, 1);
1.948 + /* make room for a pointer to the block returned by calloc */
1.949 + void *transform_start = original_block + sizeof(void*);
1.950 + /* align transform_start */
1.951 + qcms_transform *transform_aligned = (qcms_transform*)(((uintptr_t)transform_start + 15) & ~0xf);
1.952 +
1.953 + /* store a pointer to the block returned by calloc so that we can free it later */
1.954 + void **(original_block_ptr) = (void**)transform_aligned;
1.955 + if (!original_block)
1.956 + return NULL;
1.957 + original_block_ptr--;
1.958 + *original_block_ptr = original_block;
1.959 +
1.960 + return transform_aligned;
1.961 +}
1.962 +static void transform_free(qcms_transform *t)
1.963 +{
1.964 + /* get at the pointer to the unaligned block returned by calloc */
1.965 + void **p = (void**)t;
1.966 + p--;
1.967 + free(*p);
1.968 +}
1.969 +#endif
1.970 +
1.971 +void qcms_transform_release(qcms_transform *t)
1.972 +{
1.973 + /* ensure we only free the gamma tables once even if there are
1.974 + * multiple references to the same data */
1.975 +
1.976 + if (t->output_table_r)
1.977 + precache_release(t->output_table_r);
1.978 + if (t->output_table_g)
1.979 + precache_release(t->output_table_g);
1.980 + if (t->output_table_b)
1.981 + precache_release(t->output_table_b);
1.982 +
1.983 + free(t->input_gamma_table_r);
1.984 + if (t->input_gamma_table_g != t->input_gamma_table_r)
1.985 + free(t->input_gamma_table_g);
1.986 + if (t->input_gamma_table_g != t->input_gamma_table_r &&
1.987 + t->input_gamma_table_g != t->input_gamma_table_b)
1.988 + free(t->input_gamma_table_b);
1.989 +
1.990 + free(t->input_gamma_table_gray);
1.991 +
1.992 + free(t->output_gamma_lut_r);
1.993 + free(t->output_gamma_lut_g);
1.994 + free(t->output_gamma_lut_b);
1.995 +
1.996 + transform_free(t);
1.997 +}
1.998 +
1.999 +#ifdef X86
1.1000 +// Determine if we can build with SSE2 (this was partly copied from jmorecfg.h in
1.1001 +// mozilla/jpeg)
1.1002 + // -------------------------------------------------------------------------
1.1003 +#if defined(_M_IX86) && defined(_MSC_VER)
1.1004 +#define HAS_CPUID
1.1005 +/* Get us a CPUID function. Avoid clobbering EBX because sometimes it's the PIC
1.1006 + register - I'm not sure if that ever happens on windows, but cpuid isn't
1.1007 + on the critical path so we just preserve the register to be safe and to be
1.1008 + consistent with the non-windows version. */
1.1009 +static void cpuid(uint32_t fxn, uint32_t *a, uint32_t *b, uint32_t *c, uint32_t *d) {
1.1010 + uint32_t a_, b_, c_, d_;
1.1011 + __asm {
1.1012 + xchg ebx, esi
1.1013 + mov eax, fxn
1.1014 + cpuid
1.1015 + mov a_, eax
1.1016 + mov b_, ebx
1.1017 + mov c_, ecx
1.1018 + mov d_, edx
1.1019 + xchg ebx, esi
1.1020 + }
1.1021 + *a = a_;
1.1022 + *b = b_;
1.1023 + *c = c_;
1.1024 + *d = d_;
1.1025 +}
1.1026 +#elif (defined(__GNUC__) || defined(__SUNPRO_C)) && (defined(__i386__) || defined(__i386))
1.1027 +#define HAS_CPUID
1.1028 +/* Get us a CPUID function. We can't use ebx because it's the PIC register on
1.1029 + some platforms, so we use ESI instead and save ebx to avoid clobbering it. */
1.1030 +static void cpuid(uint32_t fxn, uint32_t *a, uint32_t *b, uint32_t *c, uint32_t *d) {
1.1031 +
1.1032 + uint32_t a_, b_, c_, d_;
1.1033 + __asm__ __volatile__ ("xchgl %%ebx, %%esi; cpuid; xchgl %%ebx, %%esi;"
1.1034 + : "=a" (a_), "=S" (b_), "=c" (c_), "=d" (d_) : "a" (fxn));
1.1035 + *a = a_;
1.1036 + *b = b_;
1.1037 + *c = c_;
1.1038 + *d = d_;
1.1039 +}
1.1040 +#endif
1.1041 +
1.1042 +// -------------------------Runtime SSEx Detection-----------------------------
1.1043 +
1.1044 +/* MMX is always supported per
1.1045 + * Gecko v1.9.1 minimum CPU requirements */
1.1046 +#define SSE1_EDX_MASK (1UL << 25)
1.1047 +#define SSE2_EDX_MASK (1UL << 26)
1.1048 +#define SSE3_ECX_MASK (1UL << 0)
1.1049 +
1.1050 +static int sse_version_available(void)
1.1051 +{
1.1052 +#if defined(__x86_64__) || defined(__x86_64) || defined(_M_AMD64)
1.1053 + /* we know at build time that 64-bit CPUs always have SSE2
1.1054 + * this tells the compiler that non-SSE2 branches will never be
1.1055 + * taken (i.e. OK to optimze away the SSE1 and non-SIMD code */
1.1056 + return 2;
1.1057 +#elif defined(HAS_CPUID)
1.1058 + static int sse_version = -1;
1.1059 + uint32_t a, b, c, d;
1.1060 + uint32_t function = 0x00000001;
1.1061 +
1.1062 + if (sse_version == -1) {
1.1063 + sse_version = 0;
1.1064 + cpuid(function, &a, &b, &c, &d);
1.1065 + if (c & SSE3_ECX_MASK)
1.1066 + sse_version = 3;
1.1067 + else if (d & SSE2_EDX_MASK)
1.1068 + sse_version = 2;
1.1069 + else if (d & SSE1_EDX_MASK)
1.1070 + sse_version = 1;
1.1071 + }
1.1072 +
1.1073 + return sse_version;
1.1074 +#else
1.1075 + return 0;
1.1076 +#endif
1.1077 +}
1.1078 +#endif
1.1079 +
1.1080 +static const struct matrix bradford_matrix = {{ { 0.8951f, 0.2664f,-0.1614f},
1.1081 + {-0.7502f, 1.7135f, 0.0367f},
1.1082 + { 0.0389f,-0.0685f, 1.0296f}},
1.1083 + false};
1.1084 +
1.1085 +static const struct matrix bradford_matrix_inv = {{ { 0.9869929f,-0.1470543f, 0.1599627f},
1.1086 + { 0.4323053f, 0.5183603f, 0.0492912f},
1.1087 + {-0.0085287f, 0.0400428f, 0.9684867f}},
1.1088 + false};
1.1089 +
1.1090 +// See ICCv4 E.3
1.1091 +struct matrix compute_whitepoint_adaption(float X, float Y, float Z) {
1.1092 + float p = (0.96422f*bradford_matrix.m[0][0] + 1.000f*bradford_matrix.m[1][0] + 0.82521f*bradford_matrix.m[2][0]) /
1.1093 + (X*bradford_matrix.m[0][0] + Y*bradford_matrix.m[1][0] + Z*bradford_matrix.m[2][0] );
1.1094 + float y = (0.96422f*bradford_matrix.m[0][1] + 1.000f*bradford_matrix.m[1][1] + 0.82521f*bradford_matrix.m[2][1]) /
1.1095 + (X*bradford_matrix.m[0][1] + Y*bradford_matrix.m[1][1] + Z*bradford_matrix.m[2][1] );
1.1096 + float b = (0.96422f*bradford_matrix.m[0][2] + 1.000f*bradford_matrix.m[1][2] + 0.82521f*bradford_matrix.m[2][2]) /
1.1097 + (X*bradford_matrix.m[0][2] + Y*bradford_matrix.m[1][2] + Z*bradford_matrix.m[2][2] );
1.1098 + struct matrix white_adaption = {{ {p,0,0}, {0,y,0}, {0,0,b}}, false};
1.1099 + return matrix_multiply( bradford_matrix_inv, matrix_multiply(white_adaption, bradford_matrix) );
1.1100 +}
1.1101 +
1.1102 +void qcms_profile_precache_output_transform(qcms_profile *profile)
1.1103 +{
1.1104 + /* we only support precaching on rgb profiles */
1.1105 + if (profile->color_space != RGB_SIGNATURE)
1.1106 + return;
1.1107 +
1.1108 + if (qcms_supports_iccv4) {
1.1109 + /* don't precache since we will use the B2A LUT */
1.1110 + if (profile->B2A0)
1.1111 + return;
1.1112 +
1.1113 + /* don't precache since we will use the mBA LUT */
1.1114 + if (profile->mBA)
1.1115 + return;
1.1116 + }
1.1117 +
1.1118 + /* don't precache if we do not have the TRC curves */
1.1119 + if (!profile->redTRC || !profile->greenTRC || !profile->blueTRC)
1.1120 + return;
1.1121 +
1.1122 + if (!profile->output_table_r) {
1.1123 + profile->output_table_r = precache_create();
1.1124 + if (profile->output_table_r &&
1.1125 + !compute_precache(profile->redTRC, profile->output_table_r->data)) {
1.1126 + precache_release(profile->output_table_r);
1.1127 + profile->output_table_r = NULL;
1.1128 + }
1.1129 + }
1.1130 + if (!profile->output_table_g) {
1.1131 + profile->output_table_g = precache_create();
1.1132 + if (profile->output_table_g &&
1.1133 + !compute_precache(profile->greenTRC, profile->output_table_g->data)) {
1.1134 + precache_release(profile->output_table_g);
1.1135 + profile->output_table_g = NULL;
1.1136 + }
1.1137 + }
1.1138 + if (!profile->output_table_b) {
1.1139 + profile->output_table_b = precache_create();
1.1140 + if (profile->output_table_b &&
1.1141 + !compute_precache(profile->blueTRC, profile->output_table_b->data)) {
1.1142 + precache_release(profile->output_table_b);
1.1143 + profile->output_table_b = NULL;
1.1144 + }
1.1145 + }
1.1146 +}
1.1147 +
1.1148 +/* Replace the current transformation with a LUT transformation using a given number of sample points */
1.1149 +qcms_transform* qcms_transform_precacheLUT_float(qcms_transform *transform, qcms_profile *in, qcms_profile *out,
1.1150 + int samples, qcms_data_type in_type)
1.1151 +{
1.1152 + /* The range between which 2 consecutive sample points can be used to interpolate */
1.1153 + uint16_t x,y,z;
1.1154 + uint32_t l;
1.1155 + uint32_t lutSize = 3 * samples * samples * samples;
1.1156 + float* src = NULL;
1.1157 + float* dest = NULL;
1.1158 + float* lut = NULL;
1.1159 +
1.1160 + src = malloc(lutSize*sizeof(float));
1.1161 + dest = malloc(lutSize*sizeof(float));
1.1162 +
1.1163 + if (src && dest) {
1.1164 + /* Prepare a list of points we want to sample */
1.1165 + l = 0;
1.1166 + for (x = 0; x < samples; x++) {
1.1167 + for (y = 0; y < samples; y++) {
1.1168 + for (z = 0; z < samples; z++) {
1.1169 + src[l++] = x / (float)(samples-1);
1.1170 + src[l++] = y / (float)(samples-1);
1.1171 + src[l++] = z / (float)(samples-1);
1.1172 + }
1.1173 + }
1.1174 + }
1.1175 +
1.1176 + lut = qcms_chain_transform(in, out, src, dest, lutSize);
1.1177 + if (lut) {
1.1178 + transform->r_clut = &lut[0];
1.1179 + transform->g_clut = &lut[1];
1.1180 + transform->b_clut = &lut[2];
1.1181 + transform->grid_size = samples;
1.1182 + if (in_type == QCMS_DATA_RGBA_8) {
1.1183 + transform->transform_fn = qcms_transform_data_tetra_clut_rgba;
1.1184 + } else {
1.1185 + transform->transform_fn = qcms_transform_data_tetra_clut;
1.1186 + }
1.1187 + }
1.1188 + }
1.1189 +
1.1190 +
1.1191 + //XXX: qcms_modular_transform_data may return either the src or dest buffer. If so it must not be free-ed
1.1192 + if (src && lut != src) {
1.1193 + free(src);
1.1194 + }
1.1195 + if (dest && lut != dest) {
1.1196 + free(dest);
1.1197 + }
1.1198 +
1.1199 + if (lut == NULL) {
1.1200 + return NULL;
1.1201 + }
1.1202 + return transform;
1.1203 +}
1.1204 +
1.1205 +#define NO_MEM_TRANSFORM NULL
1.1206 +
1.1207 +qcms_transform* qcms_transform_create(
1.1208 + qcms_profile *in, qcms_data_type in_type,
1.1209 + qcms_profile *out, qcms_data_type out_type,
1.1210 + qcms_intent intent)
1.1211 +{
1.1212 + bool precache = false;
1.1213 +
1.1214 + qcms_transform *transform = transform_alloc();
1.1215 + if (!transform) {
1.1216 + return NULL;
1.1217 + }
1.1218 + if (out_type != QCMS_DATA_RGB_8 &&
1.1219 + out_type != QCMS_DATA_RGBA_8) {
1.1220 + assert(0 && "output type");
1.1221 + transform_free(transform);
1.1222 + return NULL;
1.1223 + }
1.1224 +
1.1225 + if (out->output_table_r &&
1.1226 + out->output_table_g &&
1.1227 + out->output_table_b) {
1.1228 + precache = true;
1.1229 + }
1.1230 +
1.1231 + // This precache assumes RGB_SIGNATURE (fails on GRAY_SIGNATURE, for instance)
1.1232 + if (qcms_supports_iccv4 &&
1.1233 + (in_type == QCMS_DATA_RGB_8 || in_type == QCMS_DATA_RGBA_8) &&
1.1234 + (in->A2B0 || out->B2A0 || in->mAB || out->mAB))
1.1235 + {
1.1236 + // Precache the transformation to a CLUT 33x33x33 in size.
1.1237 + // 33 is used by many profiles and works well in pratice.
1.1238 + // This evenly divides 256 into blocks of 8x8x8.
1.1239 + // TODO For transforming small data sets of about 200x200 or less
1.1240 + // precaching should be avoided.
1.1241 + qcms_transform *result = qcms_transform_precacheLUT_float(transform, in, out, 33, in_type);
1.1242 + if (!result) {
1.1243 + assert(0 && "precacheLUT failed");
1.1244 + transform_free(transform);
1.1245 + return NULL;
1.1246 + }
1.1247 + return result;
1.1248 + }
1.1249 +
1.1250 + if (precache) {
1.1251 + transform->output_table_r = precache_reference(out->output_table_r);
1.1252 + transform->output_table_g = precache_reference(out->output_table_g);
1.1253 + transform->output_table_b = precache_reference(out->output_table_b);
1.1254 + } else {
1.1255 + if (!out->redTRC || !out->greenTRC || !out->blueTRC) {
1.1256 + qcms_transform_release(transform);
1.1257 + return NO_MEM_TRANSFORM;
1.1258 + }
1.1259 + build_output_lut(out->redTRC, &transform->output_gamma_lut_r, &transform->output_gamma_lut_r_length);
1.1260 + build_output_lut(out->greenTRC, &transform->output_gamma_lut_g, &transform->output_gamma_lut_g_length);
1.1261 + build_output_lut(out->blueTRC, &transform->output_gamma_lut_b, &transform->output_gamma_lut_b_length);
1.1262 + if (!transform->output_gamma_lut_r || !transform->output_gamma_lut_g || !transform->output_gamma_lut_b) {
1.1263 + qcms_transform_release(transform);
1.1264 + return NO_MEM_TRANSFORM;
1.1265 + }
1.1266 + }
1.1267 +
1.1268 + if (in->color_space == RGB_SIGNATURE) {
1.1269 + struct matrix in_matrix, out_matrix, result;
1.1270 +
1.1271 + if (in_type != QCMS_DATA_RGB_8 &&
1.1272 + in_type != QCMS_DATA_RGBA_8){
1.1273 + assert(0 && "input type");
1.1274 + transform_free(transform);
1.1275 + return NULL;
1.1276 + }
1.1277 + if (precache) {
1.1278 +#ifdef X86
1.1279 + if (sse_version_available() >= 2) {
1.1280 + if (in_type == QCMS_DATA_RGB_8)
1.1281 + transform->transform_fn = qcms_transform_data_rgb_out_lut_sse2;
1.1282 + else
1.1283 + transform->transform_fn = qcms_transform_data_rgba_out_lut_sse2;
1.1284 +
1.1285 +#if !(defined(_MSC_VER) && defined(_M_AMD64))
1.1286 + /* Microsoft Compiler for x64 doesn't support MMX.
1.1287 + * SSE code uses MMX so that we disable on x64 */
1.1288 + } else
1.1289 + if (sse_version_available() >= 1) {
1.1290 + if (in_type == QCMS_DATA_RGB_8)
1.1291 + transform->transform_fn = qcms_transform_data_rgb_out_lut_sse1;
1.1292 + else
1.1293 + transform->transform_fn = qcms_transform_data_rgba_out_lut_sse1;
1.1294 +#endif
1.1295 + } else
1.1296 +#endif
1.1297 +#if (defined(__POWERPC__) || defined(__powerpc__))
1.1298 + if (have_altivec()) {
1.1299 + if (in_type == QCMS_DATA_RGB_8)
1.1300 + transform->transform_fn = qcms_transform_data_rgb_out_lut_altivec;
1.1301 + else
1.1302 + transform->transform_fn = qcms_transform_data_rgba_out_lut_altivec;
1.1303 + } else
1.1304 +#endif
1.1305 + {
1.1306 + if (in_type == QCMS_DATA_RGB_8)
1.1307 + transform->transform_fn = qcms_transform_data_rgb_out_lut_precache;
1.1308 + else
1.1309 + transform->transform_fn = qcms_transform_data_rgba_out_lut_precache;
1.1310 + }
1.1311 + } else {
1.1312 + if (in_type == QCMS_DATA_RGB_8)
1.1313 + transform->transform_fn = qcms_transform_data_rgb_out_lut;
1.1314 + else
1.1315 + transform->transform_fn = qcms_transform_data_rgba_out_lut;
1.1316 + }
1.1317 +
1.1318 + //XXX: avoid duplicating tables if we can
1.1319 + transform->input_gamma_table_r = build_input_gamma_table(in->redTRC);
1.1320 + transform->input_gamma_table_g = build_input_gamma_table(in->greenTRC);
1.1321 + transform->input_gamma_table_b = build_input_gamma_table(in->blueTRC);
1.1322 + if (!transform->input_gamma_table_r || !transform->input_gamma_table_g || !transform->input_gamma_table_b) {
1.1323 + qcms_transform_release(transform);
1.1324 + return NO_MEM_TRANSFORM;
1.1325 + }
1.1326 +
1.1327 +
1.1328 + /* build combined colorant matrix */
1.1329 + in_matrix = build_colorant_matrix(in);
1.1330 + out_matrix = build_colorant_matrix(out);
1.1331 + out_matrix = matrix_invert(out_matrix);
1.1332 + if (out_matrix.invalid) {
1.1333 + qcms_transform_release(transform);
1.1334 + return NULL;
1.1335 + }
1.1336 + result = matrix_multiply(out_matrix, in_matrix);
1.1337 +
1.1338 + /* store the results in column major mode
1.1339 + * this makes doing the multiplication with sse easier */
1.1340 + transform->matrix[0][0] = result.m[0][0];
1.1341 + transform->matrix[1][0] = result.m[0][1];
1.1342 + transform->matrix[2][0] = result.m[0][2];
1.1343 + transform->matrix[0][1] = result.m[1][0];
1.1344 + transform->matrix[1][1] = result.m[1][1];
1.1345 + transform->matrix[2][1] = result.m[1][2];
1.1346 + transform->matrix[0][2] = result.m[2][0];
1.1347 + transform->matrix[1][2] = result.m[2][1];
1.1348 + transform->matrix[2][2] = result.m[2][2];
1.1349 +
1.1350 + } else if (in->color_space == GRAY_SIGNATURE) {
1.1351 + if (in_type != QCMS_DATA_GRAY_8 &&
1.1352 + in_type != QCMS_DATA_GRAYA_8){
1.1353 + assert(0 && "input type");
1.1354 + transform_free(transform);
1.1355 + return NULL;
1.1356 + }
1.1357 +
1.1358 + transform->input_gamma_table_gray = build_input_gamma_table(in->grayTRC);
1.1359 + if (!transform->input_gamma_table_gray) {
1.1360 + qcms_transform_release(transform);
1.1361 + return NO_MEM_TRANSFORM;
1.1362 + }
1.1363 +
1.1364 + if (precache) {
1.1365 + if (in_type == QCMS_DATA_GRAY_8) {
1.1366 + transform->transform_fn = qcms_transform_data_gray_out_precache;
1.1367 + } else {
1.1368 + transform->transform_fn = qcms_transform_data_graya_out_precache;
1.1369 + }
1.1370 + } else {
1.1371 + if (in_type == QCMS_DATA_GRAY_8) {
1.1372 + transform->transform_fn = qcms_transform_data_gray_out_lut;
1.1373 + } else {
1.1374 + transform->transform_fn = qcms_transform_data_graya_out_lut;
1.1375 + }
1.1376 + }
1.1377 + } else {
1.1378 + assert(0 && "unexpected colorspace");
1.1379 + transform_free(transform);
1.1380 + return NULL;
1.1381 + }
1.1382 + return transform;
1.1383 +}
1.1384 +
1.1385 +#if defined(__GNUC__) && !defined(__x86_64__) && !defined(__amd64__)
1.1386 +/* we need this to avoid crashes when gcc assumes the stack is 128bit aligned */
1.1387 +__attribute__((__force_align_arg_pointer__))
1.1388 +#endif
1.1389 +void qcms_transform_data(qcms_transform *transform, void *src, void *dest, size_t length)
1.1390 +{
1.1391 + transform->transform_fn(transform, src, dest, length);
1.1392 +}
1.1393 +
1.1394 +qcms_bool qcms_supports_iccv4;
1.1395 +void qcms_enable_iccv4()
1.1396 +{
1.1397 + qcms_supports_iccv4 = true;
1.1398 +}
