1 /*
2 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
3 *
4 * This code is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License version 2 only, as
6 * published by the Free Software Foundation. Oracle designates this
7 * particular file as subject to the "Classpath" exception as provided
8 * by Oracle in the LICENSE file that accompanied this code.
9 *
10 * This code is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
13 * version 2 for more details (a copy is included in the LICENSE file that
14 * accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License version
17 * 2 along with this work; if not, write to the Free Software Foundation,
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
19 *
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
21 * or visit www.oracle.com if you need additional information or have any
22 * questions.
23 */
24
25 // This file is available under and governed by the GNU General Public
26 // License version 2 only, as published by the Free Software Foundation.
27 // However, the following notice accompanied the original version of this
28 // file:
29 //
30 //---------------------------------------------------------------------------------
31 //
32 // Little Color Management System
33 // Copyright (c) 1998-2020 Marti Maria Saguer
34 //
35 // Permission is hereby granted, free of charge, to any person obtaining
36 // a copy of this software and associated documentation files (the "Software"),
37 // to deal in the Software without restriction, including without limitation
38 // the rights to use, copy, modify, merge, publish, distribute, sublicense,
39 // and/or sell copies of the Software, and to permit persons to whom the Software
40 // is furnished to do so, subject to the following conditions:
41 //
42 // The above copyright notice and this permission notice shall be included in
43 // all copies or substantial portions of the Software.
44 //
45 // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
46 // EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
47 // THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
48 // NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
49 // LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
50 // OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
51 // WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
52 //
53 //---------------------------------------------------------------------------------
54 //
55
56 #include "lcms2_internal.h"
57
58
59 //----------------------------------------------------------------------------------
60
61 // Optimization for 8 bits, Shaper-CLUT (3 inputs only)
62 typedef struct {
63
64 cmsContext ContextID;
65
66 const cmsInterpParams* p; // Tetrahedrical interpolation parameters. This is a not-owned pointer.
67
68 cmsUInt16Number rx[256], ry[256], rz[256];
69 cmsUInt32Number X0[256], Y0[256], Z0[256]; // Precomputed nodes and offsets for 8-bit input data
70
71
72 } Prelin8Data;
73
74
75 // Generic optimization for 16 bits Shaper-CLUT-Shaper (any inputs)
76 typedef struct {
77
78 cmsContext ContextID;
79
80 // Number of channels
81 cmsUInt32Number nInputs;
82 cmsUInt32Number nOutputs;
83
84 _cmsInterpFn16 EvalCurveIn16[MAX_INPUT_DIMENSIONS]; // The maximum number of input channels is known in advance
85 cmsInterpParams* ParamsCurveIn16[MAX_INPUT_DIMENSIONS];
86
87 _cmsInterpFn16 EvalCLUT; // The evaluator for 3D grid
88 const cmsInterpParams* CLUTparams; // (not-owned pointer)
89
90
91 _cmsInterpFn16* EvalCurveOut16; // Points to an array of curve evaluators in 16 bits (not-owned pointer)
92 cmsInterpParams** ParamsCurveOut16; // Points to an array of references to interpolation params (not-owned pointer)
93
94
95 } Prelin16Data;
96
97
98 // Optimization for matrix-shaper in 8 bits. Numbers are operated in n.14 signed, tables are stored in 1.14 fixed
99
100 typedef cmsInt32Number cmsS1Fixed14Number; // Note that this may hold more than 16 bits!
101
102 #define DOUBLE_TO_1FIXED14(x) ((cmsS1Fixed14Number) floor((x) * 16384.0 + 0.5))
103
104 typedef struct {
105
106 cmsContext ContextID;
107
108 cmsS1Fixed14Number Shaper1R[256]; // from 0..255 to 1.14 (0.0...1.0)
109 cmsS1Fixed14Number Shaper1G[256];
110 cmsS1Fixed14Number Shaper1B[256];
111
112 cmsS1Fixed14Number Mat[3][3]; // n.14 to n.14 (needs a saturation after that)
113 cmsS1Fixed14Number Off[3];
114
115 cmsUInt16Number Shaper2R[16385]; // 1.14 to 0..255
116 cmsUInt16Number Shaper2G[16385];
117 cmsUInt16Number Shaper2B[16385];
118
119 } MatShaper8Data;
120
121 // Curves, optimization is shared between 8 and 16 bits
122 typedef struct {
123
124 cmsContext ContextID;
125
126 cmsUInt32Number nCurves; // Number of curves
127 cmsUInt32Number nElements; // Elements in curves
128 cmsUInt16Number** Curves; // Points to a dynamically allocated array
129
130 } Curves16Data;
131
132
133 // Simple optimizations ----------------------------------------------------------------------------------------------------------
134
135
136 // Remove an element in linked chain
137 static
_RemoveElement(cmsStage ** head)138 void _RemoveElement(cmsStage** head)
139 {
140 cmsStage* mpe = *head;
141 cmsStage* next = mpe ->Next;
142 *head = next;
143 cmsStageFree(mpe);
144 }
145
146 // Remove all identities in chain. Note that pt actually is a double pointer to the element that holds the pointer.
147 static
_Remove1Op(cmsPipeline * Lut,cmsStageSignature UnaryOp)148 cmsBool _Remove1Op(cmsPipeline* Lut, cmsStageSignature UnaryOp)
149 {
150 cmsStage** pt = &Lut ->Elements;
151 cmsBool AnyOpt = FALSE;
152
153 while (*pt != NULL) {
154
155 if ((*pt) ->Implements == UnaryOp) {
156 _RemoveElement(pt);
157 AnyOpt = TRUE;
158 }
159 else
160 pt = &((*pt) -> Next);
161 }
162
163 return AnyOpt;
164 }
165
166 // Same, but only if two adjacent elements are found
167 static
_Remove2Op(cmsPipeline * Lut,cmsStageSignature Op1,cmsStageSignature Op2)168 cmsBool _Remove2Op(cmsPipeline* Lut, cmsStageSignature Op1, cmsStageSignature Op2)
169 {
170 cmsStage** pt1;
171 cmsStage** pt2;
172 cmsBool AnyOpt = FALSE;
173
174 pt1 = &Lut ->Elements;
175 if (*pt1 == NULL) return AnyOpt;
176
177 while (*pt1 != NULL) {
178
179 pt2 = &((*pt1) -> Next);
180 if (*pt2 == NULL) return AnyOpt;
181
182 if ((*pt1) ->Implements == Op1 && (*pt2) ->Implements == Op2) {
183 _RemoveElement(pt2);
184 _RemoveElement(pt1);
185 AnyOpt = TRUE;
186 }
187 else
188 pt1 = &((*pt1) -> Next);
189 }
190
191 return AnyOpt;
192 }
193
194
195 static
CloseEnoughFloat(cmsFloat64Number a,cmsFloat64Number b)196 cmsBool CloseEnoughFloat(cmsFloat64Number a, cmsFloat64Number b)
197 {
198 return fabs(b - a) < 0.00001f;
199 }
200
201 static
isFloatMatrixIdentity(const cmsMAT3 * a)202 cmsBool isFloatMatrixIdentity(const cmsMAT3* a)
203 {
204 cmsMAT3 Identity;
205 int i, j;
206
207 _cmsMAT3identity(&Identity);
208
209 for (i = 0; i < 3; i++)
210 for (j = 0; j < 3; j++)
211 if (!CloseEnoughFloat(a->v[i].n[j], Identity.v[i].n[j])) return FALSE;
212
213 return TRUE;
214 }
215 // if two adjacent matrices are found, multiply them.
216 static
_MultiplyMatrix(cmsPipeline * Lut)217 cmsBool _MultiplyMatrix(cmsPipeline* Lut)
218 {
219 cmsStage** pt1;
220 cmsStage** pt2;
221 cmsStage* chain;
222 cmsBool AnyOpt = FALSE;
223
224 pt1 = &Lut->Elements;
225 if (*pt1 == NULL) return AnyOpt;
226
227 while (*pt1 != NULL) {
228
229 pt2 = &((*pt1)->Next);
230 if (*pt2 == NULL) return AnyOpt;
231
232 if ((*pt1)->Implements == cmsSigMatrixElemType && (*pt2)->Implements == cmsSigMatrixElemType) {
233
234 // Get both matrices
235 _cmsStageMatrixData* m1 = (_cmsStageMatrixData*) cmsStageData(*pt1);
236 _cmsStageMatrixData* m2 = (_cmsStageMatrixData*) cmsStageData(*pt2);
237 cmsMAT3 res;
238
239 // Input offset and output offset should be zero to use this optimization
240 if (m1->Offset != NULL || m2 ->Offset != NULL ||
241 cmsStageInputChannels(*pt1) != 3 || cmsStageOutputChannels(*pt1) != 3 ||
242 cmsStageInputChannels(*pt2) != 3 || cmsStageOutputChannels(*pt2) != 3)
243 return FALSE;
244
245 // Multiply both matrices to get the result
246 _cmsMAT3per(&res, (cmsMAT3*)m2->Double, (cmsMAT3*)m1->Double);
247
248 // Get the next in chain after the matrices
249 chain = (*pt2)->Next;
250
251 // Remove both matrices
252 _RemoveElement(pt2);
253 _RemoveElement(pt1);
254
255 // Now what if the result is a plain identity?
256 if (!isFloatMatrixIdentity(&res)) {
257
258 // We can not get rid of full matrix
259 cmsStage* Multmat = cmsStageAllocMatrix(Lut->ContextID, 3, 3, (const cmsFloat64Number*) &res, NULL);
260 if (Multmat == NULL) return FALSE; // Should never happen
261
262 // Recover the chain
263 Multmat->Next = chain;
264 *pt1 = Multmat;
265 }
266
267 AnyOpt = TRUE;
268 }
269 else
270 pt1 = &((*pt1)->Next);
271 }
272
273 return AnyOpt;
274 }
275
276
277 // Preoptimize just gets rif of no-ops coming paired. Conversion from v2 to v4 followed
278 // by a v4 to v2 and vice-versa. The elements are then discarded.
279 static
PreOptimize(cmsPipeline * Lut)280 cmsBool PreOptimize(cmsPipeline* Lut)
281 {
282 cmsBool AnyOpt = FALSE, Opt;
283
284 do {
285
286 Opt = FALSE;
287
288 // Remove all identities
289 Opt |= _Remove1Op(Lut, cmsSigIdentityElemType);
290
291 // Remove XYZ2Lab followed by Lab2XYZ
292 Opt |= _Remove2Op(Lut, cmsSigXYZ2LabElemType, cmsSigLab2XYZElemType);
293
294 // Remove Lab2XYZ followed by XYZ2Lab
295 Opt |= _Remove2Op(Lut, cmsSigLab2XYZElemType, cmsSigXYZ2LabElemType);
296
297 // Remove V4 to V2 followed by V2 to V4
298 Opt |= _Remove2Op(Lut, cmsSigLabV4toV2, cmsSigLabV2toV4);
299
300 // Remove V2 to V4 followed by V4 to V2
301 Opt |= _Remove2Op(Lut, cmsSigLabV2toV4, cmsSigLabV4toV2);
302
303 // Remove float pcs Lab conversions
304 Opt |= _Remove2Op(Lut, cmsSigLab2FloatPCS, cmsSigFloatPCS2Lab);
305
306 // Remove float pcs Lab conversions
307 Opt |= _Remove2Op(Lut, cmsSigXYZ2FloatPCS, cmsSigFloatPCS2XYZ);
308
309 // Simplify matrix.
310 Opt |= _MultiplyMatrix(Lut);
311
312 if (Opt) AnyOpt = TRUE;
313
314 } while (Opt);
315
316 return AnyOpt;
317 }
318
319 static
Eval16nop1D(CMSREGISTER const cmsUInt16Number Input[],CMSREGISTER cmsUInt16Number Output[],CMSREGISTER const struct _cms_interp_struc * p)320 void Eval16nop1D(CMSREGISTER const cmsUInt16Number Input[],
321 CMSREGISTER cmsUInt16Number Output[],
322 CMSREGISTER const struct _cms_interp_struc* p)
323 {
324 Output[0] = Input[0];
325
326 cmsUNUSED_PARAMETER(p);
327 }
328
329 static
PrelinEval16(CMSREGISTER const cmsUInt16Number Input[],CMSREGISTER cmsUInt16Number Output[],CMSREGISTER const void * D)330 void PrelinEval16(CMSREGISTER const cmsUInt16Number Input[],
331 CMSREGISTER cmsUInt16Number Output[],
332 CMSREGISTER const void* D)
333 {
334 Prelin16Data* p16 = (Prelin16Data*) D;
335 cmsUInt16Number StageABC[MAX_INPUT_DIMENSIONS];
336 cmsUInt16Number StageDEF[cmsMAXCHANNELS];
337 cmsUInt32Number i;
338
339 for (i=0; i < p16 ->nInputs; i++) {
340
341 p16 ->EvalCurveIn16[i](&Input[i], &StageABC[i], p16 ->ParamsCurveIn16[i]);
342 }
343
344 p16 ->EvalCLUT(StageABC, StageDEF, p16 ->CLUTparams);
345
346 for (i=0; i < p16 ->nOutputs; i++) {
347
348 p16 ->EvalCurveOut16[i](&StageDEF[i], &Output[i], p16 ->ParamsCurveOut16[i]);
349 }
350 }
351
352
353 static
PrelinOpt16free(cmsContext ContextID,void * ptr)354 void PrelinOpt16free(cmsContext ContextID, void* ptr)
355 {
356 Prelin16Data* p16 = (Prelin16Data*) ptr;
357
358 _cmsFree(ContextID, p16 ->EvalCurveOut16);
359 _cmsFree(ContextID, p16 ->ParamsCurveOut16);
360
361 _cmsFree(ContextID, p16);
362 }
363
364 static
Prelin16dup(cmsContext ContextID,const void * ptr)365 void* Prelin16dup(cmsContext ContextID, const void* ptr)
366 {
367 Prelin16Data* p16 = (Prelin16Data*) ptr;
368 Prelin16Data* Duped = (Prelin16Data*) _cmsDupMem(ContextID, p16, sizeof(Prelin16Data));
369
370 if (Duped == NULL) return NULL;
371
372 Duped->EvalCurveOut16 = (_cmsInterpFn16*) _cmsDupMem(ContextID, p16->EvalCurveOut16, p16->nOutputs * sizeof(_cmsInterpFn16));
373 Duped->ParamsCurveOut16 = (cmsInterpParams**)_cmsDupMem(ContextID, p16->ParamsCurveOut16, p16->nOutputs * sizeof(cmsInterpParams*));
374
375 return Duped;
376 }
377
378
379 static
PrelinOpt16alloc(cmsContext ContextID,const cmsInterpParams * ColorMap,cmsUInt32Number nInputs,cmsToneCurve ** In,cmsUInt32Number nOutputs,cmsToneCurve ** Out)380 Prelin16Data* PrelinOpt16alloc(cmsContext ContextID,
381 const cmsInterpParams* ColorMap,
382 cmsUInt32Number nInputs, cmsToneCurve** In,
383 cmsUInt32Number nOutputs, cmsToneCurve** Out )
384 {
385 cmsUInt32Number i;
386 Prelin16Data* p16 = (Prelin16Data*)_cmsMallocZero(ContextID, sizeof(Prelin16Data));
387 if (p16 == NULL) return NULL;
388
389 p16 ->nInputs = nInputs;
390 p16 ->nOutputs = nOutputs;
391
392
393 for (i=0; i < nInputs; i++) {
394
395 if (In == NULL) {
396 p16 -> ParamsCurveIn16[i] = NULL;
397 p16 -> EvalCurveIn16[i] = Eval16nop1D;
398
399 }
400 else {
401 p16 -> ParamsCurveIn16[i] = In[i] ->InterpParams;
402 p16 -> EvalCurveIn16[i] = p16 ->ParamsCurveIn16[i]->Interpolation.Lerp16;
403 }
404 }
405
406 p16 ->CLUTparams = ColorMap;
407 p16 ->EvalCLUT = ColorMap ->Interpolation.Lerp16;
408
409
410 p16 -> EvalCurveOut16 = (_cmsInterpFn16*) _cmsCalloc(ContextID, nOutputs, sizeof(_cmsInterpFn16));
411 if (p16->EvalCurveOut16 == NULL)
412 {
413 _cmsFree(ContextID, p16);
414 return NULL;
415 }
416
417 p16 -> ParamsCurveOut16 = (cmsInterpParams**) _cmsCalloc(ContextID, nOutputs, sizeof(cmsInterpParams* ));
418 if (p16->ParamsCurveOut16 == NULL)
419 {
420
421 _cmsFree(ContextID, p16->EvalCurveOut16);
422 _cmsFree(ContextID, p16);
423 return NULL;
424 }
425
426 for (i=0; i < nOutputs; i++) {
427
428 if (Out == NULL) {
429 p16 ->ParamsCurveOut16[i] = NULL;
430 p16 -> EvalCurveOut16[i] = Eval16nop1D;
431 }
432 else {
433
434 p16 ->ParamsCurveOut16[i] = Out[i] ->InterpParams;
435 p16 -> EvalCurveOut16[i] = p16 ->ParamsCurveOut16[i]->Interpolation.Lerp16;
436 }
437 }
438
439 return p16;
440 }
441
442
443
444 // Resampling ---------------------------------------------------------------------------------
445
446 #define PRELINEARIZATION_POINTS 4096
447
448 // Sampler implemented by another LUT. This is a clean way to precalculate the devicelink 3D CLUT for
449 // almost any transform. We use floating point precision and then convert from floating point to 16 bits.
450 static
XFormSampler16(CMSREGISTER const cmsUInt16Number In[],CMSREGISTER cmsUInt16Number Out[],CMSREGISTER void * Cargo)451 cmsInt32Number XFormSampler16(CMSREGISTER const cmsUInt16Number In[],
452 CMSREGISTER cmsUInt16Number Out[],
453 CMSREGISTER void* Cargo)
454 {
455 cmsPipeline* Lut = (cmsPipeline*) Cargo;
456 cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
457 cmsUInt32Number i;
458
459 _cmsAssert(Lut -> InputChannels < cmsMAXCHANNELS);
460 _cmsAssert(Lut -> OutputChannels < cmsMAXCHANNELS);
461
462 // From 16 bit to floating point
463 for (i=0; i < Lut ->InputChannels; i++)
464 InFloat[i] = (cmsFloat32Number) (In[i] / 65535.0);
465
466 // Evaluate in floating point
467 cmsPipelineEvalFloat(InFloat, OutFloat, Lut);
468
469 // Back to 16 bits representation
470 for (i=0; i < Lut ->OutputChannels; i++)
471 Out[i] = _cmsQuickSaturateWord(OutFloat[i] * 65535.0);
472
473 // Always succeed
474 return TRUE;
475 }
476
477 // Try to see if the curves of a given MPE are linear
478 static
AllCurvesAreLinear(cmsStage * mpe)479 cmsBool AllCurvesAreLinear(cmsStage* mpe)
480 {
481 cmsToneCurve** Curves;
482 cmsUInt32Number i, n;
483
484 Curves = _cmsStageGetPtrToCurveSet(mpe);
485 if (Curves == NULL) return FALSE;
486
487 n = cmsStageOutputChannels(mpe);
488
489 for (i=0; i < n; i++) {
490 if (!cmsIsToneCurveLinear(Curves[i])) return FALSE;
491 }
492
493 return TRUE;
494 }
495
496 // This function replaces a specific node placed in "At" by the "Value" numbers. Its purpose
497 // is to fix scum dot on broken profiles/transforms. Works on 1, 3 and 4 channels
498 static
PatchLUT(cmsStage * CLUT,cmsUInt16Number At[],cmsUInt16Number Value[],cmsUInt32Number nChannelsOut,cmsUInt32Number nChannelsIn)499 cmsBool PatchLUT(cmsStage* CLUT, cmsUInt16Number At[], cmsUInt16Number Value[],
500 cmsUInt32Number nChannelsOut, cmsUInt32Number nChannelsIn)
501 {
502 _cmsStageCLutData* Grid = (_cmsStageCLutData*) CLUT ->Data;
503 cmsInterpParams* p16 = Grid ->Params;
504 cmsFloat64Number px, py, pz, pw;
505 int x0, y0, z0, w0;
506 int i, index;
507
508 if (CLUT -> Type != cmsSigCLutElemType) {
509 cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) Attempt to PatchLUT on non-lut stage");
510 return FALSE;
511 }
512
513 if (nChannelsIn == 4) {
514
515 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
516 py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
517 pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
518 pw = ((cmsFloat64Number) At[3] * (p16->Domain[3])) / 65535.0;
519
520 x0 = (int) floor(px);
521 y0 = (int) floor(py);
522 z0 = (int) floor(pz);
523 w0 = (int) floor(pw);
524
525 if (((px - x0) != 0) ||
526 ((py - y0) != 0) ||
527 ((pz - z0) != 0) ||
528 ((pw - w0) != 0)) return FALSE; // Not on exact node
529
530 index = (int) p16 -> opta[3] * x0 +
531 (int) p16 -> opta[2] * y0 +
532 (int) p16 -> opta[1] * z0 +
533 (int) p16 -> opta[0] * w0;
534 }
535 else
536 if (nChannelsIn == 3) {
537
538 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
539 py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
540 pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
541
542 x0 = (int) floor(px);
543 y0 = (int) floor(py);
544 z0 = (int) floor(pz);
545
546 if (((px - x0) != 0) ||
547 ((py - y0) != 0) ||
548 ((pz - z0) != 0)) return FALSE; // Not on exact node
549
550 index = (int) p16 -> opta[2] * x0 +
551 (int) p16 -> opta[1] * y0 +
552 (int) p16 -> opta[0] * z0;
553 }
554 else
555 if (nChannelsIn == 1) {
556
557 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
558
559 x0 = (int) floor(px);
560
561 if (((px - x0) != 0)) return FALSE; // Not on exact node
562
563 index = (int) p16 -> opta[0] * x0;
564 }
565 else {
566 cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) %d Channels are not supported on PatchLUT", nChannelsIn);
567 return FALSE;
568 }
569
570 for (i = 0; i < (int) nChannelsOut; i++)
571 Grid->Tab.T[index + i] = Value[i];
572
573 return TRUE;
574 }
575
576 // Auxiliary, to see if two values are equal or very different
577 static
WhitesAreEqual(cmsUInt32Number n,cmsUInt16Number White1[],cmsUInt16Number White2[])578 cmsBool WhitesAreEqual(cmsUInt32Number n, cmsUInt16Number White1[], cmsUInt16Number White2[] )
579 {
580 cmsUInt32Number i;
581
582 for (i=0; i < n; i++) {
583
584 if (abs(White1[i] - White2[i]) > 0xf000) return TRUE; // Values are so extremely different that the fixup should be avoided
585 if (White1[i] != White2[i]) return FALSE;
586 }
587 return TRUE;
588 }
589
590
591 // Locate the node for the white point and fix it to pure white in order to avoid scum dot.
592 static
FixWhiteMisalignment(cmsPipeline * Lut,cmsColorSpaceSignature EntryColorSpace,cmsColorSpaceSignature ExitColorSpace)593 cmsBool FixWhiteMisalignment(cmsPipeline* Lut, cmsColorSpaceSignature EntryColorSpace, cmsColorSpaceSignature ExitColorSpace)
594 {
595 cmsUInt16Number *WhitePointIn, *WhitePointOut;
596 cmsUInt16Number WhiteIn[cmsMAXCHANNELS], WhiteOut[cmsMAXCHANNELS], ObtainedOut[cmsMAXCHANNELS];
597 cmsUInt32Number i, nOuts, nIns;
598 cmsStage *PreLin = NULL, *CLUT = NULL, *PostLin = NULL;
599
600 if (!_cmsEndPointsBySpace(EntryColorSpace,
601 &WhitePointIn, NULL, &nIns)) return FALSE;
602
603 if (!_cmsEndPointsBySpace(ExitColorSpace,
604 &WhitePointOut, NULL, &nOuts)) return FALSE;
605
606 // It needs to be fixed?
607 if (Lut ->InputChannels != nIns) return FALSE;
608 if (Lut ->OutputChannels != nOuts) return FALSE;
609
610 cmsPipelineEval16(WhitePointIn, ObtainedOut, Lut);
611
612 if (WhitesAreEqual(nOuts, WhitePointOut, ObtainedOut)) return TRUE; // whites already match
613
614 // Check if the LUT comes as Prelin, CLUT or Postlin. We allow all combinations
615 if (!cmsPipelineCheckAndRetreiveStages(Lut, 3, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, &PreLin, &CLUT, &PostLin))
616 if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCurveSetElemType, cmsSigCLutElemType, &PreLin, &CLUT))
617 if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCLutElemType, cmsSigCurveSetElemType, &CLUT, &PostLin))
618 if (!cmsPipelineCheckAndRetreiveStages(Lut, 1, cmsSigCLutElemType, &CLUT))
619 return FALSE;
620
621 // We need to interpolate white points of both, pre and post curves
622 if (PreLin) {
623
624 cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PreLin);
625
626 for (i=0; i < nIns; i++) {
627 WhiteIn[i] = cmsEvalToneCurve16(Curves[i], WhitePointIn[i]);
628 }
629 }
630 else {
631 for (i=0; i < nIns; i++)
632 WhiteIn[i] = WhitePointIn[i];
633 }
634
635 // If any post-linearization, we need to find how is represented white before the curve, do
636 // a reverse interpolation in this case.
637 if (PostLin) {
638
639 cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PostLin);
640
641 for (i=0; i < nOuts; i++) {
642
643 cmsToneCurve* InversePostLin = cmsReverseToneCurve(Curves[i]);
644 if (InversePostLin == NULL) {
645 WhiteOut[i] = WhitePointOut[i];
646
647 } else {
648
649 WhiteOut[i] = cmsEvalToneCurve16(InversePostLin, WhitePointOut[i]);
650 cmsFreeToneCurve(InversePostLin);
651 }
652 }
653 }
654 else {
655 for (i=0; i < nOuts; i++)
656 WhiteOut[i] = WhitePointOut[i];
657 }
658
659 // Ok, proceed with patching. May fail and we don't care if it fails
660 PatchLUT(CLUT, WhiteIn, WhiteOut, nOuts, nIns);
661
662 return TRUE;
663 }
664
665 // -----------------------------------------------------------------------------------------------------------------------------------------------
666 // This function creates simple LUT from complex ones. The generated LUT has an optional set of
667 // prelinearization curves, a CLUT of nGridPoints and optional postlinearization tables.
668 // These curves have to exist in the original LUT in order to be used in the simplified output.
669 // Caller may also use the flags to allow this feature.
670 // LUTS with all curves will be simplified to a single curve. Parametric curves are lost.
671 // This function should be used on 16-bits LUTS only, as floating point losses precision when simplified
672 // -----------------------------------------------------------------------------------------------------------------------------------------------
673
674 static
OptimizeByResampling(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)675 cmsBool OptimizeByResampling(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
676 {
677 cmsPipeline* Src = NULL;
678 cmsPipeline* Dest = NULL;
679 cmsStage* mpe;
680 cmsStage* CLUT;
681 cmsStage *KeepPreLin = NULL, *KeepPostLin = NULL;
682 cmsUInt32Number nGridPoints;
683 cmsColorSpaceSignature ColorSpace, OutputColorSpace;
684 cmsStage *NewPreLin = NULL;
685 cmsStage *NewPostLin = NULL;
686 _cmsStageCLutData* DataCLUT;
687 cmsToneCurve** DataSetIn;
688 cmsToneCurve** DataSetOut;
689 Prelin16Data* p16;
690
691 // This is a lossy optimization! does not apply in floating-point cases
692 if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
693
694 ColorSpace = _cmsICCcolorSpace((int) T_COLORSPACE(*InputFormat));
695 OutputColorSpace = _cmsICCcolorSpace((int) T_COLORSPACE(*OutputFormat));
696
697 // Color space must be specified
698 if (ColorSpace == (cmsColorSpaceSignature)0 ||
699 OutputColorSpace == (cmsColorSpaceSignature)0) return FALSE;
700
701 nGridPoints = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);
702
703 // For empty LUTs, 2 points are enough
704 if (cmsPipelineStageCount(*Lut) == 0)
705 nGridPoints = 2;
706
707 Src = *Lut;
708
709 // Named color pipelines cannot be optimized either
710 for (mpe = cmsPipelineGetPtrToFirstStage(Src);
711 mpe != NULL;
712 mpe = cmsStageNext(mpe)) {
713 if (cmsStageType(mpe) == cmsSigNamedColorElemType) return FALSE;
714 }
715
716 // Allocate an empty LUT
717 Dest = cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
718 if (!Dest) return FALSE;
719
720 // Prelinearization tables are kept unless indicated by flags
721 if (*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION) {
722
723 // Get a pointer to the prelinearization element
724 cmsStage* PreLin = cmsPipelineGetPtrToFirstStage(Src);
725
726 // Check if suitable
727 if (PreLin && PreLin ->Type == cmsSigCurveSetElemType) {
728
729 // Maybe this is a linear tram, so we can avoid the whole stuff
730 if (!AllCurvesAreLinear(PreLin)) {
731
732 // All seems ok, proceed.
733 NewPreLin = cmsStageDup(PreLin);
734 if(!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, NewPreLin))
735 goto Error;
736
737 // Remove prelinearization. Since we have duplicated the curve
738 // in destination LUT, the sampling should be applied after this stage.
739 cmsPipelineUnlinkStage(Src, cmsAT_BEGIN, &KeepPreLin);
740 }
741 }
742 }
743
744 // Allocate the CLUT
745 CLUT = cmsStageAllocCLut16bit(Src ->ContextID, nGridPoints, Src ->InputChannels, Src->OutputChannels, NULL);
746 if (CLUT == NULL) goto Error;
747
748 // Add the CLUT to the destination LUT
749 if (!cmsPipelineInsertStage(Dest, cmsAT_END, CLUT)) {
750 goto Error;
751 }
752
753 // Postlinearization tables are kept unless indicated by flags
754 if (*dwFlags & cmsFLAGS_CLUT_POST_LINEARIZATION) {
755
756 // Get a pointer to the postlinearization if present
757 cmsStage* PostLin = cmsPipelineGetPtrToLastStage(Src);
758
759 // Check if suitable
760 if (PostLin && cmsStageType(PostLin) == cmsSigCurveSetElemType) {
761
762 // Maybe this is a linear tram, so we can avoid the whole stuff
763 if (!AllCurvesAreLinear(PostLin)) {
764
765 // All seems ok, proceed.
766 NewPostLin = cmsStageDup(PostLin);
767 if (!cmsPipelineInsertStage(Dest, cmsAT_END, NewPostLin))
768 goto Error;
769
770 // In destination LUT, the sampling should be applied after this stage.
771 cmsPipelineUnlinkStage(Src, cmsAT_END, &KeepPostLin);
772 }
773 }
774 }
775
776 // Now its time to do the sampling. We have to ignore pre/post linearization
777 // The source LUT without pre/post curves is passed as parameter.
778 if (!cmsStageSampleCLut16bit(CLUT, XFormSampler16, (void*) Src, 0)) {
779 Error:
780 // Ops, something went wrong, Restore stages
781 if (KeepPreLin != NULL) {
782 if (!cmsPipelineInsertStage(Src, cmsAT_BEGIN, KeepPreLin)) {
783 _cmsAssert(0); // This never happens
784 }
785 }
786 if (KeepPostLin != NULL) {
787 if (!cmsPipelineInsertStage(Src, cmsAT_END, KeepPostLin)) {
788 _cmsAssert(0); // This never happens
789 }
790 }
791 cmsPipelineFree(Dest);
792 return FALSE;
793 }
794
795 // Done.
796
797 if (KeepPreLin != NULL) cmsStageFree(KeepPreLin);
798 if (KeepPostLin != NULL) cmsStageFree(KeepPostLin);
799 cmsPipelineFree(Src);
800
801 DataCLUT = (_cmsStageCLutData*) CLUT ->Data;
802
803 if (NewPreLin == NULL) DataSetIn = NULL;
804 else DataSetIn = ((_cmsStageToneCurvesData*) NewPreLin ->Data) ->TheCurves;
805
806 if (NewPostLin == NULL) DataSetOut = NULL;
807 else DataSetOut = ((_cmsStageToneCurvesData*) NewPostLin ->Data) ->TheCurves;
808
809
810 if (DataSetIn == NULL && DataSetOut == NULL) {
811
812 _cmsPipelineSetOptimizationParameters(Dest, (_cmsPipelineEval16Fn) DataCLUT->Params->Interpolation.Lerp16, DataCLUT->Params, NULL, NULL);
813 }
814 else {
815
816 p16 = PrelinOpt16alloc(Dest ->ContextID,
817 DataCLUT ->Params,
818 Dest ->InputChannels,
819 DataSetIn,
820 Dest ->OutputChannels,
821 DataSetOut);
822
823 _cmsPipelineSetOptimizationParameters(Dest, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
824 }
825
826
827 // Don't fix white on absolute colorimetric
828 if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
829 *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
830
831 if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
832
833 FixWhiteMisalignment(Dest, ColorSpace, OutputColorSpace);
834 }
835
836 *Lut = Dest;
837 return TRUE;
838
839 cmsUNUSED_PARAMETER(Intent);
840 }
841
842
843 // -----------------------------------------------------------------------------------------------------------------------------------------------
844 // Fixes the gamma balancing of transform. This is described in my paper "Prelinearization Stages on
845 // Color-Management Application-Specific Integrated Circuits (ASICs)" presented at NIP24. It only works
846 // for RGB transforms. See the paper for more details
847 // -----------------------------------------------------------------------------------------------------------------------------------------------
848
849
850 // Normalize endpoints by slope limiting max and min. This assures endpoints as well.
851 // Descending curves are handled as well.
852 static
SlopeLimiting(cmsToneCurve * g)853 void SlopeLimiting(cmsToneCurve* g)
854 {
855 int BeginVal, EndVal;
856 int AtBegin = (int) floor((cmsFloat64Number) g ->nEntries * 0.02 + 0.5); // Cutoff at 2%
857 int AtEnd = (int) g ->nEntries - AtBegin - 1; // And 98%
858 cmsFloat64Number Val, Slope, beta;
859 int i;
860
861 if (cmsIsToneCurveDescending(g)) {
862 BeginVal = 0xffff; EndVal = 0;
863 }
864 else {
865 BeginVal = 0; EndVal = 0xffff;
866 }
867
868 // Compute slope and offset for begin of curve
869 Val = g ->Table16[AtBegin];
870 Slope = (Val - BeginVal) / AtBegin;
871 beta = Val - Slope * AtBegin;
872
873 for (i=0; i < AtBegin; i++)
874 g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
875
876 // Compute slope and offset for the end
877 Val = g ->Table16[AtEnd];
878 Slope = (EndVal - Val) / AtBegin; // AtBegin holds the X interval, which is same in both cases
879 beta = Val - Slope * AtEnd;
880
881 for (i = AtEnd; i < (int) g ->nEntries; i++)
882 g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
883 }
884
885
886 // Precomputes tables for 8-bit on input devicelink.
887 static
PrelinOpt8alloc(cmsContext ContextID,const cmsInterpParams * p,cmsToneCurve * G[3])888 Prelin8Data* PrelinOpt8alloc(cmsContext ContextID, const cmsInterpParams* p, cmsToneCurve* G[3])
889 {
890 int i;
891 cmsUInt16Number Input[3];
892 cmsS15Fixed16Number v1, v2, v3;
893 Prelin8Data* p8;
894
895 p8 = (Prelin8Data*)_cmsMallocZero(ContextID, sizeof(Prelin8Data));
896 if (p8 == NULL) return NULL;
897
898 // Since this only works for 8 bit input, values comes always as x * 257,
899 // we can safely take msb byte (x << 8 + x)
900
901 for (i=0; i < 256; i++) {
902
903 if (G != NULL) {
904
905 // Get 16-bit representation
906 Input[0] = cmsEvalToneCurve16(G[0], FROM_8_TO_16(i));
907 Input[1] = cmsEvalToneCurve16(G[1], FROM_8_TO_16(i));
908 Input[2] = cmsEvalToneCurve16(G[2], FROM_8_TO_16(i));
909 }
910 else {
911 Input[0] = FROM_8_TO_16(i);
912 Input[1] = FROM_8_TO_16(i);
913 Input[2] = FROM_8_TO_16(i);
914 }
915
916
917 // Move to 0..1.0 in fixed domain
918 v1 = _cmsToFixedDomain((int) (Input[0] * p -> Domain[0]));
919 v2 = _cmsToFixedDomain((int) (Input[1] * p -> Domain[1]));
920 v3 = _cmsToFixedDomain((int) (Input[2] * p -> Domain[2]));
921
922 // Store the precalculated table of nodes
923 p8 ->X0[i] = (p->opta[2] * FIXED_TO_INT(v1));
924 p8 ->Y0[i] = (p->opta[1] * FIXED_TO_INT(v2));
925 p8 ->Z0[i] = (p->opta[0] * FIXED_TO_INT(v3));
926
927 // Store the precalculated table of offsets
928 p8 ->rx[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v1);
929 p8 ->ry[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v2);
930 p8 ->rz[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v3);
931 }
932
933 p8 ->ContextID = ContextID;
934 p8 ->p = p;
935
936 return p8;
937 }
938
939 static
Prelin8free(cmsContext ContextID,void * ptr)940 void Prelin8free(cmsContext ContextID, void* ptr)
941 {
942 _cmsFree(ContextID, ptr);
943 }
944
945 static
Prelin8dup(cmsContext ContextID,const void * ptr)946 void* Prelin8dup(cmsContext ContextID, const void* ptr)
947 {
948 return _cmsDupMem(ContextID, ptr, sizeof(Prelin8Data));
949 }
950
951
952
953 // A optimized interpolation for 8-bit input.
954 #define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
955 static CMS_NO_SANITIZE
PrelinEval8(CMSREGISTER const cmsUInt16Number Input[],CMSREGISTER cmsUInt16Number Output[],CMSREGISTER const void * D)956 void PrelinEval8(CMSREGISTER const cmsUInt16Number Input[],
957 CMSREGISTER cmsUInt16Number Output[],
958 CMSREGISTER const void* D)
959 {
960
961 cmsUInt8Number r, g, b;
962 cmsS15Fixed16Number rx, ry, rz;
963 cmsS15Fixed16Number c0, c1, c2, c3, Rest;
964 int OutChan;
965 CMSREGISTER cmsS15Fixed16Number X0, X1, Y0, Y1, Z0, Z1;
966 Prelin8Data* p8 = (Prelin8Data*) D;
967 CMSREGISTER const cmsInterpParams* p = p8 ->p;
968 int TotalOut = (int) p -> nOutputs;
969 const cmsUInt16Number* LutTable = (const cmsUInt16Number*) p->Table;
970
971 r = (cmsUInt8Number) (Input[0] >> 8);
972 g = (cmsUInt8Number) (Input[1] >> 8);
973 b = (cmsUInt8Number) (Input[2] >> 8);
974
975 X0 = (cmsS15Fixed16Number) p8->X0[r];
976 Y0 = (cmsS15Fixed16Number) p8->Y0[g];
977 Z0 = (cmsS15Fixed16Number) p8->Z0[b];
978
979 rx = p8 ->rx[r];
980 ry = p8 ->ry[g];
981 rz = p8 ->rz[b];
982
983 X1 = X0 + (cmsS15Fixed16Number)((rx == 0) ? 0 : p ->opta[2]);
984 Y1 = Y0 + (cmsS15Fixed16Number)((ry == 0) ? 0 : p ->opta[1]);
985 Z1 = Z0 + (cmsS15Fixed16Number)((rz == 0) ? 0 : p ->opta[0]);
986
987
988 // These are the 6 Tetrahedral
989 for (OutChan=0; OutChan < TotalOut; OutChan++) {
990
991 c0 = DENS(X0, Y0, Z0);
992
993 if (rx >= ry && ry >= rz)
994 {
995 c1 = DENS(X1, Y0, Z0) - c0;
996 c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
997 c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
998 }
999 else
1000 if (rx >= rz && rz >= ry)
1001 {
1002 c1 = DENS(X1, Y0, Z0) - c0;
1003 c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
1004 c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
1005 }
1006 else
1007 if (rz >= rx && rx >= ry)
1008 {
1009 c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
1010 c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
1011 c3 = DENS(X0, Y0, Z1) - c0;
1012 }
1013 else
1014 if (ry >= rx && rx >= rz)
1015 {
1016 c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
1017 c2 = DENS(X0, Y1, Z0) - c0;
1018 c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
1019 }
1020 else
1021 if (ry >= rz && rz >= rx)
1022 {
1023 c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
1024 c2 = DENS(X0, Y1, Z0) - c0;
1025 c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);
1026 }
1027 else
1028 if (rz >= ry && ry >= rx)
1029 {
1030 c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
1031 c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
1032 c3 = DENS(X0, Y0, Z1) - c0;
1033 }
1034 else {
1035 c1 = c2 = c3 = 0;
1036 }
1037
1038 Rest = c1 * rx + c2 * ry + c3 * rz + 0x8001;
1039 Output[OutChan] = (cmsUInt16Number) (c0 + ((Rest + (Rest >> 16)) >> 16));
1040
1041 }
1042 }
1043
1044 #undef DENS
1045
1046
1047 // Curves that contain wide empty areas are not optimizeable
1048 static
IsDegenerated(const cmsToneCurve * g)1049 cmsBool IsDegenerated(const cmsToneCurve* g)
1050 {
1051 cmsUInt32Number i, Zeros = 0, Poles = 0;
1052 cmsUInt32Number nEntries = g ->nEntries;
1053
1054 for (i=0; i < nEntries; i++) {
1055
1056 if (g ->Table16[i] == 0x0000) Zeros++;
1057 if (g ->Table16[i] == 0xffff) Poles++;
1058 }
1059
1060 if (Zeros == 1 && Poles == 1) return FALSE; // For linear tables
1061 if (Zeros > (nEntries / 20)) return TRUE; // Degenerated, many zeros
1062 if (Poles > (nEntries / 20)) return TRUE; // Degenerated, many poles
1063
1064 return FALSE;
1065 }
1066
1067 // --------------------------------------------------------------------------------------------------------------
1068 // We need xput over here
1069
1070 static
OptimizeByComputingLinearization(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1071 cmsBool OptimizeByComputingLinearization(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1072 {
1073 cmsPipeline* OriginalLut;
1074 cmsUInt32Number nGridPoints;
1075 cmsToneCurve *Trans[cmsMAXCHANNELS], *TransReverse[cmsMAXCHANNELS];
1076 cmsUInt32Number t, i;
1077 cmsFloat32Number v, In[cmsMAXCHANNELS], Out[cmsMAXCHANNELS];
1078 cmsBool lIsSuitable, lIsLinear;
1079 cmsPipeline* OptimizedLUT = NULL, *LutPlusCurves = NULL;
1080 cmsStage* OptimizedCLUTmpe;
1081 cmsColorSpaceSignature ColorSpace, OutputColorSpace;
1082 cmsStage* OptimizedPrelinMpe;
1083 cmsStage* mpe;
1084 cmsToneCurve** OptimizedPrelinCurves;
1085 _cmsStageCLutData* OptimizedPrelinCLUT;
1086
1087
1088 // This is a lossy optimization! does not apply in floating-point cases
1089 if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
1090
1091 // Only on chunky RGB
1092 if (T_COLORSPACE(*InputFormat) != PT_RGB) return FALSE;
1093 if (T_PLANAR(*InputFormat)) return FALSE;
1094
1095 if (T_COLORSPACE(*OutputFormat) != PT_RGB) return FALSE;
1096 if (T_PLANAR(*OutputFormat)) return FALSE;
1097
1098 // On 16 bits, user has to specify the feature
1099 if (!_cmsFormatterIs8bit(*InputFormat)) {
1100 if (!(*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION)) return FALSE;
1101 }
1102
1103 OriginalLut = *Lut;
1104
1105 // Named color pipelines cannot be optimized either
1106 for (mpe = cmsPipelineGetPtrToFirstStage(OriginalLut);
1107 mpe != NULL;
1108 mpe = cmsStageNext(mpe)) {
1109 if (cmsStageType(mpe) == cmsSigNamedColorElemType) return FALSE;
1110 }
1111
1112 ColorSpace = _cmsICCcolorSpace((int) T_COLORSPACE(*InputFormat));
1113 OutputColorSpace = _cmsICCcolorSpace((int) T_COLORSPACE(*OutputFormat));
1114
1115 // Color space must be specified
1116 if (ColorSpace == (cmsColorSpaceSignature)0 ||
1117 OutputColorSpace == (cmsColorSpaceSignature)0) return FALSE;
1118
1119 nGridPoints = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);
1120
1121 // Empty gamma containers
1122 memset(Trans, 0, sizeof(Trans));
1123 memset(TransReverse, 0, sizeof(TransReverse));
1124
1125 // If the last stage of the original lut are curves, and those curves are
1126 // degenerated, it is likely the transform is squeezing and clipping
1127 // the output from previous CLUT. We cannot optimize this case
1128 {
1129 cmsStage* last = cmsPipelineGetPtrToLastStage(OriginalLut);
1130
1131 if (last == NULL) goto Error;
1132 if (cmsStageType(last) == cmsSigCurveSetElemType) {
1133
1134 _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*)cmsStageData(last);
1135 for (i = 0; i < Data->nCurves; i++) {
1136 if (IsDegenerated(Data->TheCurves[i]))
1137 goto Error;
1138 }
1139 }
1140 }
1141
1142 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1143 Trans[t] = cmsBuildTabulatedToneCurve16(OriginalLut ->ContextID, PRELINEARIZATION_POINTS, NULL);
1144 if (Trans[t] == NULL) goto Error;
1145 }
1146
1147 // Populate the curves
1148 for (i=0; i < PRELINEARIZATION_POINTS; i++) {
1149
1150 v = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
1151
1152 // Feed input with a gray ramp
1153 for (t=0; t < OriginalLut ->InputChannels; t++)
1154 In[t] = v;
1155
1156 // Evaluate the gray value
1157 cmsPipelineEvalFloat(In, Out, OriginalLut);
1158
1159 // Store result in curve
1160 for (t=0; t < OriginalLut ->InputChannels; t++)
1161 Trans[t] ->Table16[i] = _cmsQuickSaturateWord(Out[t] * 65535.0);
1162 }
1163
1164 // Slope-limit the obtained curves
1165 for (t = 0; t < OriginalLut ->InputChannels; t++)
1166 SlopeLimiting(Trans[t]);
1167
1168 // Check for validity
1169 lIsSuitable = TRUE;
1170 lIsLinear = TRUE;
1171 for (t=0; (lIsSuitable && (t < OriginalLut ->InputChannels)); t++) {
1172
1173 // Exclude if already linear
1174 if (!cmsIsToneCurveLinear(Trans[t]))
1175 lIsLinear = FALSE;
1176
1177 // Exclude if non-monotonic
1178 if (!cmsIsToneCurveMonotonic(Trans[t]))
1179 lIsSuitable = FALSE;
1180
1181 if (IsDegenerated(Trans[t]))
1182 lIsSuitable = FALSE;
1183 }
1184
1185 // If it is not suitable, just quit
1186 if (!lIsSuitable) goto Error;
1187
1188 // Invert curves if possible
1189 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1190 TransReverse[t] = cmsReverseToneCurveEx(PRELINEARIZATION_POINTS, Trans[t]);
1191 if (TransReverse[t] == NULL) goto Error;
1192 }
1193
1194 // Now inset the reversed curves at the begin of transform
1195 LutPlusCurves = cmsPipelineDup(OriginalLut);
1196 if (LutPlusCurves == NULL) goto Error;
1197
1198 if (!cmsPipelineInsertStage(LutPlusCurves, cmsAT_BEGIN, cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, TransReverse)))
1199 goto Error;
1200
1201 // Create the result LUT
1202 OptimizedLUT = cmsPipelineAlloc(OriginalLut ->ContextID, OriginalLut ->InputChannels, OriginalLut ->OutputChannels);
1203 if (OptimizedLUT == NULL) goto Error;
1204
1205 OptimizedPrelinMpe = cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, Trans);
1206
1207 // Create and insert the curves at the beginning
1208 if (!cmsPipelineInsertStage(OptimizedLUT, cmsAT_BEGIN, OptimizedPrelinMpe))
1209 goto Error;
1210
1211 // Allocate the CLUT for result
1212 OptimizedCLUTmpe = cmsStageAllocCLut16bit(OriginalLut ->ContextID, nGridPoints, OriginalLut ->InputChannels, OriginalLut ->OutputChannels, NULL);
1213
1214 // Add the CLUT to the destination LUT
1215 if (!cmsPipelineInsertStage(OptimizedLUT, cmsAT_END, OptimizedCLUTmpe))
1216 goto Error;
1217
1218 // Resample the LUT
1219 if (!cmsStageSampleCLut16bit(OptimizedCLUTmpe, XFormSampler16, (void*) LutPlusCurves, 0)) goto Error;
1220
1221 // Free resources
1222 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1223
1224 if (Trans[t]) cmsFreeToneCurve(Trans[t]);
1225 if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
1226 }
1227
1228 cmsPipelineFree(LutPlusCurves);
1229
1230
1231 OptimizedPrelinCurves = _cmsStageGetPtrToCurveSet(OptimizedPrelinMpe);
1232 OptimizedPrelinCLUT = (_cmsStageCLutData*) OptimizedCLUTmpe ->Data;
1233
1234 // Set the evaluator if 8-bit
1235 if (_cmsFormatterIs8bit(*InputFormat)) {
1236
1237 Prelin8Data* p8 = PrelinOpt8alloc(OptimizedLUT ->ContextID,
1238 OptimizedPrelinCLUT ->Params,
1239 OptimizedPrelinCurves);
1240 if (p8 == NULL) return FALSE;
1241
1242 _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval8, (void*) p8, Prelin8free, Prelin8dup);
1243
1244 }
1245 else
1246 {
1247 Prelin16Data* p16 = PrelinOpt16alloc(OptimizedLUT ->ContextID,
1248 OptimizedPrelinCLUT ->Params,
1249 3, OptimizedPrelinCurves, 3, NULL);
1250 if (p16 == NULL) return FALSE;
1251
1252 _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
1253
1254 }
1255
1256 // Don't fix white on absolute colorimetric
1257 if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
1258 *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
1259
1260 if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
1261
1262 if (!FixWhiteMisalignment(OptimizedLUT, ColorSpace, OutputColorSpace)) {
1263
1264 return FALSE;
1265 }
1266 }
1267
1268 // And return the obtained LUT
1269
1270 cmsPipelineFree(OriginalLut);
1271 *Lut = OptimizedLUT;
1272 return TRUE;
1273
1274 Error:
1275
1276 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1277
1278 if (Trans[t]) cmsFreeToneCurve(Trans[t]);
1279 if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
1280 }
1281
1282 if (LutPlusCurves != NULL) cmsPipelineFree(LutPlusCurves);
1283 if (OptimizedLUT != NULL) cmsPipelineFree(OptimizedLUT);
1284
1285 return FALSE;
1286
1287 cmsUNUSED_PARAMETER(Intent);
1288 cmsUNUSED_PARAMETER(lIsLinear);
1289 }
1290
1291
1292 // Curves optimizer ------------------------------------------------------------------------------------------------------------------
1293
1294 static
CurvesFree(cmsContext ContextID,void * ptr)1295 void CurvesFree(cmsContext ContextID, void* ptr)
1296 {
1297 Curves16Data* Data = (Curves16Data*) ptr;
1298 cmsUInt32Number i;
1299
1300 for (i=0; i < Data -> nCurves; i++) {
1301
1302 _cmsFree(ContextID, Data ->Curves[i]);
1303 }
1304
1305 _cmsFree(ContextID, Data ->Curves);
1306 _cmsFree(ContextID, ptr);
1307 }
1308
1309 static
CurvesDup(cmsContext ContextID,const void * ptr)1310 void* CurvesDup(cmsContext ContextID, const void* ptr)
1311 {
1312 Curves16Data* Data = (Curves16Data*)_cmsDupMem(ContextID, ptr, sizeof(Curves16Data));
1313 cmsUInt32Number i;
1314
1315 if (Data == NULL) return NULL;
1316
1317 Data->Curves = (cmsUInt16Number**) _cmsDupMem(ContextID, Data->Curves, Data->nCurves * sizeof(cmsUInt16Number*));
1318
1319 for (i=0; i < Data -> nCurves; i++) {
1320 Data->Curves[i] = (cmsUInt16Number*) _cmsDupMem(ContextID, Data->Curves[i], Data->nElements * sizeof(cmsUInt16Number));
1321 }
1322
1323 return (void*) Data;
1324 }
1325
1326 // Precomputes tables for 8-bit on input devicelink.
1327 static
CurvesAlloc(cmsContext ContextID,cmsUInt32Number nCurves,cmsUInt32Number nElements,cmsToneCurve ** G)1328 Curves16Data* CurvesAlloc(cmsContext ContextID, cmsUInt32Number nCurves, cmsUInt32Number nElements, cmsToneCurve** G)
1329 {
1330 cmsUInt32Number i, j;
1331 Curves16Data* c16;
1332
1333 c16 = (Curves16Data*)_cmsMallocZero(ContextID, sizeof(Curves16Data));
1334 if (c16 == NULL) return NULL;
1335
1336 c16 ->nCurves = nCurves;
1337 c16 ->nElements = nElements;
1338
1339 c16->Curves = (cmsUInt16Number**) _cmsCalloc(ContextID, nCurves, sizeof(cmsUInt16Number*));
1340 if (c16->Curves == NULL) {
1341 _cmsFree(ContextID, c16);
1342 return NULL;
1343 }
1344
1345 for (i=0; i < nCurves; i++) {
1346
1347 c16->Curves[i] = (cmsUInt16Number*) _cmsCalloc(ContextID, nElements, sizeof(cmsUInt16Number));
1348
1349 if (c16->Curves[i] == NULL) {
1350
1351 for (j=0; j < i; j++) {
1352 _cmsFree(ContextID, c16->Curves[j]);
1353 }
1354 _cmsFree(ContextID, c16->Curves);
1355 _cmsFree(ContextID, c16);
1356 return NULL;
1357 }
1358
1359 if (nElements == 256U) {
1360
1361 for (j=0; j < nElements; j++) {
1362
1363 c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], FROM_8_TO_16(j));
1364 }
1365 }
1366 else {
1367
1368 for (j=0; j < nElements; j++) {
1369 c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], (cmsUInt16Number) j);
1370 }
1371 }
1372 }
1373
1374 return c16;
1375 }
1376
1377 static
FastEvaluateCurves8(CMSREGISTER const cmsUInt16Number In[],CMSREGISTER cmsUInt16Number Out[],CMSREGISTER const void * D)1378 void FastEvaluateCurves8(CMSREGISTER const cmsUInt16Number In[],
1379 CMSREGISTER cmsUInt16Number Out[],
1380 CMSREGISTER const void* D)
1381 {
1382 Curves16Data* Data = (Curves16Data*) D;
1383 int x;
1384 cmsUInt32Number i;
1385
1386 for (i=0; i < Data ->nCurves; i++) {
1387
1388 x = (In[i] >> 8);
1389 Out[i] = Data -> Curves[i][x];
1390 }
1391 }
1392
1393
1394 static
FastEvaluateCurves16(CMSREGISTER const cmsUInt16Number In[],CMSREGISTER cmsUInt16Number Out[],CMSREGISTER const void * D)1395 void FastEvaluateCurves16(CMSREGISTER const cmsUInt16Number In[],
1396 CMSREGISTER cmsUInt16Number Out[],
1397 CMSREGISTER const void* D)
1398 {
1399 Curves16Data* Data = (Curves16Data*) D;
1400 cmsUInt32Number i;
1401
1402 for (i=0; i < Data ->nCurves; i++) {
1403 Out[i] = Data -> Curves[i][In[i]];
1404 }
1405 }
1406
1407
1408 static
FastIdentity16(CMSREGISTER const cmsUInt16Number In[],CMSREGISTER cmsUInt16Number Out[],CMSREGISTER const void * D)1409 void FastIdentity16(CMSREGISTER const cmsUInt16Number In[],
1410 CMSREGISTER cmsUInt16Number Out[],
1411 CMSREGISTER const void* D)
1412 {
1413 cmsPipeline* Lut = (cmsPipeline*) D;
1414 cmsUInt32Number i;
1415
1416 for (i=0; i < Lut ->InputChannels; i++) {
1417 Out[i] = In[i];
1418 }
1419 }
1420
1421
1422 // If the target LUT holds only curves, the optimization procedure is to join all those
1423 // curves together. That only works on curves and does not work on matrices.
1424 static
OptimizeByJoiningCurves(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1425 cmsBool OptimizeByJoiningCurves(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1426 {
1427 cmsToneCurve** GammaTables = NULL;
1428 cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
1429 cmsUInt32Number i, j;
1430 cmsPipeline* Src = *Lut;
1431 cmsPipeline* Dest = NULL;
1432 cmsStage* mpe;
1433 cmsStage* ObtainedCurves = NULL;
1434
1435
1436 // This is a lossy optimization! does not apply in floating-point cases
1437 if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
1438
1439 // Only curves in this LUT?
1440 for (mpe = cmsPipelineGetPtrToFirstStage(Src);
1441 mpe != NULL;
1442 mpe = cmsStageNext(mpe)) {
1443 if (cmsStageType(mpe) != cmsSigCurveSetElemType) return FALSE;
1444 }
1445
1446 // Allocate an empty LUT
1447 Dest = cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
1448 if (Dest == NULL) return FALSE;
1449
1450 // Create target curves
1451 GammaTables = (cmsToneCurve**) _cmsCalloc(Src ->ContextID, Src ->InputChannels, sizeof(cmsToneCurve*));
1452 if (GammaTables == NULL) goto Error;
1453
1454 for (i=0; i < Src ->InputChannels; i++) {
1455 GammaTables[i] = cmsBuildTabulatedToneCurve16(Src ->ContextID, PRELINEARIZATION_POINTS, NULL);
1456 if (GammaTables[i] == NULL) goto Error;
1457 }
1458
1459 // Compute 16 bit result by using floating point
1460 for (i=0; i < PRELINEARIZATION_POINTS; i++) {
1461
1462 for (j=0; j < Src ->InputChannels; j++)
1463 InFloat[j] = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
1464
1465 cmsPipelineEvalFloat(InFloat, OutFloat, Src);
1466
1467 for (j=0; j < Src ->InputChannels; j++)
1468 GammaTables[j] -> Table16[i] = _cmsQuickSaturateWord(OutFloat[j] * 65535.0);
1469 }
1470
1471 ObtainedCurves = cmsStageAllocToneCurves(Src ->ContextID, Src ->InputChannels, GammaTables);
1472 if (ObtainedCurves == NULL) goto Error;
1473
1474 for (i=0; i < Src ->InputChannels; i++) {
1475 cmsFreeToneCurve(GammaTables[i]);
1476 GammaTables[i] = NULL;
1477 }
1478
1479 if (GammaTables != NULL) {
1480 _cmsFree(Src->ContextID, GammaTables);
1481 GammaTables = NULL;
1482 }
1483
1484 // Maybe the curves are linear at the end
1485 if (!AllCurvesAreLinear(ObtainedCurves)) {
1486 _cmsStageToneCurvesData* Data;
1487
1488 if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, ObtainedCurves))
1489 goto Error;
1490 Data = (_cmsStageToneCurvesData*) cmsStageData(ObtainedCurves);
1491 ObtainedCurves = NULL;
1492
1493 // If the curves are to be applied in 8 bits, we can save memory
1494 if (_cmsFormatterIs8bit(*InputFormat)) {
1495 Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 256, Data ->TheCurves);
1496
1497 if (c16 == NULL) goto Error;
1498 *dwFlags |= cmsFLAGS_NOCACHE;
1499 _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves8, c16, CurvesFree, CurvesDup);
1500
1501 }
1502 else {
1503 Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 65536, Data ->TheCurves);
1504
1505 if (c16 == NULL) goto Error;
1506 *dwFlags |= cmsFLAGS_NOCACHE;
1507 _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves16, c16, CurvesFree, CurvesDup);
1508 }
1509 }
1510 else {
1511
1512 // LUT optimizes to nothing. Set the identity LUT
1513 cmsStageFree(ObtainedCurves);
1514 ObtainedCurves = NULL;
1515
1516 if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageAllocIdentity(Dest ->ContextID, Src ->InputChannels)))
1517 goto Error;
1518
1519 *dwFlags |= cmsFLAGS_NOCACHE;
1520 _cmsPipelineSetOptimizationParameters(Dest, FastIdentity16, (void*) Dest, NULL, NULL);
1521 }
1522
1523 // We are done.
1524 cmsPipelineFree(Src);
1525 *Lut = Dest;
1526 return TRUE;
1527
1528 Error:
1529
1530 if (ObtainedCurves != NULL) cmsStageFree(ObtainedCurves);
1531 if (GammaTables != NULL) {
1532 for (i=0; i < Src ->InputChannels; i++) {
1533 if (GammaTables[i] != NULL) cmsFreeToneCurve(GammaTables[i]);
1534 }
1535
1536 _cmsFree(Src ->ContextID, GammaTables);
1537 }
1538
1539 if (Dest != NULL) cmsPipelineFree(Dest);
1540 return FALSE;
1541
1542 cmsUNUSED_PARAMETER(Intent);
1543 cmsUNUSED_PARAMETER(InputFormat);
1544 cmsUNUSED_PARAMETER(OutputFormat);
1545 cmsUNUSED_PARAMETER(dwFlags);
1546 }
1547
1548 // -------------------------------------------------------------------------------------------------------------------------------------
1549 // LUT is Shaper - Matrix - Matrix - Shaper, which is very frequent when combining two matrix-shaper profiles
1550
1551
1552 static
FreeMatShaper(cmsContext ContextID,void * Data)1553 void FreeMatShaper(cmsContext ContextID, void* Data)
1554 {
1555 if (Data != NULL) _cmsFree(ContextID, Data);
1556 }
1557
1558 static
DupMatShaper(cmsContext ContextID,const void * Data)1559 void* DupMatShaper(cmsContext ContextID, const void* Data)
1560 {
1561 return _cmsDupMem(ContextID, Data, sizeof(MatShaper8Data));
1562 }
1563
1564
1565 // A fast matrix-shaper evaluator for 8 bits. This is a bit ticky since I'm using 1.14 signed fixed point
1566 // to accomplish some performance. Actually it takes 256x3 16 bits tables and 16385 x 3 tables of 8 bits,
1567 // in total about 50K, and the performance boost is huge!
1568 static
MatShaperEval16(CMSREGISTER const cmsUInt16Number In[],CMSREGISTER cmsUInt16Number Out[],CMSREGISTER const void * D)1569 void MatShaperEval16(CMSREGISTER const cmsUInt16Number In[],
1570 CMSREGISTER cmsUInt16Number Out[],
1571 CMSREGISTER const void* D)
1572 {
1573 MatShaper8Data* p = (MatShaper8Data*) D;
1574 cmsS1Fixed14Number l1, l2, l3, r, g, b;
1575 cmsUInt32Number ri, gi, bi;
1576
1577 // In this case (and only in this case!) we can use this simplification since
1578 // In[] is assured to come from a 8 bit number. (a << 8 | a)
1579 ri = In[0] & 0xFFU;
1580 gi = In[1] & 0xFFU;
1581 bi = In[2] & 0xFFU;
1582
1583 // Across first shaper, which also converts to 1.14 fixed point
1584 r = p->Shaper1R[ri];
1585 g = p->Shaper1G[gi];
1586 b = p->Shaper1B[bi];
1587
1588 // Evaluate the matrix in 1.14 fixed point
1589 l1 = (p->Mat[0][0] * r + p->Mat[0][1] * g + p->Mat[0][2] * b + p->Off[0] + 0x2000) >> 14;
1590 l2 = (p->Mat[1][0] * r + p->Mat[1][1] * g + p->Mat[1][2] * b + p->Off[1] + 0x2000) >> 14;
1591 l3 = (p->Mat[2][0] * r + p->Mat[2][1] * g + p->Mat[2][2] * b + p->Off[2] + 0x2000) >> 14;
1592
1593 // Now we have to clip to 0..1.0 range
1594 ri = (l1 < 0) ? 0 : ((l1 > 16384) ? 16384U : (cmsUInt32Number) l1);
1595 gi = (l2 < 0) ? 0 : ((l2 > 16384) ? 16384U : (cmsUInt32Number) l2);
1596 bi = (l3 < 0) ? 0 : ((l3 > 16384) ? 16384U : (cmsUInt32Number) l3);
1597
1598 // And across second shaper,
1599 Out[0] = p->Shaper2R[ri];
1600 Out[1] = p->Shaper2G[gi];
1601 Out[2] = p->Shaper2B[bi];
1602
1603 }
1604
1605 // This table converts from 8 bits to 1.14 after applying the curve
1606 static
FillFirstShaper(cmsS1Fixed14Number * Table,cmsToneCurve * Curve)1607 void FillFirstShaper(cmsS1Fixed14Number* Table, cmsToneCurve* Curve)
1608 {
1609 int i;
1610 cmsFloat32Number R, y;
1611
1612 for (i=0; i < 256; i++) {
1613
1614 R = (cmsFloat32Number) (i / 255.0);
1615 y = cmsEvalToneCurveFloat(Curve, R);
1616
1617 if (y < 131072.0)
1618 Table[i] = DOUBLE_TO_1FIXED14(y);
1619 else
1620 Table[i] = 0x7fffffff;
1621 }
1622 }
1623
1624 // This table converts form 1.14 (being 0x4000 the last entry) to 8 bits after applying the curve
1625 static
FillSecondShaper(cmsUInt16Number * Table,cmsToneCurve * Curve,cmsBool Is8BitsOutput)1626 void FillSecondShaper(cmsUInt16Number* Table, cmsToneCurve* Curve, cmsBool Is8BitsOutput)
1627 {
1628 int i;
1629 cmsFloat32Number R, Val;
1630
1631 for (i=0; i < 16385; i++) {
1632
1633 R = (cmsFloat32Number) (i / 16384.0);
1634 Val = cmsEvalToneCurveFloat(Curve, R); // Val comes 0..1.0
1635
1636 if (Val < 0)
1637 Val = 0;
1638
1639 if (Val > 1.0)
1640 Val = 1.0;
1641
1642 if (Is8BitsOutput) {
1643
1644 // If 8 bits output, we can optimize further by computing the / 257 part.
1645 // first we compute the resulting byte and then we store the byte times
1646 // 257. This quantization allows to round very quick by doing a >> 8, but
1647 // since the low byte is always equal to msb, we can do a & 0xff and this works!
1648 cmsUInt16Number w = _cmsQuickSaturateWord(Val * 65535.0);
1649 cmsUInt8Number b = FROM_16_TO_8(w);
1650
1651 Table[i] = FROM_8_TO_16(b);
1652 }
1653 else Table[i] = _cmsQuickSaturateWord(Val * 65535.0);
1654 }
1655 }
1656
1657 // Compute the matrix-shaper structure
1658 static
SetMatShaper(cmsPipeline * Dest,cmsToneCurve * Curve1[3],cmsMAT3 * Mat,cmsVEC3 * Off,cmsToneCurve * Curve2[3],cmsUInt32Number * OutputFormat)1659 cmsBool SetMatShaper(cmsPipeline* Dest, cmsToneCurve* Curve1[3], cmsMAT3* Mat, cmsVEC3* Off, cmsToneCurve* Curve2[3], cmsUInt32Number* OutputFormat)
1660 {
1661 MatShaper8Data* p;
1662 int i, j;
1663 cmsBool Is8Bits = _cmsFormatterIs8bit(*OutputFormat);
1664
1665 // Allocate a big chuck of memory to store precomputed tables
1666 p = (MatShaper8Data*) _cmsMalloc(Dest ->ContextID, sizeof(MatShaper8Data));
1667 if (p == NULL) return FALSE;
1668
1669 p -> ContextID = Dest -> ContextID;
1670
1671 // Precompute tables
1672 FillFirstShaper(p ->Shaper1R, Curve1[0]);
1673 FillFirstShaper(p ->Shaper1G, Curve1[1]);
1674 FillFirstShaper(p ->Shaper1B, Curve1[2]);
1675
1676 FillSecondShaper(p ->Shaper2R, Curve2[0], Is8Bits);
1677 FillSecondShaper(p ->Shaper2G, Curve2[1], Is8Bits);
1678 FillSecondShaper(p ->Shaper2B, Curve2[2], Is8Bits);
1679
1680 // Convert matrix to nFixed14. Note that those values may take more than 16 bits
1681 for (i=0; i < 3; i++) {
1682 for (j=0; j < 3; j++) {
1683 p ->Mat[i][j] = DOUBLE_TO_1FIXED14(Mat->v[i].n[j]);
1684 }
1685 }
1686
1687 for (i=0; i < 3; i++) {
1688
1689 if (Off == NULL) {
1690 p ->Off[i] = 0;
1691 }
1692 else {
1693 p ->Off[i] = DOUBLE_TO_1FIXED14(Off->n[i]);
1694 }
1695 }
1696
1697 // Mark as optimized for faster formatter
1698 if (Is8Bits)
1699 *OutputFormat |= OPTIMIZED_SH(1);
1700
1701 // Fill function pointers
1702 _cmsPipelineSetOptimizationParameters(Dest, MatShaperEval16, (void*) p, FreeMatShaper, DupMatShaper);
1703 return TRUE;
1704 }
1705
1706 // 8 bits on input allows matrix-shaper boot up to 25 Mpixels per second on RGB. That's fast!
1707 static
OptimizeMatrixShaper(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1708 cmsBool OptimizeMatrixShaper(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1709 {
1710 cmsStage* Curve1, *Curve2;
1711 cmsStage* Matrix1, *Matrix2;
1712 cmsMAT3 res;
1713 cmsBool IdentityMat;
1714 cmsPipeline* Dest, *Src;
1715 cmsFloat64Number* Offset;
1716
1717 // Only works on RGB to RGB
1718 if (T_CHANNELS(*InputFormat) != 3 || T_CHANNELS(*OutputFormat) != 3) return FALSE;
1719
1720 // Only works on 8 bit input
1721 if (!_cmsFormatterIs8bit(*InputFormat)) return FALSE;
1722
1723 // Seems suitable, proceed
1724 Src = *Lut;
1725
1726 // Check for:
1727 //
1728 // shaper-matrix-matrix-shaper
1729 // shaper-matrix-shaper
1730 //
1731 // Both of those constructs are possible (first because abs. colorimetric).
1732 // additionally, In the first case, the input matrix offset should be zero.
1733
1734 IdentityMat = FALSE;
1735 if (cmsPipelineCheckAndRetreiveStages(Src, 4,
1736 cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
1737 &Curve1, &Matrix1, &Matrix2, &Curve2)) {
1738
1739 // Get both matrices
1740 _cmsStageMatrixData* Data1 = (_cmsStageMatrixData*)cmsStageData(Matrix1);
1741 _cmsStageMatrixData* Data2 = (_cmsStageMatrixData*)cmsStageData(Matrix2);
1742
1743 // Input offset should be zero
1744 if (Data1->Offset != NULL) return FALSE;
1745
1746 // Multiply both matrices to get the result
1747 _cmsMAT3per(&res, (cmsMAT3*)Data2->Double, (cmsMAT3*)Data1->Double);
1748
1749 // Only 2nd matrix has offset, or it is zero
1750 Offset = Data2->Offset;
1751
1752 // Now the result is in res + Data2 -> Offset. Maybe is a plain identity?
1753 if (_cmsMAT3isIdentity(&res) && Offset == NULL) {
1754
1755 // We can get rid of full matrix
1756 IdentityMat = TRUE;
1757 }
1758
1759 }
1760 else {
1761
1762 if (cmsPipelineCheckAndRetreiveStages(Src, 3,
1763 cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
1764 &Curve1, &Matrix1, &Curve2)) {
1765
1766 _cmsStageMatrixData* Data = (_cmsStageMatrixData*)cmsStageData(Matrix1);
1767
1768 // Copy the matrix to our result
1769 memcpy(&res, Data->Double, sizeof(res));
1770
1771 // Preserve the Odffset (may be NULL as a zero offset)
1772 Offset = Data->Offset;
1773
1774 if (_cmsMAT3isIdentity(&res) && Offset == NULL) {
1775
1776 // We can get rid of full matrix
1777 IdentityMat = TRUE;
1778 }
1779 }
1780 else
1781 return FALSE; // Not optimizeable this time
1782
1783 }
1784
1785 // Allocate an empty LUT
1786 Dest = cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
1787 if (!Dest) return FALSE;
1788
1789 // Assamble the new LUT
1790 if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageDup(Curve1)))
1791 goto Error;
1792
1793 if (!IdentityMat) {
1794
1795 if (!cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageAllocMatrix(Dest->ContextID, 3, 3, (const cmsFloat64Number*)&res, Offset)))
1796 goto Error;
1797 }
1798
1799 if (!cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageDup(Curve2)))
1800 goto Error;
1801
1802 // If identity on matrix, we can further optimize the curves, so call the join curves routine
1803 if (IdentityMat) {
1804
1805 OptimizeByJoiningCurves(&Dest, Intent, InputFormat, OutputFormat, dwFlags);
1806 }
1807 else {
1808 _cmsStageToneCurvesData* mpeC1 = (_cmsStageToneCurvesData*) cmsStageData(Curve1);
1809 _cmsStageToneCurvesData* mpeC2 = (_cmsStageToneCurvesData*) cmsStageData(Curve2);
1810
1811 // In this particular optimization, cache does not help as it takes more time to deal with
1812 // the cache that with the pixel handling
1813 *dwFlags |= cmsFLAGS_NOCACHE;
1814
1815 // Setup the optimizarion routines
1816 SetMatShaper(Dest, mpeC1 ->TheCurves, &res, (cmsVEC3*) Offset, mpeC2->TheCurves, OutputFormat);
1817 }
1818
1819 cmsPipelineFree(Src);
1820 *Lut = Dest;
1821 return TRUE;
1822 Error:
1823 // Leave Src unchanged
1824 cmsPipelineFree(Dest);
1825 return FALSE;
1826 }
1827
1828
1829 // -------------------------------------------------------------------------------------------------------------------------------------
1830 // Optimization plug-ins
1831
1832 // List of optimizations
1833 typedef struct _cmsOptimizationCollection_st {
1834
1835 _cmsOPToptimizeFn OptimizePtr;
1836
1837 struct _cmsOptimizationCollection_st *Next;
1838
1839 } _cmsOptimizationCollection;
1840
1841
1842 // The built-in list. We currently implement 4 types of optimizations. Joining of curves, matrix-shaper, linearization and resampling
1843 static _cmsOptimizationCollection DefaultOptimization[] = {
1844
1845 { OptimizeByJoiningCurves, &DefaultOptimization[1] },
1846 { OptimizeMatrixShaper, &DefaultOptimization[2] },
1847 { OptimizeByComputingLinearization, &DefaultOptimization[3] },
1848 { OptimizeByResampling, NULL }
1849 };
1850
1851 // The linked list head
1852 _cmsOptimizationPluginChunkType _cmsOptimizationPluginChunk = { NULL };
1853
1854
1855 // Duplicates the zone of memory used by the plug-in in the new context
1856 static
DupPluginOptimizationList(struct _cmsContext_struct * ctx,const struct _cmsContext_struct * src)1857 void DupPluginOptimizationList(struct _cmsContext_struct* ctx,
1858 const struct _cmsContext_struct* src)
1859 {
1860 _cmsOptimizationPluginChunkType newHead = { NULL };
1861 _cmsOptimizationCollection* entry;
1862 _cmsOptimizationCollection* Anterior = NULL;
1863 _cmsOptimizationPluginChunkType* head = (_cmsOptimizationPluginChunkType*) src->chunks[OptimizationPlugin];
1864
1865 _cmsAssert(ctx != NULL);
1866 _cmsAssert(head != NULL);
1867
1868 // Walk the list copying all nodes
1869 for (entry = head->OptimizationCollection;
1870 entry != NULL;
1871 entry = entry ->Next) {
1872
1873 _cmsOptimizationCollection *newEntry = ( _cmsOptimizationCollection *) _cmsSubAllocDup(ctx ->MemPool, entry, sizeof(_cmsOptimizationCollection));
1874
1875 if (newEntry == NULL)
1876 return;
1877
1878 // We want to keep the linked list order, so this is a little bit tricky
1879 newEntry -> Next = NULL;
1880 if (Anterior)
1881 Anterior -> Next = newEntry;
1882
1883 Anterior = newEntry;
1884
1885 if (newHead.OptimizationCollection == NULL)
1886 newHead.OptimizationCollection = newEntry;
1887 }
1888
1889 ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx->MemPool, &newHead, sizeof(_cmsOptimizationPluginChunkType));
1890 }
1891
_cmsAllocOptimizationPluginChunk(struct _cmsContext_struct * ctx,const struct _cmsContext_struct * src)1892 void _cmsAllocOptimizationPluginChunk(struct _cmsContext_struct* ctx,
1893 const struct _cmsContext_struct* src)
1894 {
1895 if (src != NULL) {
1896
1897 // Copy all linked list
1898 DupPluginOptimizationList(ctx, src);
1899 }
1900 else {
1901 static _cmsOptimizationPluginChunkType OptimizationPluginChunkType = { NULL };
1902 ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx ->MemPool, &OptimizationPluginChunkType, sizeof(_cmsOptimizationPluginChunkType));
1903 }
1904 }
1905
1906
1907 // Register new ways to optimize
_cmsRegisterOptimizationPlugin(cmsContext ContextID,cmsPluginBase * Data)1908 cmsBool _cmsRegisterOptimizationPlugin(cmsContext ContextID, cmsPluginBase* Data)
1909 {
1910 cmsPluginOptimization* Plugin = (cmsPluginOptimization*) Data;
1911 _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
1912 _cmsOptimizationCollection* fl;
1913
1914 if (Data == NULL) {
1915
1916 ctx->OptimizationCollection = NULL;
1917 return TRUE;
1918 }
1919
1920 // Optimizer callback is required
1921 if (Plugin ->OptimizePtr == NULL) return FALSE;
1922
1923 fl = (_cmsOptimizationCollection*) _cmsPluginMalloc(ContextID, sizeof(_cmsOptimizationCollection));
1924 if (fl == NULL) return FALSE;
1925
1926 // Copy the parameters
1927 fl ->OptimizePtr = Plugin ->OptimizePtr;
1928
1929 // Keep linked list
1930 fl ->Next = ctx->OptimizationCollection;
1931
1932 // Set the head
1933 ctx ->OptimizationCollection = fl;
1934
1935 // All is ok
1936 return TRUE;
1937 }
1938
1939 // The entry point for LUT optimization
_cmsOptimizePipeline(cmsContext ContextID,cmsPipeline ** PtrLut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1940 cmsBool CMSEXPORT _cmsOptimizePipeline(cmsContext ContextID,
1941 cmsPipeline** PtrLut,
1942 cmsUInt32Number Intent,
1943 cmsUInt32Number* InputFormat,
1944 cmsUInt32Number* OutputFormat,
1945 cmsUInt32Number* dwFlags)
1946 {
1947 _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
1948 _cmsOptimizationCollection* Opts;
1949 cmsBool AnySuccess = FALSE;
1950
1951 // A CLUT is being asked, so force this specific optimization
1952 if (*dwFlags & cmsFLAGS_FORCE_CLUT) {
1953
1954 PreOptimize(*PtrLut);
1955 return OptimizeByResampling(PtrLut, Intent, InputFormat, OutputFormat, dwFlags);
1956 }
1957
1958 // Anything to optimize?
1959 if ((*PtrLut) ->Elements == NULL) {
1960 _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
1961 return TRUE;
1962 }
1963
1964 // Try to get rid of identities and trivial conversions.
1965 AnySuccess = PreOptimize(*PtrLut);
1966
1967 // After removal do we end with an identity?
1968 if ((*PtrLut) ->Elements == NULL) {
1969 _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
1970 return TRUE;
1971 }
1972
1973 // Do not optimize, keep all precision
1974 if (*dwFlags & cmsFLAGS_NOOPTIMIZE)
1975 return FALSE;
1976
1977 // Try plug-in optimizations
1978 for (Opts = ctx->OptimizationCollection;
1979 Opts != NULL;
1980 Opts = Opts ->Next) {
1981
1982 // If one schema succeeded, we are done
1983 if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
1984
1985 return TRUE; // Optimized!
1986 }
1987 }
1988
1989 // Try built-in optimizations
1990 for (Opts = DefaultOptimization;
1991 Opts != NULL;
1992 Opts = Opts ->Next) {
1993
1994 if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
1995
1996 return TRUE;
1997 }
1998 }
1999
2000 // Only simple optimizations succeeded
2001 return AnySuccess;
2002 }
2003
2004
2005
2006