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 //
33 // Little Color Management System
34 // Copyright (c) 1998-2011 Marti Maria Saguer
35 //
36 // Permission is hereby granted, free of charge, to any person obtaining
37 // a copy of this software and associated documentation files (the "Software"),
38 // to deal in the Software without restriction, including without limitation
39 // the rights to use, copy, modify, merge, publish, distribute, sublicense,
40 // and/or sell copies of the Software, and to permit persons to whom the Software
41 // is furnished to do so, subject to the following conditions:
42 //
43 // The above copyright notice and this permission notice shall be included in
44 // all copies or substantial portions of the Software.
45 //
46 // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
47 // EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
48 // THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
49 // NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
50 // LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
51 // OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
52 // WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
53 //
54 //---------------------------------------------------------------------------------
55 //
56
57 #include "lcms2_internal.h"
58
59
60 //----------------------------------------------------------------------------------
61
62 // Optimization for 8 bits, Shaper-CLUT (3 inputs only)
63 typedef struct {
64
65 cmsContext ContextID;
66
67 const cmsInterpParams* p; // Tetrahedrical interpolation parameters. This is a not-owned pointer.
68
69 cmsUInt16Number rx[256], ry[256], rz[256];
70 cmsUInt32Number X0[256], Y0[256], Z0[256]; // Precomputed nodes and offsets for 8-bit input data
71
72
73 } Prelin8Data;
74
75
76 // Generic optimization for 16 bits Shaper-CLUT-Shaper (any inputs)
77 typedef struct {
78
79 cmsContext ContextID;
80
81 // Number of channels
82 int nInputs;
83 int nOutputs;
84
85 _cmsInterpFn16 EvalCurveIn16[MAX_INPUT_DIMENSIONS]; // The maximum number of input channels is known in advance
86 cmsInterpParams* ParamsCurveIn16[MAX_INPUT_DIMENSIONS];
87
88 _cmsInterpFn16 EvalCLUT; // The evaluator for 3D grid
89 const cmsInterpParams* CLUTparams; // (not-owned pointer)
90
91
92 _cmsInterpFn16* EvalCurveOut16; // Points to an array of curve evaluators in 16 bits (not-owned pointer)
93 cmsInterpParams** ParamsCurveOut16; // Points to an array of references to interpolation params (not-owned pointer)
94
95
96 } Prelin16Data;
97
98
99 // Optimization for matrix-shaper in 8 bits. Numbers are operated in n.14 signed, tables are stored in 1.14 fixed
100
101 typedef cmsInt32Number cmsS1Fixed14Number; // Note that this may hold more than 16 bits!
102
103 #define DOUBLE_TO_1FIXED14(x) ((cmsS1Fixed14Number) floor((x) * 16384.0 + 0.5))
104
105 typedef struct {
106
107 cmsContext ContextID;
108
109 cmsS1Fixed14Number Shaper1R[256]; // from 0..255 to 1.14 (0.0...1.0)
110 cmsS1Fixed14Number Shaper1G[256];
111 cmsS1Fixed14Number Shaper1B[256];
112
113 cmsS1Fixed14Number Mat[3][3]; // n.14 to n.14 (needs a saturation after that)
114 cmsS1Fixed14Number Off[3];
115
116 cmsUInt16Number Shaper2R[16385]; // 1.14 to 0..255
117 cmsUInt16Number Shaper2G[16385];
118 cmsUInt16Number Shaper2B[16385];
119
120 } MatShaper8Data;
121
122 // Curves, optimization is shared between 8 and 16 bits
123 typedef struct {
124
125 cmsContext ContextID;
126
127 int nCurves; // Number of curves
128 int nElements; // Elements in curves
129 cmsUInt16Number** Curves; // Points to a dynamically allocated array
130
131 } Curves16Data;
132
133
134 // Simple optimizations ----------------------------------------------------------------------------------------------------------
135
136
137 // Remove an element in linked chain
138 static
139 void _RemoveElement(cmsStage** head)
140 {
141 cmsStage* mpe = *head;
142 cmsStage* next = mpe ->Next;
143 *head = next;
144 cmsStageFree(mpe);
145 }
146
147 // Remove all identities in chain. Note that pt actually is a double pointer to the element that holds the pointer.
148 static
149 cmsBool _Remove1Op(cmsPipeline* Lut, cmsStageSignature UnaryOp)
150 {
151 cmsStage** pt = &Lut ->Elements;
152 cmsBool AnyOpt = FALSE;
153
154 while (*pt != NULL) {
155
156 if ((*pt) ->Implements == UnaryOp) {
157 _RemoveElement(pt);
158 AnyOpt = TRUE;
159 }
160 else
161 pt = &((*pt) -> Next);
162 }
163
164 return AnyOpt;
165 }
166
167 // Same, but only if two adjacent elements are found
168 static
169 cmsBool _Remove2Op(cmsPipeline* Lut, cmsStageSignature Op1, cmsStageSignature Op2)
170 {
171 cmsStage** pt1;
172 cmsStage** pt2;
173 cmsBool AnyOpt = FALSE;
174
175 pt1 = &Lut ->Elements;
176 if (*pt1 == NULL) return AnyOpt;
177
178 while (*pt1 != NULL) {
179
180 pt2 = &((*pt1) -> Next);
181 if (*pt2 == NULL) return AnyOpt;
182
183 if ((*pt1) ->Implements == Op1 && (*pt2) ->Implements == Op2) {
184 _RemoveElement(pt2);
185 _RemoveElement(pt1);
186 AnyOpt = TRUE;
187 }
188 else
189 pt1 = &((*pt1) -> Next);
190 }
191
192 return AnyOpt;
193 }
194
195
196 static
197 cmsBool CloseEnoughFloat(cmsFloat64Number a, cmsFloat64Number b)
198 {
199 return fabs(b - a) < 0.00001f;
200 }
201
202 static
203 cmsBool isFloatMatrixIdentity(const cmsMAT3* a)
204 {
205 cmsMAT3 Identity;
206 int i, j;
207
208 _cmsMAT3identity(&Identity);
209
210 for (i = 0; i < 3; i++)
211 for (j = 0; j < 3; j++)
212 if (!CloseEnoughFloat(a->v[i].n[j], Identity.v[i].n[j])) return FALSE;
213
214 return TRUE;
215 }
216 // if two adjacent matrices are found, multiply them.
217 static
218 cmsBool _MultiplyMatrix(cmsPipeline* Lut)
219 {
220 cmsStage** pt1;
221 cmsStage** pt2;
222 cmsStage* chain;
223 cmsBool AnyOpt = FALSE;
224
225 pt1 = &Lut->Elements;
226 if (*pt1 == NULL) return AnyOpt;
227
228 while (*pt1 != NULL) {
229
230 pt2 = &((*pt1)->Next);
231 if (*pt2 == NULL) return AnyOpt;
232
233 if ((*pt1)->Implements == cmsSigMatrixElemType && (*pt2)->Implements == cmsSigMatrixElemType) {
234
235 // Get both matrices
236 _cmsStageMatrixData* m1 = (_cmsStageMatrixData*) cmsStageData(*pt1);
237 _cmsStageMatrixData* m2 = (_cmsStageMatrixData*) cmsStageData(*pt2);
238 cmsMAT3 res;
239
240 // Input offset and output offset should be zero to use this optimization
241 if (m1->Offset != NULL || m2 ->Offset != NULL ||
242 cmsStageInputChannels(*pt1) != 3 || cmsStageOutputChannels(*pt1) != 3 ||
243 cmsStageInputChannels(*pt2) != 3 || cmsStageOutputChannels(*pt2) != 3)
244 return FALSE;
245
246 // Multiply both matrices to get the result
247 _cmsMAT3per(&res, (cmsMAT3*)m2->Double, (cmsMAT3*)m1->Double);
248
249 // Get the next in chain afer the matrices
250 chain = (*pt2)->Next;
251
252 // Remove both matrices
253 _RemoveElement(pt2);
254 _RemoveElement(pt1);
255
256 // Now what if the result is a plain identity?
257 if (!isFloatMatrixIdentity(&res)) {
258
259 // We can not get rid of full matrix
260 cmsStage* Multmat = cmsStageAllocMatrix(Lut->ContextID, 3, 3, (const cmsFloat64Number*) &res, NULL);
261
262 // Recover the chain
263 if (Multmat != NULL) {
264 Multmat->Next = chain;
265 }
266 *pt1 = Multmat;
267 }
268
269 AnyOpt = TRUE;
270 }
271 else
272 pt1 = &((*pt1)->Next);
273 }
274
275 return AnyOpt;
276 }
277
278
279 // Preoptimize just gets rif of no-ops coming paired. Conversion from v2 to v4 followed
280 // by a v4 to v2 and vice-versa. The elements are then discarded.
281 static
282 cmsBool PreOptimize(cmsPipeline* Lut)
283 {
284 cmsBool AnyOpt = FALSE, Opt;
285
286 do {
287
288 Opt = FALSE;
289
290 // Remove all identities
291 Opt |= _Remove1Op(Lut, cmsSigIdentityElemType);
292
293 // Remove XYZ2Lab followed by Lab2XYZ
294 Opt |= _Remove2Op(Lut, cmsSigXYZ2LabElemType, cmsSigLab2XYZElemType);
295
296 // Remove Lab2XYZ followed by XYZ2Lab
297 Opt |= _Remove2Op(Lut, cmsSigLab2XYZElemType, cmsSigXYZ2LabElemType);
298
299 // Remove V4 to V2 followed by V2 to V4
300 Opt |= _Remove2Op(Lut, cmsSigLabV4toV2, cmsSigLabV2toV4);
301
302 // Remove V2 to V4 followed by V4 to V2
303 Opt |= _Remove2Op(Lut, cmsSigLabV2toV4, cmsSigLabV4toV2);
304
305 // Remove float pcs Lab conversions
306 Opt |= _Remove2Op(Lut, cmsSigLab2FloatPCS, cmsSigFloatPCS2Lab);
307
308 // Remove float pcs Lab conversions
309 Opt |= _Remove2Op(Lut, cmsSigXYZ2FloatPCS, cmsSigFloatPCS2XYZ);
310
311 // Simplify matrix.
312 Opt |= _MultiplyMatrix(Lut);
313
314 if (Opt) AnyOpt = TRUE;
315
316 } while (Opt);
317
318 return AnyOpt;
319 }
320
321 static
322 void Eval16nop1D(register const cmsUInt16Number Input[],
323 register cmsUInt16Number Output[],
324 register const struct _cms_interp_struc* p)
325 {
326 Output[0] = Input[0];
327
328 cmsUNUSED_PARAMETER(p);
329 }
330
331 static
332 void PrelinEval16(register const cmsUInt16Number Input[],
333 register cmsUInt16Number Output[],
334 register const void* D)
335 {
336 Prelin16Data* p16 = (Prelin16Data*) D;
337 cmsUInt16Number StageABC[MAX_INPUT_DIMENSIONS];
338 cmsUInt16Number StageDEF[cmsMAXCHANNELS];
339 int i;
340
341 for (i=0; i < p16 ->nInputs; i++) {
342
343 p16 ->EvalCurveIn16[i](&Input[i], &StageABC[i], p16 ->ParamsCurveIn16[i]);
344 }
345
346 p16 ->EvalCLUT(StageABC, StageDEF, p16 ->CLUTparams);
347
348 for (i=0; i < p16 ->nOutputs; i++) {
349
350 p16 ->EvalCurveOut16[i](&StageDEF[i], &Output[i], p16 ->ParamsCurveOut16[i]);
351 }
352 }
353
354
355 static
356 void PrelinOpt16free(cmsContext ContextID, void* ptr)
357 {
358 Prelin16Data* p16 = (Prelin16Data*) ptr;
359
360 _cmsFree(ContextID, p16 ->EvalCurveOut16);
361 _cmsFree(ContextID, p16 ->ParamsCurveOut16);
362
363 _cmsFree(ContextID, p16);
364 }
365
366 static
367 void* Prelin16dup(cmsContext ContextID, const void* ptr)
368 {
369 Prelin16Data* p16 = (Prelin16Data*) ptr;
370 Prelin16Data* Duped = (Prelin16Data*) _cmsDupMem(ContextID, p16, sizeof(Prelin16Data));
371
372 if (Duped == NULL) return NULL;
373
374 Duped->EvalCurveOut16 = (_cmsInterpFn16*) _cmsDupMem(ContextID, p16->EvalCurveOut16, p16->nOutputs * sizeof(_cmsInterpFn16));
375 Duped->ParamsCurveOut16 = (cmsInterpParams**)_cmsDupMem(ContextID, p16->ParamsCurveOut16, p16->nOutputs * sizeof(cmsInterpParams*));
376
377 return Duped;
378 }
379
380
381 static
382 Prelin16Data* PrelinOpt16alloc(cmsContext ContextID,
383 const cmsInterpParams* ColorMap,
384 int nInputs, cmsToneCurve** In,
385 int nOutputs, cmsToneCurve** Out )
386 {
387 int i;
388 Prelin16Data* p16 = (Prelin16Data*)_cmsMallocZero(ContextID, sizeof(Prelin16Data));
389 if (p16 == NULL) return NULL;
390
391 p16 ->nInputs = nInputs;
392 p16 -> nOutputs = nOutputs;
393
394
395 for (i=0; i < nInputs; i++) {
396
397 if (In == NULL) {
398 p16 -> ParamsCurveIn16[i] = NULL;
399 p16 -> EvalCurveIn16[i] = Eval16nop1D;
400
401 }
402 else {
403 p16 -> ParamsCurveIn16[i] = In[i] ->InterpParams;
404 p16 -> EvalCurveIn16[i] = p16 ->ParamsCurveIn16[i]->Interpolation.Lerp16;
405 }
406 }
407
408 p16 ->CLUTparams = ColorMap;
409 p16 ->EvalCLUT = ColorMap ->Interpolation.Lerp16;
410
411
412 p16 -> EvalCurveOut16 = (_cmsInterpFn16*) _cmsCalloc(ContextID, nOutputs, sizeof(_cmsInterpFn16));
413 p16 -> ParamsCurveOut16 = (cmsInterpParams**) _cmsCalloc(ContextID, nOutputs, sizeof(cmsInterpParams* ));
414
415 for (i=0; i < nOutputs; i++) {
416
417 if (Out == NULL) {
418 p16 ->ParamsCurveOut16[i] = NULL;
419 p16 -> EvalCurveOut16[i] = Eval16nop1D;
420 }
421 else {
422
423 p16 ->ParamsCurveOut16[i] = Out[i] ->InterpParams;
424 p16 -> EvalCurveOut16[i] = p16 ->ParamsCurveOut16[i]->Interpolation.Lerp16;
425 }
426 }
427
428 return p16;
429 }
430
431
432
433 // Resampling ---------------------------------------------------------------------------------
434
435 #define PRELINEARIZATION_POINTS 4096
436
437 // Sampler implemented by another LUT. This is a clean way to precalculate the devicelink 3D CLUT for
438 // almost any transform. We use floating point precision and then convert from floating point to 16 bits.
439 static
440 int XFormSampler16(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void* Cargo)
441 {
442 cmsPipeline* Lut = (cmsPipeline*) Cargo;
443 cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
444 cmsUInt32Number i;
445
446 _cmsAssert(Lut -> InputChannels < cmsMAXCHANNELS);
447 _cmsAssert(Lut -> OutputChannels < cmsMAXCHANNELS);
448
449 // From 16 bit to floating point
450 for (i=0; i < Lut ->InputChannels; i++)
451 InFloat[i] = (cmsFloat32Number) (In[i] / 65535.0);
452
453 // Evaluate in floating point
454 cmsPipelineEvalFloat(InFloat, OutFloat, Lut);
455
456 // Back to 16 bits representation
457 for (i=0; i < Lut ->OutputChannels; i++)
458 Out[i] = _cmsQuickSaturateWord(OutFloat[i] * 65535.0);
459
460 // Always succeed
461 return TRUE;
462 }
463
464 // Try to see if the curves of a given MPE are linear
465 static
466 cmsBool AllCurvesAreLinear(cmsStage* mpe)
467 {
468 cmsToneCurve** Curves;
469 cmsUInt32Number i, n;
470
471 Curves = _cmsStageGetPtrToCurveSet(mpe);
472 if (Curves == NULL) return FALSE;
473
474 n = cmsStageOutputChannels(mpe);
475
476 for (i=0; i < n; i++) {
477 if (!cmsIsToneCurveLinear(Curves[i])) return FALSE;
478 }
479
480 return TRUE;
481 }
482
483 // This function replaces a specific node placed in "At" by the "Value" numbers. Its purpose
484 // is to fix scum dot on broken profiles/transforms. Works on 1, 3 and 4 channels
485 static
486 cmsBool PatchLUT(cmsStage* CLUT, cmsUInt16Number At[], cmsUInt16Number Value[],
487 int nChannelsOut, int nChannelsIn)
488 {
489 _cmsStageCLutData* Grid = (_cmsStageCLutData*) CLUT ->Data;
490 cmsInterpParams* p16 = Grid ->Params;
491 cmsFloat64Number px, py, pz, pw;
492 int x0, y0, z0, w0;
493 int i, index;
494
495 if (CLUT -> Type != cmsSigCLutElemType) {
496 cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) Attempt to PatchLUT on non-lut stage");
497 return FALSE;
498 }
499
500 if (nChannelsIn == 4) {
501
502 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
503 py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
504 pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
505 pw = ((cmsFloat64Number) At[3] * (p16->Domain[3])) / 65535.0;
506
507 x0 = (int) floor(px);
508 y0 = (int) floor(py);
509 z0 = (int) floor(pz);
510 w0 = (int) floor(pw);
511
512 if (((px - x0) != 0) ||
513 ((py - y0) != 0) ||
514 ((pz - z0) != 0) ||
515 ((pw - w0) != 0)) return FALSE; // Not on exact node
516
517 index = p16 -> opta[3] * x0 +
518 p16 -> opta[2] * y0 +
519 p16 -> opta[1] * z0 +
520 p16 -> opta[0] * w0;
521 }
522 else
523 if (nChannelsIn == 3) {
524
525 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
526 py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
527 pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
528
529 x0 = (int) floor(px);
530 y0 = (int) floor(py);
531 z0 = (int) floor(pz);
532
533 if (((px - x0) != 0) ||
534 ((py - y0) != 0) ||
535 ((pz - z0) != 0)) return FALSE; // Not on exact node
536
537 index = p16 -> opta[2] * x0 +
538 p16 -> opta[1] * y0 +
539 p16 -> opta[0] * z0;
540 }
541 else
542 if (nChannelsIn == 1) {
543
544 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
545
546 x0 = (int) floor(px);
547
548 if (((px - x0) != 0)) return FALSE; // Not on exact node
549
550 index = p16 -> opta[0] * x0;
551 }
552 else {
553 cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) %d Channels are not supported on PatchLUT", nChannelsIn);
554 return FALSE;
555 }
556
557 for (i=0; i < nChannelsOut; i++)
558 Grid -> Tab.T[index + i] = Value[i];
559
560 return TRUE;
561 }
562
563 // Auxiliar, to see if two values are equal or very different
564 static
565 cmsBool WhitesAreEqual(int n, cmsUInt16Number White1[], cmsUInt16Number White2[] )
566 {
567 int i;
568
569 for (i=0; i < n; i++) {
570
571 if (abs(White1[i] - White2[i]) > 0xf000) return TRUE; // Values are so extremly different that the fixup should be avoided
572 if (White1[i] != White2[i]) return FALSE;
573 }
574 return TRUE;
575 }
576
577
578 // Locate the node for the white point and fix it to pure white in order to avoid scum dot.
579 static
580 cmsBool FixWhiteMisalignment(cmsPipeline* Lut, cmsColorSpaceSignature EntryColorSpace, cmsColorSpaceSignature ExitColorSpace)
581 {
582 cmsUInt16Number *WhitePointIn, *WhitePointOut;
583 cmsUInt16Number WhiteIn[cmsMAXCHANNELS], WhiteOut[cmsMAXCHANNELS], ObtainedOut[cmsMAXCHANNELS];
584 cmsUInt32Number i, nOuts, nIns;
585 cmsStage *PreLin = NULL, *CLUT = NULL, *PostLin = NULL;
586
587 if (!_cmsEndPointsBySpace(EntryColorSpace,
588 &WhitePointIn, NULL, &nIns)) return FALSE;
589
590 if (!_cmsEndPointsBySpace(ExitColorSpace,
591 &WhitePointOut, NULL, &nOuts)) return FALSE;
592
593 // It needs to be fixed?
594 if (Lut ->InputChannels != nIns) return FALSE;
595 if (Lut ->OutputChannels != nOuts) return FALSE;
596
597 cmsPipelineEval16(WhitePointIn, ObtainedOut, Lut);
598
599 if (WhitesAreEqual(nOuts, WhitePointOut, ObtainedOut)) return TRUE; // whites already match
600
601 // Check if the LUT comes as Prelin, CLUT or Postlin. We allow all combinations
602 if (!cmsPipelineCheckAndRetreiveStages(Lut, 3, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, &PreLin, &CLUT, &PostLin))
603 if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCurveSetElemType, cmsSigCLutElemType, &PreLin, &CLUT))
604 if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCLutElemType, cmsSigCurveSetElemType, &CLUT, &PostLin))
605 if (!cmsPipelineCheckAndRetreiveStages(Lut, 1, cmsSigCLutElemType, &CLUT))
606 return FALSE;
607
608 // We need to interpolate white points of both, pre and post curves
609 if (PreLin) {
610
611 cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PreLin);
612
613 for (i=0; i < nIns; i++) {
614 WhiteIn[i] = cmsEvalToneCurve16(Curves[i], WhitePointIn[i]);
615 }
616 }
617 else {
618 for (i=0; i < nIns; i++)
619 WhiteIn[i] = WhitePointIn[i];
620 }
621
622 // If any post-linearization, we need to find how is represented white before the curve, do
623 // a reverse interpolation in this case.
624 if (PostLin) {
625
626 cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PostLin);
627
628 for (i=0; i < nOuts; i++) {
629
630 cmsToneCurve* InversePostLin = cmsReverseToneCurve(Curves[i]);
631 if (InversePostLin == NULL) {
632 WhiteOut[i] = WhitePointOut[i];
633
634 } else {
635
636 WhiteOut[i] = cmsEvalToneCurve16(InversePostLin, WhitePointOut[i]);
637 cmsFreeToneCurve(InversePostLin);
638 }
639 }
640 }
641 else {
642 for (i=0; i < nOuts; i++)
643 WhiteOut[i] = WhitePointOut[i];
644 }
645
646 // Ok, proceed with patching. May fail and we don't care if it fails
647 PatchLUT(CLUT, WhiteIn, WhiteOut, nOuts, nIns);
648
649 return TRUE;
650 }
651
652 // -----------------------------------------------------------------------------------------------------------------------------------------------
653 // This function creates simple LUT from complex ones. The generated LUT has an optional set of
654 // prelinearization curves, a CLUT of nGridPoints and optional postlinearization tables.
655 // These curves have to exist in the original LUT in order to be used in the simplified output.
656 // Caller may also use the flags to allow this feature.
657 // LUTS with all curves will be simplified to a single curve. Parametric curves are lost.
658 // This function should be used on 16-bits LUTS only, as floating point losses precision when simplified
659 // -----------------------------------------------------------------------------------------------------------------------------------------------
660
661 static
662 cmsBool OptimizeByResampling(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
663 {
664 cmsPipeline* Src = NULL;
665 cmsPipeline* Dest = NULL;
666 cmsStage* mpe;
667 cmsStage* CLUT;
668 cmsStage *KeepPreLin = NULL, *KeepPostLin = NULL;
669 int nGridPoints;
670 cmsColorSpaceSignature ColorSpace, OutputColorSpace;
671 cmsStage *NewPreLin = NULL;
672 cmsStage *NewPostLin = NULL;
673 _cmsStageCLutData* DataCLUT;
674 cmsToneCurve** DataSetIn;
675 cmsToneCurve** DataSetOut;
676 Prelin16Data* p16;
677
678 // This is a loosy optimization! does not apply in floating-point cases
679 if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
680
681 ColorSpace = _cmsICCcolorSpace(T_COLORSPACE(*InputFormat));
682 OutputColorSpace = _cmsICCcolorSpace(T_COLORSPACE(*OutputFormat));
683 nGridPoints = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);
684
685 // For empty LUTs, 2 points are enough
686 if (cmsPipelineStageCount(*Lut) == 0)
687 nGridPoints = 2;
688
689 Src = *Lut;
690
691 // Named color pipelines cannot be optimized either
692 for (mpe = cmsPipelineGetPtrToFirstStage(Src);
693 mpe != NULL;
694 mpe = cmsStageNext(mpe)) {
695 if (cmsStageType(mpe) == cmsSigNamedColorElemType) return FALSE;
696 }
697
698 // Allocate an empty LUT
699 Dest = cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
700 if (!Dest) return FALSE;
701
702 // Prelinearization tables are kept unless indicated by flags
703 if (*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION) {
704
705 // Get a pointer to the prelinearization element
706 cmsStage* PreLin = cmsPipelineGetPtrToFirstStage(Src);
707
708 // Check if suitable
709 if (PreLin ->Type == cmsSigCurveSetElemType) {
710
711 // Maybe this is a linear tram, so we can avoid the whole stuff
712 if (!AllCurvesAreLinear(PreLin)) {
713
714 // All seems ok, proceed.
715 NewPreLin = cmsStageDup(PreLin);
716 if(!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, NewPreLin))
717 goto Error;
718
719 // Remove prelinearization. Since we have duplicated the curve
720 // in destination LUT, the sampling shoud be applied after this stage.
721 cmsPipelineUnlinkStage(Src, cmsAT_BEGIN, &KeepPreLin);
722 }
723 }
724 }
725
726 // Allocate the CLUT
727 CLUT = cmsStageAllocCLut16bit(Src ->ContextID, nGridPoints, Src ->InputChannels, Src->OutputChannels, NULL);
728 if (CLUT == NULL) return FALSE;
729
730 // Add the CLUT to the destination LUT
731 if (!cmsPipelineInsertStage(Dest, cmsAT_END, CLUT)) {
732 goto Error;
733 }
734
735 // Postlinearization tables are kept unless indicated by flags
736 if (*dwFlags & cmsFLAGS_CLUT_POST_LINEARIZATION) {
737
738 // Get a pointer to the postlinearization if present
739 cmsStage* PostLin = cmsPipelineGetPtrToLastStage(Src);
740
741 // Check if suitable
742 if (cmsStageType(PostLin) == cmsSigCurveSetElemType) {
743
744 // Maybe this is a linear tram, so we can avoid the whole stuff
745 if (!AllCurvesAreLinear(PostLin)) {
746
747 // All seems ok, proceed.
748 NewPostLin = cmsStageDup(PostLin);
749 if (!cmsPipelineInsertStage(Dest, cmsAT_END, NewPostLin))
750 goto Error;
751
752 // In destination LUT, the sampling shoud be applied after this stage.
753 cmsPipelineUnlinkStage(Src, cmsAT_END, &KeepPostLin);
754 }
755 }
756 }
757
758 // Now its time to do the sampling. We have to ignore pre/post linearization
759 // The source LUT whithout pre/post curves is passed as parameter.
760 if (!cmsStageSampleCLut16bit(CLUT, XFormSampler16, (void*) Src, 0)) {
761 Error:
762 // Ops, something went wrong, Restore stages
763 if (KeepPreLin != NULL) {
764 if (!cmsPipelineInsertStage(Src, cmsAT_BEGIN, KeepPreLin)) {
765 _cmsAssert(0); // This never happens
766 }
767 }
768 if (KeepPostLin != NULL) {
769 if (!cmsPipelineInsertStage(Src, cmsAT_END, KeepPostLin)) {
770 _cmsAssert(0); // This never happens
771 }
772 }
773 cmsPipelineFree(Dest);
774 return FALSE;
775 }
776
777 // Done.
778
779 if (KeepPreLin != NULL) cmsStageFree(KeepPreLin);
780 if (KeepPostLin != NULL) cmsStageFree(KeepPostLin);
781 cmsPipelineFree(Src);
782
783 DataCLUT = (_cmsStageCLutData*) CLUT ->Data;
784
785 if (NewPreLin == NULL) DataSetIn = NULL;
786 else DataSetIn = ((_cmsStageToneCurvesData*) NewPreLin ->Data) ->TheCurves;
787
788 if (NewPostLin == NULL) DataSetOut = NULL;
789 else DataSetOut = ((_cmsStageToneCurvesData*) NewPostLin ->Data) ->TheCurves;
790
791
792 if (DataSetIn == NULL && DataSetOut == NULL) {
793
794 _cmsPipelineSetOptimizationParameters(Dest, (_cmsOPTeval16Fn) DataCLUT->Params->Interpolation.Lerp16, DataCLUT->Params, NULL, NULL);
795 }
796 else {
797
798 p16 = PrelinOpt16alloc(Dest ->ContextID,
799 DataCLUT ->Params,
800 Dest ->InputChannels,
801 DataSetIn,
802 Dest ->OutputChannels,
803 DataSetOut);
804
805 _cmsPipelineSetOptimizationParameters(Dest, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
806 }
807
808
809 // Don't fix white on absolute colorimetric
810 if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
811 *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
812
813 if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
814
815 FixWhiteMisalignment(Dest, ColorSpace, OutputColorSpace);
816 }
817
818 *Lut = Dest;
819 return TRUE;
820
821 cmsUNUSED_PARAMETER(Intent);
822 }
823
824
825 // -----------------------------------------------------------------------------------------------------------------------------------------------
826 // Fixes the gamma balancing of transform. This is described in my paper "Prelinearization Stages on
827 // Color-Management Application-Specific Integrated Circuits (ASICs)" presented at NIP24. It only works
828 // for RGB transforms. See the paper for more details
829 // -----------------------------------------------------------------------------------------------------------------------------------------------
830
831
832 // Normalize endpoints by slope limiting max and min. This assures endpoints as well.
833 // Descending curves are handled as well.
834 static
835 void SlopeLimiting(cmsToneCurve* g)
836 {
837 int BeginVal, EndVal;
838 int AtBegin = (int) floor((cmsFloat64Number) g ->nEntries * 0.02 + 0.5); // Cutoff at 2%
839 int AtEnd = g ->nEntries - AtBegin - 1; // And 98%
840 cmsFloat64Number Val, Slope, beta;
841 int i;
842
843 if (cmsIsToneCurveDescending(g)) {
844 BeginVal = 0xffff; EndVal = 0;
845 }
846 else {
847 BeginVal = 0; EndVal = 0xffff;
848 }
849
850 // Compute slope and offset for begin of curve
851 Val = g ->Table16[AtBegin];
852 Slope = (Val - BeginVal) / AtBegin;
853 beta = Val - Slope * AtBegin;
854
855 for (i=0; i < AtBegin; i++)
856 g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
857
858 // Compute slope and offset for the end
859 Val = g ->Table16[AtEnd];
860 Slope = (EndVal - Val) / AtBegin; // AtBegin holds the X interval, which is same in both cases
861 beta = Val - Slope * AtEnd;
862
863 for (i = AtEnd; i < (int) g ->nEntries; i++)
864 g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
865 }
866
867
868 // Precomputes tables for 8-bit on input devicelink.
869 static
870 Prelin8Data* PrelinOpt8alloc(cmsContext ContextID, const cmsInterpParams* p, cmsToneCurve* G[3])
871 {
872 int i;
873 cmsUInt16Number Input[3];
874 cmsS15Fixed16Number v1, v2, v3;
875 Prelin8Data* p8;
876
877 p8 = (Prelin8Data*)_cmsMallocZero(ContextID, sizeof(Prelin8Data));
878 if (p8 == NULL) return NULL;
879
880 // Since this only works for 8 bit input, values comes always as x * 257,
881 // we can safely take msb byte (x << 8 + x)
882
883 for (i=0; i < 256; i++) {
884
885 if (G != NULL) {
886
887 // Get 16-bit representation
888 Input[0] = cmsEvalToneCurve16(G[0], FROM_8_TO_16(i));
889 Input[1] = cmsEvalToneCurve16(G[1], FROM_8_TO_16(i));
890 Input[2] = cmsEvalToneCurve16(G[2], FROM_8_TO_16(i));
891 }
892 else {
893 Input[0] = FROM_8_TO_16(i);
894 Input[1] = FROM_8_TO_16(i);
895 Input[2] = FROM_8_TO_16(i);
896 }
897
898
899 // Move to 0..1.0 in fixed domain
900 v1 = _cmsToFixedDomain(Input[0] * p -> Domain[0]);
901 v2 = _cmsToFixedDomain(Input[1] * p -> Domain[1]);
902 v3 = _cmsToFixedDomain(Input[2] * p -> Domain[2]);
903
904 // Store the precalculated table of nodes
905 p8 ->X0[i] = (p->opta[2] * FIXED_TO_INT(v1));
906 p8 ->Y0[i] = (p->opta[1] * FIXED_TO_INT(v2));
907 p8 ->Z0[i] = (p->opta[0] * FIXED_TO_INT(v3));
908
909 // Store the precalculated table of offsets
910 p8 ->rx[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v1);
911 p8 ->ry[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v2);
912 p8 ->rz[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v3);
913 }
914
915 p8 ->ContextID = ContextID;
916 p8 ->p = p;
917
918 return p8;
919 }
920
921 static
922 void Prelin8free(cmsContext ContextID, void* ptr)
923 {
924 _cmsFree(ContextID, ptr);
925 }
926
927 static
928 void* Prelin8dup(cmsContext ContextID, const void* ptr)
929 {
930 return _cmsDupMem(ContextID, ptr, sizeof(Prelin8Data));
931 }
932
933
934
935 // A optimized interpolation for 8-bit input.
936 #define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
937 static
938 void PrelinEval8(register const cmsUInt16Number Input[],
939 register cmsUInt16Number Output[],
940 register const void* D)
941 {
942
943 cmsUInt8Number r, g, b;
944 cmsS15Fixed16Number rx, ry, rz;
945 cmsS15Fixed16Number c0, c1, c2, c3, Rest;
946 int OutChan;
947 register cmsS15Fixed16Number X0, X1, Y0, Y1, Z0, Z1;
948 Prelin8Data* p8 = (Prelin8Data*) D;
949 register const cmsInterpParams* p = p8 ->p;
950 int TotalOut = p -> nOutputs;
951 const cmsUInt16Number* LutTable = (const cmsUInt16Number*) p->Table;
952
953 r = Input[0] >> 8;
954 g = Input[1] >> 8;
955 b = Input[2] >> 8;
956
957 X0 = X1 = p8->X0[r];
958 Y0 = Y1 = p8->Y0[g];
959 Z0 = Z1 = p8->Z0[b];
960
961 rx = p8 ->rx[r];
962 ry = p8 ->ry[g];
963 rz = p8 ->rz[b];
964
965 X1 = X0 + ((rx == 0) ? 0 : p ->opta[2]);
966 Y1 = Y0 + ((ry == 0) ? 0 : p ->opta[1]);
967 Z1 = Z0 + ((rz == 0) ? 0 : p ->opta[0]);
968
969
970 // These are the 6 Tetrahedral
971 for (OutChan=0; OutChan < TotalOut; OutChan++) {
972
973 c0 = DENS(X0, Y0, Z0);
974
975 if (rx >= ry && ry >= rz)
976 {
977 c1 = DENS(X1, Y0, Z0) - c0;
978 c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
979 c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
980 }
981 else
982 if (rx >= rz && rz >= ry)
983 {
984 c1 = DENS(X1, Y0, Z0) - c0;
985 c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
986 c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
987 }
988 else
989 if (rz >= rx && rx >= ry)
990 {
991 c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
992 c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
993 c3 = DENS(X0, Y0, Z1) - c0;
994 }
995 else
996 if (ry >= rx && rx >= rz)
997 {
998 c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
999 c2 = DENS(X0, Y1, Z0) - c0;
1000 c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
1001 }
1002 else
1003 if (ry >= rz && rz >= rx)
1004 {
1005 c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
1006 c2 = DENS(X0, Y1, Z0) - c0;
1007 c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);
1008 }
1009 else
1010 if (rz >= ry && ry >= rx)
1011 {
1012 c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
1013 c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
1014 c3 = DENS(X0, Y0, Z1) - c0;
1015 }
1016 else {
1017 c1 = c2 = c3 = 0;
1018 }
1019
1020
1021 Rest = c1 * rx + c2 * ry + c3 * rz + 0x8001;
1022 Output[OutChan] = (cmsUInt16Number)c0 + ((Rest + (Rest>>16))>>16);
1023
1024 }
1025 }
1026
1027 #undef DENS
1028
1029
1030 // Curves that contain wide empty areas are not optimizeable
1031 static
1032 cmsBool IsDegenerated(const cmsToneCurve* g)
1033 {
1034 int i, Zeros = 0, Poles = 0;
1035 int nEntries = g ->nEntries;
1036
1037 for (i=0; i < nEntries; i++) {
1038
1039 if (g ->Table16[i] == 0x0000) Zeros++;
1040 if (g ->Table16[i] == 0xffff) Poles++;
1041 }
1042
1043 if (Zeros == 1 && Poles == 1) return FALSE; // For linear tables
1044 if (Zeros > (nEntries / 4)) return TRUE; // Degenerated, mostly zeros
1045 if (Poles > (nEntries / 4)) return TRUE; // Degenerated, mostly poles
1046
1047 return FALSE;
1048 }
1049
1050 // --------------------------------------------------------------------------------------------------------------
1051 // We need xput over here
1052
1053 static
1054 cmsBool OptimizeByComputingLinearization(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1055 {
1056 cmsPipeline* OriginalLut;
1057 int nGridPoints;
1058 cmsToneCurve *Trans[cmsMAXCHANNELS], *TransReverse[cmsMAXCHANNELS];
1059 cmsUInt32Number t, i;
1060 cmsFloat32Number v, In[cmsMAXCHANNELS], Out[cmsMAXCHANNELS];
1061 cmsBool lIsSuitable, lIsLinear;
1062 cmsPipeline* OptimizedLUT = NULL, *LutPlusCurves = NULL;
1063 cmsStage* OptimizedCLUTmpe;
1064 cmsColorSpaceSignature ColorSpace, OutputColorSpace;
1065 cmsStage* OptimizedPrelinMpe;
1066 cmsStage* mpe;
1067 cmsToneCurve** OptimizedPrelinCurves;
1068 _cmsStageCLutData* OptimizedPrelinCLUT;
1069
1070
1071 // This is a loosy optimization! does not apply in floating-point cases
1072 if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
1073
1074 // Only on RGB
1075 if (T_COLORSPACE(*InputFormat) != PT_RGB) return FALSE;
1076 if (T_COLORSPACE(*OutputFormat) != PT_RGB) return FALSE;
1077
1078
1079 // On 16 bits, user has to specify the feature
1080 if (!_cmsFormatterIs8bit(*InputFormat)) {
1081 if (!(*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION)) return FALSE;
1082 }
1083
1084 OriginalLut = *Lut;
1085
1086 // Named color pipelines cannot be optimized either
1087 for (mpe = cmsPipelineGetPtrToFirstStage(OriginalLut);
1088 mpe != NULL;
1089 mpe = cmsStageNext(mpe)) {
1090 if (cmsStageType(mpe) == cmsSigNamedColorElemType) return FALSE;
1091 }
1092
1093 ColorSpace = _cmsICCcolorSpace(T_COLORSPACE(*InputFormat));
1094 OutputColorSpace = _cmsICCcolorSpace(T_COLORSPACE(*OutputFormat));
1095 nGridPoints = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);
1096
1097 // Empty gamma containers
1098 memset(Trans, 0, sizeof(Trans));
1099 memset(TransReverse, 0, sizeof(TransReverse));
1100
1101 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1102 Trans[t] = cmsBuildTabulatedToneCurve16(OriginalLut ->ContextID, PRELINEARIZATION_POINTS, NULL);
1103 if (Trans[t] == NULL) goto Error;
1104 }
1105
1106 // Populate the curves
1107 for (i=0; i < PRELINEARIZATION_POINTS; i++) {
1108
1109 v = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
1110
1111 // Feed input with a gray ramp
1112 for (t=0; t < OriginalLut ->InputChannels; t++)
1113 In[t] = v;
1114
1115 // Evaluate the gray value
1116 cmsPipelineEvalFloat(In, Out, OriginalLut);
1117
1118 // Store result in curve
1119 for (t=0; t < OriginalLut ->InputChannels; t++)
1120 Trans[t] ->Table16[i] = _cmsQuickSaturateWord(Out[t] * 65535.0);
1121 }
1122
1123 // Slope-limit the obtained curves
1124 for (t = 0; t < OriginalLut ->InputChannels; t++)
1125 SlopeLimiting(Trans[t]);
1126
1127 // Check for validity
1128 lIsSuitable = TRUE;
1129 lIsLinear = TRUE;
1130 for (t=0; (lIsSuitable && (t < OriginalLut ->InputChannels)); t++) {
1131
1132 // Exclude if already linear
1133 if (!cmsIsToneCurveLinear(Trans[t]))
1134 lIsLinear = FALSE;
1135
1136 // Exclude if non-monotonic
1137 if (!cmsIsToneCurveMonotonic(Trans[t]))
1138 lIsSuitable = FALSE;
1139
1140 if (IsDegenerated(Trans[t]))
1141 lIsSuitable = FALSE;
1142 }
1143
1144 // If it is not suitable, just quit
1145 if (!lIsSuitable) goto Error;
1146
1147 // Invert curves if possible
1148 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1149 TransReverse[t] = cmsReverseToneCurveEx(PRELINEARIZATION_POINTS, Trans[t]);
1150 if (TransReverse[t] == NULL) goto Error;
1151 }
1152
1153 // Now inset the reversed curves at the begin of transform
1154 LutPlusCurves = cmsPipelineDup(OriginalLut);
1155 if (LutPlusCurves == NULL) goto Error;
1156
1157 if (!cmsPipelineInsertStage(LutPlusCurves, cmsAT_BEGIN, cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, TransReverse)))
1158 goto Error;
1159
1160 // Create the result LUT
1161 OptimizedLUT = cmsPipelineAlloc(OriginalLut ->ContextID, OriginalLut ->InputChannels, OriginalLut ->OutputChannels);
1162 if (OptimizedLUT == NULL) goto Error;
1163
1164 OptimizedPrelinMpe = cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, Trans);
1165
1166 // Create and insert the curves at the beginning
1167 if (!cmsPipelineInsertStage(OptimizedLUT, cmsAT_BEGIN, OptimizedPrelinMpe))
1168 goto Error;
1169
1170 // Allocate the CLUT for result
1171 OptimizedCLUTmpe = cmsStageAllocCLut16bit(OriginalLut ->ContextID, nGridPoints, OriginalLut ->InputChannels, OriginalLut ->OutputChannels, NULL);
1172
1173 // Add the CLUT to the destination LUT
1174 if (!cmsPipelineInsertStage(OptimizedLUT, cmsAT_END, OptimizedCLUTmpe))
1175 goto Error;
1176
1177 // Resample the LUT
1178 if (!cmsStageSampleCLut16bit(OptimizedCLUTmpe, XFormSampler16, (void*) LutPlusCurves, 0)) goto Error;
1179
1180 // Free resources
1181 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1182
1183 if (Trans[t]) cmsFreeToneCurve(Trans[t]);
1184 if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
1185 }
1186
1187 cmsPipelineFree(LutPlusCurves);
1188
1189
1190 OptimizedPrelinCurves = _cmsStageGetPtrToCurveSet(OptimizedPrelinMpe);
1191 OptimizedPrelinCLUT = (_cmsStageCLutData*) OptimizedCLUTmpe ->Data;
1192
1193 // Set the evaluator if 8-bit
1194 if (_cmsFormatterIs8bit(*InputFormat)) {
1195
1196 Prelin8Data* p8 = PrelinOpt8alloc(OptimizedLUT ->ContextID,
1197 OptimizedPrelinCLUT ->Params,
1198 OptimizedPrelinCurves);
1199 if (p8 == NULL) return FALSE;
1200
1201 _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval8, (void*) p8, Prelin8free, Prelin8dup);
1202
1203 }
1204 else
1205 {
1206 Prelin16Data* p16 = PrelinOpt16alloc(OptimizedLUT ->ContextID,
1207 OptimizedPrelinCLUT ->Params,
1208 3, OptimizedPrelinCurves, 3, NULL);
1209 if (p16 == NULL) return FALSE;
1210
1211 _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
1212
1213 }
1214
1215 // Don't fix white on absolute colorimetric
1216 if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
1217 *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
1218
1219 if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
1220
1221 if (!FixWhiteMisalignment(OptimizedLUT, ColorSpace, OutputColorSpace)) {
1222
1223 return FALSE;
1224 }
1225 }
1226
1227 // And return the obtained LUT
1228
1229 cmsPipelineFree(OriginalLut);
1230 *Lut = OptimizedLUT;
1231 return TRUE;
1232
1233 Error:
1234
1235 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1236
1237 if (Trans[t]) cmsFreeToneCurve(Trans[t]);
1238 if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
1239 }
1240
1241 if (LutPlusCurves != NULL) cmsPipelineFree(LutPlusCurves);
1242 if (OptimizedLUT != NULL) cmsPipelineFree(OptimizedLUT);
1243
1244 return FALSE;
1245
1246 cmsUNUSED_PARAMETER(Intent);
1247 }
1248
1249
1250 // Curves optimizer ------------------------------------------------------------------------------------------------------------------
1251
1252 static
1253 void CurvesFree(cmsContext ContextID, void* ptr)
1254 {
1255 Curves16Data* Data = (Curves16Data*) ptr;
1256 int i;
1257
1258 for (i=0; i < Data -> nCurves; i++) {
1259
1260 _cmsFree(ContextID, Data ->Curves[i]);
1261 }
1262
1263 _cmsFree(ContextID, Data ->Curves);
1264 _cmsFree(ContextID, ptr);
1265 }
1266
1267 static
1268 void* CurvesDup(cmsContext ContextID, const void* ptr)
1269 {
1270 Curves16Data* Data = (Curves16Data*)_cmsDupMem(ContextID, ptr, sizeof(Curves16Data));
1271 int i;
1272
1273 if (Data == NULL) return NULL;
1274
1275 Data->Curves = (cmsUInt16Number**) _cmsDupMem(ContextID, Data->Curves, Data->nCurves * sizeof(cmsUInt16Number*));
1276
1277 for (i=0; i < Data -> nCurves; i++) {
1278 Data->Curves[i] = (cmsUInt16Number*) _cmsDupMem(ContextID, Data->Curves[i], Data->nElements * sizeof(cmsUInt16Number));
1279 }
1280
1281 return (void*) Data;
1282 }
1283
1284 // Precomputes tables for 8-bit on input devicelink.
1285 static
1286 Curves16Data* CurvesAlloc(cmsContext ContextID, int nCurves, int nElements, cmsToneCurve** G)
1287 {
1288 int i, j;
1289 Curves16Data* c16;
1290
1291 c16 = (Curves16Data*)_cmsMallocZero(ContextID, sizeof(Curves16Data));
1292 if (c16 == NULL) return NULL;
1293
1294 c16 ->nCurves = nCurves;
1295 c16 ->nElements = nElements;
1296
1297 c16->Curves = (cmsUInt16Number**) _cmsCalloc(ContextID, nCurves, sizeof(cmsUInt16Number*));
1298 if (c16 ->Curves == NULL) return NULL;
1299
1300 for (i=0; i < nCurves; i++) {
1301
1302 c16->Curves[i] = (cmsUInt16Number*) _cmsCalloc(ContextID, nElements, sizeof(cmsUInt16Number));
1303
1304 if (c16->Curves[i] == NULL) {
1305
1306 for (j=0; j < i; j++) {
1307 _cmsFree(ContextID, c16->Curves[j]);
1308 }
1309 _cmsFree(ContextID, c16->Curves);
1310 _cmsFree(ContextID, c16);
1311 return NULL;
1312 }
1313
1314 if (nElements == 256) {
1315
1316 for (j=0; j < nElements; j++) {
1317
1318 c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], FROM_8_TO_16(j));
1319 }
1320 }
1321 else {
1322
1323 for (j=0; j < nElements; j++) {
1324 c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], (cmsUInt16Number) j);
1325 }
1326 }
1327 }
1328
1329 return c16;
1330 }
1331
1332 static
1333 void FastEvaluateCurves8(register const cmsUInt16Number In[],
1334 register cmsUInt16Number Out[],
1335 register const void* D)
1336 {
1337 Curves16Data* Data = (Curves16Data*) D;
1338 cmsUInt8Number x;
1339 int i;
1340
1341 for (i=0; i < Data ->nCurves; i++) {
1342
1343 x = (In[i] >> 8);
1344 Out[i] = Data -> Curves[i][x];
1345 }
1346 }
1347
1348
1349 static
1350 void FastEvaluateCurves16(register const cmsUInt16Number In[],
1351 register cmsUInt16Number Out[],
1352 register const void* D)
1353 {
1354 Curves16Data* Data = (Curves16Data*) D;
1355 int i;
1356
1357 for (i=0; i < Data ->nCurves; i++) {
1358 Out[i] = Data -> Curves[i][In[i]];
1359 }
1360 }
1361
1362
1363 static
1364 void FastIdentity16(register const cmsUInt16Number In[],
1365 register cmsUInt16Number Out[],
1366 register const void* D)
1367 {
1368 cmsPipeline* Lut = (cmsPipeline*) D;
1369 cmsUInt32Number i;
1370
1371 for (i=0; i < Lut ->InputChannels; i++) {
1372 Out[i] = In[i];
1373 }
1374 }
1375
1376
1377 // If the target LUT holds only curves, the optimization procedure is to join all those
1378 // curves together. That only works on curves and does not work on matrices.
1379 static
1380 cmsBool OptimizeByJoiningCurves(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1381 {
1382 cmsToneCurve** GammaTables = NULL;
1383 cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
1384 cmsUInt32Number i, j;
1385 cmsPipeline* Src = *Lut;
1386 cmsPipeline* Dest = NULL;
1387 cmsStage* mpe;
1388 cmsStage* ObtainedCurves = NULL;
1389
1390
1391 // This is a loosy optimization! does not apply in floating-point cases
1392 if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
1393
1394 // Only curves in this LUT?
1395 for (mpe = cmsPipelineGetPtrToFirstStage(Src);
1396 mpe != NULL;
1397 mpe = cmsStageNext(mpe)) {
1398 if (cmsStageType(mpe) != cmsSigCurveSetElemType) return FALSE;
1399 }
1400
1401 // Allocate an empty LUT
1402 Dest = cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
1403 if (Dest == NULL) return FALSE;
1404
1405 // Create target curves
1406 GammaTables = (cmsToneCurve**) _cmsCalloc(Src ->ContextID, Src ->InputChannels, sizeof(cmsToneCurve*));
1407 if (GammaTables == NULL) goto Error;
1408
1409 for (i=0; i < Src ->InputChannels; i++) {
1410 GammaTables[i] = cmsBuildTabulatedToneCurve16(Src ->ContextID, PRELINEARIZATION_POINTS, NULL);
1411 if (GammaTables[i] == NULL) goto Error;
1412 }
1413
1414 // Compute 16 bit result by using floating point
1415 for (i=0; i < PRELINEARIZATION_POINTS; i++) {
1416
1417 for (j=0; j < Src ->InputChannels; j++)
1418 InFloat[j] = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
1419
1420 cmsPipelineEvalFloat(InFloat, OutFloat, Src);
1421
1422 for (j=0; j < Src ->InputChannels; j++)
1423 GammaTables[j] -> Table16[i] = _cmsQuickSaturateWord(OutFloat[j] * 65535.0);
1424 }
1425
1426 ObtainedCurves = cmsStageAllocToneCurves(Src ->ContextID, Src ->InputChannels, GammaTables);
1427 if (ObtainedCurves == NULL) goto Error;
1428
1429 for (i=0; i < Src ->InputChannels; i++) {
1430 cmsFreeToneCurve(GammaTables[i]);
1431 GammaTables[i] = NULL;
1432 }
1433
1434 if (GammaTables != NULL) _cmsFree(Src ->ContextID, GammaTables);
1435
1436 // Maybe the curves are linear at the end
1437 if (!AllCurvesAreLinear(ObtainedCurves)) {
1438
1439 if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, ObtainedCurves))
1440 goto Error;
1441
1442 // If the curves are to be applied in 8 bits, we can save memory
1443 if (_cmsFormatterIs8bit(*InputFormat)) {
1444
1445 _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*) ObtainedCurves ->Data;
1446 Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 256, Data ->TheCurves);
1447
1448 if (c16 == NULL) goto Error;
1449 *dwFlags |= cmsFLAGS_NOCACHE;
1450 _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves8, c16, CurvesFree, CurvesDup);
1451
1452 }
1453 else {
1454
1455 _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*) cmsStageData(ObtainedCurves);
1456 Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 65536, Data ->TheCurves);
1457
1458 if (c16 == NULL) goto Error;
1459 *dwFlags |= cmsFLAGS_NOCACHE;
1460 _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves16, c16, CurvesFree, CurvesDup);
1461 }
1462 }
1463 else {
1464
1465 // LUT optimizes to nothing. Set the identity LUT
1466 cmsStageFree(ObtainedCurves);
1467
1468 if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageAllocIdentity(Dest ->ContextID, Src ->InputChannels)))
1469 goto Error;
1470
1471 *dwFlags |= cmsFLAGS_NOCACHE;
1472 _cmsPipelineSetOptimizationParameters(Dest, FastIdentity16, (void*) Dest, NULL, NULL);
1473 }
1474
1475 // We are done.
1476 cmsPipelineFree(Src);
1477 *Lut = Dest;
1478 return TRUE;
1479
1480 Error:
1481
1482 if (ObtainedCurves != NULL) cmsStageFree(ObtainedCurves);
1483 if (GammaTables != NULL) {
1484 for (i=0; i < Src ->InputChannels; i++) {
1485 if (GammaTables[i] != NULL) cmsFreeToneCurve(GammaTables[i]);
1486 }
1487
1488 _cmsFree(Src ->ContextID, GammaTables);
1489 }
1490
1491 if (Dest != NULL) cmsPipelineFree(Dest);
1492 return FALSE;
1493
1494 cmsUNUSED_PARAMETER(Intent);
1495 cmsUNUSED_PARAMETER(InputFormat);
1496 cmsUNUSED_PARAMETER(OutputFormat);
1497 cmsUNUSED_PARAMETER(dwFlags);
1498 }
1499
1500 // -------------------------------------------------------------------------------------------------------------------------------------
1501 // LUT is Shaper - Matrix - Matrix - Shaper, which is very frequent when combining two matrix-shaper profiles
1502
1503
1504 static
1505 void FreeMatShaper(cmsContext ContextID, void* Data)
1506 {
1507 if (Data != NULL) _cmsFree(ContextID, Data);
1508 }
1509
1510 static
1511 void* DupMatShaper(cmsContext ContextID, const void* Data)
1512 {
1513 return _cmsDupMem(ContextID, Data, sizeof(MatShaper8Data));
1514 }
1515
1516
1517 // A fast matrix-shaper evaluator for 8 bits. This is a bit ticky since I'm using 1.14 signed fixed point
1518 // to accomplish some performance. Actually it takes 256x3 16 bits tables and 16385 x 3 tables of 8 bits,
1519 // in total about 50K, and the performance boost is huge!
1520 static
1521 void MatShaperEval16(register const cmsUInt16Number In[],
1522 register cmsUInt16Number Out[],
1523 register const void* D)
1524 {
1525 MatShaper8Data* p = (MatShaper8Data*) D;
1526 cmsS1Fixed14Number l1, l2, l3, r, g, b;
1527 cmsUInt32Number ri, gi, bi;
1528
1529 // In this case (and only in this case!) we can use this simplification since
1530 // In[] is assured to come from a 8 bit number. (a << 8 | a)
1531 ri = In[0] & 0xFF;
1532 gi = In[1] & 0xFF;
1533 bi = In[2] & 0xFF;
1534
1535 // Across first shaper, which also converts to 1.14 fixed point
1536 r = p->Shaper1R[ri];
1537 g = p->Shaper1G[gi];
1538 b = p->Shaper1B[bi];
1539
1540 // Evaluate the matrix in 1.14 fixed point
1541 l1 = (p->Mat[0][0] * r + p->Mat[0][1] * g + p->Mat[0][2] * b + p->Off[0] + 0x2000) >> 14;
1542 l2 = (p->Mat[1][0] * r + p->Mat[1][1] * g + p->Mat[1][2] * b + p->Off[1] + 0x2000) >> 14;
1543 l3 = (p->Mat[2][0] * r + p->Mat[2][1] * g + p->Mat[2][2] * b + p->Off[2] + 0x2000) >> 14;
1544
1545 // Now we have to clip to 0..1.0 range
1546 ri = (l1 < 0) ? 0 : ((l1 > 16384) ? 16384 : l1);
1547 gi = (l2 < 0) ? 0 : ((l2 > 16384) ? 16384 : l2);
1548 bi = (l3 < 0) ? 0 : ((l3 > 16384) ? 16384 : l3);
1549
1550 // And across second shaper,
1551 Out[0] = p->Shaper2R[ri];
1552 Out[1] = p->Shaper2G[gi];
1553 Out[2] = p->Shaper2B[bi];
1554
1555 }
1556
1557 // This table converts from 8 bits to 1.14 after applying the curve
1558 static
1559 void FillFirstShaper(cmsS1Fixed14Number* Table, cmsToneCurve* Curve)
1560 {
1561 int i;
1562 cmsFloat32Number R, y;
1563
1564 for (i=0; i < 256; i++) {
1565
1566 R = (cmsFloat32Number) (i / 255.0);
1567 y = cmsEvalToneCurveFloat(Curve, R);
1568
1569 Table[i] = DOUBLE_TO_1FIXED14(y);
1570 }
1571 }
1572
1573 // This table converts form 1.14 (being 0x4000 the last entry) to 8 bits after applying the curve
1574 static
1575 void FillSecondShaper(cmsUInt16Number* Table, cmsToneCurve* Curve, cmsBool Is8BitsOutput)
1576 {
1577 int i;
1578 cmsFloat32Number R, Val;
1579
1580 for (i=0; i < 16385; i++) {
1581
1582 R = (cmsFloat32Number) (i / 16384.0);
1583 Val = cmsEvalToneCurveFloat(Curve, R); // Val comes 0..1.0
1584
1585 if (Is8BitsOutput) {
1586
1587 // If 8 bits output, we can optimize further by computing the / 257 part.
1588 // first we compute the resulting byte and then we store the byte times
1589 // 257. This quantization allows to round very quick by doing a >> 8, but
1590 // since the low byte is always equal to msb, we can do a & 0xff and this works!
1591 cmsUInt16Number w = _cmsQuickSaturateWord(Val * 65535.0);
1592 cmsUInt8Number b = FROM_16_TO_8(w);
1593
1594 Table[i] = FROM_8_TO_16(b);
1595 }
1596 else Table[i] = _cmsQuickSaturateWord(Val * 65535.0);
1597 }
1598 }
1599
1600 // Compute the matrix-shaper structure
1601 static
1602 cmsBool SetMatShaper(cmsPipeline* Dest, cmsToneCurve* Curve1[3], cmsMAT3* Mat, cmsVEC3* Off, cmsToneCurve* Curve2[3], cmsUInt32Number* OutputFormat)
1603 {
1604 MatShaper8Data* p;
1605 int i, j;
1606 cmsBool Is8Bits = _cmsFormatterIs8bit(*OutputFormat);
1607
1608 // Allocate a big chuck of memory to store precomputed tables
1609 p = (MatShaper8Data*) _cmsMalloc(Dest ->ContextID, sizeof(MatShaper8Data));
1610 if (p == NULL) return FALSE;
1611
1612 p -> ContextID = Dest -> ContextID;
1613
1614 // Precompute tables
1615 FillFirstShaper(p ->Shaper1R, Curve1[0]);
1616 FillFirstShaper(p ->Shaper1G, Curve1[1]);
1617 FillFirstShaper(p ->Shaper1B, Curve1[2]);
1618
1619 FillSecondShaper(p ->Shaper2R, Curve2[0], Is8Bits);
1620 FillSecondShaper(p ->Shaper2G, Curve2[1], Is8Bits);
1621 FillSecondShaper(p ->Shaper2B, Curve2[2], Is8Bits);
1622
1623 // Convert matrix to nFixed14. Note that those values may take more than 16 bits as
1624 for (i=0; i < 3; i++) {
1625 for (j=0; j < 3; j++) {
1626 p ->Mat[i][j] = DOUBLE_TO_1FIXED14(Mat->v[i].n[j]);
1627 }
1628 }
1629
1630 for (i=0; i < 3; i++) {
1631
1632 if (Off == NULL) {
1633 p ->Off[i] = 0;
1634 }
1635 else {
1636 p ->Off[i] = DOUBLE_TO_1FIXED14(Off->n[i]);
1637 }
1638 }
1639
1640 // Mark as optimized for faster formatter
1641 if (Is8Bits)
1642 *OutputFormat |= OPTIMIZED_SH(1);
1643
1644 // Fill function pointers
1645 _cmsPipelineSetOptimizationParameters(Dest, MatShaperEval16, (void*) p, FreeMatShaper, DupMatShaper);
1646 return TRUE;
1647 }
1648
1649 // 8 bits on input allows matrix-shaper boot up to 25 Mpixels per second on RGB. That's fast!
1650 static
1651 cmsBool OptimizeMatrixShaper(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1652 {
1653 cmsStage* Curve1, *Curve2;
1654 cmsStage* Matrix1, *Matrix2;
1655 cmsMAT3 res;
1656 cmsBool IdentityMat;
1657 cmsPipeline* Dest, *Src;
1658 cmsFloat64Number* Offset;
1659
1660 // Only works on RGB to RGB
1661 if (T_CHANNELS(*InputFormat) != 3 || T_CHANNELS(*OutputFormat) != 3) return FALSE;
1662
1663 // Only works on 8 bit input
1664 if (!_cmsFormatterIs8bit(*InputFormat)) return FALSE;
1665
1666 // Seems suitable, proceed
1667 Src = *Lut;
1668
1669 // Check for:
1670 //
1671 // shaper-matrix-matrix-shaper
1672 // shaper-matrix-shaper
1673 //
1674 // Both of those constructs are possible (first because abs. colorimetric).
1675 // additionally, In the first case, the input matrix offset should be zero.
1676
1677 IdentityMat = FALSE;
1678 if (cmsPipelineCheckAndRetreiveStages(Src, 4,
1679 cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
1680 &Curve1, &Matrix1, &Matrix2, &Curve2)) {
1681
1682 // Get both matrices
1683 _cmsStageMatrixData* Data1 = (_cmsStageMatrixData*)cmsStageData(Matrix1);
1684 _cmsStageMatrixData* Data2 = (_cmsStageMatrixData*)cmsStageData(Matrix2);
1685
1686 // Input offset should be zero
1687 if (Data1->Offset != NULL) return FALSE;
1688
1689 // Multiply both matrices to get the result
1690 _cmsMAT3per(&res, (cmsMAT3*)Data2->Double, (cmsMAT3*)Data1->Double);
1691
1692 // Only 2nd matrix has offset, or it is zero
1693 Offset = Data2->Offset;
1694
1695 // Now the result is in res + Data2 -> Offset. Maybe is a plain identity?
1696 if (_cmsMAT3isIdentity(&res) && Offset == NULL) {
1697
1698 // We can get rid of full matrix
1699 IdentityMat = TRUE;
1700 }
1701
1702 }
1703 else {
1704
1705 if (cmsPipelineCheckAndRetreiveStages(Src, 3,
1706 cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
1707 &Curve1, &Matrix1, &Curve2)) {
1708
1709 _cmsStageMatrixData* Data = (_cmsStageMatrixData*)cmsStageData(Matrix1);
1710
1711 // Copy the matrix to our result
1712 memcpy(&res, Data->Double, sizeof(res));
1713
1714 // Preserve the Odffset (may be NULL as a zero offset)
1715 Offset = Data->Offset;
1716
1717 if (_cmsMAT3isIdentity(&res) && Offset == NULL) {
1718
1719 // We can get rid of full matrix
1720 IdentityMat = TRUE;
1721 }
1722 }
1723 else
1724 return FALSE; // Not optimizeable this time
1725
1726 }
1727
1728 // Allocate an empty LUT
1729 Dest = cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
1730 if (!Dest) return FALSE;
1731
1732 // Assamble the new LUT
1733 if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageDup(Curve1)))
1734 goto Error;
1735
1736 if (!IdentityMat) {
1737
1738 if (!cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageAllocMatrix(Dest->ContextID, 3, 3, (const cmsFloat64Number*)&res, Offset)))
1739 goto Error;
1740 }
1741
1742 if (!cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageDup(Curve2)))
1743 goto Error;
1744
1745 // If identity on matrix, we can further optimize the curves, so call the join curves routine
1746 if (IdentityMat) {
1747
1748 OptimizeByJoiningCurves(&Dest, Intent, InputFormat, OutputFormat, dwFlags);
1749 }
1750 else {
1751 _cmsStageToneCurvesData* mpeC1 = (_cmsStageToneCurvesData*) cmsStageData(Curve1);
1752 _cmsStageToneCurvesData* mpeC2 = (_cmsStageToneCurvesData*) cmsStageData(Curve2);
1753
1754 // In this particular optimization, caché does not help as it takes more time to deal with
1755 // the caché that with the pixel handling
1756 *dwFlags |= cmsFLAGS_NOCACHE;
1757
1758 // Setup the optimizarion routines
1759 SetMatShaper(Dest, mpeC1 ->TheCurves, &res, (cmsVEC3*) Offset, mpeC2->TheCurves, OutputFormat);
1760 }
1761
1762 cmsPipelineFree(Src);
1763 *Lut = Dest;
1764 return TRUE;
1765 Error:
1766 // Leave Src unchanged
1767 cmsPipelineFree(Dest);
1768 return FALSE;
1769 }
1770
1771
1772 // -------------------------------------------------------------------------------------------------------------------------------------
1773 // Optimization plug-ins
1774
1775 // List of optimizations
1776 typedef struct _cmsOptimizationCollection_st {
1777
1778 _cmsOPToptimizeFn OptimizePtr;
1779
1780 struct _cmsOptimizationCollection_st *Next;
1781
1782 } _cmsOptimizationCollection;
1783
1784
1785 // The built-in list. We currently implement 4 types of optimizations. Joining of curves, matrix-shaper, linearization and resampling
1786 static _cmsOptimizationCollection DefaultOptimization[] = {
1787
1788 { OptimizeByJoiningCurves, &DefaultOptimization[1] },
1789 { OptimizeMatrixShaper, &DefaultOptimization[2] },
1790 { OptimizeByComputingLinearization, &DefaultOptimization[3] },
1791 { OptimizeByResampling, NULL }
1792 };
1793
1794 // The linked list head
1795 _cmsOptimizationPluginChunkType _cmsOptimizationPluginChunk = { NULL };
1796
1797
1798 // Duplicates the zone of memory used by the plug-in in the new context
1799 static
1800 void DupPluginOptimizationList(struct _cmsContext_struct* ctx,
1801 const struct _cmsContext_struct* src)
1802 {
1803 _cmsOptimizationPluginChunkType newHead = { NULL };
1804 _cmsOptimizationCollection* entry;
1805 _cmsOptimizationCollection* Anterior = NULL;
1806 _cmsOptimizationPluginChunkType* head = (_cmsOptimizationPluginChunkType*) src->chunks[OptimizationPlugin];
1807
1808 _cmsAssert(ctx != NULL);
1809 _cmsAssert(head != NULL);
1810
1811 // Walk the list copying all nodes
1812 for (entry = head->OptimizationCollection;
1813 entry != NULL;
1814 entry = entry ->Next) {
1815
1816 _cmsOptimizationCollection *newEntry = ( _cmsOptimizationCollection *) _cmsSubAllocDup(ctx ->MemPool, entry, sizeof(_cmsOptimizationCollection));
1817
1818 if (newEntry == NULL)
1819 return;
1820
1821 // We want to keep the linked list order, so this is a little bit tricky
1822 newEntry -> Next = NULL;
1823 if (Anterior)
1824 Anterior -> Next = newEntry;
1825
1826 Anterior = newEntry;
1827
1828 if (newHead.OptimizationCollection == NULL)
1829 newHead.OptimizationCollection = newEntry;
1830 }
1831
1832 ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx->MemPool, &newHead, sizeof(_cmsOptimizationPluginChunkType));
1833 }
1834
1835 void _cmsAllocOptimizationPluginChunk(struct _cmsContext_struct* ctx,
1836 const struct _cmsContext_struct* src)
1837 {
1838 if (src != NULL) {
1839
1840 // Copy all linked list
1841 DupPluginOptimizationList(ctx, src);
1842 }
1843 else {
1844 static _cmsOptimizationPluginChunkType OptimizationPluginChunkType = { NULL };
1845 ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx ->MemPool, &OptimizationPluginChunkType, sizeof(_cmsOptimizationPluginChunkType));
1846 }
1847 }
1848
1849
1850 // Register new ways to optimize
1851 cmsBool _cmsRegisterOptimizationPlugin(cmsContext ContextID, cmsPluginBase* Data)
1852 {
1853 cmsPluginOptimization* Plugin = (cmsPluginOptimization*) Data;
1854 _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
1855 _cmsOptimizationCollection* fl;
1856
1857 if (Data == NULL) {
1858
1859 ctx->OptimizationCollection = NULL;
1860 return TRUE;
1861 }
1862
1863 // Optimizer callback is required
1864 if (Plugin ->OptimizePtr == NULL) return FALSE;
1865
1866 fl = (_cmsOptimizationCollection*) _cmsPluginMalloc(ContextID, sizeof(_cmsOptimizationCollection));
1867 if (fl == NULL) return FALSE;
1868
1869 // Copy the parameters
1870 fl ->OptimizePtr = Plugin ->OptimizePtr;
1871
1872 // Keep linked list
1873 fl ->Next = ctx->OptimizationCollection;
1874
1875 // Set the head
1876 ctx ->OptimizationCollection = fl;
1877
1878 // All is ok
1879 return TRUE;
1880 }
1881
1882 // The entry point for LUT optimization
1883 cmsBool _cmsOptimizePipeline(cmsContext ContextID,
1884 cmsPipeline** PtrLut,
1885 int Intent,
1886 cmsUInt32Number* InputFormat,
1887 cmsUInt32Number* OutputFormat,
1888 cmsUInt32Number* dwFlags)
1889 {
1890 _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
1891 _cmsOptimizationCollection* Opts;
1892 cmsBool AnySuccess = FALSE;
1893
1894 // A CLUT is being asked, so force this specific optimization
1895 if (*dwFlags & cmsFLAGS_FORCE_CLUT) {
1896
1897 PreOptimize(*PtrLut);
1898 return OptimizeByResampling(PtrLut, Intent, InputFormat, OutputFormat, dwFlags);
1899 }
1900
1901 // Anything to optimize?
1902 if ((*PtrLut) ->Elements == NULL) {
1903 _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
1904 return TRUE;
1905 }
1906
1907 // Try to get rid of identities and trivial conversions.
1908 AnySuccess = PreOptimize(*PtrLut);
1909
1910 // After removal do we end with an identity?
1911 if ((*PtrLut) ->Elements == NULL) {
1912 _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
1913 return TRUE;
1914 }
1915
1916 // Do not optimize, keep all precision
1917 if (*dwFlags & cmsFLAGS_NOOPTIMIZE)
1918 return FALSE;
1919
1920 // Try plug-in optimizations
1921 for (Opts = ctx->OptimizationCollection;
1922 Opts != NULL;
1923 Opts = Opts ->Next) {
1924
1925 // If one schema succeeded, we are done
1926 if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
1927
1928 return TRUE; // Optimized!
1929 }
1930 }
1931
1932 // Try built-in optimizations
1933 for (Opts = DefaultOptimization;
1934 Opts != NULL;
1935 Opts = Opts ->Next) {
1936
1937 if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
1938
1939 return TRUE;
1940 }
1941 }
1942
1943 // Only simple optimizations succeeded
1944 return AnySuccess;
1945 }
1946
1947
1948