1 //---------------------------------------------------------------------------------
2 //
3 // Little Color Management System
4 // Copyright (c) 1998-2017 Marti Maria Saguer
5 //
6 // Permission is hereby granted, free of charge, to any person obtaining
7 // a copy of this software and associated documentation files (the "Software"),
8 // to deal in the Software without restriction, including without limitation
9 // the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 // and/or sell copies of the Software, and to permit persons to whom the Software
11 // is furnished to do so, subject to the following conditions:
12 //
13 // The above copyright notice and this permission notice shall be included in
14 // all copies or substantial portions of the Software.
15 //
16 // THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
17 // EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
18 // THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
19 // NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
20 // LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
21 // OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
22 // WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
23 //
24 //---------------------------------------------------------------------------------
25 //
26
27 #include "lcms2_internal.h"
28
29
30 //----------------------------------------------------------------------------------
31
32 // Optimization for 8 bits, Shaper-CLUT (3 inputs only)
33 typedef struct {
34
35 cmsContext ContextID;
36
37 const cmsInterpParams* p; // Tetrahedrical interpolation parameters. This is a not-owned pointer.
38
39 cmsUInt16Number rx[256], ry[256], rz[256];
40 cmsUInt32Number X0[256], Y0[256], Z0[256]; // Precomputed nodes and offsets for 8-bit input data
41
42
43 } Prelin8Data;
44
45
46 // Generic optimization for 16 bits Shaper-CLUT-Shaper (any inputs)
47 typedef struct {
48
49 cmsContext ContextID;
50
51 // Number of channels
52 cmsUInt32Number nInputs;
53 cmsUInt32Number nOutputs;
54
55 _cmsInterpFn16 EvalCurveIn16[MAX_INPUT_DIMENSIONS]; // The maximum number of input channels is known in advance
56 cmsInterpParams* ParamsCurveIn16[MAX_INPUT_DIMENSIONS];
57
58 _cmsInterpFn16 EvalCLUT; // The evaluator for 3D grid
59 const cmsInterpParams* CLUTparams; // (not-owned pointer)
60
61
62 _cmsInterpFn16* EvalCurveOut16; // Points to an array of curve evaluators in 16 bits (not-owned pointer)
63 cmsInterpParams** ParamsCurveOut16; // Points to an array of references to interpolation params (not-owned pointer)
64
65
66 } Prelin16Data;
67
68
69 // Optimization for matrix-shaper in 8 bits. Numbers are operated in n.14 signed, tables are stored in 1.14 fixed
70
71 typedef cmsInt32Number cmsS1Fixed14Number; // Note that this may hold more than 16 bits!
72
73 #define DOUBLE_TO_1FIXED14(x) ((cmsS1Fixed14Number) floor((x) * 16384.0 + 0.5))
74
75 typedef struct {
76
77 cmsContext ContextID;
78
79 cmsS1Fixed14Number Shaper1R[256]; // from 0..255 to 1.14 (0.0...1.0)
80 cmsS1Fixed14Number Shaper1G[256];
81 cmsS1Fixed14Number Shaper1B[256];
82
83 cmsS1Fixed14Number Mat[3][3]; // n.14 to n.14 (needs a saturation after that)
84 cmsS1Fixed14Number Off[3];
85
86 cmsUInt16Number Shaper2R[16385]; // 1.14 to 0..255
87 cmsUInt16Number Shaper2G[16385];
88 cmsUInt16Number Shaper2B[16385];
89
90 } MatShaper8Data;
91
92 // Curves, optimization is shared between 8 and 16 bits
93 typedef struct {
94 cmsUInt32Number nCurves; // Number of curves
95 cmsUInt32Number nElements; // Elements in curves
96 cmsUInt16Number** Curves; // Points to a dynamically allocated array
97
98 } Curves16Data;
99
100
101 // Simple optimizations ----------------------------------------------------------------------------------------------------------
102
103
104 // Remove an element in linked chain
105 static
_RemoveElement(cmsContext ContextID,cmsStage ** head)106 void _RemoveElement(cmsContext ContextID, cmsStage** head)
107 {
108 cmsStage* mpe = *head;
109 cmsStage* next = mpe ->Next;
110 *head = next;
111 cmsStageFree(ContextID, mpe);
112 }
113
114 // Remove all identities in chain. Note that pt actually is a double pointer to the element that holds the pointer.
115 static
_Remove1Op(cmsContext ContextID,cmsPipeline * Lut,cmsStageSignature UnaryOp)116 cmsBool _Remove1Op(cmsContext ContextID, cmsPipeline* Lut, cmsStageSignature UnaryOp)
117 {
118 cmsStage** pt = &Lut ->Elements;
119 cmsBool AnyOpt = FALSE;
120
121 while (*pt != NULL) {
122
123 if ((*pt) ->Implements == UnaryOp) {
124 _RemoveElement(ContextID, pt);
125 AnyOpt = TRUE;
126 }
127 else
128 pt = &((*pt) -> Next);
129 }
130
131 return AnyOpt;
132 }
133
134 // Same, but only if two adjacent elements are found
135 static
_Remove2Op(cmsContext ContextID,cmsPipeline * Lut,cmsStageSignature Op1,cmsStageSignature Op2)136 cmsBool _Remove2Op(cmsContext ContextID, cmsPipeline* Lut, cmsStageSignature Op1, cmsStageSignature Op2)
137 {
138 cmsStage** pt1;
139 cmsStage** pt2;
140 cmsBool AnyOpt = FALSE;
141
142 pt1 = &Lut ->Elements;
143 if (*pt1 == NULL) return AnyOpt;
144
145 while (*pt1 != NULL) {
146
147 pt2 = &((*pt1) -> Next);
148 if (*pt2 == NULL) return AnyOpt;
149
150 if ((*pt1) ->Implements == Op1 && (*pt2) ->Implements == Op2) {
151 _RemoveElement(ContextID, pt2);
152 _RemoveElement(ContextID, pt1);
153 AnyOpt = TRUE;
154 }
155 else
156 pt1 = &((*pt1) -> Next);
157 }
158
159 return AnyOpt;
160 }
161
162
163 static
CloseEnoughFloat(cmsFloat64Number a,cmsFloat64Number b)164 cmsBool CloseEnoughFloat(cmsFloat64Number a, cmsFloat64Number b)
165 {
166 return fabs(b - a) < 0.00001f;
167 }
168
169 static
isFloatMatrixIdentity(cmsContext ContextID,const cmsMAT3 * a)170 cmsBool isFloatMatrixIdentity(cmsContext ContextID, const cmsMAT3* a)
171 {
172 cmsMAT3 Identity;
173 int i, j;
174
175 _cmsMAT3identity(ContextID, &Identity);
176
177 for (i = 0; i < 3; i++)
178 for (j = 0; j < 3; j++)
179 if (!CloseEnoughFloat(a->v[i].n[j], Identity.v[i].n[j])) return FALSE;
180
181 return TRUE;
182 }
183 // if two adjacent matrices are found, multiply them.
184 static
_MultiplyMatrix(cmsContext ContextID,cmsPipeline * Lut)185 cmsBool _MultiplyMatrix(cmsContext ContextID, cmsPipeline* Lut)
186 {
187 cmsStage** pt1;
188 cmsStage** pt2;
189 cmsStage* chain;
190 cmsBool AnyOpt = FALSE;
191
192 pt1 = &Lut->Elements;
193 if (*pt1 == NULL) return AnyOpt;
194
195 while (*pt1 != NULL) {
196
197 pt2 = &((*pt1)->Next);
198 if (*pt2 == NULL) return AnyOpt;
199
200 if ((*pt1)->Implements == cmsSigMatrixElemType && (*pt2)->Implements == cmsSigMatrixElemType) {
201
202 // Get both matrices
203 _cmsStageMatrixData* m1 = (_cmsStageMatrixData*) cmsStageData(ContextID, *pt1);
204 _cmsStageMatrixData* m2 = (_cmsStageMatrixData*) cmsStageData(ContextID, *pt2);
205 cmsMAT3 res;
206
207 // Input offset and output offset should be zero to use this optimization
208 if (m1->Offset != NULL || m2 ->Offset != NULL ||
209 cmsStageInputChannels(ContextID, *pt1) != 3 || cmsStageOutputChannels(ContextID, *pt1) != 3 ||
210 cmsStageInputChannels(ContextID, *pt2) != 3 || cmsStageOutputChannels(ContextID, *pt2) != 3)
211 return FALSE;
212
213 // Multiply both matrices to get the result
214 _cmsMAT3per(ContextID, &res, (cmsMAT3*)m2->Double, (cmsMAT3*)m1->Double);
215
216 // Get the next in chain after the matrices
217 chain = (*pt2)->Next;
218
219 // Remove both matrices
220 _RemoveElement(ContextID, pt2);
221 _RemoveElement(ContextID, pt1);
222
223 // Now what if the result is a plain identity?
224 if (!isFloatMatrixIdentity(ContextID, &res)) {
225
226 // We can not get rid of full matrix
227 cmsStage* Multmat = cmsStageAllocMatrix(ContextID, 3, 3, (const cmsFloat64Number*) &res, NULL);
228 if (Multmat == NULL) return FALSE; // Should never happen
229
230 // Recover the chain
231 Multmat->Next = chain;
232 *pt1 = Multmat;
233 }
234
235 AnyOpt = TRUE;
236 }
237 else
238 pt1 = &((*pt1)->Next);
239 }
240
241 return AnyOpt;
242 }
243
244
245 // Preoptimize just gets rif of no-ops coming paired. Conversion from v2 to v4 followed
246 // by a v4 to v2 and vice-versa. The elements are then discarded.
247 static
PreOptimize(cmsContext ContextID,cmsPipeline * Lut)248 cmsBool PreOptimize(cmsContext ContextID, cmsPipeline* Lut)
249 {
250 cmsBool AnyOpt = FALSE, Opt;
251
252 do {
253
254 Opt = FALSE;
255
256 // Remove all identities
257 Opt |= _Remove1Op(ContextID, Lut, cmsSigIdentityElemType);
258
259 // Remove XYZ2Lab followed by Lab2XYZ
260 Opt |= _Remove2Op(ContextID, Lut, cmsSigXYZ2LabElemType, cmsSigLab2XYZElemType);
261
262 // Remove Lab2XYZ followed by XYZ2Lab
263 Opt |= _Remove2Op(ContextID, Lut, cmsSigLab2XYZElemType, cmsSigXYZ2LabElemType);
264
265 // Remove V4 to V2 followed by V2 to V4
266 Opt |= _Remove2Op(ContextID, Lut, cmsSigLabV4toV2, cmsSigLabV2toV4);
267
268 // Remove V2 to V4 followed by V4 to V2
269 Opt |= _Remove2Op(ContextID, Lut, cmsSigLabV2toV4, cmsSigLabV4toV2);
270
271 // Remove float pcs Lab conversions
272 Opt |= _Remove2Op(ContextID, Lut, cmsSigLab2FloatPCS, cmsSigFloatPCS2Lab);
273
274 // Remove float pcs Lab conversions
275 Opt |= _Remove2Op(ContextID, Lut, cmsSigXYZ2FloatPCS, cmsSigFloatPCS2XYZ);
276
277 // Simplify matrix.
278 Opt |= _MultiplyMatrix(ContextID, Lut);
279
280 if (Opt) AnyOpt = TRUE;
281
282 } while (Opt);
283
284 return AnyOpt;
285 }
286
287 static
Eval16nop1D(cmsContext ContextID,register const cmsUInt16Number Input[],register cmsUInt16Number Output[],register const struct _cms_interp_struc * p)288 void Eval16nop1D(cmsContext ContextID, register const cmsUInt16Number Input[],
289 register cmsUInt16Number Output[],
290 register const struct _cms_interp_struc* p)
291 {
292 cmsUNUSED_PARAMETER(ContextID);
293 Output[0] = Input[0];
294
295 cmsUNUSED_PARAMETER(p);
296 }
297
298 static
PrelinEval16(cmsContext ContextID,register const cmsUInt16Number Input[],register cmsUInt16Number Output[],register const void * D)299 void PrelinEval16(cmsContext ContextID, register const cmsUInt16Number Input[],
300 register cmsUInt16Number Output[],
301 register const void* D)
302 {
303 Prelin16Data* p16 = (Prelin16Data*) D;
304 cmsUInt16Number StageABC[MAX_INPUT_DIMENSIONS];
305 cmsUInt16Number StageDEF[cmsMAXCHANNELS];
306 cmsUInt32Number i;
307
308 for (i=0; i < p16 ->nInputs; i++) {
309
310 p16 ->EvalCurveIn16[i](ContextID, &Input[i], &StageABC[i], p16 ->ParamsCurveIn16[i]);
311 }
312
313 p16 ->EvalCLUT(ContextID, StageABC, StageDEF, p16 ->CLUTparams);
314
315 for (i=0; i < p16 ->nOutputs; i++) {
316
317 p16 ->EvalCurveOut16[i](ContextID, &StageDEF[i], &Output[i], p16 ->ParamsCurveOut16[i]);
318 }
319 }
320
321
322 static
PrelinOpt16free(cmsContext ContextID,void * ptr)323 void PrelinOpt16free(cmsContext ContextID, void* ptr)
324 {
325 Prelin16Data* p16 = (Prelin16Data*) ptr;
326
327 _cmsFree(ContextID, p16 ->EvalCurveOut16);
328 _cmsFree(ContextID, p16 ->ParamsCurveOut16);
329
330 _cmsFree(ContextID, p16);
331 }
332
333 static
Prelin16dup(cmsContext ContextID,const void * ptr)334 void* Prelin16dup(cmsContext ContextID, const void* ptr)
335 {
336 Prelin16Data* p16 = (Prelin16Data*) ptr;
337 Prelin16Data* Duped = (Prelin16Data*) _cmsDupMem(ContextID, p16, sizeof(Prelin16Data));
338
339 if (Duped == NULL) return NULL;
340
341 Duped->EvalCurveOut16 = (_cmsInterpFn16*) _cmsDupMem(ContextID, p16->EvalCurveOut16, p16->nOutputs * sizeof(_cmsInterpFn16));
342 Duped->ParamsCurveOut16 = (cmsInterpParams**)_cmsDupMem(ContextID, p16->ParamsCurveOut16, p16->nOutputs * sizeof(cmsInterpParams*));
343
344 return Duped;
345 }
346
347
348 static
PrelinOpt16alloc(cmsContext ContextID,const cmsInterpParams * ColorMap,cmsUInt32Number nInputs,cmsToneCurve ** In,cmsUInt32Number nOutputs,cmsToneCurve ** Out)349 Prelin16Data* PrelinOpt16alloc(cmsContext ContextID,
350 const cmsInterpParams* ColorMap,
351 cmsUInt32Number nInputs, cmsToneCurve** In,
352 cmsUInt32Number nOutputs, cmsToneCurve** Out )
353 {
354 cmsUInt32Number i;
355 Prelin16Data* p16 = (Prelin16Data*)_cmsMallocZero(ContextID, sizeof(Prelin16Data));
356 if (p16 == NULL) return NULL;
357
358 p16 ->nInputs = nInputs;
359 p16 ->nOutputs = nOutputs;
360
361
362 for (i=0; i < nInputs; i++) {
363
364 if (In == NULL) {
365 p16 -> ParamsCurveIn16[i] = NULL;
366 p16 -> EvalCurveIn16[i] = Eval16nop1D;
367
368 }
369 else {
370 p16 -> ParamsCurveIn16[i] = In[i] ->InterpParams;
371 p16 -> EvalCurveIn16[i] = p16 ->ParamsCurveIn16[i]->Interpolation.Lerp16;
372 }
373 }
374
375 p16 ->CLUTparams = ColorMap;
376 p16 ->EvalCLUT = ColorMap ->Interpolation.Lerp16;
377
378
379 p16 -> EvalCurveOut16 = (_cmsInterpFn16*) _cmsCalloc(ContextID, nOutputs, sizeof(_cmsInterpFn16));
380 p16 -> ParamsCurveOut16 = (cmsInterpParams**) _cmsCalloc(ContextID, nOutputs, sizeof(cmsInterpParams* ));
381
382 for (i=0; i < nOutputs; i++) {
383
384 if (Out == NULL) {
385 p16 ->ParamsCurveOut16[i] = NULL;
386 p16 -> EvalCurveOut16[i] = Eval16nop1D;
387 }
388 else {
389
390 p16 ->ParamsCurveOut16[i] = Out[i] ->InterpParams;
391 p16 -> EvalCurveOut16[i] = p16 ->ParamsCurveOut16[i]->Interpolation.Lerp16;
392 }
393 }
394
395 return p16;
396 }
397
398
399
400 // Resampling ---------------------------------------------------------------------------------
401
402 #define PRELINEARIZATION_POINTS 4096
403
404 // Sampler implemented by another LUT. This is a clean way to precalculate the devicelink 3D CLUT for
405 // almost any transform. We use floating point precision and then convert from floating point to 16 bits.
406 static
XFormSampler16(cmsContext ContextID,register const cmsUInt16Number In[],register cmsUInt16Number Out[],register void * Cargo)407 cmsInt32Number XFormSampler16(cmsContext ContextID, register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void* Cargo)
408 {
409 cmsPipeline* Lut = (cmsPipeline*) Cargo;
410 cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
411 cmsUInt32Number i;
412
413 _cmsAssert(Lut -> InputChannels < cmsMAXCHANNELS);
414 _cmsAssert(Lut -> OutputChannels < cmsMAXCHANNELS);
415
416 // From 16 bit to floating point
417 for (i=0; i < Lut ->InputChannels; i++)
418 InFloat[i] = (cmsFloat32Number) (In[i] / 65535.0);
419
420 // Evaluate in floating point
421 cmsPipelineEvalFloat(ContextID, InFloat, OutFloat, Lut);
422
423 // Back to 16 bits representation
424 for (i=0; i < Lut ->OutputChannels; i++)
425 Out[i] = _cmsQuickSaturateWord(OutFloat[i] * 65535.0);
426
427 // Always succeed
428 return TRUE;
429 }
430
431 // Try to see if the curves of a given MPE are linear
432 static
AllCurvesAreLinear(cmsContext ContextID,cmsStage * mpe)433 cmsBool AllCurvesAreLinear(cmsContext ContextID, cmsStage* mpe)
434 {
435 cmsToneCurve** Curves;
436 cmsUInt32Number i, n;
437
438 Curves = _cmsStageGetPtrToCurveSet(mpe);
439 if (Curves == NULL) return FALSE;
440
441 n = cmsStageOutputChannels(ContextID, mpe);
442
443 for (i=0; i < n; i++) {
444 if (!cmsIsToneCurveLinear(ContextID, Curves[i])) return FALSE;
445 }
446
447 return TRUE;
448 }
449
450 // This function replaces a specific node placed in "At" by the "Value" numbers. Its purpose
451 // is to fix scum dot on broken profiles/transforms. Works on 1, 3 and 4 channels
452 static
PatchLUT(cmsContext ContextID,cmsStage * CLUT,cmsUInt16Number At[],cmsUInt16Number Value[],cmsUInt32Number nChannelsOut,cmsUInt32Number nChannelsIn)453 cmsBool PatchLUT(cmsContext ContextID, cmsStage* CLUT, cmsUInt16Number At[], cmsUInt16Number Value[],
454 cmsUInt32Number nChannelsOut, cmsUInt32Number nChannelsIn)
455 {
456 _cmsStageCLutData* Grid = (_cmsStageCLutData*) CLUT ->Data;
457 cmsInterpParams* p16 = Grid ->Params;
458 cmsFloat64Number px, py, pz, pw;
459 int x0, y0, z0, w0;
460 int i, index;
461
462 if (CLUT -> Type != cmsSigCLutElemType) {
463 cmsSignalError(ContextID, cmsERROR_INTERNAL, "(internal) Attempt to PatchLUT on non-lut stage");
464 return FALSE;
465 }
466
467 if (nChannelsIn == 4) {
468
469 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
470 py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
471 pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
472 pw = ((cmsFloat64Number) At[3] * (p16->Domain[3])) / 65535.0;
473
474 x0 = (int) floor(px);
475 y0 = (int) floor(py);
476 z0 = (int) floor(pz);
477 w0 = (int) floor(pw);
478
479 if (((px - x0) != 0) ||
480 ((py - y0) != 0) ||
481 ((pz - z0) != 0) ||
482 ((pw - w0) != 0)) return FALSE; // Not on exact node
483
484 index = (int) p16 -> opta[3] * x0 +
485 (int) p16 -> opta[2] * y0 +
486 (int) p16 -> opta[1] * z0 +
487 (int) p16 -> opta[0] * w0;
488 }
489 else
490 if (nChannelsIn == 3) {
491
492 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
493 py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
494 pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
495
496 x0 = (int) floor(px);
497 y0 = (int) floor(py);
498 z0 = (int) floor(pz);
499
500 if (((px - x0) != 0) ||
501 ((py - y0) != 0) ||
502 ((pz - z0) != 0)) return FALSE; // Not on exact node
503
504 index = (int) p16 -> opta[2] * x0 +
505 (int) p16 -> opta[1] * y0 +
506 (int) p16 -> opta[0] * z0;
507 }
508 else
509 if (nChannelsIn == 1) {
510
511 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
512
513 x0 = (int) floor(px);
514
515 if (((px - x0) != 0)) return FALSE; // Not on exact node
516
517 index = (int) p16 -> opta[0] * x0;
518 }
519 else {
520 cmsSignalError(ContextID, cmsERROR_INTERNAL, "(internal) %d Channels are not supported on PatchLUT", nChannelsIn);
521 return FALSE;
522 }
523
524 for (i = 0; i < (int) nChannelsOut; i++)
525 Grid->Tab.T[index + i] = Value[i];
526
527 return TRUE;
528 }
529
530 // Auxiliary, to see if two values are equal or very different
531 static
WhitesAreEqual(cmsUInt32Number n,cmsUInt16Number White1[],cmsUInt16Number White2[])532 cmsBool WhitesAreEqual(cmsUInt32Number n, cmsUInt16Number White1[], cmsUInt16Number White2[] )
533 {
534 cmsUInt32Number i;
535
536 for (i=0; i < n; i++) {
537
538 if (abs(White1[i] - White2[i]) > 0xf000) return TRUE; // Values are so extremely different that the fixup should be avoided
539 if (White1[i] != White2[i]) return FALSE;
540 }
541 return TRUE;
542 }
543
544
545 // Locate the node for the white point and fix it to pure white in order to avoid scum dot.
546 static
FixWhiteMisalignment(cmsContext ContextID,cmsPipeline * Lut,cmsColorSpaceSignature EntryColorSpace,cmsColorSpaceSignature ExitColorSpace)547 cmsBool FixWhiteMisalignment(cmsContext ContextID, cmsPipeline* Lut, cmsColorSpaceSignature EntryColorSpace, cmsColorSpaceSignature ExitColorSpace)
548 {
549 cmsUInt16Number *WhitePointIn, *WhitePointOut;
550 cmsUInt16Number WhiteIn[cmsMAXCHANNELS], WhiteOut[cmsMAXCHANNELS], ObtainedOut[cmsMAXCHANNELS];
551 cmsUInt32Number i, nOuts, nIns;
552 cmsStage *PreLin = NULL, *CLUT = NULL, *PostLin = NULL;
553
554 if (!_cmsEndPointsBySpace(EntryColorSpace,
555 &WhitePointIn, NULL, &nIns)) return FALSE;
556
557 if (!_cmsEndPointsBySpace(ExitColorSpace,
558 &WhitePointOut, NULL, &nOuts)) return FALSE;
559
560 // It needs to be fixed?
561 if (Lut ->InputChannels != nIns) return FALSE;
562 if (Lut ->OutputChannels != nOuts) return FALSE;
563
564 cmsPipelineEval16(ContextID, WhitePointIn, ObtainedOut, Lut);
565
566 if (WhitesAreEqual(nOuts, WhitePointOut, ObtainedOut)) return TRUE; // whites already match
567
568 // Check if the LUT comes as Prelin, CLUT or Postlin. We allow all combinations
569 if (!cmsPipelineCheckAndRetreiveStages(ContextID, Lut, 3, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, &PreLin, &CLUT, &PostLin))
570 if (!cmsPipelineCheckAndRetreiveStages(ContextID, Lut, 2, cmsSigCurveSetElemType, cmsSigCLutElemType, &PreLin, &CLUT))
571 if (!cmsPipelineCheckAndRetreiveStages(ContextID, Lut, 2, cmsSigCLutElemType, cmsSigCurveSetElemType, &CLUT, &PostLin))
572 if (!cmsPipelineCheckAndRetreiveStages(ContextID, Lut, 1, cmsSigCLutElemType, &CLUT))
573 return FALSE;
574
575 // We need to interpolate white points of both, pre and post curves
576 if (PreLin) {
577
578 cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PreLin);
579
580 for (i=0; i < nIns; i++) {
581 WhiteIn[i] = cmsEvalToneCurve16(ContextID, Curves[i], WhitePointIn[i]);
582 }
583 }
584 else {
585 for (i=0; i < nIns; i++)
586 WhiteIn[i] = WhitePointIn[i];
587 }
588
589 // If any post-linearization, we need to find how is represented white before the curve, do
590 // a reverse interpolation in this case.
591 if (PostLin) {
592
593 cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PostLin);
594
595 for (i=0; i < nOuts; i++) {
596
597 cmsToneCurve* InversePostLin = cmsReverseToneCurve(ContextID, Curves[i]);
598 if (InversePostLin == NULL) {
599 WhiteOut[i] = WhitePointOut[i];
600
601 } else {
602
603 WhiteOut[i] = cmsEvalToneCurve16(ContextID, InversePostLin, WhitePointOut[i]);
604 cmsFreeToneCurve(ContextID, InversePostLin);
605 }
606 }
607 }
608 else {
609 for (i=0; i < nOuts; i++)
610 WhiteOut[i] = WhitePointOut[i];
611 }
612
613 // Ok, proceed with patching. May fail and we don't care if it fails
614 PatchLUT(ContextID, CLUT, WhiteIn, WhiteOut, nOuts, nIns);
615
616 return TRUE;
617 }
618
619 // -----------------------------------------------------------------------------------------------------------------------------------------------
620 // This function creates simple LUT from complex ones. The generated LUT has an optional set of
621 // prelinearization curves, a CLUT of nGridPoints and optional postlinearization tables.
622 // These curves have to exist in the original LUT in order to be used in the simplified output.
623 // Caller may also use the flags to allow this feature.
624 // LUTS with all curves will be simplified to a single curve. Parametric curves are lost.
625 // This function should be used on 16-bits LUTS only, as floating point losses precision when simplified
626 // -----------------------------------------------------------------------------------------------------------------------------------------------
627
628 static
OptimizeByResampling(cmsContext ContextID,cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)629 cmsBool OptimizeByResampling(cmsContext ContextID, cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
630 {
631 cmsPipeline* Src = NULL;
632 cmsPipeline* Dest = NULL;
633 cmsStage* mpe;
634 cmsStage* CLUT;
635 cmsStage *KeepPreLin = NULL, *KeepPostLin = NULL;
636 cmsUInt32Number nGridPoints;
637 cmsColorSpaceSignature ColorSpace, OutputColorSpace;
638 cmsStage *NewPreLin = NULL;
639 cmsStage *NewPostLin = NULL;
640 _cmsStageCLutData* DataCLUT;
641 cmsToneCurve** DataSetIn;
642 cmsToneCurve** DataSetOut;
643 Prelin16Data* p16;
644
645 // This is a lossy optimization! does not apply in floating-point cases
646 if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
647
648 ColorSpace = _cmsICCcolorSpace(ContextID, (int) T_COLORSPACE(*InputFormat));
649 OutputColorSpace = _cmsICCcolorSpace(ContextID, (int) T_COLORSPACE(*OutputFormat));
650
651 // Color space must be specified
652 if (ColorSpace == (cmsColorSpaceSignature)0 ||
653 OutputColorSpace == (cmsColorSpaceSignature)0) return FALSE;
654
655 nGridPoints = _cmsReasonableGridpointsByColorspace(ContextID, ColorSpace, *dwFlags);
656
657 // For empty LUTs, 2 points are enough
658 if (cmsPipelineStageCount(ContextID, *Lut) == 0)
659 nGridPoints = 2;
660
661 Src = *Lut;
662
663 // Named color pipelines cannot be optimized either
664 for (mpe = cmsPipelineGetPtrToFirstStage(ContextID, Src);
665 mpe != NULL;
666 mpe = cmsStageNext(ContextID, mpe)) {
667 if (cmsStageType(ContextID, mpe) == cmsSigNamedColorElemType) return FALSE;
668 }
669
670 // Allocate an empty LUT
671 Dest = cmsPipelineAlloc(ContextID, Src ->InputChannels, Src ->OutputChannels);
672 if (!Dest) return FALSE;
673
674 // Prelinearization tables are kept unless indicated by flags
675 if (*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION) {
676
677 // Get a pointer to the prelinearization element
678 cmsStage* PreLin = cmsPipelineGetPtrToFirstStage(ContextID, Src);
679
680 // Check if suitable
681 if (PreLin && PreLin ->Type == cmsSigCurveSetElemType) {
682
683 // Maybe this is a linear tram, so we can avoid the whole stuff
684 if (!AllCurvesAreLinear(ContextID, PreLin)) {
685
686 // All seems ok, proceed.
687 NewPreLin = cmsStageDup(ContextID, PreLin);
688 if(!cmsPipelineInsertStage(ContextID, Dest, cmsAT_BEGIN, NewPreLin))
689 goto Error;
690
691 // Remove prelinearization. Since we have duplicated the curve
692 // in destination LUT, the sampling should be applied after this stage.
693 cmsPipelineUnlinkStage(ContextID, Src, cmsAT_BEGIN, &KeepPreLin);
694 }
695 }
696 }
697
698 ContextID->dwFlags = *dwFlags;
699 // Allocate the CLUT
700 CLUT = cmsStageAllocCLut16bit(ContextID, nGridPoints, Src ->InputChannels, Src->OutputChannels, NULL);
701 ContextID->dwFlags = 0;
702 if (CLUT == NULL) goto Error;
703
704 // Add the CLUT to the destination LUT
705 if (!cmsPipelineInsertStage(ContextID, Dest, cmsAT_END, CLUT)) {
706 goto Error;
707 }
708
709 // Postlinearization tables are kept unless indicated by flags
710 if (*dwFlags & cmsFLAGS_CLUT_POST_LINEARIZATION) {
711
712 // Get a pointer to the postlinearization if present
713 cmsStage* PostLin = cmsPipelineGetPtrToLastStage(ContextID, Src);
714
715 // Check if suitable
716 if (PostLin && cmsStageType(ContextID, PostLin) == cmsSigCurveSetElemType) {
717
718 // Maybe this is a linear tram, so we can avoid the whole stuff
719 if (!AllCurvesAreLinear(ContextID, PostLin)) {
720
721 // All seems ok, proceed.
722 NewPostLin = cmsStageDup(ContextID, PostLin);
723 if (!cmsPipelineInsertStage(ContextID, Dest, cmsAT_END, NewPostLin))
724 goto Error;
725
726 // In destination LUT, the sampling should be applied after this stage.
727 cmsPipelineUnlinkStage(ContextID, Src, cmsAT_END, &KeepPostLin);
728 }
729 }
730 }
731
732 // Now its time to do the sampling. We have to ignore pre/post linearization
733 // The source LUT without pre/post curves is passed as parameter.
734 if (!cmsStageSampleCLut16bit(ContextID, CLUT, XFormSampler16, (void*) Src, 0)) {
735 Error:
736 // Ops, something went wrong, Restore stages
737 if (KeepPreLin != NULL) {
738 if (!cmsPipelineInsertStage(ContextID, Src, cmsAT_BEGIN, KeepPreLin)) {
739 _cmsAssert(0); // This never happens
740 }
741 }
742 if (KeepPostLin != NULL) {
743 if (!cmsPipelineInsertStage(ContextID, Src, cmsAT_END, KeepPostLin)) {
744 _cmsAssert(0); // This never happens
745 }
746 }
747 cmsPipelineFree(ContextID, Dest);
748 return FALSE;
749 }
750
751 // Done.
752
753 if (KeepPreLin != NULL) cmsStageFree(ContextID, KeepPreLin);
754 if (KeepPostLin != NULL) cmsStageFree(ContextID, KeepPostLin);
755 cmsPipelineFree(ContextID, Src);
756
757 DataCLUT = (_cmsStageCLutData*) CLUT ->Data;
758
759 if (NewPreLin == NULL) DataSetIn = NULL;
760 else DataSetIn = ((_cmsStageToneCurvesData*) NewPreLin ->Data) ->TheCurves;
761
762 if (NewPostLin == NULL) DataSetOut = NULL;
763 else DataSetOut = ((_cmsStageToneCurvesData*) NewPostLin ->Data) ->TheCurves;
764
765
766 if (DataSetIn == NULL && DataSetOut == NULL) {
767
768 _cmsPipelineSetOptimizationParameters(ContextID, Dest, (_cmsOPTeval16Fn) DataCLUT->Params->Interpolation.Lerp16, DataCLUT->Params, NULL, NULL);
769 }
770 else {
771
772 p16 = PrelinOpt16alloc(ContextID,
773 DataCLUT ->Params,
774 Dest ->InputChannels,
775 DataSetIn,
776 Dest ->OutputChannels,
777 DataSetOut);
778
779 _cmsPipelineSetOptimizationParameters(ContextID, Dest, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
780 }
781
782
783 // Don't fix white on absolute colorimetric
784 if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
785 *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
786
787 if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
788
789 FixWhiteMisalignment(ContextID, Dest, ColorSpace, OutputColorSpace);
790 }
791
792 *Lut = Dest;
793 return TRUE;
794
795 cmsUNUSED_PARAMETER(Intent);
796 }
797
798
799 // -----------------------------------------------------------------------------------------------------------------------------------------------
800 // Fixes the gamma balancing of transform. This is described in my paper "Prelinearization Stages on
801 // Color-Management Application-Specific Integrated Circuits (ASICs)" presented at NIP24. It only works
802 // for RGB transforms. See the paper for more details
803 // -----------------------------------------------------------------------------------------------------------------------------------------------
804
805
806 // Normalize endpoints by slope limiting max and min. This assures endpoints as well.
807 // Descending curves are handled as well.
808 static
SlopeLimiting(cmsContext ContextID,cmsToneCurve * g)809 void SlopeLimiting(cmsContext ContextID, cmsToneCurve* g)
810 {
811 int BeginVal, EndVal;
812 int AtBegin = (int) floor((cmsFloat64Number) g ->nEntries * 0.02 + 0.5); // Cutoff at 2%
813 int AtEnd = (int) g ->nEntries - AtBegin - 1; // And 98%
814 cmsFloat64Number Val, Slope, beta;
815 int i;
816
817 if (cmsIsToneCurveDescending(ContextID, g)) {
818 BeginVal = 0xffff; EndVal = 0;
819 }
820 else {
821 BeginVal = 0; EndVal = 0xffff;
822 }
823
824 // Compute slope and offset for begin of curve
825 Val = g ->Table16[AtBegin];
826 Slope = (Val - BeginVal) / AtBegin;
827 beta = Val - Slope * AtBegin;
828
829 for (i=0; i < AtBegin; i++)
830 g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
831
832 // Compute slope and offset for the end
833 Val = g ->Table16[AtEnd];
834 Slope = (EndVal - Val) / AtBegin; // AtBegin holds the X interval, which is same in both cases
835 beta = Val - Slope * AtEnd;
836
837 for (i = AtEnd; i < (int) g ->nEntries; i++)
838 g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
839 }
840
841
842 // Precomputes tables for 8-bit on input devicelink.
843 static
PrelinOpt8alloc(cmsContext ContextID,const cmsInterpParams * p,cmsToneCurve * G[3])844 Prelin8Data* PrelinOpt8alloc(cmsContext ContextID, const cmsInterpParams* p, cmsToneCurve* G[3])
845 {
846 int i;
847 cmsUInt16Number Input[3];
848 cmsS15Fixed16Number v1, v2, v3;
849 Prelin8Data* p8;
850
851 p8 = (Prelin8Data*)_cmsMallocZero(ContextID, sizeof(Prelin8Data));
852 if (p8 == NULL) return NULL;
853
854 // Since this only works for 8 bit input, values comes always as x * 257,
855 // we can safely take msb byte (x << 8 + x)
856
857 for (i=0; i < 256; i++) {
858
859 if (G != NULL) {
860
861 // Get 16-bit representation
862 Input[0] = cmsEvalToneCurve16(ContextID, G[0], FROM_8_TO_16(i));
863 Input[1] = cmsEvalToneCurve16(ContextID, G[1], FROM_8_TO_16(i));
864 Input[2] = cmsEvalToneCurve16(ContextID, G[2], FROM_8_TO_16(i));
865 }
866 else {
867 Input[0] = FROM_8_TO_16(i);
868 Input[1] = FROM_8_TO_16(i);
869 Input[2] = FROM_8_TO_16(i);
870 }
871
872
873 // Move to 0..1.0 in fixed domain
874 v1 = _cmsToFixedDomain((int) (Input[0] * p -> Domain[0]));
875 v2 = _cmsToFixedDomain((int) (Input[1] * p -> Domain[1]));
876 v3 = _cmsToFixedDomain((int) (Input[2] * p -> Domain[2]));
877
878 // Store the precalculated table of nodes
879 p8 ->X0[i] = (p->opta[2] * FIXED_TO_INT(v1));
880 p8 ->Y0[i] = (p->opta[1] * FIXED_TO_INT(v2));
881 p8 ->Z0[i] = (p->opta[0] * FIXED_TO_INT(v3));
882
883 // Store the precalculated table of offsets
884 p8 ->rx[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v1);
885 p8 ->ry[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v2);
886 p8 ->rz[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v3);
887 }
888
889 p8 ->ContextID = ContextID;
890 p8 ->p = p;
891
892 return p8;
893 }
894
895 static
Prelin8free(cmsContext ContextID,void * ptr)896 void Prelin8free(cmsContext ContextID, void* ptr)
897 {
898 _cmsFree(ContextID, ptr);
899 }
900
901 static
Prelin8dup(cmsContext ContextID,const void * ptr)902 void* Prelin8dup(cmsContext ContextID, const void* ptr)
903 {
904 return _cmsDupMem(ContextID, ptr, sizeof(Prelin8Data));
905 }
906
907
908
909 // A optimized interpolation for 8-bit input.
910 #define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
911 static CMS_NO_SANITIZE
PrelinEval8(cmsContext ContextID,register const cmsUInt16Number Input[],register cmsUInt16Number Output[],register const void * D)912 void PrelinEval8(cmsContext ContextID, register const cmsUInt16Number Input[],
913 register cmsUInt16Number Output[],
914 register const void* D)
915 {
916 cmsUInt8Number r, g, b;
917 cmsS15Fixed16Number rx, ry, rz;
918 cmsS15Fixed16Number c0, c1, c2, c3, Rest;
919 int OutChan;
920 register cmsS15Fixed16Number X0, X1, Y0, Y1, Z0, Z1;
921 Prelin8Data* p8 = (Prelin8Data*) D;
922 register const cmsInterpParams* p = p8 ->p;
923 int TotalOut = (int) p -> nOutputs;
924 const cmsUInt16Number* LutTable = (const cmsUInt16Number*) p->Table;
925 cmsUNUSED_PARAMETER(ContextID);
926
927 r = (cmsUInt8Number) (Input[0] >> 8);
928 g = (cmsUInt8Number) (Input[1] >> 8);
929 b = (cmsUInt8Number) (Input[2] >> 8);
930
931 X0 = X1 = (cmsS15Fixed16Number) p8->X0[r];
932 Y0 = Y1 = (cmsS15Fixed16Number) p8->Y0[g];
933 Z0 = Z1 = (cmsS15Fixed16Number) p8->Z0[b];
934
935 rx = p8 ->rx[r];
936 ry = p8 ->ry[g];
937 rz = p8 ->rz[b];
938
939 X1 = X0 + (cmsS15Fixed16Number)((rx == 0) ? 0 : p ->opta[2]);
940 Y1 = Y0 + (cmsS15Fixed16Number)((ry == 0) ? 0 : p ->opta[1]);
941 Z1 = Z0 + (cmsS15Fixed16Number)((rz == 0) ? 0 : p ->opta[0]);
942
943
944 // These are the 6 Tetrahedral
945 for (OutChan=0; OutChan < TotalOut; OutChan++) {
946
947 c0 = DENS(X0, Y0, Z0);
948
949 if (rx >= ry && ry >= rz)
950 {
951 c1 = DENS(X1, Y0, Z0) - c0;
952 c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
953 c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
954 }
955 else
956 if (rx >= rz && rz >= ry)
957 {
958 c1 = DENS(X1, Y0, Z0) - c0;
959 c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
960 c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
961 }
962 else
963 if (rz >= rx && rx >= ry)
964 {
965 c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
966 c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
967 c3 = DENS(X0, Y0, Z1) - c0;
968 }
969 else
970 if (ry >= rx && rx >= rz)
971 {
972 c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
973 c2 = DENS(X0, Y1, Z0) - c0;
974 c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
975 }
976 else
977 if (ry >= rz && rz >= rx)
978 {
979 c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
980 c2 = DENS(X0, Y1, Z0) - c0;
981 c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);
982 }
983 else
984 if (rz >= ry && ry >= rx)
985 {
986 c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
987 c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
988 c3 = DENS(X0, Y0, Z1) - c0;
989 }
990 else {
991 c1 = c2 = c3 = 0;
992 }
993
994 Rest = c1 * rx + c2 * ry + c3 * rz + 0x8001;
995 Output[OutChan] = (cmsUInt16Number) (c0 + ((Rest + (Rest >> 16)) >> 16));
996
997 }
998 }
999
1000 #undef DENS
1001
1002
1003 // Curves that contain wide empty areas are not optimizeable
1004 static
IsDegenerated(const cmsToneCurve * g)1005 cmsBool IsDegenerated(const cmsToneCurve* g)
1006 {
1007 cmsUInt32Number i, Zeros = 0, Poles = 0;
1008 cmsUInt32Number nEntries = g ->nEntries;
1009
1010 for (i=0; i < nEntries; i++) {
1011
1012 if (g ->Table16[i] == 0x0000) Zeros++;
1013 if (g ->Table16[i] == 0xffff) Poles++;
1014 }
1015
1016 if (Zeros == 1 && Poles == 1) return FALSE; // For linear tables
1017 if (Zeros > (nEntries / 20)) return TRUE; // Degenerated, many zeros
1018 if (Poles > (nEntries / 20)) return TRUE; // Degenerated, many poles
1019
1020 return FALSE;
1021 }
1022
1023 // --------------------------------------------------------------------------------------------------------------
1024 // We need xput over here
1025
1026 static
OptimizeByComputingLinearization(cmsContext ContextID,cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1027 cmsBool OptimizeByComputingLinearization(cmsContext ContextID, cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1028 {
1029 cmsPipeline* OriginalLut;
1030 cmsUInt32Number nGridPoints;
1031 cmsToneCurve *Trans[cmsMAXCHANNELS], *TransReverse[cmsMAXCHANNELS];
1032 cmsUInt32Number t, i;
1033 cmsFloat32Number v, In[cmsMAXCHANNELS], Out[cmsMAXCHANNELS];
1034 cmsBool lIsSuitable, lIsLinear;
1035 cmsPipeline* OptimizedLUT = NULL, *LutPlusCurves = NULL;
1036 cmsStage* OptimizedCLUTmpe;
1037 cmsColorSpaceSignature ColorSpace, OutputColorSpace;
1038 cmsStage* OptimizedPrelinMpe;
1039 cmsStage* mpe;
1040 cmsToneCurve** OptimizedPrelinCurves;
1041 _cmsStageCLutData* OptimizedPrelinCLUT;
1042
1043
1044 // This is a lossy optimization! does not apply in floating-point cases
1045 if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
1046
1047 // Only on chunky RGB
1048 if (T_COLORSPACE(*InputFormat) != PT_RGB) return FALSE;
1049 if (T_PLANAR(*InputFormat)) return FALSE;
1050
1051 if (T_COLORSPACE(*OutputFormat) != PT_RGB) return FALSE;
1052 if (T_PLANAR(*OutputFormat)) return FALSE;
1053
1054 // On 16 bits, user has to specify the feature
1055 if (!_cmsFormatterIs8bit(*InputFormat)) {
1056 if (!(*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION)) return FALSE;
1057 }
1058
1059 OriginalLut = *Lut;
1060
1061 // Named color pipelines cannot be optimized either
1062 for (mpe = cmsPipelineGetPtrToFirstStage(ContextID, OriginalLut);
1063 mpe != NULL;
1064 mpe = cmsStageNext(ContextID, mpe)) {
1065 if (cmsStageType(ContextID, mpe) == cmsSigNamedColorElemType) return FALSE;
1066 }
1067
1068 ColorSpace = _cmsICCcolorSpace(ContextID, (int) T_COLORSPACE(*InputFormat));
1069 OutputColorSpace = _cmsICCcolorSpace(ContextID, (int) T_COLORSPACE(*OutputFormat));
1070
1071 // Color space must be specified
1072 if (ColorSpace == (cmsColorSpaceSignature)0 ||
1073 OutputColorSpace == (cmsColorSpaceSignature)0) return FALSE;
1074
1075 nGridPoints = _cmsReasonableGridpointsByColorspace(ContextID, ColorSpace, *dwFlags);
1076
1077 // Empty gamma containers
1078 memset(Trans, 0, sizeof(Trans));
1079 memset(TransReverse, 0, sizeof(TransReverse));
1080
1081 // If the last stage of the original lut are curves, and those curves are
1082 // degenerated, it is likely the transform is squeezing and clipping
1083 // the output from previous CLUT. We cannot optimize this case
1084 {
1085 cmsStage* last = cmsPipelineGetPtrToLastStage(ContextID, OriginalLut);
1086
1087 if (cmsStageType(ContextID, last) == cmsSigCurveSetElemType) {
1088
1089 _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*)cmsStageData(ContextID, last);
1090 for (i = 0; i < Data->nCurves; i++) {
1091 if (IsDegenerated(Data->TheCurves[i]))
1092 goto Error;
1093 }
1094 }
1095 }
1096
1097 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1098 Trans[t] = cmsBuildTabulatedToneCurve16(ContextID, PRELINEARIZATION_POINTS, NULL);
1099 if (Trans[t] == NULL) goto Error;
1100 }
1101
1102 // Populate the curves
1103 for (i=0; i < PRELINEARIZATION_POINTS; i++) {
1104
1105 v = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
1106
1107 // Feed input with a gray ramp
1108 for (t=0; t < OriginalLut ->InputChannels; t++)
1109 In[t] = v;
1110
1111 // Evaluate the gray value
1112 cmsPipelineEvalFloat(ContextID, In, Out, OriginalLut);
1113
1114 // Store result in curve
1115 for (t=0; t < OriginalLut ->InputChannels; t++)
1116 Trans[t] ->Table16[i] = _cmsQuickSaturateWord(Out[t] * 65535.0);
1117 }
1118
1119 // Slope-limit the obtained curves
1120 for (t = 0; t < OriginalLut ->InputChannels; t++)
1121 SlopeLimiting(ContextID, Trans[t]);
1122
1123 // Check for validity
1124 lIsSuitable = TRUE;
1125 lIsLinear = TRUE;
1126 for (t=0; (lIsSuitable && (t < OriginalLut ->InputChannels)); t++) {
1127
1128 // Exclude if already linear
1129 if (!cmsIsToneCurveLinear(ContextID, Trans[t]))
1130 lIsLinear = FALSE;
1131
1132 // Exclude if non-monotonic
1133 if (!cmsIsToneCurveMonotonic(ContextID, Trans[t]))
1134 lIsSuitable = FALSE;
1135
1136 if (IsDegenerated(Trans[t]))
1137 lIsSuitable = FALSE;
1138 }
1139
1140 // If it is not suitable, just quit
1141 if (!lIsSuitable) goto Error;
1142
1143 // Invert curves if possible
1144 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1145 TransReverse[t] = cmsReverseToneCurveEx(ContextID, PRELINEARIZATION_POINTS, Trans[t]);
1146 if (TransReverse[t] == NULL) goto Error;
1147 }
1148
1149 // Now inset the reversed curves at the begin of transform
1150 LutPlusCurves = cmsPipelineDup(ContextID, OriginalLut);
1151 if (LutPlusCurves == NULL) goto Error;
1152
1153 if (!cmsPipelineInsertStage(ContextID, LutPlusCurves, cmsAT_BEGIN, cmsStageAllocToneCurves(ContextID, OriginalLut ->InputChannels, TransReverse)))
1154 goto Error;
1155
1156 // Create the result LUT
1157 OptimizedLUT = cmsPipelineAlloc(ContextID, OriginalLut ->InputChannels, OriginalLut ->OutputChannels);
1158 if (OptimizedLUT == NULL) goto Error;
1159
1160 OptimizedPrelinMpe = cmsStageAllocToneCurves(ContextID, OriginalLut ->InputChannels, Trans);
1161
1162 // Create and insert the curves at the beginning
1163 if (!cmsPipelineInsertStage(ContextID, OptimizedLUT, cmsAT_BEGIN, OptimizedPrelinMpe))
1164 goto Error;
1165
1166 // Allocate the CLUT for result
1167 OptimizedCLUTmpe = cmsStageAllocCLut16bit(ContextID, nGridPoints, OriginalLut ->InputChannels, OriginalLut ->OutputChannels, NULL);
1168
1169 // Add the CLUT to the destination LUT
1170 if (!cmsPipelineInsertStage(ContextID, OptimizedLUT, cmsAT_END, OptimizedCLUTmpe))
1171 goto Error;
1172
1173 // Resample the LUT
1174 if (!cmsStageSampleCLut16bit(ContextID, OptimizedCLUTmpe, XFormSampler16, (void*) LutPlusCurves, 0)) goto Error;
1175
1176 // Free resources
1177 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1178
1179 if (Trans[t]) cmsFreeToneCurve(ContextID, Trans[t]);
1180 if (TransReverse[t]) cmsFreeToneCurve(ContextID, TransReverse[t]);
1181 }
1182
1183 cmsPipelineFree(ContextID, LutPlusCurves);
1184
1185
1186 OptimizedPrelinCurves = _cmsStageGetPtrToCurveSet(OptimizedPrelinMpe);
1187 OptimizedPrelinCLUT = (_cmsStageCLutData*) OptimizedCLUTmpe ->Data;
1188
1189 // Set the evaluator if 8-bit
1190 if (_cmsFormatterIs8bit(*InputFormat)) {
1191
1192 Prelin8Data* p8 = PrelinOpt8alloc(ContextID,
1193 OptimizedPrelinCLUT ->Params,
1194 OptimizedPrelinCurves);
1195 if (p8 == NULL) return FALSE;
1196
1197 _cmsPipelineSetOptimizationParameters(ContextID, OptimizedLUT, PrelinEval8, (void*) p8, Prelin8free, Prelin8dup);
1198
1199 }
1200 else
1201 {
1202 Prelin16Data* p16 = PrelinOpt16alloc(ContextID,
1203 OptimizedPrelinCLUT ->Params,
1204 3, OptimizedPrelinCurves, 3, NULL);
1205 if (p16 == NULL) return FALSE;
1206
1207 _cmsPipelineSetOptimizationParameters(ContextID, OptimizedLUT, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
1208
1209 }
1210
1211 // Don't fix white on absolute colorimetric
1212 if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
1213 *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
1214
1215 if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
1216
1217 if (!FixWhiteMisalignment(ContextID, OptimizedLUT, ColorSpace, OutputColorSpace)) {
1218
1219 return FALSE;
1220 }
1221 }
1222
1223 // And return the obtained LUT
1224
1225 cmsPipelineFree(ContextID, OriginalLut);
1226 *Lut = OptimizedLUT;
1227 return TRUE;
1228
1229 Error:
1230
1231 for (t = 0; t < OriginalLut ->InputChannels; t++) {
1232
1233 if (Trans[t]) cmsFreeToneCurve(ContextID, Trans[t]);
1234 if (TransReverse[t]) cmsFreeToneCurve(ContextID, TransReverse[t]);
1235 }
1236
1237 if (LutPlusCurves != NULL) cmsPipelineFree(ContextID, LutPlusCurves);
1238 if (OptimizedLUT != NULL) cmsPipelineFree(ContextID, OptimizedLUT);
1239
1240 return FALSE;
1241
1242 cmsUNUSED_PARAMETER(Intent);
1243 cmsUNUSED_PARAMETER(lIsLinear);
1244 }
1245
1246
1247 // Curves optimizer ------------------------------------------------------------------------------------------------------------------
1248
1249 static
CurvesFree(cmsContext ContextID,void * ptr)1250 void CurvesFree(cmsContext ContextID, void* ptr)
1251 {
1252 Curves16Data* Data = (Curves16Data*) ptr;
1253 cmsUInt32Number i;
1254
1255 for (i=0; i < Data -> nCurves; i++) {
1256
1257 _cmsFree(ContextID, Data ->Curves[i]);
1258 }
1259
1260 _cmsFree(ContextID, Data ->Curves);
1261 _cmsFree(ContextID, ptr);
1262 }
1263
1264 static
CurvesDup(cmsContext ContextID,const void * ptr)1265 void* CurvesDup(cmsContext ContextID, const void* ptr)
1266 {
1267 Curves16Data* Data = (Curves16Data*)_cmsDupMem(ContextID, ptr, sizeof(Curves16Data));
1268 cmsUInt32Number i;
1269
1270 if (Data == NULL) return NULL;
1271
1272 Data->Curves = (cmsUInt16Number**) _cmsDupMem(ContextID, Data->Curves, Data->nCurves * sizeof(cmsUInt16Number*));
1273
1274 for (i=0; i < Data -> nCurves; i++) {
1275 Data->Curves[i] = (cmsUInt16Number*) _cmsDupMem(ContextID, Data->Curves[i], Data->nElements * sizeof(cmsUInt16Number));
1276 }
1277
1278 return (void*) Data;
1279 }
1280
1281 // Precomputes tables for 8-bit on input devicelink.
1282 static
CurvesAlloc(cmsContext ContextID,cmsUInt32Number nCurves,cmsUInt32Number nElements,cmsToneCurve ** G)1283 Curves16Data* CurvesAlloc(cmsContext ContextID, cmsUInt32Number nCurves, cmsUInt32Number nElements, cmsToneCurve** G)
1284 {
1285 cmsUInt32Number i, j;
1286 Curves16Data* c16;
1287
1288 c16 = (Curves16Data*)_cmsMallocZero(ContextID, sizeof(Curves16Data));
1289 if (c16 == NULL) return NULL;
1290
1291 c16 ->nCurves = nCurves;
1292 c16 ->nElements = nElements;
1293
1294 c16->Curves = (cmsUInt16Number**) _cmsCalloc(ContextID, nCurves, sizeof(cmsUInt16Number*));
1295 if (c16->Curves == NULL) {
1296 _cmsFree(ContextID, c16);
1297 return NULL;
1298 }
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 == 256U) {
1315
1316 for (j=0; j < nElements; j++) {
1317
1318 c16 ->Curves[i][j] = cmsEvalToneCurve16(ContextID, 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(ContextID, G[i], (cmsUInt16Number) j);
1325 }
1326 }
1327 }
1328
1329 return c16;
1330 }
1331
1332 static
FastEvaluateCurves8(cmsContext ContextID,register const cmsUInt16Number In[],register cmsUInt16Number Out[],register const void * D)1333 void FastEvaluateCurves8(cmsContext ContextID, register const cmsUInt16Number In[],
1334 register cmsUInt16Number Out[],
1335 register const void* D)
1336 {
1337 Curves16Data* Data = (Curves16Data*) D;
1338 int x;
1339 cmsUInt32Number i;
1340 cmsUNUSED_PARAMETER(ContextID);
1341
1342 for (i=0; i < Data ->nCurves; i++) {
1343
1344 x = (In[i] >> 8);
1345 Out[i] = Data -> Curves[i][x];
1346 }
1347 }
1348
1349
1350 static
FastEvaluateCurves16(cmsContext ContextID,register const cmsUInt16Number In[],register cmsUInt16Number Out[],register const void * D)1351 void FastEvaluateCurves16(cmsContext ContextID, register const cmsUInt16Number In[],
1352 register cmsUInt16Number Out[],
1353 register const void* D)
1354 {
1355 Curves16Data* Data = (Curves16Data*) D;
1356 cmsUInt32Number i;
1357 cmsUNUSED_PARAMETER(ContextID);
1358
1359 for (i=0; i < Data ->nCurves; i++) {
1360 Out[i] = Data -> Curves[i][In[i]];
1361 }
1362 }
1363
1364
1365 static
FastIdentity16(cmsContext ContextID,register const cmsUInt16Number In[],register cmsUInt16Number Out[],register const void * D)1366 void FastIdentity16(cmsContext ContextID, register const cmsUInt16Number In[],
1367 register cmsUInt16Number Out[],
1368 register const void* D)
1369 {
1370 cmsPipeline* Lut = (cmsPipeline*) D;
1371 cmsUInt32Number i;
1372 cmsUNUSED_PARAMETER(ContextID);
1373
1374 for (i=0; i < Lut ->InputChannels; i++) {
1375 Out[i] = In[i];
1376 }
1377 }
1378
1379
1380 // If the target LUT holds only curves, the optimization procedure is to join all those
1381 // curves together. That only works on curves and does not work on matrices.
1382 static
OptimizeByJoiningCurves(cmsContext ContextID,cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1383 cmsBool OptimizeByJoiningCurves(cmsContext ContextID, cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1384 {
1385 cmsToneCurve** GammaTables = NULL;
1386 cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
1387 cmsUInt32Number i, j;
1388 cmsPipeline* Src = *Lut;
1389 cmsPipeline* Dest = NULL;
1390 cmsStage* mpe;
1391 cmsStage* ObtainedCurves = NULL;
1392
1393
1394 // This is a lossy optimization! does not apply in floating-point cases
1395 if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
1396
1397 // Only curves in this LUT?
1398 for (mpe = cmsPipelineGetPtrToFirstStage(ContextID, Src);
1399 mpe != NULL;
1400 mpe = cmsStageNext(ContextID, mpe)) {
1401 if (cmsStageType(ContextID, mpe) != cmsSigCurveSetElemType) return FALSE;
1402 }
1403
1404 // Allocate an empty LUT
1405 Dest = cmsPipelineAlloc(ContextID, Src ->InputChannels, Src ->OutputChannels);
1406 if (Dest == NULL) return FALSE;
1407
1408 // Create target curves
1409 GammaTables = (cmsToneCurve**) _cmsCalloc(ContextID, Src ->InputChannels, sizeof(cmsToneCurve*));
1410 if (GammaTables == NULL) goto Error;
1411
1412 for (i=0; i < Src ->InputChannels; i++) {
1413 GammaTables[i] = cmsBuildTabulatedToneCurve16(ContextID, PRELINEARIZATION_POINTS, NULL);
1414 if (GammaTables[i] == NULL) goto Error;
1415 }
1416
1417 // Compute 16 bit result by using floating point
1418 for (i=0; i < PRELINEARIZATION_POINTS; i++) {
1419
1420 for (j=0; j < Src ->InputChannels; j++)
1421 InFloat[j] = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
1422
1423 cmsPipelineEvalFloat(ContextID, InFloat, OutFloat, Src);
1424
1425 for (j=0; j < Src ->InputChannels; j++)
1426 GammaTables[j] -> Table16[i] = _cmsQuickSaturateWord(OutFloat[j] * 65535.0);
1427 }
1428
1429 ObtainedCurves = cmsStageAllocToneCurves(ContextID, Src ->InputChannels, GammaTables);
1430 if (ObtainedCurves == NULL) goto Error;
1431
1432 for (i=0; i < Src ->InputChannels; i++) {
1433 cmsFreeToneCurve(ContextID, GammaTables[i]);
1434 GammaTables[i] = NULL;
1435 }
1436
1437 if (GammaTables != NULL) {
1438 _cmsFree(ContextID, GammaTables);
1439 GammaTables = NULL;
1440 }
1441
1442 // Maybe the curves are linear at the end
1443 if (!AllCurvesAreLinear(ContextID, ObtainedCurves)) {
1444 _cmsStageToneCurvesData* Data;
1445
1446 if (!cmsPipelineInsertStage(ContextID, Dest, cmsAT_BEGIN, ObtainedCurves))
1447 goto Error;
1448 Data = (_cmsStageToneCurvesData*) cmsStageData(ContextID, ObtainedCurves);
1449 ObtainedCurves = NULL;
1450
1451 // If the curves are to be applied in 8 bits, we can save memory
1452 if (_cmsFormatterIs8bit(*InputFormat)) {
1453 Curves16Data* c16 = CurvesAlloc(ContextID, Data ->nCurves, 256, Data ->TheCurves);
1454
1455 if (c16 == NULL) goto Error;
1456 *dwFlags |= cmsFLAGS_NOCACHE;
1457 _cmsPipelineSetOptimizationParameters(ContextID, Dest, FastEvaluateCurves8, c16, CurvesFree, CurvesDup);
1458
1459 }
1460 else {
1461 Curves16Data* c16 = CurvesAlloc(ContextID, Data ->nCurves, 65536, Data ->TheCurves);
1462
1463 if (c16 == NULL) goto Error;
1464 *dwFlags |= cmsFLAGS_NOCACHE;
1465 _cmsPipelineSetOptimizationParameters(ContextID, Dest, FastEvaluateCurves16, c16, CurvesFree, CurvesDup);
1466 }
1467 }
1468 else {
1469
1470 // LUT optimizes to nothing. Set the identity LUT
1471 cmsStageFree(ContextID, ObtainedCurves);
1472 ObtainedCurves = NULL;
1473
1474 if (!cmsPipelineInsertStage(ContextID, Dest, cmsAT_BEGIN, cmsStageAllocIdentity(ContextID, Src ->InputChannels)))
1475 goto Error;
1476
1477 *dwFlags |= cmsFLAGS_NOCACHE;
1478 _cmsPipelineSetOptimizationParameters(ContextID, Dest, FastIdentity16, (void*) Dest, NULL, NULL);
1479 }
1480
1481 // We are done.
1482 cmsPipelineFree(ContextID, Src);
1483 *Lut = Dest;
1484 return TRUE;
1485
1486 Error:
1487
1488 if (ObtainedCurves != NULL) cmsStageFree(ContextID, ObtainedCurves);
1489 if (GammaTables != NULL) {
1490 for (i=0; i < Src ->InputChannels; i++) {
1491 if (GammaTables[i] != NULL) cmsFreeToneCurve(ContextID, GammaTables[i]);
1492 }
1493
1494 _cmsFree(ContextID, GammaTables);
1495 }
1496
1497 if (Dest != NULL) cmsPipelineFree(ContextID, Dest);
1498 return FALSE;
1499
1500 cmsUNUSED_PARAMETER(Intent);
1501 cmsUNUSED_PARAMETER(InputFormat);
1502 cmsUNUSED_PARAMETER(OutputFormat);
1503 cmsUNUSED_PARAMETER(dwFlags);
1504 }
1505
1506 // -------------------------------------------------------------------------------------------------------------------------------------
1507 // LUT is Shaper - Matrix - Matrix - Shaper, which is very frequent when combining two matrix-shaper profiles
1508
1509
1510 static
FreeMatShaper(cmsContext ContextID,void * Data)1511 void FreeMatShaper(cmsContext ContextID, void* Data)
1512 {
1513 if (Data != NULL) _cmsFree(ContextID, Data);
1514 }
1515
1516 static
DupMatShaper(cmsContext ContextID,const void * Data)1517 void* DupMatShaper(cmsContext ContextID, const void* Data)
1518 {
1519 return _cmsDupMem(ContextID, Data, sizeof(MatShaper8Data));
1520 }
1521
1522
1523 // A fast matrix-shaper evaluator for 8 bits. This is a bit ticky since I'm using 1.14 signed fixed point
1524 // to accomplish some performance. Actually it takes 256x3 16 bits tables and 16385 x 3 tables of 8 bits,
1525 // in total about 50K, and the performance boost is huge!
1526 static
MatShaperEval16(cmsContext ContextID,register const cmsUInt16Number In[],register cmsUInt16Number Out[],register const void * D)1527 void MatShaperEval16(cmsContext ContextID, register const cmsUInt16Number In[],
1528 register cmsUInt16Number Out[],
1529 register const void* D)
1530 {
1531 MatShaper8Data* p = (MatShaper8Data*) D;
1532 cmsS1Fixed14Number l1, l2, l3, r, g, b;
1533 cmsUInt32Number ri, gi, bi;
1534 cmsUNUSED_PARAMETER(ContextID);
1535
1536 // In this case (and only in this case!) we can use this simplification since
1537 // In[] is assured to come from a 8 bit number. (a << 8 | a)
1538 ri = In[0] & 0xFFU;
1539 gi = In[1] & 0xFFU;
1540 bi = In[2] & 0xFFU;
1541
1542 // Across first shaper, which also converts to 1.14 fixed point
1543 r = p->Shaper1R[ri];
1544 g = p->Shaper1G[gi];
1545 b = p->Shaper1B[bi];
1546
1547 // Evaluate the matrix in 1.14 fixed point
1548 l1 = (p->Mat[0][0] * r + p->Mat[0][1] * g + p->Mat[0][2] * b + p->Off[0] + 0x2000) >> 14;
1549 l2 = (p->Mat[1][0] * r + p->Mat[1][1] * g + p->Mat[1][2] * b + p->Off[1] + 0x2000) >> 14;
1550 l3 = (p->Mat[2][0] * r + p->Mat[2][1] * g + p->Mat[2][2] * b + p->Off[2] + 0x2000) >> 14;
1551
1552 // Now we have to clip to 0..1.0 range
1553 ri = (l1 < 0) ? 0 : ((l1 > 16384) ? 16384U : (cmsUInt32Number) l1);
1554 gi = (l2 < 0) ? 0 : ((l2 > 16384) ? 16384U : (cmsUInt32Number) l2);
1555 bi = (l3 < 0) ? 0 : ((l3 > 16384) ? 16384U : (cmsUInt32Number) l3);
1556
1557 // And across second shaper,
1558 Out[0] = p->Shaper2R[ri];
1559 Out[1] = p->Shaper2G[gi];
1560 Out[2] = p->Shaper2B[bi];
1561
1562 }
1563
1564 // This table converts from 8 bits to 1.14 after applying the curve
1565 static
FillFirstShaper(cmsContext ContextID,cmsS1Fixed14Number * Table,cmsToneCurve * Curve)1566 void FillFirstShaper(cmsContext ContextID, cmsS1Fixed14Number* Table, cmsToneCurve* Curve)
1567 {
1568 int i;
1569 cmsFloat32Number R, y;
1570
1571 for (i=0; i < 256; i++) {
1572
1573 R = (cmsFloat32Number) (i / 255.0);
1574 y = cmsEvalToneCurveFloat(ContextID, Curve, R);
1575
1576 if (y < 131072.0)
1577 Table[i] = DOUBLE_TO_1FIXED14(y);
1578 else
1579 Table[i] = 0x7fffffff;
1580 }
1581 }
1582
1583 // This table converts form 1.14 (being 0x4000 the last entry) to 8 bits after applying the curve
1584 static
FillSecondShaper(cmsContext ContextID,cmsUInt16Number * Table,cmsToneCurve * Curve,cmsBool Is8BitsOutput)1585 void FillSecondShaper(cmsContext ContextID, cmsUInt16Number* Table, cmsToneCurve* Curve, cmsBool Is8BitsOutput)
1586 {
1587 int i;
1588 cmsFloat32Number R, Val;
1589
1590 for (i=0; i < 16385; i++) {
1591
1592 R = (cmsFloat32Number) (i / 16384.0);
1593 Val = cmsEvalToneCurveFloat(ContextID, Curve, R); // Val comes 0..1.0
1594
1595 if (Val < 0)
1596 Val = 0;
1597
1598 if (Val > 1.0)
1599 Val = 1.0;
1600
1601 if (Is8BitsOutput) {
1602
1603 // If 8 bits output, we can optimize further by computing the / 257 part.
1604 // first we compute the resulting byte and then we store the byte times
1605 // 257. This quantization allows to round very quick by doing a >> 8, but
1606 // since the low byte is always equal to msb, we can do a & 0xff and this works!
1607 cmsUInt16Number w = _cmsQuickSaturateWord(Val * 65535.0);
1608 cmsUInt8Number b = FROM_16_TO_8(w);
1609
1610 Table[i] = FROM_8_TO_16(b);
1611 }
1612 else Table[i] = _cmsQuickSaturateWord(Val * 65535.0);
1613 }
1614 }
1615
1616 // Compute the matrix-shaper structure
1617 static
SetMatShaper(cmsContext ContextID,cmsPipeline * Dest,cmsToneCurve * Curve1[3],cmsMAT3 * Mat,cmsVEC3 * Off,cmsToneCurve * Curve2[3],cmsUInt32Number * OutputFormat)1618 cmsBool SetMatShaper(cmsContext ContextID, cmsPipeline* Dest, cmsToneCurve* Curve1[3], cmsMAT3* Mat, cmsVEC3* Off, cmsToneCurve* Curve2[3], cmsUInt32Number* OutputFormat)
1619 {
1620 MatShaper8Data* p;
1621 int i, j;
1622 cmsBool Is8Bits = _cmsFormatterIs8bit(*OutputFormat);
1623
1624 // Allocate a big chuck of memory to store precomputed tables
1625 p = (MatShaper8Data*) _cmsMalloc(ContextID, sizeof(MatShaper8Data));
1626 if (p == NULL) return FALSE;
1627
1628 // Precompute tables
1629 FillFirstShaper(ContextID, p ->Shaper1R, Curve1[0]);
1630 FillFirstShaper(ContextID, p ->Shaper1G, Curve1[1]);
1631 FillFirstShaper(ContextID, p ->Shaper1B, Curve1[2]);
1632
1633 FillSecondShaper(ContextID, p ->Shaper2R, Curve2[0], Is8Bits);
1634 FillSecondShaper(ContextID, p ->Shaper2G, Curve2[1], Is8Bits);
1635 FillSecondShaper(ContextID, p ->Shaper2B, Curve2[2], Is8Bits);
1636
1637 // Convert matrix to nFixed14. Note that those values may take more than 16 bits
1638 for (i=0; i < 3; i++) {
1639 for (j=0; j < 3; j++) {
1640 p ->Mat[i][j] = DOUBLE_TO_1FIXED14(Mat->v[i].n[j]);
1641 }
1642 }
1643
1644 for (i=0; i < 3; i++) {
1645
1646 if (Off == NULL) {
1647 p ->Off[i] = 0;
1648 }
1649 else {
1650 p ->Off[i] = DOUBLE_TO_1FIXED14(Off->n[i]);
1651 }
1652 }
1653
1654 // Mark as optimized for faster formatter
1655 if (Is8Bits)
1656 *OutputFormat |= OPTIMIZED_SH(1);
1657
1658 // Fill function pointers
1659 _cmsPipelineSetOptimizationParameters(ContextID, Dest, MatShaperEval16, (void*) p, FreeMatShaper, DupMatShaper);
1660 return TRUE;
1661 }
1662
1663 // 8 bits on input allows matrix-shaper boot up to 25 Mpixels per second on RGB. That's fast!
1664 static
OptimizeMatrixShaper(cmsContext ContextID,cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1665 cmsBool OptimizeMatrixShaper(cmsContext ContextID, cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1666 {
1667 cmsStage* Curve1, *Curve2;
1668 cmsStage* Matrix1, *Matrix2;
1669 cmsMAT3 res;
1670 cmsBool IdentityMat;
1671 cmsPipeline* Dest, *Src;
1672 cmsFloat64Number* Offset;
1673
1674 // Only works on RGB to RGB
1675 if (T_CHANNELS(*InputFormat) != 3 || T_CHANNELS(*OutputFormat) != 3) return FALSE;
1676
1677 // Only works on 8 bit input
1678 if (!_cmsFormatterIs8bit(*InputFormat)) return FALSE;
1679
1680 // Seems suitable, proceed
1681 Src = *Lut;
1682
1683 // Check for:
1684 //
1685 // shaper-matrix-matrix-shaper
1686 // shaper-matrix-shaper
1687 //
1688 // Both of those constructs are possible (first because abs. colorimetric).
1689 // additionally, In the first case, the input matrix offset should be zero.
1690
1691 IdentityMat = FALSE;
1692 if (cmsPipelineCheckAndRetreiveStages(ContextID, Src, 4,
1693 cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
1694 &Curve1, &Matrix1, &Matrix2, &Curve2)) {
1695
1696 // Get both matrices
1697 _cmsStageMatrixData* Data1 = (_cmsStageMatrixData*)cmsStageData(ContextID, Matrix1);
1698 _cmsStageMatrixData* Data2 = (_cmsStageMatrixData*)cmsStageData(ContextID, Matrix2);
1699
1700 // Input offset should be zero
1701 if (Data1->Offset != NULL) return FALSE;
1702
1703 // Multiply both matrices to get the result
1704 _cmsMAT3per(ContextID, &res, (cmsMAT3*)Data2->Double, (cmsMAT3*)Data1->Double);
1705
1706 // Only 2nd matrix has offset, or it is zero
1707 Offset = Data2->Offset;
1708
1709 // Now the result is in res + Data2 -> Offset. Maybe is a plain identity?
1710 if (_cmsMAT3isIdentity(ContextID, &res) && Offset == NULL) {
1711
1712 // We can get rid of full matrix
1713 IdentityMat = TRUE;
1714 }
1715
1716 }
1717 else {
1718
1719 if (cmsPipelineCheckAndRetreiveStages(ContextID, Src, 3,
1720 cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
1721 &Curve1, &Matrix1, &Curve2)) {
1722
1723 _cmsStageMatrixData* Data = (_cmsStageMatrixData*)cmsStageData(ContextID, Matrix1);
1724
1725 // Copy the matrix to our result
1726 memcpy(&res, Data->Double, sizeof(res));
1727
1728 // Preserve the Odffset (may be NULL as a zero offset)
1729 Offset = Data->Offset;
1730
1731 if (_cmsMAT3isIdentity(ContextID, &res) && Offset == NULL) {
1732
1733 // We can get rid of full matrix
1734 IdentityMat = TRUE;
1735 }
1736 }
1737 else
1738 return FALSE; // Not optimizeable this time
1739
1740 }
1741
1742 // Allocate an empty LUT
1743 Dest = cmsPipelineAlloc(ContextID, Src ->InputChannels, Src ->OutputChannels);
1744 if (!Dest) return FALSE;
1745
1746 // Assamble the new LUT
1747 if (!cmsPipelineInsertStage(ContextID, Dest, cmsAT_BEGIN, cmsStageDup(ContextID, Curve1)))
1748 goto Error;
1749
1750 if (!IdentityMat) {
1751
1752 if (!cmsPipelineInsertStage(ContextID, Dest, cmsAT_END, cmsStageAllocMatrix(ContextID, 3, 3, (const cmsFloat64Number*)&res, Offset)))
1753 goto Error;
1754 }
1755
1756 if (!cmsPipelineInsertStage(ContextID, Dest, cmsAT_END, cmsStageDup(ContextID, Curve2)))
1757 goto Error;
1758
1759 // If identity on matrix, we can further optimize the curves, so call the join curves routine
1760 if (IdentityMat) {
1761
1762 OptimizeByJoiningCurves(ContextID, &Dest, Intent, InputFormat, OutputFormat, dwFlags);
1763 }
1764 else {
1765 _cmsStageToneCurvesData* mpeC1 = (_cmsStageToneCurvesData*) cmsStageData(ContextID, Curve1);
1766 _cmsStageToneCurvesData* mpeC2 = (_cmsStageToneCurvesData*) cmsStageData(ContextID, Curve2);
1767
1768 // In this particular optimization, cache does not help as it takes more time to deal with
1769 // the cache that with the pixel handling
1770 *dwFlags |= cmsFLAGS_NOCACHE;
1771
1772 // Setup the optimizarion routines
1773 SetMatShaper(ContextID, Dest, mpeC1 ->TheCurves, &res, (cmsVEC3*) Offset, mpeC2->TheCurves, OutputFormat);
1774 }
1775
1776 cmsPipelineFree(ContextID, Src);
1777 *Lut = Dest;
1778 return TRUE;
1779 Error:
1780 // Leave Src unchanged
1781 cmsPipelineFree(ContextID, Dest);
1782 return FALSE;
1783 }
1784
1785
1786 // -------------------------------------------------------------------------------------------------------------------------------------
1787 // Optimization plug-ins
1788
1789 // List of optimizations
1790 typedef struct _cmsOptimizationCollection_st {
1791
1792 _cmsOPToptimizeFn OptimizePtr;
1793
1794 struct _cmsOptimizationCollection_st *Next;
1795
1796 } _cmsOptimizationCollection;
1797
1798
1799 // The built-in list. We currently implement 4 types of optimizations. Joining of curves, matrix-shaper, linearization and resampling
1800 static _cmsOptimizationCollection DefaultOptimization[] = {
1801
1802 { OptimizeByJoiningCurves, &DefaultOptimization[1] },
1803 { OptimizeMatrixShaper, &DefaultOptimization[2] },
1804 { OptimizeByComputingLinearization, &DefaultOptimization[3] },
1805 { OptimizeByResampling, NULL }
1806 };
1807
1808 // The linked list head
1809 _cmsOptimizationPluginChunkType _cmsOptimizationPluginChunk = { NULL };
1810
1811
1812 // Duplicates the zone of memory used by the plug-in in the new context
1813 static
DupPluginOptimizationList(struct _cmsContext_struct * ctx,const struct _cmsContext_struct * src)1814 void DupPluginOptimizationList(struct _cmsContext_struct* ctx,
1815 const struct _cmsContext_struct* src)
1816 {
1817 _cmsOptimizationPluginChunkType newHead = { NULL };
1818 _cmsOptimizationCollection* entry;
1819 _cmsOptimizationCollection* Anterior = NULL;
1820 _cmsOptimizationPluginChunkType* head = (_cmsOptimizationPluginChunkType*) src->chunks[OptimizationPlugin];
1821
1822 _cmsAssert(ctx != NULL);
1823 _cmsAssert(head != NULL);
1824
1825 // Walk the list copying all nodes
1826 for (entry = head->OptimizationCollection;
1827 entry != NULL;
1828 entry = entry ->Next) {
1829
1830 _cmsOptimizationCollection *newEntry = ( _cmsOptimizationCollection *) _cmsSubAllocDup(ctx ->MemPool, entry, sizeof(_cmsOptimizationCollection));
1831
1832 if (newEntry == NULL)
1833 return;
1834
1835 // We want to keep the linked list order, so this is a little bit tricky
1836 newEntry -> Next = NULL;
1837 if (Anterior)
1838 Anterior -> Next = newEntry;
1839
1840 Anterior = newEntry;
1841
1842 if (newHead.OptimizationCollection == NULL)
1843 newHead.OptimizationCollection = newEntry;
1844 }
1845
1846 ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx->MemPool, &newHead, sizeof(_cmsOptimizationPluginChunkType));
1847 }
1848
_cmsAllocOptimizationPluginChunk(struct _cmsContext_struct * ctx,const struct _cmsContext_struct * src)1849 void _cmsAllocOptimizationPluginChunk(struct _cmsContext_struct* ctx,
1850 const struct _cmsContext_struct* src)
1851 {
1852 if (src != NULL) {
1853
1854 // Copy all linked list
1855 DupPluginOptimizationList(ctx, src);
1856 }
1857 else {
1858 static _cmsOptimizationPluginChunkType OptimizationPluginChunkType = { NULL };
1859 ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx ->MemPool, &OptimizationPluginChunkType, sizeof(_cmsOptimizationPluginChunkType));
1860 }
1861 }
1862
1863
1864 // Register new ways to optimize
_cmsRegisterOptimizationPlugin(cmsContext ContextID,cmsPluginBase * Data)1865 cmsBool _cmsRegisterOptimizationPlugin(cmsContext ContextID, cmsPluginBase* Data)
1866 {
1867 cmsPluginOptimization* Plugin = (cmsPluginOptimization*) Data;
1868 _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
1869 _cmsOptimizationCollection* fl;
1870
1871 if (Data == NULL) {
1872
1873 ctx->OptimizationCollection = NULL;
1874 return TRUE;
1875 }
1876
1877 // Optimizer callback is required
1878 if (Plugin ->OptimizePtr == NULL) return FALSE;
1879
1880 fl = (_cmsOptimizationCollection*) _cmsPluginMalloc(ContextID, sizeof(_cmsOptimizationCollection));
1881 if (fl == NULL) return FALSE;
1882
1883 // Copy the parameters
1884 fl ->OptimizePtr = Plugin ->OptimizePtr;
1885
1886 // Keep linked list
1887 fl ->Next = ctx->OptimizationCollection;
1888
1889 // Set the head
1890 ctx ->OptimizationCollection = fl;
1891
1892 // All is ok
1893 return TRUE;
1894 }
1895
1896 // The entry point for LUT optimization
_cmsOptimizePipeline(cmsContext ContextID,cmsPipeline ** PtrLut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1897 cmsBool _cmsOptimizePipeline(cmsContext ContextID,
1898 cmsPipeline** PtrLut,
1899 cmsUInt32Number Intent,
1900 cmsUInt32Number* InputFormat,
1901 cmsUInt32Number* OutputFormat,
1902 cmsUInt32Number* dwFlags)
1903 {
1904 _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
1905 _cmsOptimizationCollection* Opts;
1906 cmsBool AnySuccess = FALSE;
1907
1908 // A CLUT is being asked, so force this specific optimization
1909 if (*dwFlags & cmsFLAGS_FORCE_CLUT) {
1910
1911 PreOptimize(ContextID, *PtrLut);
1912 return OptimizeByResampling(ContextID, PtrLut, Intent, InputFormat, OutputFormat, dwFlags);
1913 }
1914
1915 // Anything to optimize?
1916 if ((*PtrLut) ->Elements == NULL) {
1917 _cmsPipelineSetOptimizationParameters(ContextID, *PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
1918 return TRUE;
1919 }
1920
1921 // Try to get rid of identities and trivial conversions.
1922 AnySuccess = PreOptimize(ContextID, *PtrLut);
1923
1924 // After removal do we end with an identity?
1925 if ((*PtrLut) ->Elements == NULL) {
1926 _cmsPipelineSetOptimizationParameters(ContextID, *PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
1927 return TRUE;
1928 }
1929
1930 // Do not optimize, keep all precision
1931 if (*dwFlags & cmsFLAGS_NOOPTIMIZE)
1932 return FALSE;
1933
1934 // Try plug-in optimizations
1935 for (Opts = ctx->OptimizationCollection;
1936 Opts != NULL;
1937 Opts = Opts ->Next) {
1938
1939 // If one schema succeeded, we are done
1940 if (Opts ->OptimizePtr(ContextID, PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
1941
1942 return TRUE; // Optimized!
1943 }
1944 }
1945
1946 // Try built-in optimizations
1947 for (Opts = DefaultOptimization;
1948 Opts != NULL;
1949 Opts = Opts ->Next) {
1950
1951 if (Opts ->OptimizePtr(ContextID, PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
1952
1953 return TRUE;
1954 }
1955 }
1956
1957 // Only simple optimizations succeeded
1958 return AnySuccess;
1959 }
1960
_cmsLutIsIdentity(cmsPipeline * PtrLut)1961 cmsBool _cmsLutIsIdentity(cmsPipeline *PtrLut)
1962 {
1963 return !PtrLut || PtrLut->Eval16Fn == FastIdentity16;
1964 }
1965