• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //---------------------------------------------------------------------------------
2 //
3 //  Little Color Management System
4 //  Copyright (c) 1998-2016 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     int nInputs;
53     int 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 
95     cmsContext ContextID;
96 
97     int nCurves;                  // Number of curves
98     int nElements;                // Elements in curves
99     cmsUInt16Number** Curves;     // Points to a dynamically  allocated array
100 
101 } Curves16Data;
102 
103 
104 // Simple optimizations ----------------------------------------------------------------------------------------------------------
105 
106 
107 // Remove an element in linked chain
108 static
_RemoveElement(cmsStage ** head)109 void _RemoveElement(cmsStage** head)
110 {
111     cmsStage* mpe = *head;
112     cmsStage* next = mpe ->Next;
113     *head = next;
114     cmsStageFree(mpe);
115 }
116 
117 // Remove all identities in chain. Note that pt actually is a double pointer to the element that holds the pointer.
118 static
_Remove1Op(cmsPipeline * Lut,cmsStageSignature UnaryOp)119 cmsBool _Remove1Op(cmsPipeline* Lut, cmsStageSignature UnaryOp)
120 {
121     cmsStage** pt = &Lut ->Elements;
122     cmsBool AnyOpt = FALSE;
123 
124     while (*pt != NULL) {
125 
126         if ((*pt) ->Implements == UnaryOp) {
127             _RemoveElement(pt);
128             AnyOpt = TRUE;
129         }
130         else
131             pt = &((*pt) -> Next);
132     }
133 
134     return AnyOpt;
135 }
136 
137 // Same, but only if two adjacent elements are found
138 static
_Remove2Op(cmsPipeline * Lut,cmsStageSignature Op1,cmsStageSignature Op2)139 cmsBool _Remove2Op(cmsPipeline* Lut, cmsStageSignature Op1, cmsStageSignature Op2)
140 {
141     cmsStage** pt1;
142     cmsStage** pt2;
143     cmsBool AnyOpt = FALSE;
144 
145     pt1 = &Lut ->Elements;
146     if (*pt1 == NULL) return AnyOpt;
147 
148     while (*pt1 != NULL) {
149 
150         pt2 = &((*pt1) -> Next);
151         if (*pt2 == NULL) return AnyOpt;
152 
153         if ((*pt1) ->Implements == Op1 && (*pt2) ->Implements == Op2) {
154             _RemoveElement(pt2);
155             _RemoveElement(pt1);
156             AnyOpt = TRUE;
157         }
158         else
159             pt1 = &((*pt1) -> Next);
160     }
161 
162     return AnyOpt;
163 }
164 
165 
166 static
CloseEnoughFloat(cmsFloat64Number a,cmsFloat64Number b)167 cmsBool CloseEnoughFloat(cmsFloat64Number a, cmsFloat64Number b)
168 {
169        return fabs(b - a) < 0.00001f;
170 }
171 
172 static
isFloatMatrixIdentity(const cmsMAT3 * a)173 cmsBool  isFloatMatrixIdentity(const cmsMAT3* a)
174 {
175        cmsMAT3 Identity;
176        int i, j;
177 
178        _cmsMAT3identity(&Identity);
179 
180        for (i = 0; i < 3; i++)
181               for (j = 0; j < 3; j++)
182                      if (!CloseEnoughFloat(a->v[i].n[j], Identity.v[i].n[j])) return FALSE;
183 
184        return TRUE;
185 }
186 // if two adjacent matrices are found, multiply them.
187 static
_MultiplyMatrix(cmsPipeline * Lut)188 cmsBool _MultiplyMatrix(cmsPipeline* Lut)
189 {
190        cmsStage** pt1;
191        cmsStage** pt2;
192        cmsStage*  chain;
193        cmsBool AnyOpt = FALSE;
194 
195        pt1 = &Lut->Elements;
196        if (*pt1 == NULL) return AnyOpt;
197 
198        while (*pt1 != NULL) {
199 
200               pt2 = &((*pt1)->Next);
201               if (*pt2 == NULL) return AnyOpt;
202 
203               if ((*pt1)->Implements == cmsSigMatrixElemType && (*pt2)->Implements == cmsSigMatrixElemType) {
204 
205                      // Get both matrices
206                      _cmsStageMatrixData* m1 = (_cmsStageMatrixData*) cmsStageData(*pt1);
207                      _cmsStageMatrixData* m2 = (_cmsStageMatrixData*) cmsStageData(*pt2);
208                      cmsMAT3 res;
209 
210                      // Input offset and output offset should be zero to use this optimization
211                      if (m1->Offset != NULL || m2 ->Offset != NULL ||
212                             cmsStageInputChannels(*pt1) != 3 || cmsStageOutputChannels(*pt1) != 3 ||
213                             cmsStageInputChannels(*pt2) != 3 || cmsStageOutputChannels(*pt2) != 3)
214                             return FALSE;
215 
216                      // Multiply both matrices to get the result
217                      _cmsMAT3per(&res, (cmsMAT3*)m2->Double, (cmsMAT3*)m1->Double);
218 
219                      // Get the next in chain afer the matrices
220                      chain = (*pt2)->Next;
221 
222                      // Remove both matrices
223                      _RemoveElement(pt2);
224                      _RemoveElement(pt1);
225 
226                      // Now what if the result is a plain identity?
227                      if (!isFloatMatrixIdentity(&res)) {
228 
229                             // We can not get rid of full matrix
230                             cmsStage* Multmat = cmsStageAllocMatrix(Lut->ContextID, 3, 3, (const cmsFloat64Number*) &res, NULL);
231                             if (Multmat == NULL) return FALSE;  // Should never happen
232 
233                             // Recover the chain
234                             Multmat->Next = chain;
235                             *pt1 = Multmat;
236                      }
237 
238                      AnyOpt = TRUE;
239               }
240               else
241                      pt1 = &((*pt1)->Next);
242        }
243 
244        return AnyOpt;
245 }
246 
247 
248 // Preoptimize just gets rif of no-ops coming paired. Conversion from v2 to v4 followed
249 // by a v4 to v2 and vice-versa. The elements are then discarded.
250 static
PreOptimize(cmsPipeline * Lut)251 cmsBool PreOptimize(cmsPipeline* Lut)
252 {
253     cmsBool AnyOpt = FALSE, Opt;
254 
255     do {
256 
257         Opt = FALSE;
258 
259         // Remove all identities
260         Opt |= _Remove1Op(Lut, cmsSigIdentityElemType);
261 
262         // Remove XYZ2Lab followed by Lab2XYZ
263         Opt |= _Remove2Op(Lut, cmsSigXYZ2LabElemType, cmsSigLab2XYZElemType);
264 
265         // Remove Lab2XYZ followed by XYZ2Lab
266         Opt |= _Remove2Op(Lut, cmsSigLab2XYZElemType, cmsSigXYZ2LabElemType);
267 
268         // Remove V4 to V2 followed by V2 to V4
269         Opt |= _Remove2Op(Lut, cmsSigLabV4toV2, cmsSigLabV2toV4);
270 
271         // Remove V2 to V4 followed by V4 to V2
272         Opt |= _Remove2Op(Lut, cmsSigLabV2toV4, cmsSigLabV4toV2);
273 
274         // Remove float pcs Lab conversions
275         Opt |= _Remove2Op(Lut, cmsSigLab2FloatPCS, cmsSigFloatPCS2Lab);
276 
277         // Remove float pcs Lab conversions
278         Opt |= _Remove2Op(Lut, cmsSigXYZ2FloatPCS, cmsSigFloatPCS2XYZ);
279 
280         // Simplify matrix.
281         Opt |= _MultiplyMatrix(Lut);
282 
283         if (Opt) AnyOpt = TRUE;
284 
285     } while (Opt);
286 
287     return AnyOpt;
288 }
289 
290 static
Eval16nop1D(register const cmsUInt16Number Input[],register cmsUInt16Number Output[],register const struct _cms_interp_struc * p)291 void Eval16nop1D(register const cmsUInt16Number Input[],
292                  register cmsUInt16Number Output[],
293                  register const struct _cms_interp_struc* p)
294 {
295     Output[0] = Input[0];
296 
297     cmsUNUSED_PARAMETER(p);
298 }
299 
300 static
PrelinEval16(register const cmsUInt16Number Input[],register cmsUInt16Number Output[],register const void * D)301 void PrelinEval16(register const cmsUInt16Number Input[],
302                   register cmsUInt16Number Output[],
303                   register const void* D)
304 {
305     Prelin16Data* p16 = (Prelin16Data*) D;
306     cmsUInt16Number  StageABC[MAX_INPUT_DIMENSIONS];
307     cmsUInt16Number  StageDEF[cmsMAXCHANNELS];
308     int i;
309 
310     for (i=0; i < p16 ->nInputs; i++) {
311 
312         p16 ->EvalCurveIn16[i](&Input[i], &StageABC[i], p16 ->ParamsCurveIn16[i]);
313     }
314 
315     p16 ->EvalCLUT(StageABC, StageDEF, p16 ->CLUTparams);
316 
317     for (i=0; i < p16 ->nOutputs; i++) {
318 
319         p16 ->EvalCurveOut16[i](&StageDEF[i], &Output[i], p16 ->ParamsCurveOut16[i]);
320     }
321 }
322 
323 
324 static
PrelinOpt16free(cmsContext ContextID,void * ptr)325 void PrelinOpt16free(cmsContext ContextID, void* ptr)
326 {
327     Prelin16Data* p16 = (Prelin16Data*) ptr;
328 
329     _cmsFree(ContextID, p16 ->EvalCurveOut16);
330     _cmsFree(ContextID, p16 ->ParamsCurveOut16);
331 
332     _cmsFree(ContextID, p16);
333 }
334 
335 static
Prelin16dup(cmsContext ContextID,const void * ptr)336 void* Prelin16dup(cmsContext ContextID, const void* ptr)
337 {
338     Prelin16Data* p16 = (Prelin16Data*) ptr;
339     Prelin16Data* Duped = (Prelin16Data*) _cmsDupMem(ContextID, p16, sizeof(Prelin16Data));
340 
341     if (Duped == NULL) return NULL;
342 
343     Duped->EvalCurveOut16 = (_cmsInterpFn16*) _cmsDupMem(ContextID, p16->EvalCurveOut16, p16->nOutputs * sizeof(_cmsInterpFn16));
344     Duped->ParamsCurveOut16 = (cmsInterpParams**)_cmsDupMem(ContextID, p16->ParamsCurveOut16, p16->nOutputs * sizeof(cmsInterpParams*));
345 
346     return Duped;
347 }
348 
349 
350 static
PrelinOpt16alloc(cmsContext ContextID,const cmsInterpParams * ColorMap,int nInputs,cmsToneCurve ** In,int nOutputs,cmsToneCurve ** Out)351 Prelin16Data* PrelinOpt16alloc(cmsContext ContextID,
352                                const cmsInterpParams* ColorMap,
353                                int nInputs, cmsToneCurve** In,
354                                int nOutputs, cmsToneCurve** Out )
355 {
356     int i;
357     Prelin16Data* p16 = (Prelin16Data*)_cmsMallocZero(ContextID, sizeof(Prelin16Data));
358     if (p16 == NULL) return NULL;
359 
360     p16 ->nInputs = nInputs;
361     p16 -> nOutputs = nOutputs;
362 
363 
364     for (i=0; i < nInputs; i++) {
365 
366         if (In == NULL) {
367             p16 -> ParamsCurveIn16[i] = NULL;
368             p16 -> EvalCurveIn16[i] = Eval16nop1D;
369 
370         }
371         else {
372             p16 -> ParamsCurveIn16[i] = In[i] ->InterpParams;
373             p16 -> EvalCurveIn16[i] = p16 ->ParamsCurveIn16[i]->Interpolation.Lerp16;
374         }
375     }
376 
377     p16 ->CLUTparams = ColorMap;
378     p16 ->EvalCLUT   = ColorMap ->Interpolation.Lerp16;
379 
380 
381     p16 -> EvalCurveOut16 = (_cmsInterpFn16*) _cmsCalloc(ContextID, nOutputs, sizeof(_cmsInterpFn16));
382     p16 -> ParamsCurveOut16 = (cmsInterpParams**) _cmsCalloc(ContextID, nOutputs, sizeof(cmsInterpParams* ));
383 
384     for (i=0; i < nOutputs; i++) {
385 
386         if (Out == NULL) {
387             p16 ->ParamsCurveOut16[i] = NULL;
388             p16 -> EvalCurveOut16[i] = Eval16nop1D;
389         }
390         else {
391 
392             p16 ->ParamsCurveOut16[i] = Out[i] ->InterpParams;
393             p16 -> EvalCurveOut16[i] = p16 ->ParamsCurveOut16[i]->Interpolation.Lerp16;
394         }
395     }
396 
397     return p16;
398 }
399 
400 
401 
402 // Resampling ---------------------------------------------------------------------------------
403 
404 #define PRELINEARIZATION_POINTS 4096
405 
406 // Sampler implemented by another LUT. This is a clean way to precalculate the devicelink 3D CLUT for
407 // almost any transform. We use floating point precision and then convert from floating point to 16 bits.
408 static
XFormSampler16(register const cmsUInt16Number In[],register cmsUInt16Number Out[],register void * Cargo)409 int XFormSampler16(register const cmsUInt16Number In[], register cmsUInt16Number Out[], register void* Cargo)
410 {
411     cmsPipeline* Lut = (cmsPipeline*) Cargo;
412     cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
413     cmsUInt32Number i;
414 
415     _cmsAssert(Lut -> InputChannels < cmsMAXCHANNELS);
416     _cmsAssert(Lut -> OutputChannels < cmsMAXCHANNELS);
417 
418     // From 16 bit to floating point
419     for (i=0; i < Lut ->InputChannels; i++)
420         InFloat[i] = (cmsFloat32Number) (In[i] / 65535.0);
421 
422     // Evaluate in floating point
423     cmsPipelineEvalFloat(InFloat, OutFloat, Lut);
424 
425     // Back to 16 bits representation
426     for (i=0; i < Lut ->OutputChannels; i++)
427         Out[i] = _cmsQuickSaturateWord(OutFloat[i] * 65535.0);
428 
429     // Always succeed
430     return TRUE;
431 }
432 
433 // Try to see if the curves of a given MPE are linear
434 static
AllCurvesAreLinear(cmsStage * mpe)435 cmsBool AllCurvesAreLinear(cmsStage* mpe)
436 {
437     cmsToneCurve** Curves;
438     cmsUInt32Number i, n;
439 
440     Curves = _cmsStageGetPtrToCurveSet(mpe);
441     if (Curves == NULL) return FALSE;
442 
443     n = cmsStageOutputChannels(mpe);
444 
445     for (i=0; i < n; i++) {
446         if (!cmsIsToneCurveLinear(Curves[i])) return FALSE;
447     }
448 
449     return TRUE;
450 }
451 
452 // This function replaces a specific node placed in "At" by the "Value" numbers. Its purpose
453 // is to fix scum dot on broken profiles/transforms. Works on 1, 3 and 4 channels
454 static
PatchLUT(cmsStage * CLUT,cmsUInt16Number At[],cmsUInt16Number Value[],int nChannelsOut,int nChannelsIn)455 cmsBool  PatchLUT(cmsStage* CLUT, cmsUInt16Number At[], cmsUInt16Number Value[],
456                   int nChannelsOut, int nChannelsIn)
457 {
458     _cmsStageCLutData* Grid = (_cmsStageCLutData*) CLUT ->Data;
459     cmsInterpParams* p16  = Grid ->Params;
460     cmsFloat64Number px, py, pz, pw;
461     int        x0, y0, z0, w0;
462     int        i, index;
463 
464     if (CLUT -> Type != cmsSigCLutElemType) {
465         cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) Attempt to PatchLUT on non-lut stage");
466         return FALSE;
467     }
468 
469     if (nChannelsIn == 4) {
470 
471         px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
472         py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
473         pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
474         pw = ((cmsFloat64Number) At[3] * (p16->Domain[3])) / 65535.0;
475 
476         x0 = (int) floor(px);
477         y0 = (int) floor(py);
478         z0 = (int) floor(pz);
479         w0 = (int) floor(pw);
480 
481         if (((px - x0) != 0) ||
482             ((py - y0) != 0) ||
483             ((pz - z0) != 0) ||
484             ((pw - w0) != 0)) return FALSE; // Not on exact node
485 
486         index = p16 -> opta[3] * x0 +
487                 p16 -> opta[2] * y0 +
488                 p16 -> opta[1] * z0 +
489                 p16 -> opta[0] * w0;
490     }
491     else
492         if (nChannelsIn == 3) {
493 
494             px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
495             py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
496             pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
497 
498             x0 = (int) floor(px);
499             y0 = (int) floor(py);
500             z0 = (int) floor(pz);
501 
502             if (((px - x0) != 0) ||
503                 ((py - y0) != 0) ||
504                 ((pz - z0) != 0)) return FALSE;  // Not on exact node
505 
506             index = p16 -> opta[2] * x0 +
507                     p16 -> opta[1] * y0 +
508                     p16 -> opta[0] * z0;
509         }
510         else
511             if (nChannelsIn == 1) {
512 
513                 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
514 
515                 x0 = (int) floor(px);
516 
517                 if (((px - x0) != 0)) return FALSE; // Not on exact node
518 
519                 index = p16 -> opta[0] * x0;
520             }
521             else {
522                 cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) %d Channels are not supported on PatchLUT", nChannelsIn);
523                 return FALSE;
524             }
525 
526             for (i=0; i < nChannelsOut; i++)
527                 Grid -> Tab.T[index + i] = Value[i];
528 
529             return TRUE;
530 }
531 
532 // Auxiliary, to see if two values are equal or very different
533 static
WhitesAreEqual(int n,cmsUInt16Number White1[],cmsUInt16Number White2[])534 cmsBool WhitesAreEqual(int n, cmsUInt16Number White1[], cmsUInt16Number White2[] )
535 {
536     int i;
537 
538     for (i=0; i < n; i++) {
539 
540         if (abs(White1[i] - White2[i]) > 0xf000) return TRUE;  // Values are so extremely different that the fixup should be avoided
541         if (White1[i] != White2[i]) return FALSE;
542     }
543     return TRUE;
544 }
545 
546 
547 // Locate the node for the white point and fix it to pure white in order to avoid scum dot.
548 static
FixWhiteMisalignment(cmsPipeline * Lut,cmsColorSpaceSignature EntryColorSpace,cmsColorSpaceSignature ExitColorSpace)549 cmsBool FixWhiteMisalignment(cmsPipeline* Lut, cmsColorSpaceSignature EntryColorSpace, cmsColorSpaceSignature ExitColorSpace)
550 {
551     cmsUInt16Number *WhitePointIn, *WhitePointOut;
552     cmsUInt16Number  WhiteIn[cmsMAXCHANNELS], WhiteOut[cmsMAXCHANNELS], ObtainedOut[cmsMAXCHANNELS];
553     cmsUInt32Number i, nOuts, nIns;
554     cmsStage *PreLin = NULL, *CLUT = NULL, *PostLin = NULL;
555 
556     if (!_cmsEndPointsBySpace(EntryColorSpace,
557         &WhitePointIn, NULL, &nIns)) return FALSE;
558 
559     if (!_cmsEndPointsBySpace(ExitColorSpace,
560         &WhitePointOut, NULL, &nOuts)) return FALSE;
561 
562     // It needs to be fixed?
563     if (Lut ->InputChannels != nIns) return FALSE;
564     if (Lut ->OutputChannels != nOuts) return FALSE;
565 
566     cmsPipelineEval16(WhitePointIn, ObtainedOut, Lut);
567 
568     if (WhitesAreEqual(nOuts, WhitePointOut, ObtainedOut)) return TRUE; // whites already match
569 
570     // Check if the LUT comes as Prelin, CLUT or Postlin. We allow all combinations
571     if (!cmsPipelineCheckAndRetreiveStages(Lut, 3, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, &PreLin, &CLUT, &PostLin))
572         if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCurveSetElemType, cmsSigCLutElemType, &PreLin, &CLUT))
573             if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCLutElemType, cmsSigCurveSetElemType, &CLUT, &PostLin))
574                 if (!cmsPipelineCheckAndRetreiveStages(Lut, 1, cmsSigCLutElemType, &CLUT))
575                     return FALSE;
576 
577     // We need to interpolate white points of both, pre and post curves
578     if (PreLin) {
579 
580         cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PreLin);
581 
582         for (i=0; i < nIns; i++) {
583             WhiteIn[i] = cmsEvalToneCurve16(Curves[i], WhitePointIn[i]);
584         }
585     }
586     else {
587         for (i=0; i < nIns; i++)
588             WhiteIn[i] = WhitePointIn[i];
589     }
590 
591     // If any post-linearization, we need to find how is represented white before the curve, do
592     // a reverse interpolation in this case.
593     if (PostLin) {
594 
595         cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PostLin);
596 
597         for (i=0; i < nOuts; i++) {
598 
599             cmsToneCurve* InversePostLin = cmsReverseToneCurve(Curves[i]);
600             if (InversePostLin == NULL) {
601                 WhiteOut[i] = WhitePointOut[i];
602 
603             } else {
604 
605                 WhiteOut[i] = cmsEvalToneCurve16(InversePostLin, WhitePointOut[i]);
606                 cmsFreeToneCurve(InversePostLin);
607             }
608         }
609     }
610     else {
611         for (i=0; i < nOuts; i++)
612             WhiteOut[i] = WhitePointOut[i];
613     }
614 
615     // Ok, proceed with patching. May fail and we don't care if it fails
616     PatchLUT(CLUT, WhiteIn, WhiteOut, nOuts, nIns);
617 
618     return TRUE;
619 }
620 
621 // -----------------------------------------------------------------------------------------------------------------------------------------------
622 // This function creates simple LUT from complex ones. The generated LUT has an optional set of
623 // prelinearization curves, a CLUT of nGridPoints and optional postlinearization tables.
624 // These curves have to exist in the original LUT in order to be used in the simplified output.
625 // Caller may also use the flags to allow this feature.
626 // LUTS with all curves will be simplified to a single curve. Parametric curves are lost.
627 // This function should be used on 16-bits LUTS only, as floating point losses precision when simplified
628 // -----------------------------------------------------------------------------------------------------------------------------------------------
629 
630 static
OptimizeByResampling(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)631 cmsBool OptimizeByResampling(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
632 {
633     cmsPipeline* Src = NULL;
634     cmsPipeline* Dest = NULL;
635     cmsStage* mpe;
636     cmsStage* CLUT;
637     cmsStage *KeepPreLin = NULL, *KeepPostLin = NULL;
638     int nGridPoints;
639     cmsColorSpaceSignature ColorSpace, OutputColorSpace;
640     cmsStage *NewPreLin = NULL;
641     cmsStage *NewPostLin = NULL;
642     _cmsStageCLutData* DataCLUT;
643     cmsToneCurve** DataSetIn;
644     cmsToneCurve** DataSetOut;
645     Prelin16Data* p16;
646 
647     // This is a loosy optimization! does not apply in floating-point cases
648     if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
649 
650     ColorSpace       = _cmsICCcolorSpace(T_COLORSPACE(*InputFormat));
651     OutputColorSpace = _cmsICCcolorSpace(T_COLORSPACE(*OutputFormat));
652     nGridPoints      = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);
653 
654     // For empty LUTs, 2 points are enough
655     if (cmsPipelineStageCount(*Lut) == 0)
656         nGridPoints = 2;
657 
658     Src = *Lut;
659 
660     // Named color pipelines cannot be optimized either
661     for (mpe = cmsPipelineGetPtrToFirstStage(Src);
662         mpe != NULL;
663         mpe = cmsStageNext(mpe)) {
664             if (cmsStageType(mpe) == cmsSigNamedColorElemType) return FALSE;
665     }
666 
667     // Allocate an empty LUT
668     Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
669     if (!Dest) return FALSE;
670 
671     // Prelinearization tables are kept unless indicated by flags
672     if (*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION) {
673 
674         // Get a pointer to the prelinearization element
675         cmsStage* PreLin = cmsPipelineGetPtrToFirstStage(Src);
676 
677         // Check if suitable
678         if (PreLin && PreLin ->Type == cmsSigCurveSetElemType) {
679 
680             // Maybe this is a linear tram, so we can avoid the whole stuff
681             if (!AllCurvesAreLinear(PreLin)) {
682 
683                 // All seems ok, proceed.
684                 NewPreLin = cmsStageDup(PreLin);
685                 if(!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, NewPreLin))
686                     goto Error;
687 
688                 // Remove prelinearization. Since we have duplicated the curve
689                 // in destination LUT, the sampling shoud be applied after this stage.
690                 cmsPipelineUnlinkStage(Src, cmsAT_BEGIN, &KeepPreLin);
691             }
692         }
693     }
694 
695     // Allocate the CLUT
696     CLUT = cmsStageAllocCLut16bit(Src ->ContextID, nGridPoints, Src ->InputChannels, Src->OutputChannels, NULL);
697     if (CLUT == NULL) goto Error;
698 
699     // Add the CLUT to the destination LUT
700     if (!cmsPipelineInsertStage(Dest, cmsAT_END, CLUT)) {
701         goto Error;
702     }
703 
704     // Postlinearization tables are kept unless indicated by flags
705     if (*dwFlags & cmsFLAGS_CLUT_POST_LINEARIZATION) {
706 
707         // Get a pointer to the postlinearization if present
708         cmsStage* PostLin = cmsPipelineGetPtrToLastStage(Src);
709 
710         // Check if suitable
711         if (PostLin && cmsStageType(PostLin) == cmsSigCurveSetElemType) {
712 
713             // Maybe this is a linear tram, so we can avoid the whole stuff
714             if (!AllCurvesAreLinear(PostLin)) {
715 
716                 // All seems ok, proceed.
717                 NewPostLin = cmsStageDup(PostLin);
718                 if (!cmsPipelineInsertStage(Dest, cmsAT_END, NewPostLin))
719                     goto Error;
720 
721                 // In destination LUT, the sampling shoud be applied after this stage.
722                 cmsPipelineUnlinkStage(Src, cmsAT_END, &KeepPostLin);
723             }
724         }
725     }
726 
727     // Now its time to do the sampling. We have to ignore pre/post linearization
728     // The source LUT whithout pre/post curves is passed as parameter.
729     if (!cmsStageSampleCLut16bit(CLUT, XFormSampler16, (void*) Src, 0)) {
730 Error:
731         // Ops, something went wrong, Restore stages
732         if (KeepPreLin != NULL) {
733             if (!cmsPipelineInsertStage(Src, cmsAT_BEGIN, KeepPreLin)) {
734                 _cmsAssert(0); // This never happens
735             }
736         }
737         if (KeepPostLin != NULL) {
738             if (!cmsPipelineInsertStage(Src, cmsAT_END,   KeepPostLin)) {
739                 _cmsAssert(0); // This never happens
740             }
741         }
742         cmsPipelineFree(Dest);
743         return FALSE;
744     }
745 
746     // Done.
747 
748     if (KeepPreLin != NULL) cmsStageFree(KeepPreLin);
749     if (KeepPostLin != NULL) cmsStageFree(KeepPostLin);
750     cmsPipelineFree(Src);
751 
752     DataCLUT = (_cmsStageCLutData*) CLUT ->Data;
753 
754     if (NewPreLin == NULL) DataSetIn = NULL;
755     else DataSetIn = ((_cmsStageToneCurvesData*) NewPreLin ->Data) ->TheCurves;
756 
757     if (NewPostLin == NULL) DataSetOut = NULL;
758     else  DataSetOut = ((_cmsStageToneCurvesData*) NewPostLin ->Data) ->TheCurves;
759 
760 
761     if (DataSetIn == NULL && DataSetOut == NULL) {
762 
763         _cmsPipelineSetOptimizationParameters(Dest, (_cmsOPTeval16Fn) DataCLUT->Params->Interpolation.Lerp16, DataCLUT->Params, NULL, NULL);
764     }
765     else {
766 
767         p16 = PrelinOpt16alloc(Dest ->ContextID,
768             DataCLUT ->Params,
769             Dest ->InputChannels,
770             DataSetIn,
771             Dest ->OutputChannels,
772             DataSetOut);
773 
774         _cmsPipelineSetOptimizationParameters(Dest, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
775     }
776 
777 
778     // Don't fix white on absolute colorimetric
779     if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
780         *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
781 
782     if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
783 
784         FixWhiteMisalignment(Dest, ColorSpace, OutputColorSpace);
785     }
786 
787     *Lut = Dest;
788     return TRUE;
789 
790     cmsUNUSED_PARAMETER(Intent);
791 }
792 
793 
794 // -----------------------------------------------------------------------------------------------------------------------------------------------
795 // Fixes the gamma balancing of transform. This is described in my paper "Prelinearization Stages on
796 // Color-Management Application-Specific Integrated Circuits (ASICs)" presented at NIP24. It only works
797 // for RGB transforms. See the paper for more details
798 // -----------------------------------------------------------------------------------------------------------------------------------------------
799 
800 
801 // Normalize endpoints by slope limiting max and min. This assures endpoints as well.
802 // Descending curves are handled as well.
803 static
SlopeLimiting(cmsToneCurve * g)804 void SlopeLimiting(cmsToneCurve* g)
805 {
806     int BeginVal, EndVal;
807     int AtBegin = (int) floor((cmsFloat64Number) g ->nEntries * 0.02 + 0.5);   // Cutoff at 2%
808     int AtEnd   = g ->nEntries - AtBegin - 1;                                  // And 98%
809     cmsFloat64Number Val, Slope, beta;
810     int i;
811 
812     if (cmsIsToneCurveDescending(g)) {
813         BeginVal = 0xffff; EndVal = 0;
814     }
815     else {
816         BeginVal = 0; EndVal = 0xffff;
817     }
818 
819     // Compute slope and offset for begin of curve
820     Val   = g ->Table16[AtBegin];
821     Slope = (Val - BeginVal) / AtBegin;
822     beta  = Val - Slope * AtBegin;
823 
824     for (i=0; i < AtBegin; i++)
825         g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
826 
827     // Compute slope and offset for the end
828     Val   = g ->Table16[AtEnd];
829     Slope = (EndVal - Val) / AtBegin;   // AtBegin holds the X interval, which is same in both cases
830     beta  = Val - Slope * AtEnd;
831 
832     for (i = AtEnd; i < (int) g ->nEntries; i++)
833         g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
834 }
835 
836 
837 // Precomputes tables for 8-bit on input devicelink.
838 static
PrelinOpt8alloc(cmsContext ContextID,const cmsInterpParams * p,cmsToneCurve * G[3])839 Prelin8Data* PrelinOpt8alloc(cmsContext ContextID, const cmsInterpParams* p, cmsToneCurve* G[3])
840 {
841     int i;
842     cmsUInt16Number Input[3];
843     cmsS15Fixed16Number v1, v2, v3;
844     Prelin8Data* p8;
845 
846     p8 = (Prelin8Data*)_cmsMallocZero(ContextID, sizeof(Prelin8Data));
847     if (p8 == NULL) return NULL;
848 
849     // Since this only works for 8 bit input, values comes always as x * 257,
850     // we can safely take msb byte (x << 8 + x)
851 
852     for (i=0; i < 256; i++) {
853 
854         if (G != NULL) {
855 
856             // Get 16-bit representation
857             Input[0] = cmsEvalToneCurve16(G[0], FROM_8_TO_16(i));
858             Input[1] = cmsEvalToneCurve16(G[1], FROM_8_TO_16(i));
859             Input[2] = cmsEvalToneCurve16(G[2], FROM_8_TO_16(i));
860         }
861         else {
862             Input[0] = FROM_8_TO_16(i);
863             Input[1] = FROM_8_TO_16(i);
864             Input[2] = FROM_8_TO_16(i);
865         }
866 
867 
868         // Move to 0..1.0 in fixed domain
869         v1 = _cmsToFixedDomain(Input[0] * p -> Domain[0]);
870         v2 = _cmsToFixedDomain(Input[1] * p -> Domain[1]);
871         v3 = _cmsToFixedDomain(Input[2] * p -> Domain[2]);
872 
873         // Store the precalculated table of nodes
874         p8 ->X0[i] = (p->opta[2] * FIXED_TO_INT(v1));
875         p8 ->Y0[i] = (p->opta[1] * FIXED_TO_INT(v2));
876         p8 ->Z0[i] = (p->opta[0] * FIXED_TO_INT(v3));
877 
878         // Store the precalculated table of offsets
879         p8 ->rx[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v1);
880         p8 ->ry[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v2);
881         p8 ->rz[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v3);
882     }
883 
884     p8 ->ContextID = ContextID;
885     p8 ->p = p;
886 
887     return p8;
888 }
889 
890 static
Prelin8free(cmsContext ContextID,void * ptr)891 void Prelin8free(cmsContext ContextID, void* ptr)
892 {
893     _cmsFree(ContextID, ptr);
894 }
895 
896 static
Prelin8dup(cmsContext ContextID,const void * ptr)897 void* Prelin8dup(cmsContext ContextID, const void* ptr)
898 {
899     return _cmsDupMem(ContextID, ptr, sizeof(Prelin8Data));
900 }
901 
902 
903 
904 // A optimized interpolation for 8-bit input.
905 #define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
906 static
PrelinEval8(register const cmsUInt16Number Input[],register cmsUInt16Number Output[],register const void * D)907 void PrelinEval8(register const cmsUInt16Number Input[],
908                   register cmsUInt16Number Output[],
909                   register const void* D)
910 {
911 
912     cmsUInt8Number         r, g, b;
913     cmsS15Fixed16Number    rx, ry, rz;
914     cmsS15Fixed16Number    c0, c1, c2, c3, Rest;
915     int                    OutChan;
916     register cmsS15Fixed16Number    X0, X1, Y0, Y1, Z0, Z1;
917     Prelin8Data* p8 = (Prelin8Data*) D;
918     register const cmsInterpParams* p = p8 ->p;
919     int                    TotalOut = p -> nOutputs;
920     const cmsUInt16Number* LutTable = (const cmsUInt16Number*) p->Table;
921 
922     r = Input[0] >> 8;
923     g = Input[1] >> 8;
924     b = Input[2] >> 8;
925 
926     X0 = X1 = p8->X0[r];
927     Y0 = Y1 = p8->Y0[g];
928     Z0 = Z1 = p8->Z0[b];
929 
930     rx = p8 ->rx[r];
931     ry = p8 ->ry[g];
932     rz = p8 ->rz[b];
933 
934     X1 = X0 + ((rx == 0) ? 0 : p ->opta[2]);
935     Y1 = Y0 + ((ry == 0) ? 0 : p ->opta[1]);
936     Z1 = Z0 + ((rz == 0) ? 0 : p ->opta[0]);
937 
938 
939     // These are the 6 Tetrahedral
940     for (OutChan=0; OutChan < TotalOut; OutChan++) {
941 
942         c0 = DENS(X0, Y0, Z0);
943 
944         if (rx >= ry && ry >= rz)
945         {
946             c1 = DENS(X1, Y0, Z0) - c0;
947             c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
948             c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
949         }
950         else
951             if (rx >= rz && rz >= ry)
952             {
953                 c1 = DENS(X1, Y0, Z0) - c0;
954                 c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
955                 c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
956             }
957             else
958                 if (rz >= rx && rx >= ry)
959                 {
960                     c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
961                     c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
962                     c3 = DENS(X0, Y0, Z1) - c0;
963                 }
964                 else
965                     if (ry >= rx && rx >= rz)
966                     {
967                         c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
968                         c2 = DENS(X0, Y1, Z0) - c0;
969                         c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
970                     }
971                     else
972                         if (ry >= rz && rz >= rx)
973                         {
974                             c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
975                             c2 = DENS(X0, Y1, Z0) - c0;
976                             c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);
977                         }
978                         else
979                             if (rz >= ry && ry >= rx)
980                             {
981                                 c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
982                                 c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
983                                 c3 = DENS(X0, Y0, Z1) - c0;
984                             }
985                             else  {
986                                 c1 = c2 = c3 = 0;
987                             }
988 
989 
990                             Rest = c1 * rx + c2 * ry + c3 * rz + 0x8001;
991                             Output[OutChan] = (cmsUInt16Number)c0 + ((Rest + (Rest>>16))>>16);
992 
993     }
994 }
995 
996 #undef DENS
997 
998 
999 // Curves that contain wide empty areas are not optimizeable
1000 static
IsDegenerated(const cmsToneCurve * g)1001 cmsBool IsDegenerated(const cmsToneCurve* g)
1002 {
1003     int i, Zeros = 0, Poles = 0;
1004     int nEntries = g ->nEntries;
1005 
1006     for (i=0; i < nEntries; i++) {
1007 
1008         if (g ->Table16[i] == 0x0000) Zeros++;
1009         if (g ->Table16[i] == 0xffff) Poles++;
1010     }
1011 
1012     if (Zeros == 1 && Poles == 1) return FALSE;  // For linear tables
1013     if (Zeros > (nEntries / 20)) return TRUE;  // Degenerated, many zeros
1014     if (Poles > (nEntries / 20)) return TRUE;  // Degenerated, many poles
1015 
1016     return FALSE;
1017 }
1018 
1019 // --------------------------------------------------------------------------------------------------------------
1020 // We need xput over here
1021 
1022 static
OptimizeByComputingLinearization(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1023 cmsBool OptimizeByComputingLinearization(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1024 {
1025     cmsPipeline* OriginalLut;
1026     int nGridPoints;
1027     cmsToneCurve *Trans[cmsMAXCHANNELS], *TransReverse[cmsMAXCHANNELS];
1028     cmsUInt32Number t, i;
1029     cmsFloat32Number v, In[cmsMAXCHANNELS], Out[cmsMAXCHANNELS];
1030     cmsBool lIsSuitable, lIsLinear;
1031     cmsPipeline* OptimizedLUT = NULL, *LutPlusCurves = NULL;
1032     cmsStage* OptimizedCLUTmpe;
1033     cmsColorSpaceSignature ColorSpace, OutputColorSpace;
1034     cmsStage* OptimizedPrelinMpe;
1035     cmsStage* mpe;
1036     cmsToneCurve** OptimizedPrelinCurves;
1037     _cmsStageCLutData* OptimizedPrelinCLUT;
1038 
1039 
1040     // This is a loosy optimization! does not apply in floating-point cases
1041     if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
1042 
1043     // Only on chunky RGB
1044     if (T_COLORSPACE(*InputFormat)  != PT_RGB) return FALSE;
1045     if (T_PLANAR(*InputFormat)) return FALSE;
1046 
1047     if (T_COLORSPACE(*OutputFormat) != PT_RGB) return FALSE;
1048     if (T_PLANAR(*OutputFormat)) return FALSE;
1049 
1050     // On 16 bits, user has to specify the feature
1051     if (!_cmsFormatterIs8bit(*InputFormat)) {
1052         if (!(*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION)) return FALSE;
1053     }
1054 
1055     OriginalLut = *Lut;
1056 
1057    // Named color pipelines cannot be optimized either
1058    for (mpe = cmsPipelineGetPtrToFirstStage(OriginalLut);
1059          mpe != NULL;
1060          mpe = cmsStageNext(mpe)) {
1061             if (cmsStageType(mpe) == cmsSigNamedColorElemType) return FALSE;
1062     }
1063 
1064     ColorSpace       = _cmsICCcolorSpace(T_COLORSPACE(*InputFormat));
1065     OutputColorSpace = _cmsICCcolorSpace(T_COLORSPACE(*OutputFormat));
1066     nGridPoints      = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);
1067 
1068     // Empty gamma containers
1069     memset(Trans, 0, sizeof(Trans));
1070     memset(TransReverse, 0, sizeof(TransReverse));
1071 
1072     // If the last stage of the original lut are curves, and those curves are
1073     // degenerated, it is likely the transform is squeezing and clipping
1074     // the output from previous CLUT. We cannot optimize this case
1075     {
1076         cmsStage* last = cmsPipelineGetPtrToLastStage(OriginalLut);
1077 
1078         if (cmsStageType(last) == cmsSigCurveSetElemType) {
1079 
1080             _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*)cmsStageData(last);
1081             for (i = 0; i < Data->nCurves; i++) {
1082                 if (IsDegenerated(Data->TheCurves[i]))
1083                     goto Error;
1084             }
1085         }
1086     }
1087 
1088     for (t = 0; t < OriginalLut ->InputChannels; t++) {
1089         Trans[t] = cmsBuildTabulatedToneCurve16(OriginalLut ->ContextID, PRELINEARIZATION_POINTS, NULL);
1090         if (Trans[t] == NULL) goto Error;
1091     }
1092 
1093     // Populate the curves
1094     for (i=0; i < PRELINEARIZATION_POINTS; i++) {
1095 
1096         v = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
1097 
1098         // Feed input with a gray ramp
1099         for (t=0; t < OriginalLut ->InputChannels; t++)
1100             In[t] = v;
1101 
1102         // Evaluate the gray value
1103         cmsPipelineEvalFloat(In, Out, OriginalLut);
1104 
1105         // Store result in curve
1106         for (t=0; t < OriginalLut ->InputChannels; t++)
1107             Trans[t] ->Table16[i] = _cmsQuickSaturateWord(Out[t] * 65535.0);
1108     }
1109 
1110     // Slope-limit the obtained curves
1111     for (t = 0; t < OriginalLut ->InputChannels; t++)
1112         SlopeLimiting(Trans[t]);
1113 
1114     // Check for validity
1115     lIsSuitable = TRUE;
1116     lIsLinear   = TRUE;
1117     for (t=0; (lIsSuitable && (t < OriginalLut ->InputChannels)); t++) {
1118 
1119         // Exclude if already linear
1120         if (!cmsIsToneCurveLinear(Trans[t]))
1121             lIsLinear = FALSE;
1122 
1123         // Exclude if non-monotonic
1124         if (!cmsIsToneCurveMonotonic(Trans[t]))
1125             lIsSuitable = FALSE;
1126 
1127         if (IsDegenerated(Trans[t]))
1128             lIsSuitable = FALSE;
1129     }
1130 
1131     // If it is not suitable, just quit
1132     if (!lIsSuitable) goto Error;
1133 
1134     // Invert curves if possible
1135     for (t = 0; t < OriginalLut ->InputChannels; t++) {
1136         TransReverse[t] = cmsReverseToneCurveEx(PRELINEARIZATION_POINTS, Trans[t]);
1137         if (TransReverse[t] == NULL) goto Error;
1138     }
1139 
1140     // Now inset the reversed curves at the begin of transform
1141     LutPlusCurves = cmsPipelineDup(OriginalLut);
1142     if (LutPlusCurves == NULL) goto Error;
1143 
1144     if (!cmsPipelineInsertStage(LutPlusCurves, cmsAT_BEGIN, cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, TransReverse)))
1145         goto Error;
1146 
1147     // Create the result LUT
1148     OptimizedLUT = cmsPipelineAlloc(OriginalLut ->ContextID, OriginalLut ->InputChannels, OriginalLut ->OutputChannels);
1149     if (OptimizedLUT == NULL) goto Error;
1150 
1151     OptimizedPrelinMpe = cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, Trans);
1152 
1153     // Create and insert the curves at the beginning
1154     if (!cmsPipelineInsertStage(OptimizedLUT, cmsAT_BEGIN, OptimizedPrelinMpe))
1155         goto Error;
1156 
1157     // Allocate the CLUT for result
1158     OptimizedCLUTmpe = cmsStageAllocCLut16bit(OriginalLut ->ContextID, nGridPoints, OriginalLut ->InputChannels, OriginalLut ->OutputChannels, NULL);
1159 
1160     // Add the CLUT to the destination LUT
1161     if (!cmsPipelineInsertStage(OptimizedLUT, cmsAT_END, OptimizedCLUTmpe))
1162         goto Error;
1163 
1164     // Resample the LUT
1165     if (!cmsStageSampleCLut16bit(OptimizedCLUTmpe, XFormSampler16, (void*) LutPlusCurves, 0)) goto Error;
1166 
1167     // Free resources
1168     for (t = 0; t < OriginalLut ->InputChannels; t++) {
1169 
1170         if (Trans[t]) cmsFreeToneCurve(Trans[t]);
1171         if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
1172     }
1173 
1174     cmsPipelineFree(LutPlusCurves);
1175 
1176 
1177     OptimizedPrelinCurves = _cmsStageGetPtrToCurveSet(OptimizedPrelinMpe);
1178     OptimizedPrelinCLUT   = (_cmsStageCLutData*) OptimizedCLUTmpe ->Data;
1179 
1180     // Set the evaluator if 8-bit
1181     if (_cmsFormatterIs8bit(*InputFormat)) {
1182 
1183         Prelin8Data* p8 = PrelinOpt8alloc(OptimizedLUT ->ContextID,
1184                                                 OptimizedPrelinCLUT ->Params,
1185                                                 OptimizedPrelinCurves);
1186         if (p8 == NULL) return FALSE;
1187 
1188         _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval8, (void*) p8, Prelin8free, Prelin8dup);
1189 
1190     }
1191     else
1192     {
1193         Prelin16Data* p16 = PrelinOpt16alloc(OptimizedLUT ->ContextID,
1194             OptimizedPrelinCLUT ->Params,
1195             3, OptimizedPrelinCurves, 3, NULL);
1196         if (p16 == NULL) return FALSE;
1197 
1198         _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
1199 
1200     }
1201 
1202     // Don't fix white on absolute colorimetric
1203     if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
1204         *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
1205 
1206     if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
1207 
1208         if (!FixWhiteMisalignment(OptimizedLUT, ColorSpace, OutputColorSpace)) {
1209 
1210             return FALSE;
1211         }
1212     }
1213 
1214     // And return the obtained LUT
1215 
1216     cmsPipelineFree(OriginalLut);
1217     *Lut = OptimizedLUT;
1218     return TRUE;
1219 
1220 Error:
1221 
1222     for (t = 0; t < OriginalLut ->InputChannels; t++) {
1223 
1224         if (Trans[t]) cmsFreeToneCurve(Trans[t]);
1225         if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
1226     }
1227 
1228     if (LutPlusCurves != NULL) cmsPipelineFree(LutPlusCurves);
1229     if (OptimizedLUT != NULL) cmsPipelineFree(OptimizedLUT);
1230 
1231     return FALSE;
1232 
1233     cmsUNUSED_PARAMETER(Intent);
1234 }
1235 
1236 
1237 // Curves optimizer ------------------------------------------------------------------------------------------------------------------
1238 
1239 static
CurvesFree(cmsContext ContextID,void * ptr)1240 void CurvesFree(cmsContext ContextID, void* ptr)
1241 {
1242      Curves16Data* Data = (Curves16Data*) ptr;
1243      int i;
1244 
1245      for (i=0; i < Data -> nCurves; i++) {
1246 
1247          _cmsFree(ContextID, Data ->Curves[i]);
1248      }
1249 
1250      _cmsFree(ContextID, Data ->Curves);
1251      _cmsFree(ContextID, ptr);
1252 }
1253 
1254 static
CurvesDup(cmsContext ContextID,const void * ptr)1255 void* CurvesDup(cmsContext ContextID, const void* ptr)
1256 {
1257     Curves16Data* Data = (Curves16Data*)_cmsDupMem(ContextID, ptr, sizeof(Curves16Data));
1258     int i;
1259 
1260     if (Data == NULL) return NULL;
1261 
1262     Data->Curves = (cmsUInt16Number**) _cmsDupMem(ContextID, Data->Curves, Data->nCurves * sizeof(cmsUInt16Number*));
1263 
1264     for (i=0; i < Data -> nCurves; i++) {
1265         Data->Curves[i] = (cmsUInt16Number*) _cmsDupMem(ContextID, Data->Curves[i], Data->nElements * sizeof(cmsUInt16Number));
1266     }
1267 
1268     return (void*) Data;
1269 }
1270 
1271 // Precomputes tables for 8-bit on input devicelink.
1272 static
CurvesAlloc(cmsContext ContextID,int nCurves,int nElements,cmsToneCurve ** G)1273 Curves16Data* CurvesAlloc(cmsContext ContextID, int nCurves, int nElements, cmsToneCurve** G)
1274 {
1275     int i, j;
1276     Curves16Data* c16;
1277 
1278     c16 = (Curves16Data*)_cmsMallocZero(ContextID, sizeof(Curves16Data));
1279     if (c16 == NULL) return NULL;
1280 
1281     c16 ->nCurves = nCurves;
1282     c16 ->nElements = nElements;
1283 
1284     c16->Curves = (cmsUInt16Number**) _cmsCalloc(ContextID, nCurves, sizeof(cmsUInt16Number*));
1285     if (c16 ->Curves == NULL) return NULL;
1286 
1287     for (i=0; i < nCurves; i++) {
1288 
1289         c16->Curves[i] = (cmsUInt16Number*) _cmsCalloc(ContextID, nElements, sizeof(cmsUInt16Number));
1290 
1291         if (c16->Curves[i] == NULL) {
1292 
1293             for (j=0; j < i; j++) {
1294                 _cmsFree(ContextID, c16->Curves[j]);
1295             }
1296             _cmsFree(ContextID, c16->Curves);
1297             _cmsFree(ContextID, c16);
1298             return NULL;
1299         }
1300 
1301         if (nElements == 256) {
1302 
1303             for (j=0; j < nElements; j++) {
1304 
1305                 c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], FROM_8_TO_16(j));
1306             }
1307         }
1308         else {
1309 
1310             for (j=0; j < nElements; j++) {
1311                 c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], (cmsUInt16Number) j);
1312             }
1313         }
1314     }
1315 
1316     return c16;
1317 }
1318 
1319 static
FastEvaluateCurves8(register const cmsUInt16Number In[],register cmsUInt16Number Out[],register const void * D)1320 void FastEvaluateCurves8(register const cmsUInt16Number In[],
1321                           register cmsUInt16Number Out[],
1322                           register const void* D)
1323 {
1324     Curves16Data* Data = (Curves16Data*) D;
1325     cmsUInt8Number x;
1326     int i;
1327 
1328     for (i=0; i < Data ->nCurves; i++) {
1329 
1330          x = (In[i] >> 8);
1331          Out[i] = Data -> Curves[i][x];
1332     }
1333 }
1334 
1335 
1336 static
FastEvaluateCurves16(register const cmsUInt16Number In[],register cmsUInt16Number Out[],register const void * D)1337 void FastEvaluateCurves16(register const cmsUInt16Number In[],
1338                           register cmsUInt16Number Out[],
1339                           register const void* D)
1340 {
1341     Curves16Data* Data = (Curves16Data*) D;
1342     int i;
1343 
1344     for (i=0; i < Data ->nCurves; i++) {
1345          Out[i] = Data -> Curves[i][In[i]];
1346     }
1347 }
1348 
1349 
1350 static
FastIdentity16(register const cmsUInt16Number In[],register cmsUInt16Number Out[],register const void * D)1351 void FastIdentity16(register const cmsUInt16Number In[],
1352                     register cmsUInt16Number Out[],
1353                     register const void* D)
1354 {
1355     cmsPipeline* Lut = (cmsPipeline*) D;
1356     cmsUInt32Number i;
1357 
1358     for (i=0; i < Lut ->InputChannels; i++) {
1359          Out[i] = In[i];
1360     }
1361 }
1362 
1363 
1364 // If the target LUT holds only curves, the optimization procedure is to join all those
1365 // curves together. That only works on curves and does not work on matrices.
1366 static
OptimizeByJoiningCurves(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1367 cmsBool OptimizeByJoiningCurves(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1368 {
1369     cmsToneCurve** GammaTables = NULL;
1370     cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
1371     cmsUInt32Number i, j;
1372     cmsPipeline* Src = *Lut;
1373     cmsPipeline* Dest = NULL;
1374     cmsStage* mpe;
1375     cmsStage* ObtainedCurves = NULL;
1376 
1377 
1378     // This is a loosy optimization! does not apply in floating-point cases
1379     if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
1380 
1381     //  Only curves in this LUT?
1382     for (mpe = cmsPipelineGetPtrToFirstStage(Src);
1383          mpe != NULL;
1384          mpe = cmsStageNext(mpe)) {
1385             if (cmsStageType(mpe) != cmsSigCurveSetElemType) return FALSE;
1386     }
1387 
1388     // Allocate an empty LUT
1389     Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
1390     if (Dest == NULL) return FALSE;
1391 
1392     // Create target curves
1393     GammaTables = (cmsToneCurve**) _cmsCalloc(Src ->ContextID, Src ->InputChannels, sizeof(cmsToneCurve*));
1394     if (GammaTables == NULL) goto Error;
1395 
1396     for (i=0; i < Src ->InputChannels; i++) {
1397         GammaTables[i] = cmsBuildTabulatedToneCurve16(Src ->ContextID, PRELINEARIZATION_POINTS, NULL);
1398         if (GammaTables[i] == NULL) goto Error;
1399     }
1400 
1401     // Compute 16 bit result by using floating point
1402     for (i=0; i < PRELINEARIZATION_POINTS; i++) {
1403 
1404         for (j=0; j < Src ->InputChannels; j++)
1405             InFloat[j] = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
1406 
1407         cmsPipelineEvalFloat(InFloat, OutFloat, Src);
1408 
1409         for (j=0; j < Src ->InputChannels; j++)
1410             GammaTables[j] -> Table16[i] = _cmsQuickSaturateWord(OutFloat[j] * 65535.0);
1411     }
1412 
1413     ObtainedCurves = cmsStageAllocToneCurves(Src ->ContextID, Src ->InputChannels, GammaTables);
1414     if (ObtainedCurves == NULL) goto Error;
1415 
1416     for (i=0; i < Src ->InputChannels; i++) {
1417         cmsFreeToneCurve(GammaTables[i]);
1418         GammaTables[i] = NULL;
1419     }
1420 
1421     if (GammaTables != NULL) {
1422         _cmsFree(Src->ContextID, GammaTables);
1423         GammaTables = NULL;
1424     }
1425 
1426     // Maybe the curves are linear at the end
1427     if (!AllCurvesAreLinear(ObtainedCurves)) {
1428 
1429         if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, ObtainedCurves))
1430             goto Error;
1431 
1432         // If the curves are to be applied in 8 bits, we can save memory
1433         if (_cmsFormatterIs8bit(*InputFormat)) {
1434 
1435             _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*) ObtainedCurves ->Data;
1436              Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 256, Data ->TheCurves);
1437 
1438              if (c16 == NULL) goto Error;
1439              *dwFlags |= cmsFLAGS_NOCACHE;
1440             _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves8, c16, CurvesFree, CurvesDup);
1441 
1442         }
1443         else {
1444 
1445             _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*) cmsStageData(ObtainedCurves);
1446              Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 65536, Data ->TheCurves);
1447 
1448              if (c16 == NULL) goto Error;
1449              *dwFlags |= cmsFLAGS_NOCACHE;
1450             _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves16, c16, CurvesFree, CurvesDup);
1451         }
1452     }
1453     else {
1454 
1455         // LUT optimizes to nothing. Set the identity LUT
1456         cmsStageFree(ObtainedCurves);
1457 
1458         if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageAllocIdentity(Dest ->ContextID, Src ->InputChannels)))
1459             goto Error;
1460 
1461         *dwFlags |= cmsFLAGS_NOCACHE;
1462         _cmsPipelineSetOptimizationParameters(Dest, FastIdentity16, (void*) Dest, NULL, NULL);
1463     }
1464 
1465     // We are done.
1466     cmsPipelineFree(Src);
1467     *Lut = Dest;
1468     return TRUE;
1469 
1470 Error:
1471 
1472     if (ObtainedCurves != NULL) cmsStageFree(ObtainedCurves);
1473     if (GammaTables != NULL) {
1474         for (i=0; i < Src ->InputChannels; i++) {
1475             if (GammaTables[i] != NULL) cmsFreeToneCurve(GammaTables[i]);
1476         }
1477 
1478         _cmsFree(Src ->ContextID, GammaTables);
1479     }
1480 
1481     if (Dest != NULL) cmsPipelineFree(Dest);
1482     return FALSE;
1483 
1484     cmsUNUSED_PARAMETER(Intent);
1485     cmsUNUSED_PARAMETER(InputFormat);
1486     cmsUNUSED_PARAMETER(OutputFormat);
1487     cmsUNUSED_PARAMETER(dwFlags);
1488 }
1489 
1490 // -------------------------------------------------------------------------------------------------------------------------------------
1491 // LUT is Shaper - Matrix - Matrix - Shaper, which is very frequent when combining two matrix-shaper profiles
1492 
1493 
1494 static
FreeMatShaper(cmsContext ContextID,void * Data)1495 void  FreeMatShaper(cmsContext ContextID, void* Data)
1496 {
1497     if (Data != NULL) _cmsFree(ContextID, Data);
1498 }
1499 
1500 static
DupMatShaper(cmsContext ContextID,const void * Data)1501 void* DupMatShaper(cmsContext ContextID, const void* Data)
1502 {
1503     return _cmsDupMem(ContextID, Data, sizeof(MatShaper8Data));
1504 }
1505 
1506 
1507 // A fast matrix-shaper evaluator for 8 bits. This is a bit ticky since I'm using 1.14 signed fixed point
1508 // to accomplish some performance. Actually it takes 256x3 16 bits tables and 16385 x 3 tables of 8 bits,
1509 // in total about 50K, and the performance boost is huge!
1510 static
MatShaperEval16(register const cmsUInt16Number In[],register cmsUInt16Number Out[],register const void * D)1511 void MatShaperEval16(register const cmsUInt16Number In[],
1512                      register cmsUInt16Number Out[],
1513                      register const void* D)
1514 {
1515     MatShaper8Data* p = (MatShaper8Data*) D;
1516     cmsS1Fixed14Number l1, l2, l3, r, g, b;
1517     cmsUInt32Number ri, gi, bi;
1518 
1519     // In this case (and only in this case!) we can use this simplification since
1520     // In[] is assured to come from a 8 bit number. (a << 8 | a)
1521     ri = In[0] & 0xFF;
1522     gi = In[1] & 0xFF;
1523     bi = In[2] & 0xFF;
1524 
1525     // Across first shaper, which also converts to 1.14 fixed point
1526     r = p->Shaper1R[ri];
1527     g = p->Shaper1G[gi];
1528     b = p->Shaper1B[bi];
1529 
1530     // Evaluate the matrix in 1.14 fixed point
1531     l1 =  (p->Mat[0][0] * r + p->Mat[0][1] * g + p->Mat[0][2] * b + p->Off[0] + 0x2000) >> 14;
1532     l2 =  (p->Mat[1][0] * r + p->Mat[1][1] * g + p->Mat[1][2] * b + p->Off[1] + 0x2000) >> 14;
1533     l3 =  (p->Mat[2][0] * r + p->Mat[2][1] * g + p->Mat[2][2] * b + p->Off[2] + 0x2000) >> 14;
1534 
1535     // Now we have to clip to 0..1.0 range
1536     ri = (l1 < 0) ? 0 : ((l1 > 16384) ? 16384 : l1);
1537     gi = (l2 < 0) ? 0 : ((l2 > 16384) ? 16384 : l2);
1538     bi = (l3 < 0) ? 0 : ((l3 > 16384) ? 16384 : l3);
1539 
1540     // And across second shaper,
1541     Out[0] = p->Shaper2R[ri];
1542     Out[1] = p->Shaper2G[gi];
1543     Out[2] = p->Shaper2B[bi];
1544 
1545 }
1546 
1547 // This table converts from 8 bits to 1.14 after applying the curve
1548 static
FillFirstShaper(cmsS1Fixed14Number * Table,cmsToneCurve * Curve)1549 cmsBool FillFirstShaper(cmsS1Fixed14Number* Table, cmsToneCurve* Curve)
1550 {
1551     int i;
1552     cmsFloat32Number R, y;
1553 
1554     for (i=0; i < 256; i++) {
1555 
1556         R   = (cmsFloat32Number) (i / 255.0);
1557         y   = cmsEvalToneCurveFloat(Curve, R);
1558         if (isinf(y))
1559             return FALSE;
1560 
1561         Table[i] = DOUBLE_TO_1FIXED14(y);
1562     }
1563     return TRUE;
1564 }
1565 
1566 // This table converts form 1.14 (being 0x4000 the last entry) to 8 bits after applying the curve
1567 static
FillSecondShaper(cmsUInt16Number * Table,cmsToneCurve * Curve,cmsBool Is8BitsOutput)1568 cmsBool FillSecondShaper(cmsUInt16Number* Table, cmsToneCurve* Curve, cmsBool Is8BitsOutput)
1569 {
1570     int i;
1571     cmsFloat32Number R, Val;
1572 
1573     for (i=0; i < 16385; i++) {
1574 
1575         R   = (cmsFloat32Number) (i / 16384.0);
1576         Val = cmsEvalToneCurveFloat(Curve, R);    // Val comes 0..1.0
1577         if (isinf(Val))
1578             return FALSE;
1579 
1580         if (Is8BitsOutput) {
1581 
1582             // If 8 bits output, we can optimize further by computing the / 257 part.
1583             // first we compute the resulting byte and then we store the byte times
1584             // 257. This quantization allows to round very quick by doing a >> 8, but
1585             // since the low byte is always equal to msb, we can do a & 0xff and this works!
1586             cmsUInt16Number w = _cmsQuickSaturateWord(Val * 65535.0);
1587             cmsUInt8Number  b = FROM_16_TO_8(w);
1588 
1589             Table[i] = FROM_8_TO_16(b);
1590         }
1591         else Table[i]  = _cmsQuickSaturateWord(Val * 65535.0);
1592     }
1593     return TRUE;
1594 }
1595 
1596 // Compute the matrix-shaper structure
1597 static
SetMatShaper(cmsPipeline * Dest,cmsToneCurve * Curve1[3],cmsMAT3 * Mat,cmsVEC3 * Off,cmsToneCurve * Curve2[3],cmsUInt32Number * OutputFormat)1598 cmsBool SetMatShaper(cmsPipeline* Dest, cmsToneCurve* Curve1[3], cmsMAT3* Mat, cmsVEC3* Off, cmsToneCurve* Curve2[3], cmsUInt32Number* OutputFormat)
1599 {
1600     MatShaper8Data* p;
1601     int i, j;
1602     cmsBool Is8Bits = _cmsFormatterIs8bit(*OutputFormat);
1603 
1604     // Allocate a big chuck of memory to store precomputed tables
1605     p = (MatShaper8Data*) _cmsMalloc(Dest ->ContextID, sizeof(MatShaper8Data));
1606     if (p == NULL) return FALSE;
1607 
1608     p -> ContextID = Dest -> ContextID;
1609 
1610     // Precompute tables
1611     if (!FillFirstShaper(p ->Shaper1R, Curve1[0]))
1612         goto Error;
1613     if (!FillFirstShaper(p ->Shaper1G, Curve1[1]))
1614         goto Error;
1615     if (!FillFirstShaper(p ->Shaper1B, Curve1[2]))
1616         goto Error;
1617 
1618     if (!FillSecondShaper(p ->Shaper2R, Curve2[0], Is8Bits))
1619         goto Error;
1620     if (!FillSecondShaper(p ->Shaper2G, Curve2[1], Is8Bits))
1621         goto Error;
1622     if (!FillSecondShaper(p ->Shaper2B, Curve2[2], Is8Bits))
1623         goto Error;
1624 
1625     // Convert matrix to nFixed14. Note that those values may take more than 16 bits as
1626     for (i=0; i < 3; i++) {
1627         for (j=0; j < 3; j++) {
1628             p ->Mat[i][j] = DOUBLE_TO_1FIXED14(Mat->v[i].n[j]);
1629         }
1630     }
1631 
1632     for (i=0; i < 3; i++) {
1633 
1634         if (Off == NULL) {
1635             p ->Off[i] = 0;
1636         }
1637         else {
1638             p ->Off[i] = DOUBLE_TO_1FIXED14(Off->n[i]);
1639         }
1640     }
1641 
1642     // Mark as optimized for faster formatter
1643     if (Is8Bits)
1644         *OutputFormat |= OPTIMIZED_SH(1);
1645 
1646     // Fill function pointers
1647     _cmsPipelineSetOptimizationParameters(Dest, MatShaperEval16, (void*) p, FreeMatShaper, DupMatShaper);
1648     return TRUE;
1649     Error:
1650         _cmsFree(Dest->ContextID, p);
1651         return FALSE;
1652 }
1653 
1654 //  8 bits on input allows matrix-shaper boot up to 25 Mpixels per second on RGB. That's fast!
1655 static
OptimizeMatrixShaper(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1656 cmsBool OptimizeMatrixShaper(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1657 {
1658        cmsStage* Curve1, *Curve2;
1659        cmsStage* Matrix1, *Matrix2;
1660        cmsMAT3 res;
1661        cmsBool IdentityMat;
1662        cmsPipeline* Dest, *Src;
1663        cmsFloat64Number* Offset;
1664 
1665        // Only works on RGB to RGB
1666        if (T_CHANNELS(*InputFormat) != 3 || T_CHANNELS(*OutputFormat) != 3) return FALSE;
1667 
1668        // Only works on 8 bit input
1669        if (!_cmsFormatterIs8bit(*InputFormat)) return FALSE;
1670 
1671        // Seems suitable, proceed
1672        Src = *Lut;
1673 
1674        // Check for:
1675        //
1676        //    shaper-matrix-matrix-shaper
1677        //    shaper-matrix-shaper
1678        //
1679        // Both of those constructs are possible (first because abs. colorimetric).
1680        // additionally, In the first case, the input matrix offset should be zero.
1681 
1682        IdentityMat = FALSE;
1683        if (cmsPipelineCheckAndRetreiveStages(Src, 4,
1684               cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
1685               &Curve1, &Matrix1, &Matrix2, &Curve2)) {
1686 
1687               // Get both matrices
1688               _cmsStageMatrixData* Data1 = (_cmsStageMatrixData*)cmsStageData(Matrix1);
1689               _cmsStageMatrixData* Data2 = (_cmsStageMatrixData*)cmsStageData(Matrix2);
1690 
1691               // Input offset should be zero
1692               if (Data1->Offset != NULL) return FALSE;
1693 
1694               // Multiply both matrices to get the result
1695               _cmsMAT3per(&res, (cmsMAT3*)Data2->Double, (cmsMAT3*)Data1->Double);
1696 
1697               // Only 2nd matrix has offset, or it is zero
1698               Offset = Data2->Offset;
1699 
1700               // Now the result is in res + Data2 -> Offset. Maybe is a plain identity?
1701               if (_cmsMAT3isIdentity(&res) && Offset == NULL) {
1702 
1703                      // We can get rid of full matrix
1704                      IdentityMat = TRUE;
1705               }
1706 
1707        }
1708        else {
1709 
1710               if (cmsPipelineCheckAndRetreiveStages(Src, 3,
1711                      cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
1712                      &Curve1, &Matrix1, &Curve2)) {
1713 
1714                      _cmsStageMatrixData* Data = (_cmsStageMatrixData*)cmsStageData(Matrix1);
1715 
1716                      // Copy the matrix to our result
1717                      memcpy(&res, Data->Double, sizeof(res));
1718 
1719                      // Preserve the Odffset (may be NULL as a zero offset)
1720                      Offset = Data->Offset;
1721 
1722                      if (_cmsMAT3isIdentity(&res) && Offset == NULL) {
1723 
1724                             // We can get rid of full matrix
1725                             IdentityMat = TRUE;
1726                      }
1727               }
1728               else
1729                      return FALSE; // Not optimizeable this time
1730 
1731        }
1732 
1733       // Allocate an empty LUT
1734     Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
1735     if (!Dest) return FALSE;
1736 
1737     // Assamble the new LUT
1738     if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageDup(Curve1)))
1739         goto Error;
1740 
1741     if (!IdentityMat) {
1742 
1743            if (!cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageAllocMatrix(Dest->ContextID, 3, 3, (const cmsFloat64Number*)&res, Offset)))
1744                   goto Error;
1745     }
1746 
1747     if (!cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageDup(Curve2)))
1748         goto Error;
1749 
1750     // If identity on matrix, we can further optimize the curves, so call the join curves routine
1751     if (IdentityMat) {
1752 
1753         OptimizeByJoiningCurves(&Dest, Intent, InputFormat, OutputFormat, dwFlags);
1754     }
1755     else {
1756         _cmsStageToneCurvesData* mpeC1 = (_cmsStageToneCurvesData*) cmsStageData(Curve1);
1757         _cmsStageToneCurvesData* mpeC2 = (_cmsStageToneCurvesData*) cmsStageData(Curve2);
1758 
1759         // In this particular optimization, cache does not help as it takes more time to deal with
1760         // the cache that with the pixel handling
1761         *dwFlags |= cmsFLAGS_NOCACHE;
1762 
1763         // Setup the optimizarion routines
1764         if (!SetMatShaper(Dest, mpeC1 ->TheCurves, &res, (cmsVEC3*) Offset, mpeC2->TheCurves, OutputFormat))
1765             goto Error;
1766     }
1767 
1768     cmsPipelineFree(Src);
1769     *Lut = Dest;
1770     return TRUE;
1771 Error:
1772     // Leave Src unchanged
1773     cmsPipelineFree(Dest);
1774     return FALSE;
1775 }
1776 
1777 
1778 // -------------------------------------------------------------------------------------------------------------------------------------
1779 // Optimization plug-ins
1780 
1781 // List of optimizations
1782 typedef struct _cmsOptimizationCollection_st {
1783 
1784     _cmsOPToptimizeFn  OptimizePtr;
1785 
1786     struct _cmsOptimizationCollection_st *Next;
1787 
1788 } _cmsOptimizationCollection;
1789 
1790 
1791 // The built-in list. We currently implement 4 types of optimizations. Joining of curves, matrix-shaper, linearization and resampling
1792 static _cmsOptimizationCollection DefaultOptimization[] = {
1793 
1794     { OptimizeByJoiningCurves,            &DefaultOptimization[1] },
1795     { OptimizeMatrixShaper,               &DefaultOptimization[2] },
1796     { OptimizeByComputingLinearization,   &DefaultOptimization[3] },
1797     { OptimizeByResampling,               NULL }
1798 };
1799 
1800 // The linked list head
1801 _cmsOptimizationPluginChunkType _cmsOptimizationPluginChunk = { NULL };
1802 
1803 
1804 // Duplicates the zone of memory used by the plug-in in the new context
1805 static
DupPluginOptimizationList(struct _cmsContext_struct * ctx,const struct _cmsContext_struct * src)1806 void DupPluginOptimizationList(struct _cmsContext_struct* ctx,
1807                                const struct _cmsContext_struct* src)
1808 {
1809    _cmsOptimizationPluginChunkType newHead = { NULL };
1810    _cmsOptimizationCollection*  entry;
1811    _cmsOptimizationCollection*  Anterior = NULL;
1812    _cmsOptimizationPluginChunkType* head = (_cmsOptimizationPluginChunkType*) src->chunks[OptimizationPlugin];
1813 
1814     _cmsAssert(ctx != NULL);
1815     _cmsAssert(head != NULL);
1816 
1817     // Walk the list copying all nodes
1818    for (entry = head->OptimizationCollection;
1819         entry != NULL;
1820         entry = entry ->Next) {
1821 
1822             _cmsOptimizationCollection *newEntry = ( _cmsOptimizationCollection *) _cmsSubAllocDup(ctx ->MemPool, entry, sizeof(_cmsOptimizationCollection));
1823 
1824             if (newEntry == NULL)
1825                 return;
1826 
1827             // We want to keep the linked list order, so this is a little bit tricky
1828             newEntry -> Next = NULL;
1829             if (Anterior)
1830                 Anterior -> Next = newEntry;
1831 
1832             Anterior = newEntry;
1833 
1834             if (newHead.OptimizationCollection == NULL)
1835                 newHead.OptimizationCollection = newEntry;
1836     }
1837 
1838   ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx->MemPool, &newHead, sizeof(_cmsOptimizationPluginChunkType));
1839 }
1840 
_cmsAllocOptimizationPluginChunk(struct _cmsContext_struct * ctx,const struct _cmsContext_struct * src)1841 void  _cmsAllocOptimizationPluginChunk(struct _cmsContext_struct* ctx,
1842                                          const struct _cmsContext_struct* src)
1843 {
1844   if (src != NULL) {
1845 
1846         // Copy all linked list
1847        DupPluginOptimizationList(ctx, src);
1848     }
1849     else {
1850         static _cmsOptimizationPluginChunkType OptimizationPluginChunkType = { NULL };
1851         ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx ->MemPool, &OptimizationPluginChunkType, sizeof(_cmsOptimizationPluginChunkType));
1852     }
1853 }
1854 
1855 
1856 // Register new ways to optimize
_cmsRegisterOptimizationPlugin(cmsContext ContextID,cmsPluginBase * Data)1857 cmsBool  _cmsRegisterOptimizationPlugin(cmsContext ContextID, cmsPluginBase* Data)
1858 {
1859     cmsPluginOptimization* Plugin = (cmsPluginOptimization*) Data;
1860     _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
1861     _cmsOptimizationCollection* fl;
1862 
1863     if (Data == NULL) {
1864 
1865         ctx->OptimizationCollection = NULL;
1866         return TRUE;
1867     }
1868 
1869     // Optimizer callback is required
1870     if (Plugin ->OptimizePtr == NULL) return FALSE;
1871 
1872     fl = (_cmsOptimizationCollection*) _cmsPluginMalloc(ContextID, sizeof(_cmsOptimizationCollection));
1873     if (fl == NULL) return FALSE;
1874 
1875     // Copy the parameters
1876     fl ->OptimizePtr = Plugin ->OptimizePtr;
1877 
1878     // Keep linked list
1879     fl ->Next = ctx->OptimizationCollection;
1880 
1881     // Set the head
1882     ctx ->OptimizationCollection = fl;
1883 
1884     // All is ok
1885     return TRUE;
1886 }
1887 
1888 // The entry point for LUT optimization
_cmsOptimizePipeline(cmsContext ContextID,cmsPipeline ** PtrLut,int Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1889 cmsBool _cmsOptimizePipeline(cmsContext ContextID,
1890                              cmsPipeline**    PtrLut,
1891                              int              Intent,
1892                              cmsUInt32Number* InputFormat,
1893                              cmsUInt32Number* OutputFormat,
1894                              cmsUInt32Number* dwFlags)
1895 {
1896     _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
1897     _cmsOptimizationCollection* Opts;
1898     cmsBool AnySuccess = FALSE;
1899 
1900     // A CLUT is being asked, so force this specific optimization
1901     if (*dwFlags & cmsFLAGS_FORCE_CLUT) {
1902 
1903         PreOptimize(*PtrLut);
1904         return OptimizeByResampling(PtrLut, Intent, InputFormat, OutputFormat, dwFlags);
1905     }
1906 
1907     // Anything to optimize?
1908     if ((*PtrLut) ->Elements == NULL) {
1909         _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
1910         return TRUE;
1911     }
1912 
1913     // Try to get rid of identities and trivial conversions.
1914     AnySuccess = PreOptimize(*PtrLut);
1915 
1916     // After removal do we end with an identity?
1917     if ((*PtrLut) ->Elements == NULL) {
1918         _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
1919         return TRUE;
1920     }
1921 
1922     // Do not optimize, keep all precision
1923     if (*dwFlags & cmsFLAGS_NOOPTIMIZE)
1924         return FALSE;
1925 
1926     // Try plug-in optimizations
1927     for (Opts = ctx->OptimizationCollection;
1928          Opts != NULL;
1929          Opts = Opts ->Next) {
1930 
1931             // If one schema succeeded, we are done
1932             if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
1933 
1934                 return TRUE;    // Optimized!
1935             }
1936     }
1937 
1938    // Try built-in optimizations
1939     for (Opts = DefaultOptimization;
1940          Opts != NULL;
1941          Opts = Opts ->Next) {
1942 
1943             if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
1944 
1945                 return TRUE;
1946             }
1947     }
1948 
1949     // Only simple optimizations succeeded
1950     return AnySuccess;
1951 }
1952 
1953 
1954 
1955