• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 //---------------------------------------------------------------------------------
2 //
3 //  Little Color Management System
4 //  Copyright (c) 1998-2023 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 
95     cmsContext ContextID;
96 
97     cmsUInt32Number nCurves;      // Number of curves
98     cmsUInt32Number nElements;    // Elements in curves
99     cmsUInt16Number** Curves;     // Points to a dynamically  allocated array
100 
101 } Curves16Data;
102 
103 
104 // Simple optimizations ----------------------------------------------------------------------------------------------------------
105 
106 
107 // Clamp a fixed point integer to signed 28 bits to avoid overflow in
108 // calculations.  Clamp is intended for use with colorants, requiring one bit
109 // for a colorant and another two bits to avoid overflow when combining the
110 // colors.
_FixedClamp(cmsS1Fixed14Number n)111 cmsINLINE cmsS1Fixed14Number _FixedClamp(cmsS1Fixed14Number n) {
112   const cmsS1Fixed14Number max_positive = 268435455;  // 0x0FFFFFFF;
113   const cmsS1Fixed14Number max_negative = -268435456; // 0xF0000000;
114   // Normally expect the provided number to be in the range [0..1] (but in
115   // fixed 1.14 format), so can perform a quick check for this typical case
116   // to reduce number of compares.
117   const cmsS1Fixed14Number typical_range_mask = 0xFFFF8000;
118 
119   if (!(n & typical_range_mask))
120     return n;
121   if (n < max_negative)
122      return max_negative;
123   if (n > max_positive)
124     return max_positive;
125   return n;
126 }
127 
128 // Perform one row of matrix multiply with translation for MatShaperEval16().
_MatShaperEvaluateRow(cmsS1Fixed14Number * mat,cmsS1Fixed14Number off,cmsS1Fixed14Number r,cmsS1Fixed14Number g,cmsS1Fixed14Number b)129 cmsINLINE cmsInt64Number _MatShaperEvaluateRow(cmsS1Fixed14Number* mat,
130                                                cmsS1Fixed14Number off,
131                                                cmsS1Fixed14Number r,
132                                                cmsS1Fixed14Number g,
133                                                cmsS1Fixed14Number b) {
134   return ((cmsInt64Number)mat[0] * r +
135           (cmsInt64Number)mat[1] * g +
136           (cmsInt64Number)mat[2] * b +
137           off + 0x2000) >> 14;
138 }
139 
140 // Remove an element in linked chain
141 static
_RemoveElement(cmsStage ** head)142 void _RemoveElement(cmsStage** head)
143 {
144     cmsStage* mpe = *head;
145     cmsStage* next = mpe ->Next;
146     *head = next;
147     cmsStageFree(mpe);
148 }
149 
150 // Remove all identities in chain. Note that pt actually is a double pointer to the element that holds the pointer.
151 static
_Remove1Op(cmsPipeline * Lut,cmsStageSignature UnaryOp)152 cmsBool _Remove1Op(cmsPipeline* Lut, cmsStageSignature UnaryOp)
153 {
154     cmsStage** pt = &Lut ->Elements;
155     cmsBool AnyOpt = FALSE;
156 
157     while (*pt != NULL) {
158 
159         if ((*pt) ->Implements == UnaryOp) {
160             _RemoveElement(pt);
161             AnyOpt = TRUE;
162         }
163         else
164             pt = &((*pt) -> Next);
165     }
166 
167     return AnyOpt;
168 }
169 
170 // Same, but only if two adjacent elements are found
171 static
_Remove2Op(cmsPipeline * Lut,cmsStageSignature Op1,cmsStageSignature Op2)172 cmsBool _Remove2Op(cmsPipeline* Lut, cmsStageSignature Op1, cmsStageSignature Op2)
173 {
174     cmsStage** pt1;
175     cmsStage** pt2;
176     cmsBool AnyOpt = FALSE;
177 
178     pt1 = &Lut ->Elements;
179     if (*pt1 == NULL) return AnyOpt;
180 
181     while (*pt1 != NULL) {
182 
183         pt2 = &((*pt1) -> Next);
184         if (*pt2 == NULL) return AnyOpt;
185 
186         if ((*pt1) ->Implements == Op1 && (*pt2) ->Implements == Op2) {
187             _RemoveElement(pt2);
188             _RemoveElement(pt1);
189             AnyOpt = TRUE;
190         }
191         else
192             pt1 = &((*pt1) -> Next);
193     }
194 
195     return AnyOpt;
196 }
197 
198 
199 static
CloseEnoughFloat(cmsFloat64Number a,cmsFloat64Number b)200 cmsBool CloseEnoughFloat(cmsFloat64Number a, cmsFloat64Number b)
201 {
202        return fabs(b - a) < 0.00001f;
203 }
204 
205 static
isFloatMatrixIdentity(const cmsMAT3 * a)206 cmsBool  isFloatMatrixIdentity(const cmsMAT3* a)
207 {
208        cmsMAT3 Identity;
209        int i, j;
210 
211        _cmsMAT3identity(&Identity);
212 
213        for (i = 0; i < 3; i++)
214               for (j = 0; j < 3; j++)
215                      if (!CloseEnoughFloat(a->v[i].n[j], Identity.v[i].n[j])) return FALSE;
216 
217        return TRUE;
218 }
219 // if two adjacent matrices are found, multiply them.
220 static
_MultiplyMatrix(cmsPipeline * Lut)221 cmsBool _MultiplyMatrix(cmsPipeline* Lut)
222 {
223        cmsStage** pt1;
224        cmsStage** pt2;
225        cmsStage*  chain;
226        cmsBool AnyOpt = FALSE;
227 
228        pt1 = &Lut->Elements;
229        if (*pt1 == NULL) return AnyOpt;
230 
231        while (*pt1 != NULL) {
232 
233               pt2 = &((*pt1)->Next);
234               if (*pt2 == NULL) return AnyOpt;
235 
236               if ((*pt1)->Implements == cmsSigMatrixElemType && (*pt2)->Implements == cmsSigMatrixElemType) {
237 
238                      // Get both matrices
239                      _cmsStageMatrixData* m1 = (_cmsStageMatrixData*) cmsStageData(*pt1);
240                      _cmsStageMatrixData* m2 = (_cmsStageMatrixData*) cmsStageData(*pt2);
241                      cmsMAT3 res;
242 
243                      // Input offset and output offset should be zero to use this optimization
244                      if (m1->Offset != NULL || m2 ->Offset != NULL ||
245                             cmsStageInputChannels(*pt1) != 3 || cmsStageOutputChannels(*pt1) != 3 ||
246                             cmsStageInputChannels(*pt2) != 3 || cmsStageOutputChannels(*pt2) != 3)
247                             return FALSE;
248 
249                      // Multiply both matrices to get the result
250                      _cmsMAT3per(&res, (cmsMAT3*)m2->Double, (cmsMAT3*)m1->Double);
251 
252                      // Get the next in chain after the matrices
253                      chain = (*pt2)->Next;
254 
255                      // Remove both matrices
256                      _RemoveElement(pt2);
257                      _RemoveElement(pt1);
258 
259                      // Now what if the result is a plain identity?
260                      if (!isFloatMatrixIdentity(&res)) {
261 
262                             // We can not get rid of full matrix
263                             cmsStage* Multmat = cmsStageAllocMatrix(Lut->ContextID, 3, 3, (const cmsFloat64Number*) &res, NULL);
264                             if (Multmat == NULL) return FALSE;  // Should never happen
265 
266                             // Recover the chain
267                             Multmat->Next = chain;
268                             *pt1 = Multmat;
269                      }
270 
271                      AnyOpt = TRUE;
272               }
273               else
274                      pt1 = &((*pt1)->Next);
275        }
276 
277        return AnyOpt;
278 }
279 
280 
281 // Preoptimize just gets rif of no-ops coming paired. Conversion from v2 to v4 followed
282 // by a v4 to v2 and vice-versa. The elements are then discarded.
283 static
PreOptimize(cmsPipeline * Lut)284 cmsBool PreOptimize(cmsPipeline* Lut)
285 {
286     cmsBool AnyOpt = FALSE, Opt;
287 
288     do {
289 
290         Opt = FALSE;
291 
292         // Remove all identities
293         Opt |= _Remove1Op(Lut, cmsSigIdentityElemType);
294 
295         // Remove XYZ2Lab followed by Lab2XYZ
296         Opt |= _Remove2Op(Lut, cmsSigXYZ2LabElemType, cmsSigLab2XYZElemType);
297 
298         // Remove Lab2XYZ followed by XYZ2Lab
299         Opt |= _Remove2Op(Lut, cmsSigLab2XYZElemType, cmsSigXYZ2LabElemType);
300 
301         // Remove V4 to V2 followed by V2 to V4
302         Opt |= _Remove2Op(Lut, cmsSigLabV4toV2, cmsSigLabV2toV4);
303 
304         // Remove V2 to V4 followed by V4 to V2
305         Opt |= _Remove2Op(Lut, cmsSigLabV2toV4, cmsSigLabV4toV2);
306 
307         // Remove float pcs Lab conversions
308         Opt |= _Remove2Op(Lut, cmsSigLab2FloatPCS, cmsSigFloatPCS2Lab);
309 
310         // Remove float pcs Lab conversions
311         Opt |= _Remove2Op(Lut, cmsSigXYZ2FloatPCS, cmsSigFloatPCS2XYZ);
312 
313         // Simplify matrix.
314         Opt |= _MultiplyMatrix(Lut);
315 
316         if (Opt) AnyOpt = TRUE;
317 
318     } while (Opt);
319 
320     return AnyOpt;
321 }
322 
323 static
Eval16nop1D(CMSREGISTER const cmsUInt16Number Input[],CMSREGISTER cmsUInt16Number Output[],CMSREGISTER const struct _cms_interp_struc * p)324 void Eval16nop1D(CMSREGISTER const cmsUInt16Number Input[],
325                  CMSREGISTER cmsUInt16Number Output[],
326                  CMSREGISTER const struct _cms_interp_struc* p)
327 {
328     Output[0] = Input[0];
329 
330     cmsUNUSED_PARAMETER(p);
331 }
332 
333 static
PrelinEval16(CMSREGISTER const cmsUInt16Number Input[],CMSREGISTER cmsUInt16Number Output[],CMSREGISTER const void * D)334 void PrelinEval16(CMSREGISTER const cmsUInt16Number Input[],
335                   CMSREGISTER cmsUInt16Number Output[],
336                   CMSREGISTER const void* D)
337 {
338     Prelin16Data* p16 = (Prelin16Data*) D;
339     cmsUInt16Number  StageABC[MAX_INPUT_DIMENSIONS];
340     cmsUInt16Number  StageDEF[cmsMAXCHANNELS];
341     cmsUInt32Number i;
342 
343     for (i=0; i < p16 ->nInputs; i++) {
344 
345         p16 ->EvalCurveIn16[i](&Input[i], &StageABC[i], p16 ->ParamsCurveIn16[i]);
346     }
347 
348     p16 ->EvalCLUT(StageABC, StageDEF, p16 ->CLUTparams);
349 
350     for (i=0; i < p16 ->nOutputs; i++) {
351 
352         p16 ->EvalCurveOut16[i](&StageDEF[i], &Output[i], p16 ->ParamsCurveOut16[i]);
353     }
354 }
355 
356 
357 static
PrelinOpt16free(cmsContext ContextID,void * ptr)358 void PrelinOpt16free(cmsContext ContextID, void* ptr)
359 {
360     Prelin16Data* p16 = (Prelin16Data*) ptr;
361 
362     _cmsFree(ContextID, p16 ->EvalCurveOut16);
363     _cmsFree(ContextID, p16 ->ParamsCurveOut16);
364 
365     _cmsFree(ContextID, p16);
366 }
367 
368 static
Prelin16dup(cmsContext ContextID,const void * ptr)369 void* Prelin16dup(cmsContext ContextID, const void* ptr)
370 {
371     Prelin16Data* p16 = (Prelin16Data*) ptr;
372     Prelin16Data* Duped = (Prelin16Data*) _cmsDupMem(ContextID, p16, sizeof(Prelin16Data));
373 
374     if (Duped == NULL) return NULL;
375 
376     Duped->EvalCurveOut16 = (_cmsInterpFn16*) _cmsDupMem(ContextID, p16->EvalCurveOut16, p16->nOutputs * sizeof(_cmsInterpFn16));
377     Duped->ParamsCurveOut16 = (cmsInterpParams**)_cmsDupMem(ContextID, p16->ParamsCurveOut16, p16->nOutputs * sizeof(cmsInterpParams*));
378 
379     return Duped;
380 }
381 
382 
383 static
PrelinOpt16alloc(cmsContext ContextID,const cmsInterpParams * ColorMap,cmsUInt32Number nInputs,cmsToneCurve ** In,cmsUInt32Number nOutputs,cmsToneCurve ** Out)384 Prelin16Data* PrelinOpt16alloc(cmsContext ContextID,
385                                const cmsInterpParams* ColorMap,
386                                cmsUInt32Number nInputs, cmsToneCurve** In,
387                                cmsUInt32Number nOutputs, cmsToneCurve** Out )
388 {
389     cmsUInt32Number i;
390     Prelin16Data* p16 = (Prelin16Data*)_cmsMallocZero(ContextID, sizeof(Prelin16Data));
391     if (p16 == NULL) return NULL;
392 
393     p16 ->nInputs = nInputs;
394     p16 ->nOutputs = nOutputs;
395 
396 
397     for (i=0; i < nInputs; i++) {
398 
399         if (In == NULL) {
400             p16 -> ParamsCurveIn16[i] = NULL;
401             p16 -> EvalCurveIn16[i] = Eval16nop1D;
402 
403         }
404         else {
405             p16 -> ParamsCurveIn16[i] = In[i] ->InterpParams;
406             p16 -> EvalCurveIn16[i] = p16 ->ParamsCurveIn16[i]->Interpolation.Lerp16;
407         }
408     }
409 
410     p16 ->CLUTparams = ColorMap;
411     p16 ->EvalCLUT   = ColorMap ->Interpolation.Lerp16;
412 
413 
414     p16 -> EvalCurveOut16 = (_cmsInterpFn16*) _cmsCalloc(ContextID, nOutputs, sizeof(_cmsInterpFn16));
415     if (p16->EvalCurveOut16 == NULL)
416     {
417         _cmsFree(ContextID, p16);
418         return NULL;
419     }
420 
421     p16 -> ParamsCurveOut16 = (cmsInterpParams**) _cmsCalloc(ContextID, nOutputs, sizeof(cmsInterpParams* ));
422     if (p16->ParamsCurveOut16 == NULL)
423     {
424 
425         _cmsFree(ContextID, p16->EvalCurveOut16);
426         _cmsFree(ContextID, p16);
427         return NULL;
428     }
429 
430     for (i=0; i < nOutputs; i++) {
431 
432         if (Out == NULL) {
433             p16 ->ParamsCurveOut16[i] = NULL;
434             p16 -> EvalCurveOut16[i] = Eval16nop1D;
435         }
436         else {
437 
438             p16 ->ParamsCurveOut16[i] = Out[i] ->InterpParams;
439             p16 -> EvalCurveOut16[i] = p16 ->ParamsCurveOut16[i]->Interpolation.Lerp16;
440         }
441     }
442 
443     return p16;
444 }
445 
446 
447 
448 // Resampling ---------------------------------------------------------------------------------
449 
450 #define PRELINEARIZATION_POINTS 4096
451 
452 // Sampler implemented by another LUT. This is a clean way to precalculate the devicelink 3D CLUT for
453 // almost any transform. We use floating point precision and then convert from floating point to 16 bits.
454 static
XFormSampler16(CMSREGISTER const cmsUInt16Number In[],CMSREGISTER cmsUInt16Number Out[],CMSREGISTER void * Cargo)455 cmsInt32Number XFormSampler16(CMSREGISTER const cmsUInt16Number In[],
456                               CMSREGISTER cmsUInt16Number Out[],
457                               CMSREGISTER void* Cargo)
458 {
459     cmsPipeline* Lut = (cmsPipeline*) Cargo;
460     cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
461     cmsUInt32Number i;
462 
463     _cmsAssert(Lut -> InputChannels < cmsMAXCHANNELS);
464     _cmsAssert(Lut -> OutputChannels < cmsMAXCHANNELS);
465 
466     // From 16 bit to floating point
467     for (i=0; i < Lut ->InputChannels; i++)
468         InFloat[i] = (cmsFloat32Number) (In[i] / 65535.0);
469 
470     // Evaluate in floating point
471     cmsPipelineEvalFloat(InFloat, OutFloat, Lut);
472 
473     // Back to 16 bits representation
474     for (i=0; i < Lut ->OutputChannels; i++)
475         Out[i] = _cmsQuickSaturateWord(OutFloat[i] * 65535.0);
476 
477     // Always succeed
478     return TRUE;
479 }
480 
481 // Try to see if the curves of a given MPE are linear
482 static
AllCurvesAreLinear(cmsStage * mpe)483 cmsBool AllCurvesAreLinear(cmsStage* mpe)
484 {
485     cmsToneCurve** Curves;
486     cmsUInt32Number i, n;
487 
488     Curves = _cmsStageGetPtrToCurveSet(mpe);
489     if (Curves == NULL) return FALSE;
490 
491     n = cmsStageOutputChannels(mpe);
492 
493     for (i=0; i < n; i++) {
494         if (!cmsIsToneCurveLinear(Curves[i])) return FALSE;
495     }
496 
497     return TRUE;
498 }
499 
500 // This function replaces a specific node placed in "At" by the "Value" numbers. Its purpose
501 // is to fix scum dot on broken profiles/transforms. Works on 1, 3 and 4 channels
502 static
PatchLUT(cmsStage * CLUT,cmsUInt16Number At[],cmsUInt16Number Value[],cmsUInt32Number nChannelsOut,cmsUInt32Number nChannelsIn)503 cmsBool  PatchLUT(cmsStage* CLUT, cmsUInt16Number At[], cmsUInt16Number Value[],
504                   cmsUInt32Number nChannelsOut, cmsUInt32Number nChannelsIn)
505 {
506     _cmsStageCLutData* Grid = (_cmsStageCLutData*) CLUT ->Data;
507     cmsInterpParams* p16  = Grid ->Params;
508     cmsFloat64Number px, py, pz, pw;
509     int        x0, y0, z0, w0;
510     int        i, index;
511 
512     if (CLUT -> Type != cmsSigCLutElemType) {
513         cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) Attempt to PatchLUT on non-lut stage");
514         return FALSE;
515     }
516 
517     if (nChannelsIn == 4) {
518 
519         px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
520         py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
521         pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
522         pw = ((cmsFloat64Number) At[3] * (p16->Domain[3])) / 65535.0;
523 
524         x0 = (int) floor(px);
525         y0 = (int) floor(py);
526         z0 = (int) floor(pz);
527         w0 = (int) floor(pw);
528 
529         if (((px - x0) != 0) ||
530             ((py - y0) != 0) ||
531             ((pz - z0) != 0) ||
532             ((pw - w0) != 0)) return FALSE; // Not on exact node
533 
534         index = (int) p16 -> opta[3] * x0 +
535                 (int) p16 -> opta[2] * y0 +
536                 (int) p16 -> opta[1] * z0 +
537                 (int) p16 -> opta[0] * w0;
538     }
539     else
540         if (nChannelsIn == 3) {
541 
542             px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
543             py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
544             pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
545 
546             x0 = (int) floor(px);
547             y0 = (int) floor(py);
548             z0 = (int) floor(pz);
549 
550             if (((px - x0) != 0) ||
551                 ((py - y0) != 0) ||
552                 ((pz - z0) != 0)) return FALSE;  // Not on exact node
553 
554             index = (int) p16 -> opta[2] * x0 +
555                     (int) p16 -> opta[1] * y0 +
556                     (int) p16 -> opta[0] * z0;
557         }
558         else
559             if (nChannelsIn == 1) {
560 
561                 px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
562 
563                 x0 = (int) floor(px);
564 
565                 if (((px - x0) != 0)) return FALSE; // Not on exact node
566 
567                 index = (int) p16 -> opta[0] * x0;
568             }
569             else {
570                 cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) %d Channels are not supported on PatchLUT", nChannelsIn);
571                 return FALSE;
572             }
573 
574     for (i = 0; i < (int) nChannelsOut; i++)
575         Grid->Tab.T[index + i] = Value[i];
576 
577     return TRUE;
578 }
579 
580 // Auxiliary, to see if two values are equal or very different
581 static
WhitesAreEqual(cmsUInt32Number n,cmsUInt16Number White1[],cmsUInt16Number White2[])582 cmsBool WhitesAreEqual(cmsUInt32Number n, cmsUInt16Number White1[], cmsUInt16Number White2[] )
583 {
584     cmsUInt32Number i;
585 
586     for (i=0; i < n; i++) {
587 
588         if (abs(White1[i] - White2[i]) > 0xf000) return TRUE;  // Values are so extremely different that the fixup should be avoided
589         if (White1[i] != White2[i]) return FALSE;
590     }
591     return TRUE;
592 }
593 
594 
595 // Locate the node for the white point and fix it to pure white in order to avoid scum dot.
596 static
FixWhiteMisalignment(cmsPipeline * Lut,cmsColorSpaceSignature EntryColorSpace,cmsColorSpaceSignature ExitColorSpace)597 cmsBool FixWhiteMisalignment(cmsPipeline* Lut, cmsColorSpaceSignature EntryColorSpace, cmsColorSpaceSignature ExitColorSpace)
598 {
599     cmsUInt16Number *WhitePointIn, *WhitePointOut;
600     cmsUInt16Number  WhiteIn[cmsMAXCHANNELS], WhiteOut[cmsMAXCHANNELS], ObtainedOut[cmsMAXCHANNELS];
601     cmsUInt32Number i, nOuts, nIns;
602     cmsStage *PreLin = NULL, *CLUT = NULL, *PostLin = NULL;
603 
604     if (!_cmsEndPointsBySpace(EntryColorSpace,
605         &WhitePointIn, NULL, &nIns)) return FALSE;
606 
607     if (!_cmsEndPointsBySpace(ExitColorSpace,
608         &WhitePointOut, NULL, &nOuts)) return FALSE;
609 
610     // It needs to be fixed?
611     if (Lut ->InputChannels != nIns) return FALSE;
612     if (Lut ->OutputChannels != nOuts) return FALSE;
613 
614     cmsPipelineEval16(WhitePointIn, ObtainedOut, Lut);
615 
616     if (WhitesAreEqual(nOuts, WhitePointOut, ObtainedOut)) return TRUE; // whites already match
617 
618     // Check if the LUT comes as Prelin, CLUT or Postlin. We allow all combinations
619     if (!cmsPipelineCheckAndRetreiveStages(Lut, 3, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, &PreLin, &CLUT, &PostLin))
620         if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCurveSetElemType, cmsSigCLutElemType, &PreLin, &CLUT))
621             if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCLutElemType, cmsSigCurveSetElemType, &CLUT, &PostLin))
622                 if (!cmsPipelineCheckAndRetreiveStages(Lut, 1, cmsSigCLutElemType, &CLUT))
623                     return FALSE;
624 
625     // We need to interpolate white points of both, pre and post curves
626     if (PreLin) {
627 
628         cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PreLin);
629 
630         for (i=0; i < nIns; i++) {
631             WhiteIn[i] = cmsEvalToneCurve16(Curves[i], WhitePointIn[i]);
632         }
633     }
634     else {
635         for (i=0; i < nIns; i++)
636             WhiteIn[i] = WhitePointIn[i];
637     }
638 
639     // If any post-linearization, we need to find how is represented white before the curve, do
640     // a reverse interpolation in this case.
641     if (PostLin) {
642 
643         cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PostLin);
644 
645         for (i=0; i < nOuts; i++) {
646 
647             cmsToneCurve* InversePostLin = cmsReverseToneCurve(Curves[i]);
648             if (InversePostLin == NULL) {
649                 WhiteOut[i] = WhitePointOut[i];
650 
651             } else {
652 
653                 WhiteOut[i] = cmsEvalToneCurve16(InversePostLin, WhitePointOut[i]);
654                 cmsFreeToneCurve(InversePostLin);
655             }
656         }
657     }
658     else {
659         for (i=0; i < nOuts; i++)
660             WhiteOut[i] = WhitePointOut[i];
661     }
662 
663     // Ok, proceed with patching. May fail and we don't care if it fails
664     PatchLUT(CLUT, WhiteIn, WhiteOut, nOuts, nIns);
665 
666     return TRUE;
667 }
668 
669 // -----------------------------------------------------------------------------------------------------------------------------------------------
670 // This function creates simple LUT from complex ones. The generated LUT has an optional set of
671 // prelinearization curves, a CLUT of nGridPoints and optional postlinearization tables.
672 // These curves have to exist in the original LUT in order to be used in the simplified output.
673 // Caller may also use the flags to allow this feature.
674 // LUTS with all curves will be simplified to a single curve. Parametric curves are lost.
675 // This function should be used on 16-bits LUTS only, as floating point losses precision when simplified
676 // -----------------------------------------------------------------------------------------------------------------------------------------------
677 
678 static
OptimizeByResampling(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)679 cmsBool OptimizeByResampling(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
680 {
681     cmsPipeline* Src = NULL;
682     cmsPipeline* Dest = NULL;
683     cmsStage* CLUT;
684     cmsStage *KeepPreLin = NULL, *KeepPostLin = NULL;
685     cmsUInt32Number nGridPoints;
686     cmsColorSpaceSignature ColorSpace, OutputColorSpace;
687     cmsStage *NewPreLin = NULL;
688     cmsStage *NewPostLin = NULL;
689     _cmsStageCLutData* DataCLUT;
690     cmsToneCurve** DataSetIn;
691     cmsToneCurve** DataSetOut;
692     Prelin16Data* p16;
693 
694     // This is a lossy optimization! does not apply in floating-point cases
695     if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
696 
697     ColorSpace       = _cmsICCcolorSpace((int) T_COLORSPACE(*InputFormat));
698     OutputColorSpace = _cmsICCcolorSpace((int) T_COLORSPACE(*OutputFormat));
699 
700     // Color space must be specified
701     if (ColorSpace == (cmsColorSpaceSignature)0 ||
702         OutputColorSpace == (cmsColorSpaceSignature)0) return FALSE;
703 
704     nGridPoints = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);
705 
706     // For empty LUTs, 2 points are enough
707     if (cmsPipelineStageCount(*Lut) == 0)
708         nGridPoints = 2;
709 
710     Src = *Lut;
711 
712     // Allocate an empty LUT
713     Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
714     if (!Dest) return FALSE;
715 
716     // Prelinearization tables are kept unless indicated by flags
717     if (*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION) {
718 
719         // Get a pointer to the prelinearization element
720         cmsStage* PreLin = cmsPipelineGetPtrToFirstStage(Src);
721 
722         // Check if suitable
723         if (PreLin && PreLin ->Type == cmsSigCurveSetElemType) {
724 
725             // Maybe this is a linear tram, so we can avoid the whole stuff
726             if (!AllCurvesAreLinear(PreLin)) {
727 
728                 // All seems ok, proceed.
729                 NewPreLin = cmsStageDup(PreLin);
730                 if(!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, NewPreLin))
731                     goto Error;
732 
733                 // Remove prelinearization. Since we have duplicated the curve
734                 // in destination LUT, the sampling should be applied after this stage.
735                 cmsPipelineUnlinkStage(Src, cmsAT_BEGIN, &KeepPreLin);
736             }
737         }
738     }
739 
740     // Allocate the CLUT
741     CLUT = cmsStageAllocCLut16bit(Src ->ContextID, nGridPoints, Src ->InputChannels, Src->OutputChannels, NULL);
742     if (CLUT == NULL) goto Error;
743 
744     // Add the CLUT to the destination LUT
745     if (!cmsPipelineInsertStage(Dest, cmsAT_END, CLUT)) {
746         goto Error;
747     }
748 
749     // Postlinearization tables are kept unless indicated by flags
750     if (*dwFlags & cmsFLAGS_CLUT_POST_LINEARIZATION) {
751 
752         // Get a pointer to the postlinearization if present
753         cmsStage* PostLin = cmsPipelineGetPtrToLastStage(Src);
754 
755         // Check if suitable
756         if (PostLin && cmsStageType(PostLin) == cmsSigCurveSetElemType) {
757 
758             // Maybe this is a linear tram, so we can avoid the whole stuff
759             if (!AllCurvesAreLinear(PostLin)) {
760 
761                 // All seems ok, proceed.
762                 NewPostLin = cmsStageDup(PostLin);
763                 if (!cmsPipelineInsertStage(Dest, cmsAT_END, NewPostLin))
764                     goto Error;
765 
766                 // In destination LUT, the sampling should be applied after this stage.
767                 cmsPipelineUnlinkStage(Src, cmsAT_END, &KeepPostLin);
768             }
769         }
770     }
771 
772     // Now its time to do the sampling. We have to ignore pre/post linearization
773     // The source LUT without pre/post curves is passed as parameter.
774     if (!cmsStageSampleCLut16bit(CLUT, XFormSampler16, (void*) Src, 0)) {
775 Error:
776         // Ops, something went wrong, Restore stages
777         if (KeepPreLin != NULL) {
778             if (!cmsPipelineInsertStage(Src, cmsAT_BEGIN, KeepPreLin)) {
779                 _cmsAssert(0); // This never happens
780             }
781         }
782         if (KeepPostLin != NULL) {
783             if (!cmsPipelineInsertStage(Src, cmsAT_END,   KeepPostLin)) {
784                 _cmsAssert(0); // This never happens
785             }
786         }
787         cmsPipelineFree(Dest);
788         return FALSE;
789     }
790 
791     // Done.
792 
793     if (KeepPreLin != NULL) cmsStageFree(KeepPreLin);
794     if (KeepPostLin != NULL) cmsStageFree(KeepPostLin);
795     cmsPipelineFree(Src);
796 
797     DataCLUT = (_cmsStageCLutData*) CLUT ->Data;
798 
799     if (NewPreLin == NULL) DataSetIn = NULL;
800     else DataSetIn = ((_cmsStageToneCurvesData*) NewPreLin ->Data) ->TheCurves;
801 
802     if (NewPostLin == NULL) DataSetOut = NULL;
803     else  DataSetOut = ((_cmsStageToneCurvesData*) NewPostLin ->Data) ->TheCurves;
804 
805 
806     if (DataSetIn == NULL && DataSetOut == NULL) {
807 
808         _cmsPipelineSetOptimizationParameters(Dest, (_cmsPipelineEval16Fn) DataCLUT->Params->Interpolation.Lerp16, DataCLUT->Params, NULL, NULL);
809     }
810     else {
811 
812         p16 = PrelinOpt16alloc(Dest ->ContextID,
813             DataCLUT ->Params,
814             Dest ->InputChannels,
815             DataSetIn,
816             Dest ->OutputChannels,
817             DataSetOut);
818 
819         _cmsPipelineSetOptimizationParameters(Dest, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
820     }
821 
822 
823     // Don't fix white on absolute colorimetric
824     if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
825         *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
826 
827     if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
828 
829         FixWhiteMisalignment(Dest, ColorSpace, OutputColorSpace);
830     }
831 
832     *Lut = Dest;
833     return TRUE;
834 
835     cmsUNUSED_PARAMETER(Intent);
836 }
837 
838 
839 // -----------------------------------------------------------------------------------------------------------------------------------------------
840 // Fixes the gamma balancing of transform. This is described in my paper "Prelinearization Stages on
841 // Color-Management Application-Specific Integrated Circuits (ASICs)" presented at NIP24. It only works
842 // for RGB transforms. See the paper for more details
843 // -----------------------------------------------------------------------------------------------------------------------------------------------
844 
845 
846 // Normalize endpoints by slope limiting max and min. This assures endpoints as well.
847 // Descending curves are handled as well.
848 static
SlopeLimiting(cmsToneCurve * g)849 void SlopeLimiting(cmsToneCurve* g)
850 {
851     int BeginVal, EndVal;
852     int AtBegin = (int) floor((cmsFloat64Number) g ->nEntries * 0.02 + 0.5);   // Cutoff at 2%
853     int AtEnd   = (int) g ->nEntries - AtBegin - 1;                                  // And 98%
854     cmsFloat64Number Val, Slope, beta;
855     int i;
856 
857     if (cmsIsToneCurveDescending(g)) {
858         BeginVal = 0xffff; EndVal = 0;
859     }
860     else {
861         BeginVal = 0; EndVal = 0xffff;
862     }
863 
864     // Compute slope and offset for begin of curve
865     Val   = g ->Table16[AtBegin];
866     Slope = (Val - BeginVal) / AtBegin;
867     beta  = Val - Slope * AtBegin;
868 
869     for (i=0; i < AtBegin; i++)
870         g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
871 
872     // Compute slope and offset for the end
873     Val   = g ->Table16[AtEnd];
874     Slope = (EndVal - Val) / AtBegin;   // AtBegin holds the X interval, which is same in both cases
875     beta  = Val - Slope * AtEnd;
876 
877     for (i = AtEnd; i < (int) g ->nEntries; i++)
878         g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
879 }
880 
881 
882 // Precomputes tables for 8-bit on input devicelink.
883 static
PrelinOpt8alloc(cmsContext ContextID,const cmsInterpParams * p,cmsToneCurve * G[3])884 Prelin8Data* PrelinOpt8alloc(cmsContext ContextID, const cmsInterpParams* p, cmsToneCurve* G[3])
885 {
886     int i;
887     cmsUInt16Number Input[3];
888     cmsS15Fixed16Number v1, v2, v3;
889     Prelin8Data* p8;
890 
891     p8 = (Prelin8Data*)_cmsMallocZero(ContextID, sizeof(Prelin8Data));
892     if (p8 == NULL) return NULL;
893 
894     // Since this only works for 8 bit input, values comes always as x * 257,
895     // we can safely take msb byte (x << 8 + x)
896 
897     for (i=0; i < 256; i++) {
898 
899         if (G != NULL) {
900 
901             // Get 16-bit representation
902             Input[0] = cmsEvalToneCurve16(G[0], FROM_8_TO_16(i));
903             Input[1] = cmsEvalToneCurve16(G[1], FROM_8_TO_16(i));
904             Input[2] = cmsEvalToneCurve16(G[2], FROM_8_TO_16(i));
905         }
906         else {
907             Input[0] = FROM_8_TO_16(i);
908             Input[1] = FROM_8_TO_16(i);
909             Input[2] = FROM_8_TO_16(i);
910         }
911 
912 
913         // Move to 0..1.0 in fixed domain
914         v1 = _cmsToFixedDomain((int) (Input[0] * p -> Domain[0]));
915         v2 = _cmsToFixedDomain((int) (Input[1] * p -> Domain[1]));
916         v3 = _cmsToFixedDomain((int) (Input[2] * p -> Domain[2]));
917 
918         // Store the precalculated table of nodes
919         p8 ->X0[i] = (p->opta[2] * FIXED_TO_INT(v1));
920         p8 ->Y0[i] = (p->opta[1] * FIXED_TO_INT(v2));
921         p8 ->Z0[i] = (p->opta[0] * FIXED_TO_INT(v3));
922 
923         // Store the precalculated table of offsets
924         p8 ->rx[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v1);
925         p8 ->ry[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v2);
926         p8 ->rz[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v3);
927     }
928 
929     p8 ->ContextID = ContextID;
930     p8 ->p = p;
931 
932     return p8;
933 }
934 
935 static
Prelin8free(cmsContext ContextID,void * ptr)936 void Prelin8free(cmsContext ContextID, void* ptr)
937 {
938     _cmsFree(ContextID, ptr);
939 }
940 
941 static
Prelin8dup(cmsContext ContextID,const void * ptr)942 void* Prelin8dup(cmsContext ContextID, const void* ptr)
943 {
944     return _cmsDupMem(ContextID, ptr, sizeof(Prelin8Data));
945 }
946 
947 
948 
949 // A optimized interpolation for 8-bit input.
950 #define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
951 static CMS_NO_SANITIZE
PrelinEval8(CMSREGISTER const cmsUInt16Number Input[],CMSREGISTER cmsUInt16Number Output[],CMSREGISTER const void * D)952 void PrelinEval8(CMSREGISTER const cmsUInt16Number Input[],
953                  CMSREGISTER cmsUInt16Number Output[],
954                  CMSREGISTER const void* D)
955 {
956 
957     cmsUInt8Number         r, g, b;
958     cmsS15Fixed16Number    rx, ry, rz;
959     cmsS15Fixed16Number    c0, c1, c2, c3, Rest;
960     int                    OutChan;
961     CMSREGISTER cmsS15Fixed16Number X0, X1, Y0, Y1, Z0, Z1;
962     Prelin8Data* p8 = (Prelin8Data*) D;
963     CMSREGISTER const cmsInterpParams* p = p8 ->p;
964     int                    TotalOut = (int) p -> nOutputs;
965     const cmsUInt16Number* LutTable = (const cmsUInt16Number*) p->Table;
966 
967     r = (cmsUInt8Number) (Input[0] >> 8);
968     g = (cmsUInt8Number) (Input[1] >> 8);
969     b = (cmsUInt8Number) (Input[2] >> 8);
970 
971     X0 = (cmsS15Fixed16Number) p8->X0[r];
972     Y0 = (cmsS15Fixed16Number) p8->Y0[g];
973     Z0 = (cmsS15Fixed16Number) p8->Z0[b];
974 
975     rx = p8 ->rx[r];
976     ry = p8 ->ry[g];
977     rz = p8 ->rz[b];
978 
979     X1 = X0 + (cmsS15Fixed16Number)((rx == 0) ? 0 :  p ->opta[2]);
980     Y1 = Y0 + (cmsS15Fixed16Number)((ry == 0) ? 0 :  p ->opta[1]);
981     Z1 = Z0 + (cmsS15Fixed16Number)((rz == 0) ? 0 :  p ->opta[0]);
982 
983 
984     // These are the 6 Tetrahedral
985     for (OutChan=0; OutChan < TotalOut; OutChan++) {
986 
987         c0 = DENS(X0, Y0, Z0);
988 
989         if (rx >= ry && ry >= rz)
990         {
991             c1 = DENS(X1, Y0, Z0) - c0;
992             c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
993             c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
994         }
995         else
996             if (rx >= rz && rz >= ry)
997             {
998                 c1 = DENS(X1, Y0, Z0) - c0;
999                 c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
1000                 c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
1001             }
1002             else
1003                 if (rz >= rx && rx >= ry)
1004                 {
1005                     c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
1006                     c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
1007                     c3 = DENS(X0, Y0, Z1) - c0;
1008                 }
1009                 else
1010                     if (ry >= rx && rx >= rz)
1011                     {
1012                         c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
1013                         c2 = DENS(X0, Y1, Z0) - c0;
1014                         c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
1015                     }
1016                     else
1017                         if (ry >= rz && rz >= rx)
1018                         {
1019                             c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
1020                             c2 = DENS(X0, Y1, Z0) - c0;
1021                             c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);
1022                         }
1023                         else
1024                             if (rz >= ry && ry >= rx)
1025                             {
1026                                 c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
1027                                 c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
1028                                 c3 = DENS(X0, Y0, Z1) - c0;
1029                             }
1030                             else  {
1031                                 c1 = c2 = c3 = 0;
1032                             }
1033 
1034         Rest = c1 * rx + c2 * ry + c3 * rz + 0x8001;
1035         Output[OutChan] = (cmsUInt16Number) (c0 + ((Rest + (Rest >> 16)) >> 16));
1036 
1037     }
1038 }
1039 
1040 #undef DENS
1041 
1042 
1043 // Curves that contain wide empty areas are not optimizeable
1044 static
IsDegenerated(const cmsToneCurve * g)1045 cmsBool IsDegenerated(const cmsToneCurve* g)
1046 {
1047     cmsUInt32Number i, Zeros = 0, Poles = 0;
1048     cmsUInt32Number nEntries = g ->nEntries;
1049 
1050     for (i=0; i < nEntries; i++) {
1051 
1052         if (g ->Table16[i] == 0x0000) Zeros++;
1053         if (g ->Table16[i] == 0xffff) Poles++;
1054     }
1055 
1056     if (Zeros == 1 && Poles == 1) return FALSE;  // For linear tables
1057     if (Zeros > (nEntries / 20)) return TRUE;  // Degenerated, many zeros
1058     if (Poles > (nEntries / 20)) return TRUE;  // Degenerated, many poles
1059 
1060     return FALSE;
1061 }
1062 
1063 // --------------------------------------------------------------------------------------------------------------
1064 // We need xput over here
1065 
1066 static
OptimizeByComputingLinearization(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1067 cmsBool OptimizeByComputingLinearization(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1068 {
1069     cmsPipeline* OriginalLut;
1070     cmsUInt32Number nGridPoints;
1071     cmsToneCurve *Trans[cmsMAXCHANNELS], *TransReverse[cmsMAXCHANNELS];
1072     cmsUInt32Number t, i;
1073     cmsFloat32Number v, In[cmsMAXCHANNELS], Out[cmsMAXCHANNELS];
1074     cmsBool lIsSuitable, lIsLinear;
1075     cmsPipeline* OptimizedLUT = NULL, *LutPlusCurves = NULL;
1076     cmsStage* OptimizedCLUTmpe;
1077     cmsColorSpaceSignature ColorSpace, OutputColorSpace;
1078     cmsStage* OptimizedPrelinMpe;
1079     cmsToneCurve** OptimizedPrelinCurves;
1080     _cmsStageCLutData* OptimizedPrelinCLUT;
1081 
1082 
1083     // This is a lossy optimization! does not apply in floating-point cases
1084     if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
1085 
1086     // Only on chunky RGB
1087     if (T_COLORSPACE(*InputFormat)  != PT_RGB) return FALSE;
1088     if (T_PLANAR(*InputFormat)) return FALSE;
1089 
1090     if (T_COLORSPACE(*OutputFormat) != PT_RGB) return FALSE;
1091     if (T_PLANAR(*OutputFormat)) return FALSE;
1092 
1093     // On 16 bits, user has to specify the feature
1094     if (!_cmsFormatterIs8bit(*InputFormat)) {
1095         if (!(*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION)) return FALSE;
1096     }
1097 
1098     OriginalLut = *Lut;
1099 
1100     ColorSpace       = _cmsICCcolorSpace((int) T_COLORSPACE(*InputFormat));
1101     OutputColorSpace = _cmsICCcolorSpace((int) T_COLORSPACE(*OutputFormat));
1102 
1103     // Color space must be specified
1104     if (ColorSpace == (cmsColorSpaceSignature)0 ||
1105         OutputColorSpace == (cmsColorSpaceSignature)0) return FALSE;
1106 
1107     nGridPoints      = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);
1108 
1109     // Empty gamma containers
1110     memset(Trans, 0, sizeof(Trans));
1111     memset(TransReverse, 0, sizeof(TransReverse));
1112 
1113     // If the last stage of the original lut are curves, and those curves are
1114     // degenerated, it is likely the transform is squeezing and clipping
1115     // the output from previous CLUT. We cannot optimize this case
1116     {
1117         cmsStage* last = cmsPipelineGetPtrToLastStage(OriginalLut);
1118 
1119         if (last == NULL) goto Error;
1120         if (cmsStageType(last) == cmsSigCurveSetElemType) {
1121 
1122             _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*)cmsStageData(last);
1123             for (i = 0; i < Data->nCurves; i++) {
1124                 if (IsDegenerated(Data->TheCurves[i]))
1125                     goto Error;
1126             }
1127         }
1128     }
1129 
1130     for (t = 0; t < OriginalLut ->InputChannels; t++) {
1131         Trans[t] = cmsBuildTabulatedToneCurve16(OriginalLut ->ContextID, PRELINEARIZATION_POINTS, NULL);
1132         if (Trans[t] == NULL) goto Error;
1133     }
1134 
1135     // Populate the curves
1136     for (i=0; i < PRELINEARIZATION_POINTS; i++) {
1137 
1138         v = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
1139 
1140         // Feed input with a gray ramp
1141         for (t=0; t < OriginalLut ->InputChannels; t++)
1142             In[t] = v;
1143 
1144         // Evaluate the gray value
1145         cmsPipelineEvalFloat(In, Out, OriginalLut);
1146 
1147         // Store result in curve
1148         for (t=0; t < OriginalLut ->InputChannels; t++)
1149             Trans[t] ->Table16[i] = _cmsQuickSaturateWord(Out[t] * 65535.0);
1150     }
1151 
1152     // Slope-limit the obtained curves
1153     for (t = 0; t < OriginalLut ->InputChannels; t++)
1154         SlopeLimiting(Trans[t]);
1155 
1156     // Check for validity
1157     lIsSuitable = TRUE;
1158     lIsLinear   = TRUE;
1159     for (t=0; (lIsSuitable && (t < OriginalLut ->InputChannels)); t++) {
1160 
1161         // Exclude if already linear
1162         if (!cmsIsToneCurveLinear(Trans[t]))
1163             lIsLinear = FALSE;
1164 
1165         // Exclude if non-monotonic
1166         if (!cmsIsToneCurveMonotonic(Trans[t]))
1167             lIsSuitable = FALSE;
1168 
1169         if (IsDegenerated(Trans[t]))
1170             lIsSuitable = FALSE;
1171     }
1172 
1173     // If it is not suitable, just quit
1174     if (!lIsSuitable) goto Error;
1175 
1176     // Invert curves if possible
1177     for (t = 0; t < OriginalLut ->InputChannels; t++) {
1178         TransReverse[t] = cmsReverseToneCurveEx(PRELINEARIZATION_POINTS, Trans[t]);
1179         if (TransReverse[t] == NULL) goto Error;
1180     }
1181 
1182     // Now inset the reversed curves at the begin of transform
1183     LutPlusCurves = cmsPipelineDup(OriginalLut);
1184     if (LutPlusCurves == NULL) goto Error;
1185 
1186     if (!cmsPipelineInsertStage(LutPlusCurves, cmsAT_BEGIN, cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, TransReverse)))
1187         goto Error;
1188 
1189     // Create the result LUT
1190     OptimizedLUT = cmsPipelineAlloc(OriginalLut ->ContextID, OriginalLut ->InputChannels, OriginalLut ->OutputChannels);
1191     if (OptimizedLUT == NULL) goto Error;
1192 
1193     OptimizedPrelinMpe = cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, Trans);
1194 
1195     // Create and insert the curves at the beginning
1196     if (!cmsPipelineInsertStage(OptimizedLUT, cmsAT_BEGIN, OptimizedPrelinMpe))
1197         goto Error;
1198 
1199     // Allocate the CLUT for result
1200     OptimizedCLUTmpe = cmsStageAllocCLut16bit(OriginalLut ->ContextID, nGridPoints, OriginalLut ->InputChannels, OriginalLut ->OutputChannels, NULL);
1201 
1202     // Add the CLUT to the destination LUT
1203     if (!cmsPipelineInsertStage(OptimizedLUT, cmsAT_END, OptimizedCLUTmpe))
1204         goto Error;
1205 
1206     // Resample the LUT
1207     if (!cmsStageSampleCLut16bit(OptimizedCLUTmpe, XFormSampler16, (void*) LutPlusCurves, 0)) goto Error;
1208 
1209     // Free resources
1210     for (t = 0; t < OriginalLut ->InputChannels; t++) {
1211 
1212         if (Trans[t]) cmsFreeToneCurve(Trans[t]);
1213         if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
1214     }
1215 
1216     cmsPipelineFree(LutPlusCurves);
1217 
1218 
1219     OptimizedPrelinCurves = _cmsStageGetPtrToCurveSet(OptimizedPrelinMpe);
1220     OptimizedPrelinCLUT   = (_cmsStageCLutData*) OptimizedCLUTmpe ->Data;
1221 
1222     // Set the evaluator if 8-bit
1223     if (_cmsFormatterIs8bit(*InputFormat)) {
1224 
1225         Prelin8Data* p8 = PrelinOpt8alloc(OptimizedLUT ->ContextID,
1226                                                 OptimizedPrelinCLUT ->Params,
1227                                                 OptimizedPrelinCurves);
1228         if (p8 == NULL) return FALSE;
1229 
1230         _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval8, (void*) p8, Prelin8free, Prelin8dup);
1231 
1232     }
1233     else
1234     {
1235         Prelin16Data* p16 = PrelinOpt16alloc(OptimizedLUT ->ContextID,
1236             OptimizedPrelinCLUT ->Params,
1237             3, OptimizedPrelinCurves, 3, NULL);
1238         if (p16 == NULL) return FALSE;
1239 
1240         _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
1241 
1242     }
1243 
1244     // Don't fix white on absolute colorimetric
1245     if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
1246         *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;
1247 
1248     if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {
1249 
1250         if (!FixWhiteMisalignment(OptimizedLUT, ColorSpace, OutputColorSpace)) {
1251 
1252             return FALSE;
1253         }
1254     }
1255 
1256     // And return the obtained LUT
1257 
1258     cmsPipelineFree(OriginalLut);
1259     *Lut = OptimizedLUT;
1260     return TRUE;
1261 
1262 Error:
1263 
1264     for (t = 0; t < OriginalLut ->InputChannels; t++) {
1265 
1266         if (Trans[t]) cmsFreeToneCurve(Trans[t]);
1267         if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
1268     }
1269 
1270     if (LutPlusCurves != NULL) cmsPipelineFree(LutPlusCurves);
1271     if (OptimizedLUT != NULL) cmsPipelineFree(OptimizedLUT);
1272 
1273     return FALSE;
1274 
1275     cmsUNUSED_PARAMETER(Intent);
1276     cmsUNUSED_PARAMETER(lIsLinear);
1277 }
1278 
1279 
1280 // Curves optimizer ------------------------------------------------------------------------------------------------------------------
1281 
1282 static
CurvesFree(cmsContext ContextID,void * ptr)1283 void CurvesFree(cmsContext ContextID, void* ptr)
1284 {
1285      Curves16Data* Data = (Curves16Data*) ptr;
1286      cmsUInt32Number i;
1287 
1288      for (i=0; i < Data -> nCurves; i++) {
1289 
1290          _cmsFree(ContextID, Data ->Curves[i]);
1291      }
1292 
1293      _cmsFree(ContextID, Data ->Curves);
1294      _cmsFree(ContextID, ptr);
1295 }
1296 
1297 static
CurvesDup(cmsContext ContextID,const void * ptr)1298 void* CurvesDup(cmsContext ContextID, const void* ptr)
1299 {
1300     Curves16Data* Data = (Curves16Data*)_cmsDupMem(ContextID, ptr, sizeof(Curves16Data));
1301     cmsUInt32Number i;
1302 
1303     if (Data == NULL) return NULL;
1304 
1305     Data->Curves = (cmsUInt16Number**) _cmsDupMem(ContextID, Data->Curves, Data->nCurves * sizeof(cmsUInt16Number*));
1306 
1307     for (i=0; i < Data -> nCurves; i++) {
1308         Data->Curves[i] = (cmsUInt16Number*) _cmsDupMem(ContextID, Data->Curves[i], Data->nElements * sizeof(cmsUInt16Number));
1309     }
1310 
1311     return (void*) Data;
1312 }
1313 
1314 // Precomputes tables for 8-bit on input devicelink.
1315 static
CurvesAlloc(cmsContext ContextID,cmsUInt32Number nCurves,cmsUInt32Number nElements,cmsToneCurve ** G)1316 Curves16Data* CurvesAlloc(cmsContext ContextID, cmsUInt32Number nCurves, cmsUInt32Number nElements, cmsToneCurve** G)
1317 {
1318     cmsUInt32Number i, j;
1319     Curves16Data* c16;
1320 
1321     c16 = (Curves16Data*)_cmsMallocZero(ContextID, sizeof(Curves16Data));
1322     if (c16 == NULL) return NULL;
1323 
1324     c16 ->nCurves = nCurves;
1325     c16 ->nElements = nElements;
1326 
1327     c16->Curves = (cmsUInt16Number**) _cmsCalloc(ContextID, nCurves, sizeof(cmsUInt16Number*));
1328     if (c16->Curves == NULL) {
1329         _cmsFree(ContextID, c16);
1330         return NULL;
1331     }
1332 
1333     for (i=0; i < nCurves; i++) {
1334 
1335         c16->Curves[i] = (cmsUInt16Number*) _cmsCalloc(ContextID, nElements, sizeof(cmsUInt16Number));
1336 
1337         if (c16->Curves[i] == NULL) {
1338 
1339             for (j=0; j < i; j++) {
1340                 _cmsFree(ContextID, c16->Curves[j]);
1341             }
1342             _cmsFree(ContextID, c16->Curves);
1343             _cmsFree(ContextID, c16);
1344             return NULL;
1345         }
1346 
1347         if (nElements == 256U) {
1348 
1349             for (j=0; j < nElements; j++) {
1350 
1351                 c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], FROM_8_TO_16(j));
1352             }
1353         }
1354         else {
1355 
1356             for (j=0; j < nElements; j++) {
1357                 c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], (cmsUInt16Number) j);
1358             }
1359         }
1360     }
1361 
1362     return c16;
1363 }
1364 
1365 static
FastEvaluateCurves8(CMSREGISTER const cmsUInt16Number In[],CMSREGISTER cmsUInt16Number Out[],CMSREGISTER const void * D)1366 void FastEvaluateCurves8(CMSREGISTER const cmsUInt16Number In[],
1367                          CMSREGISTER cmsUInt16Number Out[],
1368                          CMSREGISTER const void* D)
1369 {
1370     Curves16Data* Data = (Curves16Data*) D;
1371     int x;
1372     cmsUInt32Number i;
1373 
1374     for (i=0; i < Data ->nCurves; i++) {
1375 
1376          x = (In[i] >> 8);
1377          Out[i] = Data -> Curves[i][x];
1378     }
1379 }
1380 
1381 
1382 static
FastEvaluateCurves16(CMSREGISTER const cmsUInt16Number In[],CMSREGISTER cmsUInt16Number Out[],CMSREGISTER const void * D)1383 void FastEvaluateCurves16(CMSREGISTER const cmsUInt16Number In[],
1384                           CMSREGISTER cmsUInt16Number Out[],
1385                           CMSREGISTER const void* D)
1386 {
1387     Curves16Data* Data = (Curves16Data*) D;
1388     cmsUInt32Number i;
1389 
1390     for (i=0; i < Data ->nCurves; i++) {
1391          Out[i] = Data -> Curves[i][In[i]];
1392     }
1393 }
1394 
1395 
1396 static
FastIdentity16(CMSREGISTER const cmsUInt16Number In[],CMSREGISTER cmsUInt16Number Out[],CMSREGISTER const void * D)1397 void FastIdentity16(CMSREGISTER const cmsUInt16Number In[],
1398                     CMSREGISTER cmsUInt16Number Out[],
1399                     CMSREGISTER const void* D)
1400 {
1401     cmsPipeline* Lut = (cmsPipeline*) D;
1402     cmsUInt32Number i;
1403 
1404     for (i=0; i < Lut ->InputChannels; i++) {
1405          Out[i] = In[i];
1406     }
1407 }
1408 
1409 
1410 // If the target LUT holds only curves, the optimization procedure is to join all those
1411 // curves together. That only works on curves and does not work on matrices.
1412 static
OptimizeByJoiningCurves(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1413 cmsBool OptimizeByJoiningCurves(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1414 {
1415     cmsToneCurve** GammaTables = NULL;
1416     cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
1417     cmsUInt32Number i, j;
1418     cmsPipeline* Src = *Lut;
1419     cmsPipeline* Dest = NULL;
1420     cmsStage* mpe;
1421     cmsStage* ObtainedCurves = NULL;
1422 
1423 
1424     // This is a lossy optimization! does not apply in floating-point cases
1425     if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;
1426 
1427     //  Only curves in this LUT?
1428     for (mpe = cmsPipelineGetPtrToFirstStage(Src);
1429          mpe != NULL;
1430          mpe = cmsStageNext(mpe)) {
1431             if (cmsStageType(mpe) != cmsSigCurveSetElemType) return FALSE;
1432     }
1433 
1434     // Allocate an empty LUT
1435     Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
1436     if (Dest == NULL) return FALSE;
1437 
1438     // Create target curves
1439     GammaTables = (cmsToneCurve**) _cmsCalloc(Src ->ContextID, Src ->InputChannels, sizeof(cmsToneCurve*));
1440     if (GammaTables == NULL) goto Error;
1441 
1442     for (i=0; i < Src ->InputChannels; i++) {
1443         GammaTables[i] = cmsBuildTabulatedToneCurve16(Src ->ContextID, PRELINEARIZATION_POINTS, NULL);
1444         if (GammaTables[i] == NULL) goto Error;
1445     }
1446 
1447     // Compute 16 bit result by using floating point
1448     for (i=0; i < PRELINEARIZATION_POINTS; i++) {
1449 
1450         for (j=0; j < Src ->InputChannels; j++)
1451             InFloat[j] = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));
1452 
1453         cmsPipelineEvalFloat(InFloat, OutFloat, Src);
1454 
1455         for (j=0; j < Src ->InputChannels; j++)
1456             GammaTables[j] -> Table16[i] = _cmsQuickSaturateWord(OutFloat[j] * 65535.0);
1457     }
1458 
1459     ObtainedCurves = cmsStageAllocToneCurves(Src ->ContextID, Src ->InputChannels, GammaTables);
1460     if (ObtainedCurves == NULL) goto Error;
1461 
1462     for (i=0; i < Src ->InputChannels; i++) {
1463         cmsFreeToneCurve(GammaTables[i]);
1464         GammaTables[i] = NULL;
1465     }
1466 
1467     if (GammaTables != NULL) {
1468         _cmsFree(Src->ContextID, GammaTables);
1469         GammaTables = NULL;
1470     }
1471 
1472     // Maybe the curves are linear at the end
1473     if (!AllCurvesAreLinear(ObtainedCurves)) {
1474        _cmsStageToneCurvesData* Data;
1475 
1476         if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, ObtainedCurves))
1477             goto Error;
1478         Data = (_cmsStageToneCurvesData*) cmsStageData(ObtainedCurves);
1479         ObtainedCurves = NULL;
1480 
1481         // If the curves are to be applied in 8 bits, we can save memory
1482         if (_cmsFormatterIs8bit(*InputFormat)) {
1483              Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 256, Data ->TheCurves);
1484 
1485              if (c16 == NULL) goto Error;
1486              *dwFlags |= cmsFLAGS_NOCACHE;
1487             _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves8, c16, CurvesFree, CurvesDup);
1488 
1489         }
1490         else {
1491              Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 65536, Data ->TheCurves);
1492 
1493              if (c16 == NULL) goto Error;
1494              *dwFlags |= cmsFLAGS_NOCACHE;
1495             _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves16, c16, CurvesFree, CurvesDup);
1496         }
1497     }
1498     else {
1499 
1500         // LUT optimizes to nothing. Set the identity LUT
1501         cmsStageFree(ObtainedCurves);
1502         ObtainedCurves = NULL;
1503 
1504         if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageAllocIdentity(Dest ->ContextID, Src ->InputChannels)))
1505             goto Error;
1506 
1507         *dwFlags |= cmsFLAGS_NOCACHE;
1508         _cmsPipelineSetOptimizationParameters(Dest, FastIdentity16, (void*) Dest, NULL, NULL);
1509     }
1510 
1511     // We are done.
1512     cmsPipelineFree(Src);
1513     *Lut = Dest;
1514     return TRUE;
1515 
1516 Error:
1517 
1518     if (ObtainedCurves != NULL) cmsStageFree(ObtainedCurves);
1519     if (GammaTables != NULL) {
1520         for (i=0; i < Src ->InputChannels; i++) {
1521             if (GammaTables[i] != NULL) cmsFreeToneCurve(GammaTables[i]);
1522         }
1523 
1524         _cmsFree(Src ->ContextID, GammaTables);
1525     }
1526 
1527     if (Dest != NULL) cmsPipelineFree(Dest);
1528     return FALSE;
1529 
1530     cmsUNUSED_PARAMETER(Intent);
1531     cmsUNUSED_PARAMETER(InputFormat);
1532     cmsUNUSED_PARAMETER(OutputFormat);
1533     cmsUNUSED_PARAMETER(dwFlags);
1534 }
1535 
1536 // -------------------------------------------------------------------------------------------------------------------------------------
1537 // LUT is Shaper - Matrix - Matrix - Shaper, which is very frequent when combining two matrix-shaper profiles
1538 
1539 
1540 static
FreeMatShaper(cmsContext ContextID,void * Data)1541 void  FreeMatShaper(cmsContext ContextID, void* Data)
1542 {
1543     if (Data != NULL) _cmsFree(ContextID, Data);
1544 }
1545 
1546 static
DupMatShaper(cmsContext ContextID,const void * Data)1547 void* DupMatShaper(cmsContext ContextID, const void* Data)
1548 {
1549     return _cmsDupMem(ContextID, Data, sizeof(MatShaper8Data));
1550 }
1551 
1552 
1553 // A fast matrix-shaper evaluator for 8 bits. This is a bit tricky since I'm using 1.14 signed fixed point
1554 // to accomplish some performance. Actually it takes 256x3 16 bits tables and 16385 x 3 tables of 8 bits,
1555 // in total about 50K, and the performance boost is huge!
1556 static CMS_NO_SANITIZE
MatShaperEval16(CMSREGISTER const cmsUInt16Number In[],CMSREGISTER cmsUInt16Number Out[],CMSREGISTER const void * D)1557 void MatShaperEval16(CMSREGISTER const cmsUInt16Number In[],
1558                      CMSREGISTER cmsUInt16Number Out[],
1559                      CMSREGISTER const void* D)
1560 {
1561     MatShaper8Data* p = (MatShaper8Data*) D;
1562     cmsS1Fixed14Number r, g, b;
1563     cmsInt64Number l1, l2, l3;
1564     cmsUInt32Number ri, gi, bi;
1565 
1566     // In this case (and only in this case!) we can use this simplification since
1567     // In[] is assured to come from a 8 bit number. (a << 8 | a)
1568     ri = In[0] & 0xFFU;
1569     gi = In[1] & 0xFFU;
1570     bi = In[2] & 0xFFU;
1571 
1572     // Across first shaper, which also converts to 1.14 fixed point
1573     r = _FixedClamp(p->Shaper1R[ri]);
1574     g = _FixedClamp(p->Shaper1G[gi]);
1575     b = _FixedClamp(p->Shaper1B[bi]);
1576 
1577     // Evaluate the matrix in 1.14 fixed point
1578     l1 = _MatShaperEvaluateRow(p->Mat[0], p->Off[0], r, g, b);
1579     l2 = _MatShaperEvaluateRow(p->Mat[1], p->Off[1], r, g, b);
1580     l3 = _MatShaperEvaluateRow(p->Mat[2], p->Off[2], r, g, b);
1581 
1582     // Now we have to clip to 0..1.0 range
1583     ri = (l1 < 0) ? 0 : ((l1 > 16384) ? 16384U : (cmsUInt32Number) l1);
1584     gi = (l2 < 0) ? 0 : ((l2 > 16384) ? 16384U : (cmsUInt32Number) l2);
1585     bi = (l3 < 0) ? 0 : ((l3 > 16384) ? 16384U : (cmsUInt32Number) l3);
1586 
1587     // And across second shaper,
1588     Out[0] = p->Shaper2R[ri];
1589     Out[1] = p->Shaper2G[gi];
1590     Out[2] = p->Shaper2B[bi];
1591 
1592 }
1593 
1594 // This table converts from 8 bits to 1.14 after applying the curve
1595 static
FillFirstShaper(cmsS1Fixed14Number * Table,cmsToneCurve * Curve)1596 void FillFirstShaper(cmsS1Fixed14Number* Table, cmsToneCurve* Curve)
1597 {
1598     int i;
1599     cmsFloat32Number R, y;
1600 
1601     for (i=0; i < 256; i++) {
1602 
1603         R   = (cmsFloat32Number) (i / 255.0);
1604         y   = cmsEvalToneCurveFloat(Curve, R);
1605 
1606         if (y < 131072.0)
1607             Table[i] = DOUBLE_TO_1FIXED14(y);
1608         else
1609             Table[i] = 0x7fffffff;
1610     }
1611 }
1612 
1613 // This table converts form 1.14 (being 0x4000 the last entry) to 8 bits after applying the curve
1614 static
FillSecondShaper(cmsUInt16Number * Table,cmsToneCurve * Curve,cmsBool Is8BitsOutput)1615 void FillSecondShaper(cmsUInt16Number* Table, cmsToneCurve* Curve, cmsBool Is8BitsOutput)
1616 {
1617     int i;
1618     cmsFloat32Number R, Val;
1619 
1620     for (i=0; i < 16385; i++) {
1621 
1622         R   = (cmsFloat32Number) (i / 16384.0);
1623         Val = cmsEvalToneCurveFloat(Curve, R);    // Val comes 0..1.0
1624 
1625         if (Val < 0)
1626             Val = 0;
1627 
1628         if (Val > 1.0)
1629             Val = 1.0;
1630 
1631         if (Is8BitsOutput) {
1632 
1633             // If 8 bits output, we can optimize further by computing the / 257 part.
1634             // first we compute the resulting byte and then we store the byte times
1635             // 257. This quantization allows to round very quick by doing a >> 8, but
1636             // since the low byte is always equal to msb, we can do a & 0xff and this works!
1637             cmsUInt16Number w = _cmsQuickSaturateWord(Val * 65535.0);
1638             cmsUInt8Number  b = FROM_16_TO_8(w);
1639 
1640             Table[i] = FROM_8_TO_16(b);
1641         }
1642         else Table[i]  = _cmsQuickSaturateWord(Val * 65535.0);
1643     }
1644 }
1645 
1646 // Compute the matrix-shaper structure
1647 static
SetMatShaper(cmsPipeline * Dest,cmsToneCurve * Curve1[3],cmsMAT3 * Mat,cmsVEC3 * Off,cmsToneCurve * Curve2[3],cmsUInt32Number * OutputFormat)1648 cmsBool SetMatShaper(cmsPipeline* Dest, cmsToneCurve* Curve1[3], cmsMAT3* Mat, cmsVEC3* Off, cmsToneCurve* Curve2[3], cmsUInt32Number* OutputFormat)
1649 {
1650     MatShaper8Data* p;
1651     int i, j;
1652     cmsBool Is8Bits = _cmsFormatterIs8bit(*OutputFormat);
1653 
1654     // Allocate a big chuck of memory to store precomputed tables
1655     p = (MatShaper8Data*) _cmsMalloc(Dest ->ContextID, sizeof(MatShaper8Data));
1656     if (p == NULL) return FALSE;
1657 
1658     p -> ContextID = Dest -> ContextID;
1659 
1660     // Precompute tables
1661     FillFirstShaper(p ->Shaper1R, Curve1[0]);
1662     FillFirstShaper(p ->Shaper1G, Curve1[1]);
1663     FillFirstShaper(p ->Shaper1B, Curve1[2]);
1664 
1665     FillSecondShaper(p ->Shaper2R, Curve2[0], Is8Bits);
1666     FillSecondShaper(p ->Shaper2G, Curve2[1], Is8Bits);
1667     FillSecondShaper(p ->Shaper2B, Curve2[2], Is8Bits);
1668 
1669     // Convert matrix to nFixed14. Note that those values may take more than 16 bits
1670     for (i=0; i < 3; i++) {
1671         for (j=0; j < 3; j++) {
1672             p ->Mat[i][j] = DOUBLE_TO_1FIXED14(Mat->v[i].n[j]);
1673         }
1674     }
1675 
1676     for (i=0; i < 3; i++) {
1677 
1678         if (Off == NULL) {
1679             p ->Off[i] = 0;
1680         }
1681         else {
1682             p ->Off[i] = DOUBLE_TO_1FIXED14(Off->n[i]);
1683         }
1684     }
1685 
1686     // Mark as optimized for faster formatter
1687     if (Is8Bits)
1688         *OutputFormat |= OPTIMIZED_SH(1);
1689 
1690     // Fill function pointers
1691     _cmsPipelineSetOptimizationParameters(Dest, MatShaperEval16, (void*) p, FreeMatShaper, DupMatShaper);
1692     return TRUE;
1693 }
1694 
1695 //  8 bits on input allows matrix-shaper boot up to 25 Mpixels per second on RGB. That's fast!
1696 static
OptimizeMatrixShaper(cmsPipeline ** Lut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1697 cmsBool OptimizeMatrixShaper(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
1698 {
1699        cmsStage* Curve1, *Curve2;
1700        cmsStage* Matrix1, *Matrix2;
1701        cmsMAT3 res;
1702        cmsBool IdentityMat;
1703        cmsPipeline* Dest, *Src;
1704        cmsFloat64Number* Offset;
1705 
1706        // Only works on RGB to RGB
1707        if (T_CHANNELS(*InputFormat) != 3 || T_CHANNELS(*OutputFormat) != 3) return FALSE;
1708 
1709        // Only works on 8 bit input
1710        if (!_cmsFormatterIs8bit(*InputFormat)) return FALSE;
1711 
1712        // Seems suitable, proceed
1713        Src = *Lut;
1714 
1715        // Check for:
1716        //
1717        //    shaper-matrix-matrix-shaper
1718        //    shaper-matrix-shaper
1719        //
1720        // Both of those constructs are possible (first because abs. colorimetric).
1721        // additionally, In the first case, the input matrix offset should be zero.
1722 
1723        IdentityMat = FALSE;
1724        if (cmsPipelineCheckAndRetreiveStages(Src, 4,
1725               cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
1726               &Curve1, &Matrix1, &Matrix2, &Curve2)) {
1727 
1728               // Get both matrices
1729               _cmsStageMatrixData* Data1 = (_cmsStageMatrixData*)cmsStageData(Matrix1);
1730               _cmsStageMatrixData* Data2 = (_cmsStageMatrixData*)cmsStageData(Matrix2);
1731 
1732               // Only RGB to RGB
1733               if (Matrix1->InputChannels != 3 || Matrix1->OutputChannels != 3 ||
1734                   Matrix2->InputChannels != 3 || Matrix2->OutputChannels != 3) return FALSE;
1735 
1736               // Input offset should be zero
1737               if (Data1->Offset != NULL) return FALSE;
1738 
1739               // Multiply both matrices to get the result
1740               _cmsMAT3per(&res, (cmsMAT3*)Data2->Double, (cmsMAT3*)Data1->Double);
1741 
1742               // Only 2nd matrix has offset, or it is zero
1743               Offset = Data2->Offset;
1744 
1745               // Now the result is in res + Data2 -> Offset. Maybe is a plain identity?
1746               if (_cmsMAT3isIdentity(&res) && Offset == NULL) {
1747 
1748                      // We can get rid of full matrix
1749                      IdentityMat = TRUE;
1750               }
1751 
1752        }
1753        else {
1754 
1755               if (cmsPipelineCheckAndRetreiveStages(Src, 3,
1756                      cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
1757                      &Curve1, &Matrix1, &Curve2)) {
1758 
1759                      _cmsStageMatrixData* Data = (_cmsStageMatrixData*)cmsStageData(Matrix1);
1760 
1761                      // Copy the matrix to our result
1762                      memcpy(&res, Data->Double, sizeof(res));
1763 
1764                      // Preserve the Odffset (may be NULL as a zero offset)
1765                      Offset = Data->Offset;
1766 
1767                      if (_cmsMAT3isIdentity(&res) && Offset == NULL) {
1768 
1769                             // We can get rid of full matrix
1770                             IdentityMat = TRUE;
1771                      }
1772               }
1773               else
1774                      return FALSE; // Not optimizeable this time
1775 
1776        }
1777 
1778       // Allocate an empty LUT
1779     Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
1780     if (!Dest) return FALSE;
1781 
1782     // Assamble the new LUT
1783     if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageDup(Curve1)))
1784         goto Error;
1785 
1786     if (!IdentityMat) {
1787 
1788            if (!cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageAllocMatrix(Dest->ContextID, 3, 3, (const cmsFloat64Number*)&res, Offset)))
1789                   goto Error;
1790     }
1791 
1792     if (!cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageDup(Curve2)))
1793         goto Error;
1794 
1795     // If identity on matrix, we can further optimize the curves, so call the join curves routine
1796     if (IdentityMat) {
1797 
1798         OptimizeByJoiningCurves(&Dest, Intent, InputFormat, OutputFormat, dwFlags);
1799     }
1800     else {
1801         _cmsStageToneCurvesData* mpeC1 = (_cmsStageToneCurvesData*) cmsStageData(Curve1);
1802         _cmsStageToneCurvesData* mpeC2 = (_cmsStageToneCurvesData*) cmsStageData(Curve2);
1803 
1804         // In this particular optimization, cache does not help as it takes more time to deal with
1805         // the cache that with the pixel handling
1806         *dwFlags |= cmsFLAGS_NOCACHE;
1807 
1808         // Setup the optimizarion routines
1809         SetMatShaper(Dest, mpeC1 ->TheCurves, &res, (cmsVEC3*) Offset, mpeC2->TheCurves, OutputFormat);
1810     }
1811 
1812     cmsPipelineFree(Src);
1813     *Lut = Dest;
1814     return TRUE;
1815 Error:
1816     // Leave Src unchanged
1817     cmsPipelineFree(Dest);
1818     return FALSE;
1819 }
1820 
1821 
1822 // -------------------------------------------------------------------------------------------------------------------------------------
1823 // Optimization plug-ins
1824 
1825 // List of optimizations
1826 typedef struct _cmsOptimizationCollection_st {
1827 
1828     _cmsOPToptimizeFn  OptimizePtr;
1829 
1830     struct _cmsOptimizationCollection_st *Next;
1831 
1832 } _cmsOptimizationCollection;
1833 
1834 
1835 // The built-in list. We currently implement 4 types of optimizations. Joining of curves, matrix-shaper, linearization and resampling
1836 static _cmsOptimizationCollection DefaultOptimization[] = {
1837 
1838     { OptimizeByJoiningCurves,            &DefaultOptimization[1] },
1839     { OptimizeMatrixShaper,               &DefaultOptimization[2] },
1840     { OptimizeByComputingLinearization,   &DefaultOptimization[3] },
1841     { OptimizeByResampling,               NULL }
1842 };
1843 
1844 // The linked list head
1845 _cmsOptimizationPluginChunkType _cmsOptimizationPluginChunk = { NULL };
1846 
1847 
1848 // Duplicates the zone of memory used by the plug-in in the new context
1849 static
DupPluginOptimizationList(struct _cmsContext_struct * ctx,const struct _cmsContext_struct * src)1850 void DupPluginOptimizationList(struct _cmsContext_struct* ctx,
1851                                const struct _cmsContext_struct* src)
1852 {
1853    _cmsOptimizationPluginChunkType newHead = { NULL };
1854    _cmsOptimizationCollection*  entry;
1855    _cmsOptimizationCollection*  Anterior = NULL;
1856    _cmsOptimizationPluginChunkType* head = (_cmsOptimizationPluginChunkType*) src->chunks[OptimizationPlugin];
1857 
1858     _cmsAssert(ctx != NULL);
1859     _cmsAssert(head != NULL);
1860 
1861     // Walk the list copying all nodes
1862    for (entry = head->OptimizationCollection;
1863         entry != NULL;
1864         entry = entry ->Next) {
1865 
1866             _cmsOptimizationCollection *newEntry = ( _cmsOptimizationCollection *) _cmsSubAllocDup(ctx ->MemPool, entry, sizeof(_cmsOptimizationCollection));
1867 
1868             if (newEntry == NULL)
1869                 return;
1870 
1871             // We want to keep the linked list order, so this is a little bit tricky
1872             newEntry -> Next = NULL;
1873             if (Anterior)
1874                 Anterior -> Next = newEntry;
1875 
1876             Anterior = newEntry;
1877 
1878             if (newHead.OptimizationCollection == NULL)
1879                 newHead.OptimizationCollection = newEntry;
1880     }
1881 
1882   ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx->MemPool, &newHead, sizeof(_cmsOptimizationPluginChunkType));
1883 }
1884 
_cmsAllocOptimizationPluginChunk(struct _cmsContext_struct * ctx,const struct _cmsContext_struct * src)1885 void  _cmsAllocOptimizationPluginChunk(struct _cmsContext_struct* ctx,
1886                                          const struct _cmsContext_struct* src)
1887 {
1888   if (src != NULL) {
1889 
1890         // Copy all linked list
1891        DupPluginOptimizationList(ctx, src);
1892     }
1893     else {
1894         static _cmsOptimizationPluginChunkType OptimizationPluginChunkType = { NULL };
1895         ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx ->MemPool, &OptimizationPluginChunkType, sizeof(_cmsOptimizationPluginChunkType));
1896     }
1897 }
1898 
1899 
1900 // Register new ways to optimize
_cmsRegisterOptimizationPlugin(cmsContext ContextID,cmsPluginBase * Data)1901 cmsBool  _cmsRegisterOptimizationPlugin(cmsContext ContextID, cmsPluginBase* Data)
1902 {
1903     cmsPluginOptimization* Plugin = (cmsPluginOptimization*) Data;
1904     _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
1905     _cmsOptimizationCollection* fl;
1906 
1907     if (Data == NULL) {
1908 
1909         ctx->OptimizationCollection = NULL;
1910         return TRUE;
1911     }
1912 
1913     // Optimizer callback is required
1914     if (Plugin ->OptimizePtr == NULL) return FALSE;
1915 
1916     fl = (_cmsOptimizationCollection*) _cmsPluginMalloc(ContextID, sizeof(_cmsOptimizationCollection));
1917     if (fl == NULL) return FALSE;
1918 
1919     // Copy the parameters
1920     fl ->OptimizePtr = Plugin ->OptimizePtr;
1921 
1922     // Keep linked list
1923     fl ->Next = ctx->OptimizationCollection;
1924 
1925     // Set the head
1926     ctx ->OptimizationCollection = fl;
1927 
1928     // All is ok
1929     return TRUE;
1930 }
1931 
1932 // The entry point for LUT optimization
_cmsOptimizePipeline(cmsContext ContextID,cmsPipeline ** PtrLut,cmsUInt32Number Intent,cmsUInt32Number * InputFormat,cmsUInt32Number * OutputFormat,cmsUInt32Number * dwFlags)1933 cmsBool CMSEXPORT _cmsOptimizePipeline(cmsContext ContextID,
1934                              cmsPipeline**    PtrLut,
1935                              cmsUInt32Number  Intent,
1936                              cmsUInt32Number* InputFormat,
1937                              cmsUInt32Number* OutputFormat,
1938                              cmsUInt32Number* dwFlags)
1939 {
1940     _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
1941     _cmsOptimizationCollection* Opts;
1942     cmsBool AnySuccess = FALSE;
1943     cmsStage* mpe;
1944 
1945     // A CLUT is being asked, so force this specific optimization
1946     if (*dwFlags & cmsFLAGS_FORCE_CLUT) {
1947 
1948         PreOptimize(*PtrLut);
1949         return OptimizeByResampling(PtrLut, Intent, InputFormat, OutputFormat, dwFlags);
1950     }
1951 
1952     // Anything to optimize?
1953     if ((*PtrLut) ->Elements == NULL) {
1954         _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
1955         return TRUE;
1956     }
1957 
1958     // Named color pipelines cannot be optimized
1959     for (mpe = cmsPipelineGetPtrToFirstStage(*PtrLut);
1960         mpe != NULL;
1961         mpe = cmsStageNext(mpe)) {
1962         if (cmsStageType(mpe) == cmsSigNamedColorElemType) return FALSE;
1963     }
1964 
1965     // Try to get rid of identities and trivial conversions.
1966     AnySuccess = PreOptimize(*PtrLut);
1967 
1968     // After removal do we end with an identity?
1969     if ((*PtrLut) ->Elements == NULL) {
1970         _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
1971         return TRUE;
1972     }
1973 
1974     // Do not optimize, keep all precision
1975     if (*dwFlags & cmsFLAGS_NOOPTIMIZE)
1976         return FALSE;
1977 
1978     // Try plug-in optimizations
1979     for (Opts = ctx->OptimizationCollection;
1980          Opts != NULL;
1981          Opts = Opts ->Next) {
1982 
1983             // If one schema succeeded, we are done
1984             if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
1985 
1986                 return TRUE;    // Optimized!
1987             }
1988     }
1989 
1990    // Try built-in optimizations
1991     for (Opts = DefaultOptimization;
1992          Opts != NULL;
1993          Opts = Opts ->Next) {
1994 
1995             if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {
1996 
1997                 return TRUE;
1998             }
1999     }
2000 
2001     // Only simple optimizations succeeded
2002     return AnySuccess;
2003 }
2004 
2005 
2006 
2007