aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/libs/lcms2/src/cmsopt.c
blob: b3d831cbd994f5d97bf1183f94cd064b206044a7 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
//---------------------------------------------------------------------------------
//
//  Little Color Management System
//  Copyright (c) 1998-2023 Marti Maria Saguer
//
// Permission is hereby granted, free of charge, to any person obtaining
// a copy of this software and associated documentation files (the "Software"),
// to deal in the Software without restriction, including without limitation
// the rights to use, copy, modify, merge, publish, distribute, sublicense,
// and/or sell copies of the Software, and to permit persons to whom the Software
// is furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO
// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
//
//---------------------------------------------------------------------------------
//

#include "lcms2_internal.h"


//----------------------------------------------------------------------------------

// Optimization for 8 bits, Shaper-CLUT (3 inputs only)
typedef struct {

    cmsContext ContextID;

    const cmsInterpParams* p;   // Tetrahedrical interpolation parameters. This is a not-owned pointer.

    cmsUInt16Number rx[256], ry[256], rz[256];
    cmsUInt32Number X0[256], Y0[256], Z0[256];  // Precomputed nodes and offsets for 8-bit input data


} Prelin8Data;


// Generic optimization for 16 bits Shaper-CLUT-Shaper (any inputs)
typedef struct {

    cmsContext ContextID;

    // Number of channels
    cmsUInt32Number nInputs;
    cmsUInt32Number nOutputs;

    _cmsInterpFn16 EvalCurveIn16[MAX_INPUT_DIMENSIONS];       // The maximum number of input channels is known in advance
    cmsInterpParams*  ParamsCurveIn16[MAX_INPUT_DIMENSIONS];

    _cmsInterpFn16 EvalCLUT;            // The evaluator for 3D grid
    const cmsInterpParams* CLUTparams;  // (not-owned pointer)


    _cmsInterpFn16* EvalCurveOut16;       // Points to an array of curve evaluators in 16 bits (not-owned pointer)
    cmsInterpParams**  ParamsCurveOut16;  // Points to an array of references to interpolation params (not-owned pointer)


} Prelin16Data;


// Optimization for matrix-shaper in 8 bits. Numbers are operated in n.14 signed, tables are stored in 1.14 fixed

typedef cmsInt32Number cmsS1Fixed14Number;   // Note that this may hold more than 16 bits!

#define DOUBLE_TO_1FIXED14(x) ((cmsS1Fixed14Number) floor((x) * 16384.0 + 0.5))

typedef struct {

    cmsContext ContextID;

    cmsS1Fixed14Number Shaper1R[256];  // from 0..255 to 1.14  (0.0...1.0)
    cmsS1Fixed14Number Shaper1G[256];
    cmsS1Fixed14Number Shaper1B[256];

    cmsS1Fixed14Number Mat[3][3];     // n.14 to n.14 (needs a saturation after that)
    cmsS1Fixed14Number Off[3];

    cmsUInt16Number Shaper2R[16385];    // 1.14 to 0..255
    cmsUInt16Number Shaper2G[16385];
    cmsUInt16Number Shaper2B[16385];

} MatShaper8Data;

// Curves, optimization is shared between 8 and 16 bits
typedef struct {

    cmsContext ContextID;

    cmsUInt32Number nCurves;      // Number of curves
    cmsUInt32Number nElements;    // Elements in curves
    cmsUInt16Number** Curves;     // Points to a dynamically  allocated array

} Curves16Data;


// Simple optimizations ----------------------------------------------------------------------------------------------------------


// Remove an element in linked chain
static
void _RemoveElement(cmsStage** head)
{
    cmsStage* mpe = *head;
    cmsStage* next = mpe ->Next;
    *head = next;
    cmsStageFree(mpe);
}

// Remove all identities in chain. Note that pt actually is a double pointer to the element that holds the pointer.
static
cmsBool _Remove1Op(cmsPipeline* Lut, cmsStageSignature UnaryOp)
{
    cmsStage** pt = &Lut ->Elements;
    cmsBool AnyOpt = FALSE;

    while (*pt != NULL) {

        if ((*pt) ->Implements == UnaryOp) {
            _RemoveElement(pt);
            AnyOpt = TRUE;
        }
        else
            pt = &((*pt) -> Next);
    }

    return AnyOpt;
}

// Same, but only if two adjacent elements are found
static
cmsBool _Remove2Op(cmsPipeline* Lut, cmsStageSignature Op1, cmsStageSignature Op2)
{
    cmsStage** pt1;
    cmsStage** pt2;
    cmsBool AnyOpt = FALSE;

    pt1 = &Lut ->Elements;
    if (*pt1 == NULL) return AnyOpt;

    while (*pt1 != NULL) {

        pt2 = &((*pt1) -> Next);
        if (*pt2 == NULL) return AnyOpt;

        if ((*pt1) ->Implements == Op1 && (*pt2) ->Implements == Op2) {
            _RemoveElement(pt2);
            _RemoveElement(pt1);
            AnyOpt = TRUE;
        }
        else
            pt1 = &((*pt1) -> Next);
    }

    return AnyOpt;
}


static
cmsBool CloseEnoughFloat(cmsFloat64Number a, cmsFloat64Number b)
{
       return fabs(b - a) < 0.00001f;
}

static
cmsBool  isFloatMatrixIdentity(const cmsMAT3* a)
{
       cmsMAT3 Identity;
       int i, j;

       _cmsMAT3identity(&Identity);

       for (i = 0; i < 3; i++)
              for (j = 0; j < 3; j++)
                     if (!CloseEnoughFloat(a->v[i].n[j], Identity.v[i].n[j])) return FALSE;

       return TRUE;
}
// if two adjacent matrices are found, multiply them. 
static
cmsBool _MultiplyMatrix(cmsPipeline* Lut)
{
       cmsStage** pt1;
       cmsStage** pt2;
       cmsStage*  chain;
       cmsBool AnyOpt = FALSE;

       pt1 = &Lut->Elements;
       if (*pt1 == NULL) return AnyOpt;

       while (*pt1 != NULL) {

              pt2 = &((*pt1)->Next);
              if (*pt2 == NULL) return AnyOpt;

              if ((*pt1)->Implements == cmsSigMatrixElemType && (*pt2)->Implements == cmsSigMatrixElemType) {

                     // Get both matrices
                     _cmsStageMatrixData* m1 = (_cmsStageMatrixData*) cmsStageData(*pt1);
                     _cmsStageMatrixData* m2 = (_cmsStageMatrixData*) cmsStageData(*pt2);
                     cmsMAT3 res;
                     
                     // Input offset and output offset should be zero to use this optimization
                     if (m1->Offset != NULL || m2 ->Offset != NULL || 
                            cmsStageInputChannels(*pt1) != 3 || cmsStageOutputChannels(*pt1) != 3 ||                            
                            cmsStageInputChannels(*pt2) != 3 || cmsStageOutputChannels(*pt2) != 3)
                            return FALSE;

                     // Multiply both matrices to get the result
                     _cmsMAT3per(&res, (cmsMAT3*)m2->Double, (cmsMAT3*)m1->Double);

                     // Get the next in chain after the matrices
                     chain = (*pt2)->Next;

                     // Remove both matrices
                     _RemoveElement(pt2);
                     _RemoveElement(pt1);

                     // Now what if the result is a plain identity?                     
                     if (!isFloatMatrixIdentity(&res)) {

                            // We can not get rid of full matrix                            
                            cmsStage* Multmat = cmsStageAllocMatrix(Lut->ContextID, 3, 3, (const cmsFloat64Number*) &res, NULL);
                            if (Multmat == NULL) return FALSE;  // Should never happen

                            // Recover the chain
                            Multmat->Next = chain;
                            *pt1 = Multmat;
                     }

                     AnyOpt = TRUE;
              }
              else
                     pt1 = &((*pt1)->Next);
       }

       return AnyOpt;
}


// Preoptimize just gets rif of no-ops coming paired. Conversion from v2 to v4 followed
// by a v4 to v2 and vice-versa. The elements are then discarded.
static
cmsBool PreOptimize(cmsPipeline* Lut)
{
    cmsBool AnyOpt = FALSE, Opt;

    do {

        Opt = FALSE;

        // Remove all identities
        Opt |= _Remove1Op(Lut, cmsSigIdentityElemType);

        // Remove XYZ2Lab followed by Lab2XYZ
        Opt |= _Remove2Op(Lut, cmsSigXYZ2LabElemType, cmsSigLab2XYZElemType);

        // Remove Lab2XYZ followed by XYZ2Lab
        Opt |= _Remove2Op(Lut, cmsSigLab2XYZElemType, cmsSigXYZ2LabElemType);

        // Remove V4 to V2 followed by V2 to V4
        Opt |= _Remove2Op(Lut, cmsSigLabV4toV2, cmsSigLabV2toV4);

        // Remove V2 to V4 followed by V4 to V2
        Opt |= _Remove2Op(Lut, cmsSigLabV2toV4, cmsSigLabV4toV2);

        // Remove float pcs Lab conversions
        Opt |= _Remove2Op(Lut, cmsSigLab2FloatPCS, cmsSigFloatPCS2Lab);

        // Remove float pcs Lab conversions
        Opt |= _Remove2Op(Lut, cmsSigXYZ2FloatPCS, cmsSigFloatPCS2XYZ);

        // Simplify matrix. 
        Opt |= _MultiplyMatrix(Lut);

        if (Opt) AnyOpt = TRUE;

    } while (Opt);

    return AnyOpt;
}

static
void Eval16nop1D(CMSREGISTER const cmsUInt16Number Input[],
                 CMSREGISTER cmsUInt16Number Output[],
                 CMSREGISTER const struct _cms_interp_struc* p)
{
    Output[0] = Input[0];

    cmsUNUSED_PARAMETER(p);
}

static
void PrelinEval16(CMSREGISTER const cmsUInt16Number Input[],
                  CMSREGISTER cmsUInt16Number Output[],
                  CMSREGISTER const void* D)
{
    Prelin16Data* p16 = (Prelin16Data*) D;
    cmsUInt16Number  StageABC[MAX_INPUT_DIMENSIONS];
    cmsUInt16Number  StageDEF[cmsMAXCHANNELS];
    cmsUInt32Number i;

    for (i=0; i < p16 ->nInputs; i++) {

        p16 ->EvalCurveIn16[i](&Input[i], &StageABC[i], p16 ->ParamsCurveIn16[i]);
    }

    p16 ->EvalCLUT(StageABC, StageDEF, p16 ->CLUTparams);

    for (i=0; i < p16 ->nOutputs; i++) {

        p16 ->EvalCurveOut16[i](&StageDEF[i], &Output[i], p16 ->ParamsCurveOut16[i]);
    }
}


static
void PrelinOpt16free(cmsContext ContextID, void* ptr)
{
    Prelin16Data* p16 = (Prelin16Data*) ptr;

    _cmsFree(ContextID, p16 ->EvalCurveOut16);
    _cmsFree(ContextID, p16 ->ParamsCurveOut16);

    _cmsFree(ContextID, p16);
}

static
void* Prelin16dup(cmsContext ContextID, const void* ptr)
{
    Prelin16Data* p16 = (Prelin16Data*) ptr;
    Prelin16Data* Duped = (Prelin16Data*) _cmsDupMem(ContextID, p16, sizeof(Prelin16Data));

    if (Duped == NULL) return NULL;

    Duped->EvalCurveOut16 = (_cmsInterpFn16*) _cmsDupMem(ContextID, p16->EvalCurveOut16, p16->nOutputs * sizeof(_cmsInterpFn16));
    Duped->ParamsCurveOut16 = (cmsInterpParams**)_cmsDupMem(ContextID, p16->ParamsCurveOut16, p16->nOutputs * sizeof(cmsInterpParams*));

    return Duped;
}


static
Prelin16Data* PrelinOpt16alloc(cmsContext ContextID,
                               const cmsInterpParams* ColorMap,
                               cmsUInt32Number nInputs, cmsToneCurve** In,
                               cmsUInt32Number nOutputs, cmsToneCurve** Out )
{
    cmsUInt32Number i;
    Prelin16Data* p16 = (Prelin16Data*)_cmsMallocZero(ContextID, sizeof(Prelin16Data));
    if (p16 == NULL) return NULL;

    p16 ->nInputs = nInputs;
    p16 ->nOutputs = nOutputs;


    for (i=0; i < nInputs; i++) {

        if (In == NULL) {
            p16 -> ParamsCurveIn16[i] = NULL;
            p16 -> EvalCurveIn16[i] = Eval16nop1D;

        }
        else {
            p16 -> ParamsCurveIn16[i] = In[i] ->InterpParams;
            p16 -> EvalCurveIn16[i] = p16 ->ParamsCurveIn16[i]->Interpolation.Lerp16;
        }
    }

    p16 ->CLUTparams = ColorMap;
    p16 ->EvalCLUT   = ColorMap ->Interpolation.Lerp16;


    p16 -> EvalCurveOut16 = (_cmsInterpFn16*) _cmsCalloc(ContextID, nOutputs, sizeof(_cmsInterpFn16));
    if (p16->EvalCurveOut16 == NULL)
    {
        _cmsFree(ContextID, p16);
        return NULL;
    }

    p16 -> ParamsCurveOut16 = (cmsInterpParams**) _cmsCalloc(ContextID, nOutputs, sizeof(cmsInterpParams* ));
    if (p16->ParamsCurveOut16 == NULL)
    {

        _cmsFree(ContextID, p16->EvalCurveOut16);
        _cmsFree(ContextID, p16);
        return NULL;
    }

    for (i=0; i < nOutputs; i++) {

        if (Out == NULL) {
            p16 ->ParamsCurveOut16[i] = NULL;
            p16 -> EvalCurveOut16[i] = Eval16nop1D;
        }
        else {

            p16 ->ParamsCurveOut16[i] = Out[i] ->InterpParams;
            p16 -> EvalCurveOut16[i] = p16 ->ParamsCurveOut16[i]->Interpolation.Lerp16;
        }
    }

    return p16;
}



// Resampling ---------------------------------------------------------------------------------

#define PRELINEARIZATION_POINTS 4096

// Sampler implemented by another LUT. This is a clean way to precalculate the devicelink 3D CLUT for
// almost any transform. We use floating point precision and then convert from floating point to 16 bits.
static
cmsInt32Number XFormSampler16(CMSREGISTER const cmsUInt16Number In[], 
                              CMSREGISTER cmsUInt16Number Out[], 
                              CMSREGISTER void* Cargo)
{
    cmsPipeline* Lut = (cmsPipeline*) Cargo;
    cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
    cmsUInt32Number i;

    _cmsAssert(Lut -> InputChannels < cmsMAXCHANNELS);
    _cmsAssert(Lut -> OutputChannels < cmsMAXCHANNELS);

    // From 16 bit to floating point
    for (i=0; i < Lut ->InputChannels; i++)
        InFloat[i] = (cmsFloat32Number) (In[i] / 65535.0);

    // Evaluate in floating point
    cmsPipelineEvalFloat(InFloat, OutFloat, Lut);

    // Back to 16 bits representation
    for (i=0; i < Lut ->OutputChannels; i++)
        Out[i] = _cmsQuickSaturateWord(OutFloat[i] * 65535.0);

    // Always succeed
    return TRUE;
}

// Try to see if the curves of a given MPE are linear
static
cmsBool AllCurvesAreLinear(cmsStage* mpe)
{
    cmsToneCurve** Curves;
    cmsUInt32Number i, n;

    Curves = _cmsStageGetPtrToCurveSet(mpe);
    if (Curves == NULL) return FALSE;

    n = cmsStageOutputChannels(mpe);

    for (i=0; i < n; i++) {
        if (!cmsIsToneCurveLinear(Curves[i])) return FALSE;
    }

    return TRUE;
}

// This function replaces a specific node placed in "At" by the "Value" numbers. Its purpose
// is to fix scum dot on broken profiles/transforms. Works on 1, 3 and 4 channels
static
cmsBool  PatchLUT(cmsStage* CLUT, cmsUInt16Number At[], cmsUInt16Number Value[],
                  cmsUInt32Number nChannelsOut, cmsUInt32Number nChannelsIn)
{
    _cmsStageCLutData* Grid = (_cmsStageCLutData*) CLUT ->Data;
    cmsInterpParams* p16  = Grid ->Params;
    cmsFloat64Number px, py, pz, pw;
    int        x0, y0, z0, w0;
    int        i, index;

    if (CLUT -> Type != cmsSigCLutElemType) {
        cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) Attempt to PatchLUT on non-lut stage");
        return FALSE;
    }

    if (nChannelsIn == 4) {

        px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
        py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
        pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
        pw = ((cmsFloat64Number) At[3] * (p16->Domain[3])) / 65535.0;

        x0 = (int) floor(px);
        y0 = (int) floor(py);
        z0 = (int) floor(pz);
        w0 = (int) floor(pw);

        if (((px - x0) != 0) ||
            ((py - y0) != 0) ||
            ((pz - z0) != 0) ||
            ((pw - w0) != 0)) return FALSE; // Not on exact node

        index = (int) p16 -> opta[3] * x0 +
                (int) p16 -> opta[2] * y0 +
                (int) p16 -> opta[1] * z0 +
                (int) p16 -> opta[0] * w0;
    }
    else
        if (nChannelsIn == 3) {

            px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
            py = ((cmsFloat64Number) At[1] * (p16->Domain[1])) / 65535.0;
            pz = ((cmsFloat64Number) At[2] * (p16->Domain[2])) / 65535.0;
           
            x0 = (int) floor(px);
            y0 = (int) floor(py);
            z0 = (int) floor(pz);
           
            if (((px - x0) != 0) ||
                ((py - y0) != 0) ||
                ((pz - z0) != 0)) return FALSE;  // Not on exact node

            index = (int) p16 -> opta[2] * x0 +
                    (int) p16 -> opta[1] * y0 +
                    (int) p16 -> opta[0] * z0;
        }
        else
            if (nChannelsIn == 1) {

                px = ((cmsFloat64Number) At[0] * (p16->Domain[0])) / 65535.0;
                
                x0 = (int) floor(px);
                
                if (((px - x0) != 0)) return FALSE; // Not on exact node

                index = (int) p16 -> opta[0] * x0;
            }
            else {
                cmsSignalError(CLUT->ContextID, cmsERROR_INTERNAL, "(internal) %d Channels are not supported on PatchLUT", nChannelsIn);
                return FALSE;
            }

    for (i = 0; i < (int) nChannelsOut; i++)
        Grid->Tab.T[index + i] = Value[i];

    return TRUE;
}

// Auxiliary, to see if two values are equal or very different
static
cmsBool WhitesAreEqual(cmsUInt32Number n, cmsUInt16Number White1[], cmsUInt16Number White2[] )
{
    cmsUInt32Number i;

    for (i=0; i < n; i++) {

        if (abs(White1[i] - White2[i]) > 0xf000) return TRUE;  // Values are so extremely different that the fixup should be avoided
        if (White1[i] != White2[i]) return FALSE;
    }
    return TRUE;
}


// Locate the node for the white point and fix it to pure white in order to avoid scum dot.
static
cmsBool FixWhiteMisalignment(cmsPipeline* Lut, cmsColorSpaceSignature EntryColorSpace, cmsColorSpaceSignature ExitColorSpace)
{
    cmsUInt16Number *WhitePointIn, *WhitePointOut;
    cmsUInt16Number  WhiteIn[cmsMAXCHANNELS], WhiteOut[cmsMAXCHANNELS], ObtainedOut[cmsMAXCHANNELS];
    cmsUInt32Number i, nOuts, nIns;
    cmsStage *PreLin = NULL, *CLUT = NULL, *PostLin = NULL;

    if (!_cmsEndPointsBySpace(EntryColorSpace,
        &WhitePointIn, NULL, &nIns)) return FALSE;

    if (!_cmsEndPointsBySpace(ExitColorSpace,
        &WhitePointOut, NULL, &nOuts)) return FALSE;

    // It needs to be fixed?
    if (Lut ->InputChannels != nIns) return FALSE;
    if (Lut ->OutputChannels != nOuts) return FALSE;

    cmsPipelineEval16(WhitePointIn, ObtainedOut, Lut);

    if (WhitesAreEqual(nOuts, WhitePointOut, ObtainedOut)) return TRUE; // whites already match

    // Check if the LUT comes as Prelin, CLUT or Postlin. We allow all combinations
    if (!cmsPipelineCheckAndRetreiveStages(Lut, 3, cmsSigCurveSetElemType, cmsSigCLutElemType, cmsSigCurveSetElemType, &PreLin, &CLUT, &PostLin))
        if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCurveSetElemType, cmsSigCLutElemType, &PreLin, &CLUT))
            if (!cmsPipelineCheckAndRetreiveStages(Lut, 2, cmsSigCLutElemType, cmsSigCurveSetElemType, &CLUT, &PostLin))
                if (!cmsPipelineCheckAndRetreiveStages(Lut, 1, cmsSigCLutElemType, &CLUT))
                    return FALSE;

    // We need to interpolate white points of both, pre and post curves
    if (PreLin) {

        cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PreLin);

        for (i=0; i < nIns; i++) {
            WhiteIn[i] = cmsEvalToneCurve16(Curves[i], WhitePointIn[i]);
        }
    }
    else {
        for (i=0; i < nIns; i++)
            WhiteIn[i] = WhitePointIn[i];
    }

    // If any post-linearization, we need to find how is represented white before the curve, do
    // a reverse interpolation in this case.
    if (PostLin) {

        cmsToneCurve** Curves = _cmsStageGetPtrToCurveSet(PostLin);

        for (i=0; i < nOuts; i++) {

            cmsToneCurve* InversePostLin = cmsReverseToneCurve(Curves[i]);
            if (InversePostLin == NULL) {
                WhiteOut[i] = WhitePointOut[i];    

            } else {

                WhiteOut[i] = cmsEvalToneCurve16(InversePostLin, WhitePointOut[i]);
                cmsFreeToneCurve(InversePostLin);
            }
        }
    }
    else {
        for (i=0; i < nOuts; i++)
            WhiteOut[i] = WhitePointOut[i];
    }

    // Ok, proceed with patching. May fail and we don't care if it fails
    PatchLUT(CLUT, WhiteIn, WhiteOut, nOuts, nIns);

    return TRUE;
}

// -----------------------------------------------------------------------------------------------------------------------------------------------
// This function creates simple LUT from complex ones. The generated LUT has an optional set of
// prelinearization curves, a CLUT of nGridPoints and optional postlinearization tables.
// These curves have to exist in the original LUT in order to be used in the simplified output.
// Caller may also use the flags to allow this feature.
// LUTS with all curves will be simplified to a single curve. Parametric curves are lost.
// This function should be used on 16-bits LUTS only, as floating point losses precision when simplified
// -----------------------------------------------------------------------------------------------------------------------------------------------

static
cmsBool OptimizeByResampling(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
{
    cmsPipeline* Src = NULL;
    cmsPipeline* Dest = NULL;
    cmsStage* CLUT;
    cmsStage *KeepPreLin = NULL, *KeepPostLin = NULL;
    cmsUInt32Number nGridPoints;
    cmsColorSpaceSignature ColorSpace, OutputColorSpace;
    cmsStage *NewPreLin = NULL;
    cmsStage *NewPostLin = NULL;
    _cmsStageCLutData* DataCLUT;
    cmsToneCurve** DataSetIn;
    cmsToneCurve** DataSetOut;
    Prelin16Data* p16;

    // This is a lossy optimization! does not apply in floating-point cases
    if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;

    ColorSpace       = _cmsICCcolorSpace((int) T_COLORSPACE(*InputFormat));
    OutputColorSpace = _cmsICCcolorSpace((int) T_COLORSPACE(*OutputFormat));

    // Color space must be specified
    if (ColorSpace == (cmsColorSpaceSignature)0 ||
        OutputColorSpace == (cmsColorSpaceSignature)0) return FALSE;

    nGridPoints = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);

    // For empty LUTs, 2 points are enough
    if (cmsPipelineStageCount(*Lut) == 0)
        nGridPoints = 2;

    Src = *Lut;

    // Allocate an empty LUT
    Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
    if (!Dest) return FALSE;

    // Prelinearization tables are kept unless indicated by flags
    if (*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION) {

        // Get a pointer to the prelinearization element
        cmsStage* PreLin = cmsPipelineGetPtrToFirstStage(Src);

        // Check if suitable
        if (PreLin && PreLin ->Type == cmsSigCurveSetElemType) {

            // Maybe this is a linear tram, so we can avoid the whole stuff
            if (!AllCurvesAreLinear(PreLin)) {

                // All seems ok, proceed.
                NewPreLin = cmsStageDup(PreLin);
                if(!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, NewPreLin))
                    goto Error;

                // Remove prelinearization. Since we have duplicated the curve
                // in destination LUT, the sampling should be applied after this stage.
                cmsPipelineUnlinkStage(Src, cmsAT_BEGIN, &KeepPreLin);
            }
        }
    }

    // Allocate the CLUT
    CLUT = cmsStageAllocCLut16bit(Src ->ContextID, nGridPoints, Src ->InputChannels, Src->OutputChannels, NULL);
    if (CLUT == NULL) goto Error;

    // Add the CLUT to the destination LUT
    if (!cmsPipelineInsertStage(Dest, cmsAT_END, CLUT)) {
        goto Error;
    }

    // Postlinearization tables are kept unless indicated by flags
    if (*dwFlags & cmsFLAGS_CLUT_POST_LINEARIZATION) {

        // Get a pointer to the postlinearization if present
        cmsStage* PostLin = cmsPipelineGetPtrToLastStage(Src);

        // Check if suitable
        if (PostLin && cmsStageType(PostLin) == cmsSigCurveSetElemType) {

            // Maybe this is a linear tram, so we can avoid the whole stuff
            if (!AllCurvesAreLinear(PostLin)) {

                // All seems ok, proceed.
                NewPostLin = cmsStageDup(PostLin);
                if (!cmsPipelineInsertStage(Dest, cmsAT_END, NewPostLin))
                    goto Error;

                // In destination LUT, the sampling should be applied after this stage.
                cmsPipelineUnlinkStage(Src, cmsAT_END, &KeepPostLin);
            }
        }
    }

    // Now its time to do the sampling. We have to ignore pre/post linearization
    // The source LUT without pre/post curves is passed as parameter.
    if (!cmsStageSampleCLut16bit(CLUT, XFormSampler16, (void*) Src, 0)) {
Error:
        // Ops, something went wrong, Restore stages
        if (KeepPreLin != NULL) {
            if (!cmsPipelineInsertStage(Src, cmsAT_BEGIN, KeepPreLin)) {
                _cmsAssert(0); // This never happens
            }
        }
        if (KeepPostLin != NULL) {
            if (!cmsPipelineInsertStage(Src, cmsAT_END,   KeepPostLin)) {
                _cmsAssert(0); // This never happens
            }
        }
        cmsPipelineFree(Dest);
        return FALSE;
    }

    // Done.

    if (KeepPreLin != NULL) cmsStageFree(KeepPreLin);
    if (KeepPostLin != NULL) cmsStageFree(KeepPostLin);
    cmsPipelineFree(Src);

    DataCLUT = (_cmsStageCLutData*) CLUT ->Data;

    if (NewPreLin == NULL) DataSetIn = NULL;
    else DataSetIn = ((_cmsStageToneCurvesData*) NewPreLin ->Data) ->TheCurves;

    if (NewPostLin == NULL) DataSetOut = NULL;
    else  DataSetOut = ((_cmsStageToneCurvesData*) NewPostLin ->Data) ->TheCurves;


    if (DataSetIn == NULL && DataSetOut == NULL) {

        _cmsPipelineSetOptimizationParameters(Dest, (_cmsPipelineEval16Fn) DataCLUT->Params->Interpolation.Lerp16, DataCLUT->Params, NULL, NULL);
    }
    else {

        p16 = PrelinOpt16alloc(Dest ->ContextID,
            DataCLUT ->Params,
            Dest ->InputChannels,
            DataSetIn,
            Dest ->OutputChannels,
            DataSetOut);

        _cmsPipelineSetOptimizationParameters(Dest, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);
    }


    // Don't fix white on absolute colorimetric
    if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
        *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;

    if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {

        FixWhiteMisalignment(Dest, ColorSpace, OutputColorSpace);
    }

    *Lut = Dest;
    return TRUE;

    cmsUNUSED_PARAMETER(Intent);
}


// -----------------------------------------------------------------------------------------------------------------------------------------------
// Fixes the gamma balancing of transform. This is described in my paper "Prelinearization Stages on
// Color-Management Application-Specific Integrated Circuits (ASICs)" presented at NIP24. It only works
// for RGB transforms. See the paper for more details
// -----------------------------------------------------------------------------------------------------------------------------------------------


// Normalize endpoints by slope limiting max and min. This assures endpoints as well.
// Descending curves are handled as well.
static
void SlopeLimiting(cmsToneCurve* g)
{
    int BeginVal, EndVal;
    int AtBegin = (int) floor((cmsFloat64Number) g ->nEntries * 0.02 + 0.5);   // Cutoff at 2%
    int AtEnd   = (int) g ->nEntries - AtBegin - 1;                                  // And 98%
    cmsFloat64Number Val, Slope, beta;
    int i;

    if (cmsIsToneCurveDescending(g)) {
        BeginVal = 0xffff; EndVal = 0;
    }
    else {
        BeginVal = 0; EndVal = 0xffff;
    }

    // Compute slope and offset for begin of curve
    Val   = g ->Table16[AtBegin];
    Slope = (Val - BeginVal) / AtBegin;
    beta  = Val - Slope * AtBegin;

    for (i=0; i < AtBegin; i++)
        g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);

    // Compute slope and offset for the end
    Val   = g ->Table16[AtEnd];
    Slope = (EndVal - Val) / AtBegin;   // AtBegin holds the X interval, which is same in both cases
    beta  = Val - Slope * AtEnd;

    for (i = AtEnd; i < (int) g ->nEntries; i++)
        g ->Table16[i] = _cmsQuickSaturateWord(i * Slope + beta);
}


// Precomputes tables for 8-bit on input devicelink.
static
Prelin8Data* PrelinOpt8alloc(cmsContext ContextID, const cmsInterpParams* p, cmsToneCurve* G[3])
{
    int i;
    cmsUInt16Number Input[3];
    cmsS15Fixed16Number v1, v2, v3;
    Prelin8Data* p8;

    p8 = (Prelin8Data*)_cmsMallocZero(ContextID, sizeof(Prelin8Data));
    if (p8 == NULL) return NULL;

    // Since this only works for 8 bit input, values comes always as x * 257,
    // we can safely take msb byte (x << 8 + x)

    for (i=0; i < 256; i++) {

        if (G != NULL) {

            // Get 16-bit representation
            Input[0] = cmsEvalToneCurve16(G[0], FROM_8_TO_16(i));
            Input[1] = cmsEvalToneCurve16(G[1], FROM_8_TO_16(i));
            Input[2] = cmsEvalToneCurve16(G[2], FROM_8_TO_16(i));
        }
        else {
            Input[0] = FROM_8_TO_16(i);
            Input[1] = FROM_8_TO_16(i);
            Input[2] = FROM_8_TO_16(i);
        }


        // Move to 0..1.0 in fixed domain
        v1 = _cmsToFixedDomain((int) (Input[0] * p -> Domain[0]));
        v2 = _cmsToFixedDomain((int) (Input[1] * p -> Domain[1]));
        v3 = _cmsToFixedDomain((int) (Input[2] * p -> Domain[2]));

        // Store the precalculated table of nodes
        p8 ->X0[i] = (p->opta[2] * FIXED_TO_INT(v1));
        p8 ->Y0[i] = (p->opta[1] * FIXED_TO_INT(v2));
        p8 ->Z0[i] = (p->opta[0] * FIXED_TO_INT(v3));

        // Store the precalculated table of offsets
        p8 ->rx[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v1);
        p8 ->ry[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v2);
        p8 ->rz[i] = (cmsUInt16Number) FIXED_REST_TO_INT(v3);
    }

    p8 ->ContextID = ContextID;
    p8 ->p = p;

    return p8;
}

static
void Prelin8free(cmsContext ContextID, void* ptr)
{
    _cmsFree(ContextID, ptr);
}

static
void* Prelin8dup(cmsContext ContextID, const void* ptr)
{
    return _cmsDupMem(ContextID, ptr, sizeof(Prelin8Data));
}



// A optimized interpolation for 8-bit input.
#define DENS(i,j,k) (LutTable[(i)+(j)+(k)+OutChan])
static CMS_NO_SANITIZE
void PrelinEval8(CMSREGISTER const cmsUInt16Number Input[],
                 CMSREGISTER cmsUInt16Number Output[],
                 CMSREGISTER const void* D)
{

    cmsUInt8Number         r, g, b;
    cmsS15Fixed16Number    rx, ry, rz;
    cmsS15Fixed16Number    c0, c1, c2, c3, Rest;
    int                    OutChan;
    CMSREGISTER cmsS15Fixed16Number X0, X1, Y0, Y1, Z0, Z1;
    Prelin8Data* p8 = (Prelin8Data*) D;
    CMSREGISTER const cmsInterpParams* p = p8 ->p;
    int                    TotalOut = (int) p -> nOutputs;
    const cmsUInt16Number* LutTable = (const cmsUInt16Number*) p->Table;

    r = (cmsUInt8Number) (Input[0] >> 8);
    g = (cmsUInt8Number) (Input[1] >> 8);
    b = (cmsUInt8Number) (Input[2] >> 8);

    X0 = (cmsS15Fixed16Number) p8->X0[r];
    Y0 = (cmsS15Fixed16Number) p8->Y0[g];
    Z0 = (cmsS15Fixed16Number) p8->Z0[b];

    rx = p8 ->rx[r];
    ry = p8 ->ry[g];
    rz = p8 ->rz[b];

    X1 = X0 + (cmsS15Fixed16Number)((rx == 0) ? 0 :  p ->opta[2]);
    Y1 = Y0 + (cmsS15Fixed16Number)((ry == 0) ? 0 :  p ->opta[1]);
    Z1 = Z0 + (cmsS15Fixed16Number)((rz == 0) ? 0 :  p ->opta[0]);


    // These are the 6 Tetrahedral
    for (OutChan=0; OutChan < TotalOut; OutChan++) {

        c0 = DENS(X0, Y0, Z0);

        if (rx >= ry && ry >= rz)
        {
            c1 = DENS(X1, Y0, Z0) - c0;
            c2 = DENS(X1, Y1, Z0) - DENS(X1, Y0, Z0);
            c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
        }
        else
            if (rx >= rz && rz >= ry)
            {
                c1 = DENS(X1, Y0, Z0) - c0;
                c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
                c3 = DENS(X1, Y0, Z1) - DENS(X1, Y0, Z0);
            }
            else
                if (rz >= rx && rx >= ry)
                {
                    c1 = DENS(X1, Y0, Z1) - DENS(X0, Y0, Z1);
                    c2 = DENS(X1, Y1, Z1) - DENS(X1, Y0, Z1);
                    c3 = DENS(X0, Y0, Z1) - c0;
                }
                else
                    if (ry >= rx && rx >= rz)
                    {
                        c1 = DENS(X1, Y1, Z0) - DENS(X0, Y1, Z0);
                        c2 = DENS(X0, Y1, Z0) - c0;
                        c3 = DENS(X1, Y1, Z1) - DENS(X1, Y1, Z0);
                    }
                    else
                        if (ry >= rz && rz >= rx)
                        {
                            c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
                            c2 = DENS(X0, Y1, Z0) - c0;
                            c3 = DENS(X0, Y1, Z1) - DENS(X0, Y1, Z0);
                        }
                        else
                            if (rz >= ry && ry >= rx)
                            {
                                c1 = DENS(X1, Y1, Z1) - DENS(X0, Y1, Z1);
                                c2 = DENS(X0, Y1, Z1) - DENS(X0, Y0, Z1);
                                c3 = DENS(X0, Y0, Z1) - c0;
                            }
                            else  {
                                c1 = c2 = c3 = 0;
                            }

        Rest = c1 * rx + c2 * ry + c3 * rz + 0x8001;
        Output[OutChan] = (cmsUInt16Number) (c0 + ((Rest + (Rest >> 16)) >> 16));

    }
}

#undef DENS


// Curves that contain wide empty areas are not optimizeable
static
cmsBool IsDegenerated(const cmsToneCurve* g)
{
    cmsUInt32Number i, Zeros = 0, Poles = 0;
    cmsUInt32Number nEntries = g ->nEntries;

    for (i=0; i < nEntries; i++) {

        if (g ->Table16[i] == 0x0000) Zeros++;
        if (g ->Table16[i] == 0xffff) Poles++;
    }

    if (Zeros == 1 && Poles == 1) return FALSE;  // For linear tables
    if (Zeros > (nEntries / 20)) return TRUE;  // Degenerated, many zeros
    if (Poles > (nEntries / 20)) return TRUE;  // Degenerated, many poles

    return FALSE;
}

// --------------------------------------------------------------------------------------------------------------
// We need xput over here

static
cmsBool OptimizeByComputingLinearization(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
{
    cmsPipeline* OriginalLut;
    cmsUInt32Number nGridPoints;
    cmsToneCurve *Trans[cmsMAXCHANNELS], *TransReverse[cmsMAXCHANNELS];
    cmsUInt32Number t, i;
    cmsFloat32Number v, In[cmsMAXCHANNELS], Out[cmsMAXCHANNELS];
    cmsBool lIsSuitable, lIsLinear;
    cmsPipeline* OptimizedLUT = NULL, *LutPlusCurves = NULL;
    cmsStage* OptimizedCLUTmpe;
    cmsColorSpaceSignature ColorSpace, OutputColorSpace;
    cmsStage* OptimizedPrelinMpe;
    cmsToneCurve** OptimizedPrelinCurves;
    _cmsStageCLutData* OptimizedPrelinCLUT;


    // This is a lossy optimization! does not apply in floating-point cases
    if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;

    // Only on chunky RGB
    if (T_COLORSPACE(*InputFormat)  != PT_RGB) return FALSE;
    if (T_PLANAR(*InputFormat)) return FALSE;

    if (T_COLORSPACE(*OutputFormat) != PT_RGB) return FALSE;
    if (T_PLANAR(*OutputFormat)) return FALSE;

    // On 16 bits, user has to specify the feature
    if (!_cmsFormatterIs8bit(*InputFormat)) {
        if (!(*dwFlags & cmsFLAGS_CLUT_PRE_LINEARIZATION)) return FALSE;
    }

    OriginalLut = *Lut;
   
    ColorSpace       = _cmsICCcolorSpace((int) T_COLORSPACE(*InputFormat));
    OutputColorSpace = _cmsICCcolorSpace((int) T_COLORSPACE(*OutputFormat));

    // Color space must be specified
    if (ColorSpace == (cmsColorSpaceSignature)0 ||
        OutputColorSpace == (cmsColorSpaceSignature)0) return FALSE;

    nGridPoints      = _cmsReasonableGridpointsByColorspace(ColorSpace, *dwFlags);

    // Empty gamma containers
    memset(Trans, 0, sizeof(Trans));
    memset(TransReverse, 0, sizeof(TransReverse));

    // If the last stage of the original lut are curves, and those curves are
    // degenerated, it is likely the transform is squeezing and clipping
    // the output from previous CLUT. We cannot optimize this case     
    {
        cmsStage* last = cmsPipelineGetPtrToLastStage(OriginalLut);

        if (last == NULL) goto Error;
        if (cmsStageType(last) == cmsSigCurveSetElemType) {

            _cmsStageToneCurvesData* Data = (_cmsStageToneCurvesData*)cmsStageData(last);
            for (i = 0; i < Data->nCurves; i++) {
                if (IsDegenerated(Data->TheCurves[i]))
                    goto Error;
            }
        }
    }

    for (t = 0; t < OriginalLut ->InputChannels; t++) {
        Trans[t] = cmsBuildTabulatedToneCurve16(OriginalLut ->ContextID, PRELINEARIZATION_POINTS, NULL);
        if (Trans[t] == NULL) goto Error;
    }

    // Populate the curves
    for (i=0; i < PRELINEARIZATION_POINTS; i++) {

        v = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));

        // Feed input with a gray ramp
        for (t=0; t < OriginalLut ->InputChannels; t++)
            In[t] = v;

        // Evaluate the gray value
        cmsPipelineEvalFloat(In, Out, OriginalLut);

        // Store result in curve
        for (t=0; t < OriginalLut ->InputChannels; t++)
            Trans[t] ->Table16[i] = _cmsQuickSaturateWord(Out[t] * 65535.0);
    }

    // Slope-limit the obtained curves
    for (t = 0; t < OriginalLut ->InputChannels; t++)
        SlopeLimiting(Trans[t]);

    // Check for validity
    lIsSuitable = TRUE;
    lIsLinear   = TRUE;
    for (t=0; (lIsSuitable && (t < OriginalLut ->InputChannels)); t++) {

        // Exclude if already linear
        if (!cmsIsToneCurveLinear(Trans[t]))
            lIsLinear = FALSE;

        // Exclude if non-monotonic
        if (!cmsIsToneCurveMonotonic(Trans[t]))
            lIsSuitable = FALSE;

        if (IsDegenerated(Trans[t]))
            lIsSuitable = FALSE;
    }

    // If it is not suitable, just quit
    if (!lIsSuitable) goto Error;

    // Invert curves if possible
    for (t = 0; t < OriginalLut ->InputChannels; t++) {
        TransReverse[t] = cmsReverseToneCurveEx(PRELINEARIZATION_POINTS, Trans[t]);
        if (TransReverse[t] == NULL) goto Error;
    }

    // Now inset the reversed curves at the begin of transform
    LutPlusCurves = cmsPipelineDup(OriginalLut);
    if (LutPlusCurves == NULL) goto Error;

    if (!cmsPipelineInsertStage(LutPlusCurves, cmsAT_BEGIN, cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, TransReverse)))
        goto Error;

    // Create the result LUT
    OptimizedLUT = cmsPipelineAlloc(OriginalLut ->ContextID, OriginalLut ->InputChannels, OriginalLut ->OutputChannels);
    if (OptimizedLUT == NULL) goto Error;

    OptimizedPrelinMpe = cmsStageAllocToneCurves(OriginalLut ->ContextID, OriginalLut ->InputChannels, Trans);

    // Create and insert the curves at the beginning
    if (!cmsPipelineInsertStage(OptimizedLUT, cmsAT_BEGIN, OptimizedPrelinMpe))
        goto Error;

    // Allocate the CLUT for result
    OptimizedCLUTmpe = cmsStageAllocCLut16bit(OriginalLut ->ContextID, nGridPoints, OriginalLut ->InputChannels, OriginalLut ->OutputChannels, NULL);

    // Add the CLUT to the destination LUT
    if (!cmsPipelineInsertStage(OptimizedLUT, cmsAT_END, OptimizedCLUTmpe))
        goto Error;

    // Resample the LUT
    if (!cmsStageSampleCLut16bit(OptimizedCLUTmpe, XFormSampler16, (void*) LutPlusCurves, 0)) goto Error;

    // Free resources
    for (t = 0; t < OriginalLut ->InputChannels; t++) {

        if (Trans[t]) cmsFreeToneCurve(Trans[t]);
        if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
    }

    cmsPipelineFree(LutPlusCurves);


    OptimizedPrelinCurves = _cmsStageGetPtrToCurveSet(OptimizedPrelinMpe);
    OptimizedPrelinCLUT   = (_cmsStageCLutData*) OptimizedCLUTmpe ->Data;

    // Set the evaluator if 8-bit
    if (_cmsFormatterIs8bit(*InputFormat)) {

        Prelin8Data* p8 = PrelinOpt8alloc(OptimizedLUT ->ContextID,
                                                OptimizedPrelinCLUT ->Params,
                                                OptimizedPrelinCurves);
        if (p8 == NULL) return FALSE;

        _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval8, (void*) p8, Prelin8free, Prelin8dup);

    }
    else
    {
        Prelin16Data* p16 = PrelinOpt16alloc(OptimizedLUT ->ContextID,
            OptimizedPrelinCLUT ->Params,
            3, OptimizedPrelinCurves, 3, NULL);
        if (p16 == NULL) return FALSE;

        _cmsPipelineSetOptimizationParameters(OptimizedLUT, PrelinEval16, (void*) p16, PrelinOpt16free, Prelin16dup);

    }

    // Don't fix white on absolute colorimetric
    if (Intent == INTENT_ABSOLUTE_COLORIMETRIC)
        *dwFlags |= cmsFLAGS_NOWHITEONWHITEFIXUP;

    if (!(*dwFlags & cmsFLAGS_NOWHITEONWHITEFIXUP)) {

        if (!FixWhiteMisalignment(OptimizedLUT, ColorSpace, OutputColorSpace)) {

            return FALSE;
        }
    }

    // And return the obtained LUT

    cmsPipelineFree(OriginalLut);
    *Lut = OptimizedLUT;
    return TRUE;

Error:

    for (t = 0; t < OriginalLut ->InputChannels; t++) {

        if (Trans[t]) cmsFreeToneCurve(Trans[t]);
        if (TransReverse[t]) cmsFreeToneCurve(TransReverse[t]);
    }

    if (LutPlusCurves != NULL) cmsPipelineFree(LutPlusCurves);
    if (OptimizedLUT != NULL) cmsPipelineFree(OptimizedLUT);

    return FALSE;

    cmsUNUSED_PARAMETER(Intent);
    cmsUNUSED_PARAMETER(lIsLinear);
}


// Curves optimizer ------------------------------------------------------------------------------------------------------------------

static
void CurvesFree(cmsContext ContextID, void* ptr)
{
     Curves16Data* Data = (Curves16Data*) ptr;
     cmsUInt32Number i;

     for (i=0; i < Data -> nCurves; i++) {

         _cmsFree(ContextID, Data ->Curves[i]);
     }

     _cmsFree(ContextID, Data ->Curves);
     _cmsFree(ContextID, ptr);
}

static
void* CurvesDup(cmsContext ContextID, const void* ptr)
{
    Curves16Data* Data = (Curves16Data*)_cmsDupMem(ContextID, ptr, sizeof(Curves16Data));
    cmsUInt32Number i;

    if (Data == NULL) return NULL;

    Data->Curves = (cmsUInt16Number**) _cmsDupMem(ContextID, Data->Curves, Data->nCurves * sizeof(cmsUInt16Number*));

    for (i=0; i < Data -> nCurves; i++) {
        Data->Curves[i] = (cmsUInt16Number*) _cmsDupMem(ContextID, Data->Curves[i], Data->nElements * sizeof(cmsUInt16Number));
    }

    return (void*) Data;
}

// Precomputes tables for 8-bit on input devicelink.
static
Curves16Data* CurvesAlloc(cmsContext ContextID, cmsUInt32Number nCurves, cmsUInt32Number nElements, cmsToneCurve** G)
{
    cmsUInt32Number i, j;
    Curves16Data* c16;

    c16 = (Curves16Data*)_cmsMallocZero(ContextID, sizeof(Curves16Data));
    if (c16 == NULL) return NULL;

    c16 ->nCurves = nCurves;
    c16 ->nElements = nElements;

    c16->Curves = (cmsUInt16Number**) _cmsCalloc(ContextID, nCurves, sizeof(cmsUInt16Number*));
    if (c16->Curves == NULL) {
        _cmsFree(ContextID, c16);
        return NULL;
    }

    for (i=0; i < nCurves; i++) {

        c16->Curves[i] = (cmsUInt16Number*) _cmsCalloc(ContextID, nElements, sizeof(cmsUInt16Number));

        if (c16->Curves[i] == NULL) {

            for (j=0; j < i; j++) {
                _cmsFree(ContextID, c16->Curves[j]);
            }
            _cmsFree(ContextID, c16->Curves);
            _cmsFree(ContextID, c16);
            return NULL;
        }

        if (nElements == 256U) {

            for (j=0; j < nElements; j++) {

                c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], FROM_8_TO_16(j));
            }
        }
        else {

            for (j=0; j < nElements; j++) {
                c16 ->Curves[i][j] = cmsEvalToneCurve16(G[i], (cmsUInt16Number) j);
            }
        }
    }

    return c16;
}

static
void FastEvaluateCurves8(CMSREGISTER const cmsUInt16Number In[],
                         CMSREGISTER cmsUInt16Number Out[],
                         CMSREGISTER const void* D)
{
    Curves16Data* Data = (Curves16Data*) D;
    int x;
    cmsUInt32Number i;

    for (i=0; i < Data ->nCurves; i++) {

         x = (In[i] >> 8);
         Out[i] = Data -> Curves[i][x];
    }
}


static
void FastEvaluateCurves16(CMSREGISTER const cmsUInt16Number In[],
                          CMSREGISTER cmsUInt16Number Out[],
                          CMSREGISTER const void* D)
{
    Curves16Data* Data = (Curves16Data*) D;
    cmsUInt32Number i;

    for (i=0; i < Data ->nCurves; i++) {
         Out[i] = Data -> Curves[i][In[i]];
    }
}


static
void FastIdentity16(CMSREGISTER const cmsUInt16Number In[],
                    CMSREGISTER cmsUInt16Number Out[],
                    CMSREGISTER const void* D)
{
    cmsPipeline* Lut = (cmsPipeline*) D;
    cmsUInt32Number i;

    for (i=0; i < Lut ->InputChannels; i++) {
         Out[i] = In[i];
    }
}


// If the target LUT holds only curves, the optimization procedure is to join all those
// curves together. That only works on curves and does not work on matrices.
static
cmsBool OptimizeByJoiningCurves(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
{
    cmsToneCurve** GammaTables = NULL;
    cmsFloat32Number InFloat[cmsMAXCHANNELS], OutFloat[cmsMAXCHANNELS];
    cmsUInt32Number i, j;
    cmsPipeline* Src = *Lut;
    cmsPipeline* Dest = NULL;
    cmsStage* mpe;
    cmsStage* ObtainedCurves = NULL;


    // This is a lossy optimization! does not apply in floating-point cases
    if (_cmsFormatterIsFloat(*InputFormat) || _cmsFormatterIsFloat(*OutputFormat)) return FALSE;

    //  Only curves in this LUT?
    for (mpe = cmsPipelineGetPtrToFirstStage(Src);
         mpe != NULL;
         mpe = cmsStageNext(mpe)) {
            if (cmsStageType(mpe) != cmsSigCurveSetElemType) return FALSE;
    }

    // Allocate an empty LUT
    Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
    if (Dest == NULL) return FALSE;

    // Create target curves
    GammaTables = (cmsToneCurve**) _cmsCalloc(Src ->ContextID, Src ->InputChannels, sizeof(cmsToneCurve*));
    if (GammaTables == NULL) goto Error;

    for (i=0; i < Src ->InputChannels; i++) {
        GammaTables[i] = cmsBuildTabulatedToneCurve16(Src ->ContextID, PRELINEARIZATION_POINTS, NULL);
        if (GammaTables[i] == NULL) goto Error;
    }

    // Compute 16 bit result by using floating point
    for (i=0; i < PRELINEARIZATION_POINTS; i++) {

        for (j=0; j < Src ->InputChannels; j++)
            InFloat[j] = (cmsFloat32Number) ((cmsFloat64Number) i / (PRELINEARIZATION_POINTS - 1));

        cmsPipelineEvalFloat(InFloat, OutFloat, Src);

        for (j=0; j < Src ->InputChannels; j++)
            GammaTables[j] -> Table16[i] = _cmsQuickSaturateWord(OutFloat[j] * 65535.0);
    }

    ObtainedCurves = cmsStageAllocToneCurves(Src ->ContextID, Src ->InputChannels, GammaTables);
    if (ObtainedCurves == NULL) goto Error;

    for (i=0; i < Src ->InputChannels; i++) {
        cmsFreeToneCurve(GammaTables[i]);
        GammaTables[i] = NULL;
    }

    if (GammaTables != NULL) {
        _cmsFree(Src->ContextID, GammaTables);
        GammaTables = NULL;
    }

    // Maybe the curves are linear at the end
    if (!AllCurvesAreLinear(ObtainedCurves)) {
       _cmsStageToneCurvesData* Data;

        if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, ObtainedCurves))
            goto Error;
        Data = (_cmsStageToneCurvesData*) cmsStageData(ObtainedCurves);
        ObtainedCurves = NULL;

        // If the curves are to be applied in 8 bits, we can save memory
        if (_cmsFormatterIs8bit(*InputFormat)) {
             Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 256, Data ->TheCurves);

             if (c16 == NULL) goto Error;
             *dwFlags |= cmsFLAGS_NOCACHE;
            _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves8, c16, CurvesFree, CurvesDup);

        }
        else {
             Curves16Data* c16 = CurvesAlloc(Dest ->ContextID, Data ->nCurves, 65536, Data ->TheCurves);

             if (c16 == NULL) goto Error;
             *dwFlags |= cmsFLAGS_NOCACHE;
            _cmsPipelineSetOptimizationParameters(Dest, FastEvaluateCurves16, c16, CurvesFree, CurvesDup);
        }
    }
    else {

        // LUT optimizes to nothing. Set the identity LUT
        cmsStageFree(ObtainedCurves);
        ObtainedCurves = NULL;

        if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageAllocIdentity(Dest ->ContextID, Src ->InputChannels)))
            goto Error;

        *dwFlags |= cmsFLAGS_NOCACHE;
        _cmsPipelineSetOptimizationParameters(Dest, FastIdentity16, (void*) Dest, NULL, NULL);
    }

    // We are done.
    cmsPipelineFree(Src);
    *Lut = Dest;
    return TRUE;

Error:

    if (ObtainedCurves != NULL) cmsStageFree(ObtainedCurves);
    if (GammaTables != NULL) {
        for (i=0; i < Src ->InputChannels; i++) {
            if (GammaTables[i] != NULL) cmsFreeToneCurve(GammaTables[i]);
        }

        _cmsFree(Src ->ContextID, GammaTables);
    }

    if (Dest != NULL) cmsPipelineFree(Dest);
    return FALSE;

    cmsUNUSED_PARAMETER(Intent);
    cmsUNUSED_PARAMETER(InputFormat);
    cmsUNUSED_PARAMETER(OutputFormat);
    cmsUNUSED_PARAMETER(dwFlags);
}

// -------------------------------------------------------------------------------------------------------------------------------------
// LUT is Shaper - Matrix - Matrix - Shaper, which is very frequent when combining two matrix-shaper profiles


static
void  FreeMatShaper(cmsContext ContextID, void* Data)
{
    if (Data != NULL) _cmsFree(ContextID, Data);
}

static
void* DupMatShaper(cmsContext ContextID, const void* Data)
{
    return _cmsDupMem(ContextID, Data, sizeof(MatShaper8Data));
}


// A fast matrix-shaper evaluator for 8 bits. This is a bit tricky since I'm using 1.14 signed fixed point
// to accomplish some performance. Actually it takes 256x3 16 bits tables and 16385 x 3 tables of 8 bits,
// in total about 50K, and the performance boost is huge!
static CMS_NO_SANITIZE
void MatShaperEval16(CMSREGISTER const cmsUInt16Number In[],
                     CMSREGISTER cmsUInt16Number Out[],
                     CMSREGISTER const void* D)
{
    MatShaper8Data* p = (MatShaper8Data*) D;
    cmsS1Fixed14Number l1, l2, l3, r, g, b;
    cmsUInt32Number ri, gi, bi;

    // In this case (and only in this case!) we can use this simplification since
    // In[] is assured to come from a 8 bit number. (a << 8 | a)
    ri = In[0] & 0xFFU;
    gi = In[1] & 0xFFU;
    bi = In[2] & 0xFFU;

    // Across first shaper, which also converts to 1.14 fixed point
    r = p->Shaper1R[ri];
    g = p->Shaper1G[gi];
    b = p->Shaper1B[bi];

    // Evaluate the matrix in 1.14 fixed point
    l1 =  (p->Mat[0][0] * r + p->Mat[0][1] * g + p->Mat[0][2] * b + p->Off[0] + 0x2000) >> 14;
    l2 =  (p->Mat[1][0] * r + p->Mat[1][1] * g + p->Mat[1][2] * b + p->Off[1] + 0x2000) >> 14;
    l3 =  (p->Mat[2][0] * r + p->Mat[2][1] * g + p->Mat[2][2] * b + p->Off[2] + 0x2000) >> 14;

    // Now we have to clip to 0..1.0 range
    ri = (l1 < 0) ? 0 : ((l1 > 16384) ? 16384U : (cmsUInt32Number) l1);
    gi = (l2 < 0) ? 0 : ((l2 > 16384) ? 16384U : (cmsUInt32Number) l2);
    bi = (l3 < 0) ? 0 : ((l3 > 16384) ? 16384U : (cmsUInt32Number) l3);

    // And across second shaper,
    Out[0] = p->Shaper2R[ri];
    Out[1] = p->Shaper2G[gi];
    Out[2] = p->Shaper2B[bi];

}

// This table converts from 8 bits to 1.14 after applying the curve
static
void FillFirstShaper(cmsS1Fixed14Number* Table, cmsToneCurve* Curve)
{
    int i;
    cmsFloat32Number R, y;

    for (i=0; i < 256; i++) {

        R   = (cmsFloat32Number) (i / 255.0);
        y   = cmsEvalToneCurveFloat(Curve, R);

        if (y < 131072.0)
            Table[i] = DOUBLE_TO_1FIXED14(y);
        else
            Table[i] = 0x7fffffff;
    }
}

// This table converts form 1.14 (being 0x4000 the last entry) to 8 bits after applying the curve
static
void FillSecondShaper(cmsUInt16Number* Table, cmsToneCurve* Curve, cmsBool Is8BitsOutput)
{
    int i;
    cmsFloat32Number R, Val;

    for (i=0; i < 16385; i++) {

        R   = (cmsFloat32Number) (i / 16384.0);
        Val = cmsEvalToneCurveFloat(Curve, R);    // Val comes 0..1.0

        if (Val < 0)
            Val = 0;

        if (Val > 1.0)
            Val = 1.0;

        if (Is8BitsOutput) {

            // If 8 bits output, we can optimize further by computing the / 257 part.
            // first we compute the resulting byte and then we store the byte times
            // 257. This quantization allows to round very quick by doing a >> 8, but
            // since the low byte is always equal to msb, we can do a & 0xff and this works!
            cmsUInt16Number w = _cmsQuickSaturateWord(Val * 65535.0);
            cmsUInt8Number  b = FROM_16_TO_8(w);

            Table[i] = FROM_8_TO_16(b);
        }
        else Table[i]  = _cmsQuickSaturateWord(Val * 65535.0);
    }
}

// Compute the matrix-shaper structure
static
cmsBool SetMatShaper(cmsPipeline* Dest, cmsToneCurve* Curve1[3], cmsMAT3* Mat, cmsVEC3* Off, cmsToneCurve* Curve2[3], cmsUInt32Number* OutputFormat)
{
    MatShaper8Data* p;
    int i, j;
    cmsBool Is8Bits = _cmsFormatterIs8bit(*OutputFormat);

    // Allocate a big chuck of memory to store precomputed tables
    p = (MatShaper8Data*) _cmsMalloc(Dest ->ContextID, sizeof(MatShaper8Data));
    if (p == NULL) return FALSE;

    p -> ContextID = Dest -> ContextID;

    // Precompute tables
    FillFirstShaper(p ->Shaper1R, Curve1[0]);
    FillFirstShaper(p ->Shaper1G, Curve1[1]);
    FillFirstShaper(p ->Shaper1B, Curve1[2]);

    FillSecondShaper(p ->Shaper2R, Curve2[0], Is8Bits);
    FillSecondShaper(p ->Shaper2G, Curve2[1], Is8Bits);
    FillSecondShaper(p ->Shaper2B, Curve2[2], Is8Bits);

    // Convert matrix to nFixed14. Note that those values may take more than 16 bits 
    for (i=0; i < 3; i++) {
        for (j=0; j < 3; j++) {
            p ->Mat[i][j] = DOUBLE_TO_1FIXED14(Mat->v[i].n[j]);
        }
    }

    for (i=0; i < 3; i++) {

        if (Off == NULL) {
            p ->Off[i] = 0;
        }
        else {
            p ->Off[i] = DOUBLE_TO_1FIXED14(Off->n[i]);
        }
    }

    // Mark as optimized for faster formatter
    if (Is8Bits)
        *OutputFormat |= OPTIMIZED_SH(1);

    // Fill function pointers
    _cmsPipelineSetOptimizationParameters(Dest, MatShaperEval16, (void*) p, FreeMatShaper, DupMatShaper);
    return TRUE;
}

//  8 bits on input allows matrix-shaper boot up to 25 Mpixels per second on RGB. That's fast!
static
cmsBool OptimizeMatrixShaper(cmsPipeline** Lut, cmsUInt32Number Intent, cmsUInt32Number* InputFormat, cmsUInt32Number* OutputFormat, cmsUInt32Number* dwFlags)
{
       cmsStage* Curve1, *Curve2;
       cmsStage* Matrix1, *Matrix2;
       cmsMAT3 res;
       cmsBool IdentityMat;
       cmsPipeline* Dest, *Src;
       cmsFloat64Number* Offset;

       // Only works on RGB to RGB
       if (T_CHANNELS(*InputFormat) != 3 || T_CHANNELS(*OutputFormat) != 3) return FALSE;

       // Only works on 8 bit input
       if (!_cmsFormatterIs8bit(*InputFormat)) return FALSE;

       // Seems suitable, proceed
       Src = *Lut;

       // Check for:
       // 
       //    shaper-matrix-matrix-shaper 
       //    shaper-matrix-shaper
       // 
       // Both of those constructs are possible (first because abs. colorimetric). 
       // additionally, In the first case, the input matrix offset should be zero.

       IdentityMat = FALSE;
       if (cmsPipelineCheckAndRetreiveStages(Src, 4,
              cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
              &Curve1, &Matrix1, &Matrix2, &Curve2)) {

              // Get both matrices
              _cmsStageMatrixData* Data1 = (_cmsStageMatrixData*)cmsStageData(Matrix1);
              _cmsStageMatrixData* Data2 = (_cmsStageMatrixData*)cmsStageData(Matrix2);

              // Only RGB to RGB
              if (Matrix1->InputChannels != 3 || Matrix1->OutputChannels != 3 ||
                  Matrix2->InputChannels != 3 || Matrix2->OutputChannels != 3) return FALSE;

              // Input offset should be zero
              if (Data1->Offset != NULL) return FALSE;

              // Multiply both matrices to get the result
              _cmsMAT3per(&res, (cmsMAT3*)Data2->Double, (cmsMAT3*)Data1->Double);

              // Only 2nd matrix has offset, or it is zero 
              Offset = Data2->Offset;

              // Now the result is in res + Data2 -> Offset. Maybe is a plain identity?
              if (_cmsMAT3isIdentity(&res) && Offset == NULL) {

                     // We can get rid of full matrix
                     IdentityMat = TRUE;
              }

       }
       else {

              if (cmsPipelineCheckAndRetreiveStages(Src, 3,
                     cmsSigCurveSetElemType, cmsSigMatrixElemType, cmsSigCurveSetElemType,
                     &Curve1, &Matrix1, &Curve2)) {

                     _cmsStageMatrixData* Data = (_cmsStageMatrixData*)cmsStageData(Matrix1);

                     // Copy the matrix to our result
                     memcpy(&res, Data->Double, sizeof(res));

                     // Preserve the Odffset (may be NULL as a zero offset)
                     Offset = Data->Offset;

                     if (_cmsMAT3isIdentity(&res) && Offset == NULL) {

                            // We can get rid of full matrix
                            IdentityMat = TRUE;
                     }
              }
              else
                     return FALSE; // Not optimizeable this time

       }

      // Allocate an empty LUT
    Dest =  cmsPipelineAlloc(Src ->ContextID, Src ->InputChannels, Src ->OutputChannels);
    if (!Dest) return FALSE;

    // Assamble the new LUT
    if (!cmsPipelineInsertStage(Dest, cmsAT_BEGIN, cmsStageDup(Curve1)))
        goto Error;

    if (!IdentityMat) {

           if (!cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageAllocMatrix(Dest->ContextID, 3, 3, (const cmsFloat64Number*)&res, Offset)))
                  goto Error;
    }

    if (!cmsPipelineInsertStage(Dest, cmsAT_END, cmsStageDup(Curve2)))
        goto Error;

    // If identity on matrix, we can further optimize the curves, so call the join curves routine
    if (IdentityMat) {

        OptimizeByJoiningCurves(&Dest, Intent, InputFormat, OutputFormat, dwFlags);
    }
    else {
        _cmsStageToneCurvesData* mpeC1 = (_cmsStageToneCurvesData*) cmsStageData(Curve1);
        _cmsStageToneCurvesData* mpeC2 = (_cmsStageToneCurvesData*) cmsStageData(Curve2);

        // In this particular optimization, cache does not help as it takes more time to deal with
        // the cache that with the pixel handling
        *dwFlags |= cmsFLAGS_NOCACHE;

        // Setup the optimizarion routines
        SetMatShaper(Dest, mpeC1 ->TheCurves, &res, (cmsVEC3*) Offset, mpeC2->TheCurves, OutputFormat);
    }

    cmsPipelineFree(Src);
    *Lut = Dest;
    return TRUE;
Error:
    // Leave Src unchanged
    cmsPipelineFree(Dest);
    return FALSE;
}


// -------------------------------------------------------------------------------------------------------------------------------------
// Optimization plug-ins

// List of optimizations
typedef struct _cmsOptimizationCollection_st {

    _cmsOPToptimizeFn  OptimizePtr;

    struct _cmsOptimizationCollection_st *Next;

} _cmsOptimizationCollection;


// The built-in list. We currently implement 4 types of optimizations. Joining of curves, matrix-shaper, linearization and resampling
static _cmsOptimizationCollection DefaultOptimization[] = {

    { OptimizeByJoiningCurves,            &DefaultOptimization[1] },
    { OptimizeMatrixShaper,               &DefaultOptimization[2] },
    { OptimizeByComputingLinearization,   &DefaultOptimization[3] },
    { OptimizeByResampling,               NULL }
};

// The linked list head
_cmsOptimizationPluginChunkType _cmsOptimizationPluginChunk = { NULL };


// Duplicates the zone of memory used by the plug-in in the new context
static
void DupPluginOptimizationList(struct _cmsContext_struct* ctx, 
                               const struct _cmsContext_struct* src)
{
   _cmsOptimizationPluginChunkType newHead = { NULL };
   _cmsOptimizationCollection*  entry;
   _cmsOptimizationCollection*  Anterior = NULL;
   _cmsOptimizationPluginChunkType* head = (_cmsOptimizationPluginChunkType*) src->chunks[OptimizationPlugin];

    _cmsAssert(ctx != NULL);
    _cmsAssert(head != NULL);

    // Walk the list copying all nodes
   for (entry = head->OptimizationCollection;
        entry != NULL;
        entry = entry ->Next) {

            _cmsOptimizationCollection *newEntry = ( _cmsOptimizationCollection *) _cmsSubAllocDup(ctx ->MemPool, entry, sizeof(_cmsOptimizationCollection));
   
            if (newEntry == NULL) 
                return;

            // We want to keep the linked list order, so this is a little bit tricky
            newEntry -> Next = NULL;
            if (Anterior)
                Anterior -> Next = newEntry;
     
            Anterior = newEntry;

            if (newHead.OptimizationCollection == NULL)
                newHead.OptimizationCollection = newEntry;
    }

  ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx->MemPool, &newHead, sizeof(_cmsOptimizationPluginChunkType));
}

void  _cmsAllocOptimizationPluginChunk(struct _cmsContext_struct* ctx, 
                                         const struct _cmsContext_struct* src)
{
  if (src != NULL) {

        // Copy all linked list
       DupPluginOptimizationList(ctx, src);
    }
    else {
        static _cmsOptimizationPluginChunkType OptimizationPluginChunkType = { NULL };
        ctx ->chunks[OptimizationPlugin] = _cmsSubAllocDup(ctx ->MemPool, &OptimizationPluginChunkType, sizeof(_cmsOptimizationPluginChunkType));
    }
}


// Register new ways to optimize
cmsBool  _cmsRegisterOptimizationPlugin(cmsContext ContextID, cmsPluginBase* Data)
{
    cmsPluginOptimization* Plugin = (cmsPluginOptimization*) Data;
    _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
    _cmsOptimizationCollection* fl;

    if (Data == NULL) {

        ctx->OptimizationCollection = NULL;
        return TRUE;
    }

    // Optimizer callback is required
    if (Plugin ->OptimizePtr == NULL) return FALSE;

    fl = (_cmsOptimizationCollection*) _cmsPluginMalloc(ContextID, sizeof(_cmsOptimizationCollection));
    if (fl == NULL) return FALSE;

    // Copy the parameters
    fl ->OptimizePtr = Plugin ->OptimizePtr;

    // Keep linked list
    fl ->Next = ctx->OptimizationCollection;

    // Set the head
    ctx ->OptimizationCollection = fl;

    // All is ok
    return TRUE;
}

// The entry point for LUT optimization
cmsBool CMSEXPORT _cmsOptimizePipeline(cmsContext ContextID,
                             cmsPipeline**    PtrLut,
                             cmsUInt32Number  Intent,
                             cmsUInt32Number* InputFormat,
                             cmsUInt32Number* OutputFormat,
                             cmsUInt32Number* dwFlags)
{
    _cmsOptimizationPluginChunkType* ctx = ( _cmsOptimizationPluginChunkType*) _cmsContextGetClientChunk(ContextID, OptimizationPlugin);
    _cmsOptimizationCollection* Opts;
    cmsBool AnySuccess = FALSE;
    cmsStage* mpe;

    // A CLUT is being asked, so force this specific optimization
    if (*dwFlags & cmsFLAGS_FORCE_CLUT) {

        PreOptimize(*PtrLut);
        return OptimizeByResampling(PtrLut, Intent, InputFormat, OutputFormat, dwFlags);
    }

    // Anything to optimize?
    if ((*PtrLut) ->Elements == NULL) {
        _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
        return TRUE;
    }

    // Named color pipelines cannot be optimized 
    for (mpe = cmsPipelineGetPtrToFirstStage(*PtrLut);
        mpe != NULL;
        mpe = cmsStageNext(mpe)) {
        if (cmsStageType(mpe) == cmsSigNamedColorElemType) return FALSE;
    }

    // Try to get rid of identities and trivial conversions.
    AnySuccess = PreOptimize(*PtrLut);

    // After removal do we end with an identity?
    if ((*PtrLut) ->Elements == NULL) {
        _cmsPipelineSetOptimizationParameters(*PtrLut, FastIdentity16, (void*) *PtrLut, NULL, NULL);
        return TRUE;
    }

    // Do not optimize, keep all precision
    if (*dwFlags & cmsFLAGS_NOOPTIMIZE)
        return FALSE;

    // Try plug-in optimizations 
    for (Opts = ctx->OptimizationCollection;
         Opts != NULL;
         Opts = Opts ->Next) {

            // If one schema succeeded, we are done
            if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {

                return TRUE;    // Optimized!
            }
    }

   // Try built-in optimizations 
    for (Opts = DefaultOptimization;
         Opts != NULL;
         Opts = Opts ->Next) {

            if (Opts ->OptimizePtr(PtrLut, Intent, InputFormat, OutputFormat, dwFlags)) {

                return TRUE;  
            }
    }

    // Only simple optimizations succeeded
    return AnySuccess;
}