Coverage Report

Created: 2025-06-10 07:27

/src/ghostpdl/base/gsfunc0.c
Line
Count
Source (jump to first uncovered line)
1
/* Copyright (C) 2001-2023 Artifex Software, Inc.
2
   All Rights Reserved.
3
4
   This software is provided AS-IS with no warranty, either express or
5
   implied.
6
7
   This software is distributed under license and may not be copied,
8
   modified or distributed except as expressly authorized under the terms
9
   of the license contained in the file LICENSE in this distribution.
10
11
   Refer to licensing information at http://www.artifex.com or contact
12
   Artifex Software, Inc.,  39 Mesa Street, Suite 108A, San Francisco,
13
   CA 94129, USA, for further information.
14
*/
15
16
17
/* Implementation of FunctionType 0 (Sampled) Functions */
18
#include "math_.h"
19
#include "gx.h"
20
#include "gserrors.h"
21
#include "gsfunc0.h"
22
#include "gsparam.h"
23
#include "gxfarith.h"
24
#include "gxfunc.h"
25
#include "stream.h"
26
#include "gsccolor.h"           /* Only for GS_CLIENT_COLOR_MAX_COMPONENTS */
27
28
#define POLE_CACHE_DEBUG 0      /* A temporary development technology need.
29
                                   Remove after the beta testing. */
30
98
#define POLE_CACHE_GENERIC_1D 1 /* A temporary development technology need.
31
                                   Didn't decide yet - see fn_Sd_evaluate_cubic_cached_1d. */
32
148
#define POLE_CACHE_IGNORE 0     /* A temporary development technology need.
33
                                   Remove after the beta testing. */
34
35
12.6k
#define MAX_FAST_COMPS 8
36
37
typedef struct gs_function_Sd_s {
38
    gs_function_head_t head;
39
    gs_function_Sd_params_t params;
40
} gs_function_Sd_t;
41
42
/* GC descriptor */
43
private_st_function_Sd();
44
static
45
50
ENUM_PTRS_WITH(function_Sd_enum_ptrs, gs_function_Sd_t *pfn)
46
20
{
47
20
    index -= 6;
48
20
    if (index < st_data_source_max_ptrs)
49
5
        return ENUM_USING(st_data_source, &pfn->params.DataSource,
50
20
                          sizeof(pfn->params.DataSource), index);
51
15
    return ENUM_USING_PREFIX(st_function, st_data_source_max_ptrs);
52
20
}
53
20
ENUM_PTR3(0, gs_function_Sd_t, params.Encode, params.Decode, params.Size);
54
50
ENUM_PTR3(3, gs_function_Sd_t, params.pole, params.array_step, params.stream_step);
55
50
ENUM_PTRS_END
56
static
57
5
RELOC_PTRS_WITH(function_Sd_reloc_ptrs, gs_function_Sd_t *pfn)
58
5
{
59
5
    RELOC_PREFIX(st_function);
60
5
    RELOC_USING(st_data_source, &pfn->params.DataSource,
61
5
                sizeof(pfn->params.DataSource));
62
5
    RELOC_PTR3(gs_function_Sd_t, params.Encode, params.Decode, params.Size);
63
5
    RELOC_PTR3(gs_function_Sd_t, params.pole, params.array_step, params.stream_step);
64
5
}
65
5
RELOC_PTRS_END
66
67
/* Define the maximum plausible number of inputs and outputs */
68
/* for a Sampled function. */
69
#ifndef GS_CLIENT_SAMPLED_FN_MAX_COMPONENTS   /* Allow override with XCFLAGS */
70
1.83k
#  define max_Sd_m GS_CLIENT_COLOR_MAX_COMPONENTS
71
#  define max_Sd_n GS_CLIENT_COLOR_MAX_COMPONENTS
72
#else
73
#  define max_Sd_m GS_CLIENT_SAMPLED_FN_MAX_COMPONENTS
74
#  define max_Sd_n GS_CLIENT_SAMPLED_FN_MAX_COMPONENTS
75
#endif
76
77
/* Get one set of sample values. */
78
#define SETUP_SAMPLES(bps, nbytes)\
79
8.79M
        int n = pfn->params.n;\
80
8.79M
        byte buf[max_Sd_n * ((bps + 7) >> 3)];\
81
8.79M
        const byte *p;\
82
8.79M
        int i;\
83
8.79M
\
84
8.79M
        data_source_access(&pfn->params.DataSource, offset >> 3,\
85
8.79M
                           nbytes, buf, &p)
86
87
static int
88
fn_gets_1(const gs_function_Sd_t * pfn, ulong offset, uint * samples)
89
0
{
90
0
    SETUP_SAMPLES(1, ((offset & 7) + n + 7) >> 3);
91
0
    for (i = 0; i < n; ++i) {
92
0
        samples[i] = (*p >> (~offset & 7)) & 1;
93
0
        if (!(++offset & 7))
94
0
            p++;
95
0
    }
96
0
    return 0;
97
0
}
98
static int
99
fn_gets_2(const gs_function_Sd_t * pfn, ulong offset, uint * samples)
100
0
{
101
0
    SETUP_SAMPLES(2, (((offset & 7) >> 1) + n + 3) >> 2);
102
0
    for (i = 0; i < n; ++i) {
103
0
        samples[i] = (*p >> (6 - (offset & 7))) & 3;
104
0
        if (!((offset += 2) & 7))
105
0
            p++;
106
0
    }
107
0
    return 0;
108
0
}
109
static int
110
fn_gets_4(const gs_function_Sd_t * pfn, ulong offset, uint * samples)
111
0
{
112
0
    SETUP_SAMPLES(4, (((offset & 7) >> 2) + n + 1) >> 1);
113
0
    for (i = 0; i < n; ++i) {
114
0
        samples[i] = ((offset ^= 4) & 4 ? *p >> 4 : *p++ & 0xf);
115
0
    }
116
0
    return 0;
117
0
}
118
static int
119
fn_gets_8(const gs_function_Sd_t * pfn, ulong offset, uint * samples)
120
8.79M
{
121
8.79M
    SETUP_SAMPLES(8, n);
122
18.1M
    for (i = 0; i < n; ++i) {
123
9.38M
        samples[i] = *p++;
124
9.38M
    }
125
8.79M
    return 0;
126
8.79M
}
127
static int
128
fn_gets_12(const gs_function_Sd_t * pfn, ulong offset, uint * samples)
129
0
{
130
0
    SETUP_SAMPLES(12, (((offset & 7) >> 2) + 3 * n + 1) >> 1);
131
0
    for (i = 0; i < n; ++i) {
132
0
        if (offset & 4)
133
0
            samples[i] = ((*p & 0xf) << 8) + p[1], p += 2;
134
0
        else
135
0
            samples[i] = (*p << 4) + (p[1] >> 4), p++;
136
0
        offset ^= 4;
137
0
    }
138
0
    return 0;
139
0
}
140
static int
141
fn_gets_16(const gs_function_Sd_t * pfn, ulong offset, uint * samples)
142
3.08k
{
143
3.08k
    SETUP_SAMPLES(16, n * 2);
144
6.23k
    for (i = 0; i < n; ++i) {
145
3.15k
        samples[i] = (*p << 8) + p[1];
146
3.15k
        p += 2;
147
3.15k
    }
148
3.08k
    return 0;
149
3.08k
}
150
static int
151
fn_gets_24(const gs_function_Sd_t * pfn, ulong offset, uint * samples)
152
0
{
153
0
    SETUP_SAMPLES(24, n * 3);
154
0
    for (i = 0; i < n; ++i) {
155
0
        samples[i] = (*p << 16) + (p[1] << 8) + p[2];
156
0
        p += 3;
157
0
    }
158
0
    return 0;
159
0
}
160
static int
161
fn_gets_32(const gs_function_Sd_t * pfn, ulong offset, uint * samples)
162
0
{
163
0
    SETUP_SAMPLES(32, n * 4);
164
0
    for (i = 0; i < n; ++i) {
165
0
        samples[i] = (*p << 24) + (p[1] << 16) + (p[2] << 8) + p[3];
166
0
        p += 4;
167
0
    }
168
0
    return 0;
169
0
}
170
171
static int (*const fn_get_samples[]) (const gs_function_Sd_t * pfn,
172
                                       ulong offset, uint * samples) =
173
{
174
    0, fn_gets_1, fn_gets_2, 0, fn_gets_4, 0, 0, 0,
175
        fn_gets_8, 0, 0, 0, fn_gets_12, 0, 0, 0,
176
        fn_gets_16, 0, 0, 0, 0, 0, 0, 0,
177
        fn_gets_24, 0, 0, 0, 0, 0, 0, 0,
178
        fn_gets_32
179
};
180
181
/*
182
 * Compute a value by cubic interpolation.
183
 * f[] = f(0), f(1), f(2), f(3); 1 < x < 2.
184
 * The formula is derived from those presented in
185
 * http://www.cs.uwa.edu.au/undergraduate/units/233.413/Handouts/Lecture04.html
186
 * (thanks to Raph Levien for the reference).
187
 */
188
static double
189
interpolate_cubic(double x, double f0, double f1, double f2, double f3)
190
0
{
191
    /*
192
     * The parameter 'a' affects the contribution of the high-frequency
193
     * components.  The abovementioned source suggests a = -0.5.
194
     */
195
0
#define a (-0.5)
196
0
#define SQR(v) ((v) * (v))
197
0
#define CUBE(v) ((v) * (v) * (v))
198
0
    const double xm1 = x - 1, m2x = 2 - x, m3x = 3 - x;
199
0
    const double c =
200
0
        (a * CUBE(x) - 5 * a * SQR(x) + 8 * a * x - 4 * a) * f0 +
201
0
        ((a+2) * CUBE(xm1) - (a+3) * SQR(xm1) + 1) * f1 +
202
0
        ((a+2) * CUBE(m2x) - (a+3) * SQR(m2x) + 1) * f2 +
203
0
        (a * CUBE(m3x) - 5 * a * SQR(m3x) + 8 * a * m3x - 4 * a) * f3;
204
205
0
    if_debug6('~', "[~](%g, %g, %g, %g)order3(%g) => %g\n",
206
0
              f0, f1, f2, f3, x, c);
207
0
    return c;
208
0
#undef a
209
0
#undef SQR
210
0
#undef CUBE
211
0
}
212
213
/*
214
 * Compute a value by quadratic interpolation.
215
 * f[] = f(0), f(1), f(2); 0 < x < 1.
216
 *
217
 * We used to use a quadratic formula for this, derived from
218
 * f(0) = f0, f(1) = f1, f'(1) = (f2 - f0) / 2, but now we
219
 * match what we believe is Acrobat Reader's behavior.
220
 */
221
static inline double
222
interpolate_quadratic(double x, double f0, double f1, double f2)
223
0
{
224
0
    return interpolate_cubic(x + 1, f0, f0, f1, f2);
225
0
}
226
227
/* Calculate a result by multicubic interpolation. */
228
static void
229
fn_interpolate_cubic(const gs_function_Sd_t *pfn, const float *fparts,
230
                     const int *iparts, const ulong *factors,
231
                     float *samples, ulong offset, int m)
232
0
{
233
0
    int j;
234
235
0
top:
236
0
    if (m == 0) {
237
0
        uint sdata[max_Sd_n];
238
239
0
        (*fn_get_samples[pfn->params.BitsPerSample])(pfn, offset, sdata);
240
0
        for (j = pfn->params.n - 1; j >= 0; --j)
241
0
            samples[j] = (float)sdata[j];
242
0
    } else {
243
0
        float fpart = *fparts++;
244
0
        int ipart = *iparts++;
245
0
        ulong delta = *factors++;
246
0
        int size = pfn->params.Size[pfn->params.m - m];
247
0
        float samples1[max_Sd_n], samplesm1[max_Sd_n], samples2[max_Sd_n];
248
249
0
        --m;
250
0
        if (is_fzero(fpart))
251
0
            goto top;
252
0
        fn_interpolate_cubic(pfn, fparts, iparts, factors, samples,
253
0
                             offset, m);
254
0
        fn_interpolate_cubic(pfn, fparts, iparts, factors, samples1,
255
0
                             offset + delta, m);
256
        /* Ensure we don't try to access out of bounds. */
257
        /*
258
         * If size == 1, the only possible value for ipart and fpart is
259
         * 0, so we've already handled this case.
260
         */
261
0
        if (size == 2) { /* ipart = 0 */
262
            /* Use linear interpolation. */
263
0
            for (j = pfn->params.n - 1; j >= 0; --j)
264
0
                samples[j] += (samples1[j] - samples[j]) * fpart;
265
0
            return;
266
0
        }
267
0
        if (ipart == 0) {
268
            /* Use quadratic interpolation. */
269
0
            fn_interpolate_cubic(pfn, fparts, iparts, factors,
270
0
                                 samples2, offset + delta * 2, m);
271
0
            for (j = pfn->params.n - 1; j >= 0; --j)
272
0
                samples[j] =
273
0
                    interpolate_quadratic(fpart, samples[j],
274
0
                                          samples1[j], samples2[j]);
275
0
            return;
276
0
        }
277
        /* At this point we know ipart > 0, size >= 3. */
278
0
        fn_interpolate_cubic(pfn, fparts, iparts, factors, samplesm1,
279
0
                             offset - delta, m);
280
0
        if (ipart == size - 2) {
281
            /* Use quadratic interpolation. */
282
0
            for (j = pfn->params.n - 1; j >= 0; --j)
283
0
                samples[j] =
284
0
                    interpolate_quadratic(1 - fpart, samples1[j],
285
0
                                          samples[j], samplesm1[j]);
286
0
            return;
287
0
        }
288
        /* Now we know 0 < ipart < size - 2, size > 3. */
289
0
        fn_interpolate_cubic(pfn, fparts, iparts, factors,
290
0
                             samples2, offset + delta * 2, m);
291
0
        for (j = pfn->params.n - 1; j >= 0; --j)
292
0
            samples[j] =
293
0
                interpolate_cubic(fpart + 1, samplesm1[j], samples[j],
294
0
                                  samples1[j], samples2[j]);
295
0
    }
296
0
}
297
298
/* Calculate a result by multilinear interpolation. */
299
static void
300
fn_interpolate_linear(const gs_function_Sd_t *pfn, const float *fparts,
301
                 const ulong *factors, float *samples, ulong offset, int m)
302
255k
{
303
255k
    int j;
304
305
271k
top:
306
271k
    if (m == 0) {
307
175k
        uint sdata[max_Sd_n];
308
309
175k
        (*fn_get_samples[pfn->params.BitsPerSample])(pfn, offset, sdata);
310
570k
        for (j = pfn->params.n - 1; j >= 0; --j)
311
395k
            samples[j] = (float)sdata[j];
312
175k
    } else {
313
95.4k
        float fpart = *fparts++;
314
95.4k
        float samples1[max_Sd_n];
315
316
95.4k
        if (is_fzero(fpart)) {
317
15.1k
            ++factors;
318
15.1k
            --m;
319
15.1k
            goto top;
320
15.1k
        }
321
80.2k
        fn_interpolate_linear(pfn, fparts, factors + 1, samples,
322
80.2k
                              offset, m - 1);
323
80.2k
        fn_interpolate_linear(pfn, fparts, factors + 1, samples1,
324
80.2k
                              offset + *factors, m - 1);
325
267k
        for (j = pfn->params.n - 1; j >= 0; --j)
326
187k
            samples[j] += (samples1[j] - samples[j]) * fpart;
327
80.2k
    }
328
271k
}
329
330
static inline double
331
fn_Sd_encode(const gs_function_Sd_t *pfn, int i, double sample)
332
9.20M
{
333
9.20M
    float d0, d1, r0, r1;
334
9.20M
    double value;
335
9.20M
    int bps = pfn->params.BitsPerSample;
336
    /* x86 machines have problems with shifts if bps >= 32 */
337
9.20M
    uint max_samp = (bps < (sizeof(uint) * 8)) ? ((1 << bps) - 1) : max_uint;
338
339
9.20M
    if (pfn->params.Range)
340
9.20M
        r0 = pfn->params.Range[2 * i], r1 = pfn->params.Range[2 * i + 1];
341
0
    else
342
0
        r0 = 0, r1 = (float)max_samp;
343
9.20M
    if (pfn->params.Decode)
344
727k
        d0 = pfn->params.Decode[2 * i], d1 = pfn->params.Decode[2 * i + 1];
345
8.47M
    else
346
8.47M
        d0 = r0, d1 = r1;
347
348
9.20M
    value = sample * (d1 - d0) / max_samp + d0;
349
9.20M
    if (value < r0)
350
0
        value = r0;
351
9.20M
    else if (value > r1)
352
0
        value = r1;
353
9.20M
    return value;
354
9.20M
}
355
356
/* Evaluate a Sampled function. */
357
/* A generic algorithm with a recursion by dimentions. */
358
static int
359
fn_Sd_evaluate_general(const gs_function_t * pfn_common, const float *in, float *out)
360
95.3k
{
361
95.3k
    const gs_function_Sd_t *pfn = (const gs_function_Sd_t *)pfn_common;
362
95.3k
    int bps = pfn->params.BitsPerSample;
363
95.3k
    ulong offset = 0;
364
95.3k
    int i;
365
95.3k
    float encoded[max_Sd_m];
366
95.3k
    int iparts[max_Sd_m]; /* only needed for cubic interpolation */
367
95.3k
    ulong factors[max_Sd_m];
368
95.3k
    float samples[max_Sd_n];
369
370
    /* Encode the input values. */
371
372
190k
    for (i = 0; i < pfn->params.m; ++i) {
373
95.4k
        float d0 = pfn->params.Domain[2 * i],
374
95.4k
            d1 = pfn->params.Domain[2 * i + 1];
375
95.4k
        float arg = in[i], enc;
376
377
95.4k
        if (arg < d0)
378
6
            arg = d0;
379
95.4k
        else if (arg > d1)
380
0
            arg = d1;
381
95.4k
        if (pfn->params.Encode) {
382
55.9k
            float e0 = pfn->params.Encode[2 * i];
383
55.9k
            float e1 = pfn->params.Encode[2 * i + 1];
384
385
55.9k
            enc = (arg - d0) * (e1 - e0) / (d1 - d0) + e0;
386
55.9k
            if (enc < 0)
387
0
                encoded[i] = 0;
388
55.9k
            else if (enc >= pfn->params.Size[i] - 1)
389
1.52k
                encoded[i] = (float)pfn->params.Size[i] - 1;
390
54.4k
            else
391
54.4k
                encoded[i] = enc;
392
55.9k
        } else {
393
            /* arg is guaranteed to be in bounds, ergo so is enc */
394
                /* TODO: possible issue here.  if (pfn->params.Size[i] == 1 */
395
39.4k
            encoded[i] = (arg - d0) * (pfn->params.Size[i] - 1) / (d1 - d0);
396
39.4k
        }
397
95.4k
    }
398
399
    /* Look up and interpolate the output values. */
400
401
95.3k
    {
402
95.3k
        ulong factor = (ulong)bps * pfn->params.n;
403
404
190k
        for (i = 0; i < pfn->params.m; factor *= pfn->params.Size[i++]) {
405
95.4k
            int ipart = (int)encoded[i];
406
407
95.4k
            offset += (factors[i] = factor) * ipart;
408
95.4k
            iparts[i] = ipart;  /* only needed for cubic interpolation */
409
95.4k
            encoded[i] -= ipart;
410
95.4k
        }
411
95.3k
    }
412
95.3k
    if (pfn->params.Order == 3)
413
0
        fn_interpolate_cubic(pfn, encoded, iparts, factors, samples,
414
0
                             offset, pfn->params.m);
415
95.3k
    else
416
95.3k
        fn_interpolate_linear(pfn, encoded, factors, samples, offset,
417
95.3k
                              pfn->params.m);
418
419
    /* Encode the output values. */
420
421
303k
    for (i = 0; i < pfn->params.n; ++i)
422
207k
        out[i] = (float)fn_Sd_encode(pfn, i, samples[i]);
423
424
95.3k
    return 0;
425
95.3k
}
426
427
static const double double_stub = 1e90;
428
429
static inline void
430
fn_make_cubic_poles(double *p, double f0, double f1, double f2, double f3,
431
            const int pole_step_minor)
432
0
{   /* The following is poles of the polinomial,
433
       which represents interpolate_cubic in [1,2]. */
434
0
    const double a = -0.5;
435
436
0
    p[pole_step_minor * 1] = (a*f0 + 3*f1 - a*f2)/3.0;
437
0
    p[pole_step_minor * 2] = (-a*f1 + 3*f2 + a*f3)/3.0;
438
0
}
439
440
static void
441
fn_make_poles(double *p, const int pole_step, int power, int bias)
442
0
{
443
0
    const int pole_step_minor = pole_step / 3;
444
0
    switch(power) {
445
0
        case 1: /* A linear 3d power curve. */
446
            /* bias must be 0. */
447
0
            p[pole_step_minor * 1] = (2 * p[pole_step * 0] + 1 * p[pole_step * 1]) / 3;
448
0
            p[pole_step_minor * 2] = (1 * p[pole_step * 0] + 2 * p[pole_step * 1]) / 3;
449
0
            break;
450
0
        case 2:
451
            /* bias may be be 0 or 1. */
452
            /* Duplicate the beginning or the ending pole (the old code compatible). */
453
0
            fn_make_cubic_poles(p + pole_step * bias,
454
0
                    p[pole_step * 0], p[pole_step * bias],
455
0
                    p[pole_step * (1 + bias)], p[pole_step * 2],
456
0
                    pole_step_minor);
457
0
            break;
458
0
        case 3:
459
            /* bias must be 1. */
460
0
            fn_make_cubic_poles(p + pole_step * bias,
461
0
                    p[pole_step * 0], p[pole_step * 1], p[pole_step * 2], p[pole_step * 3],
462
0
                    pole_step_minor);
463
0
            break;
464
0
        default: /* Must not happen. */
465
0
           DO_NOTHING;
466
0
    }
467
0
}
468
469
/* Evaluate a Sampled function.
470
   A cubic interpolation with a pole cache.
471
   Allows a fast check for extreme suspection. */
472
/* This implementation is a particular case of 1 dimension.
473
   maybe we'll use as an optimisation of the generic case,
474
   so keep it for a while. */
475
static int
476
fn_Sd_evaluate_cubic_cached_1d(const gs_function_Sd_t *pfn, const float *in, float *out)
477
0
{
478
0
    float d0 = pfn->params.Domain[2 * 0];
479
0
    float d1 = pfn->params.Domain[2 * 0 + 1];
480
0
    const int pole_step_minor = pfn->params.n;
481
0
    const int pole_step = 3 * pole_step_minor;
482
0
    int i0; /* A cell index. */
483
0
    int ib, ie, i, k;
484
0
    double *p, t0, t1, tt;
485
0
486
0
    tt = (in[0] - d0) * (pfn->params.Size[0] - 1) / (d1 - d0);
487
0
    i0 = (int)floor(tt);
488
0
    ib = max(i0 - 1, 0);
489
0
    ie = min(pfn->params.Size[0], i0 + 3);
490
0
    for (i = ib; i < ie; i++) {
491
0
        if (pfn->params.pole[i * pole_step] == double_stub) {
492
0
            uint sdata[max_Sd_n];
493
0
            int bps = pfn->params.BitsPerSample;
494
0
495
0
            p = &pfn->params.pole[i * pole_step];
496
0
            fn_get_samples[pfn->params.BitsPerSample](pfn, (ulong)i * bps * pfn->params.n, sdata);
497
0
            for (k = 0; k < pfn->params.n; k++, p++)
498
0
                *p = fn_Sd_encode(pfn, k, (double)sdata[k]);
499
0
        }
500
0
    }
501
0
    p = &pfn->params.pole[i0 * pole_step];
502
0
    t0 = tt - i0;
503
0
    if (t0 == 0) {
504
0
        for (k = 0; k < pfn->params.n; k++, p++)
505
0
            out[k] = *p;
506
0
    } else {
507
0
        if (p[1 * pole_step_minor] == double_stub) {
508
0
            for (k = 0; k < pfn->params.n; k++)
509
0
                fn_make_poles(&pfn->params.pole[ib * pole_step + k], pole_step,
510
0
                        ie - ib - 1, i0 - ib);
511
0
        }
512
0
        t1 = 1 - t0;
513
0
        for (k = 0; k < pfn->params.n; k++, p++) {
514
0
            double y = p[0 * pole_step_minor] * t1 * t1 * t1 +
515
0
                       p[1 * pole_step_minor] * t1 * t1 * t0 * 3 +
516
0
                       p[2 * pole_step_minor] * t1 * t0 * t0 * 3 +
517
0
                       p[3 * pole_step_minor] * t0 * t0 * t0;
518
0
            if (y < pfn->params.Range[0])
519
0
                y = pfn->params.Range[0];
520
0
            if (y > pfn->params.Range[1])
521
0
                y = pfn->params.Range[1];
522
0
            out[k] = y;
523
0
        }
524
0
    }
525
0
    return 0;
526
0
}
527
528
static inline void
529
decode_argument(const gs_function_Sd_t *pfn, const float *in, double T[max_Sd_m], int I[max_Sd_m])
530
49
{
531
49
    int i;
532
533
98
    for (i = 0; i < pfn->params.m; i++) {
534
49
        float xi = in[i];
535
49
        float d0 = pfn->params.Domain[2 * i + 0];
536
49
        float d1 = pfn->params.Domain[2 * i + 1];
537
49
        double t;
538
539
49
        if (xi < d0)
540
0
            xi = d0;
541
49
        if (xi > d1)
542
0
            xi = d1;
543
49
        t = (xi - d0) * (pfn->params.Size[i] - 1) / (d1 - d0);
544
49
        I[i] = (int)floor(t);
545
49
        T[i] = t - I[i];
546
49
    }
547
49
}
548
549
static inline void
550
index_span(const gs_function_Sd_t *pfn, int *I, double *T, int ii, int *Ii, int *ib, int *ie)
551
49
{
552
49
    *Ii = I[ii];
553
49
    if (T[ii] != 0) {
554
0
        *ib = max(*Ii - 1, 0);
555
0
        *ie = min(pfn->params.Size[ii], *Ii + 3);
556
49
    } else {
557
49
        *ib = *Ii;
558
49
        *ie = *Ii + 1;
559
49
    }
560
49
}
561
562
static inline int
563
load_vector_to(const gs_function_Sd_t *pfn, int s_offset, double *V)
564
8.61M
{
565
8.61M
    uint sdata[max_Sd_n];
566
8.61M
    int k, code;
567
568
8.61M
    code = fn_get_samples[pfn->params.BitsPerSample](pfn, s_offset, sdata);
569
8.61M
    if (code < 0)
570
0
        return code;
571
17.6M
    for (k = 0; k < pfn->params.n; k++)
572
8.99M
        V[k] = fn_Sd_encode(pfn, k, (double)sdata[k]);
573
8.61M
    return 0;
574
8.61M
}
575
576
static inline int
577
load_vector(const gs_function_Sd_t *pfn, int a_offset, int s_offset)
578
25
{
579
25
    if (*(pfn->params.pole + a_offset) == double_stub) {
580
25
        uint sdata[max_Sd_n];
581
25
        int k, code;
582
583
25
        code = fn_get_samples[pfn->params.BitsPerSample](pfn, s_offset, sdata);
584
25
        if (code < 0)
585
0
            return code;
586
125
        for (k = 0; k < pfn->params.n; k++)
587
100
            *(pfn->params.pole + a_offset + k) = fn_Sd_encode(pfn, k, (double)sdata[k]);
588
25
    }
589
25
    return 0;
590
25
}
591
592
static inline void
593
interpolate_vector(const gs_function_Sd_t *pfn, int offset, int pole_step, int power, int bias)
594
0
{
595
0
    int k;
596
597
0
    for (k = 0; k < pfn->params.n; k++)
598
0
        fn_make_poles(pfn->params.pole + offset + k, pole_step, power, bias);
599
0
}
600
601
static inline void
602
interpolate_tensors(const gs_function_Sd_t *pfn, int *I, double *T,
603
        int offset, int pole_step, int power, int bias, int ii)
604
0
{
605
0
    if (ii < 0)
606
0
        interpolate_vector(pfn, offset, pole_step, power, bias);
607
0
    else {
608
0
        int s = pfn->params.array_step[ii];
609
0
        int Ii = I[ii];
610
611
0
        if (T[ii] == 0) {
612
0
            interpolate_tensors(pfn, I, T, offset + Ii * s, pole_step, power, bias, ii - 1);
613
0
        } else {
614
0
            int l;
615
616
0
            for (l = 0; l < 4; l++)
617
0
                interpolate_tensors(pfn, I, T, offset + Ii * s + l * s / 3, pole_step, power, bias, ii - 1);
618
0
        }
619
0
    }
620
0
}
621
622
static inline bool
623
is_tensor_done(const gs_function_Sd_t *pfn, int *I, double *T, int a_offset, int ii)
624
49
{
625
    /* Check an inner pole of the cell. */
626
49
    int i, o = 0;
627
628
98
    for (i = ii; i >= 0; i--) {
629
49
        o += I[i] * pfn->params.array_step[i];
630
49
        if (T[i] != 0)
631
0
            o += pfn->params.array_step[i] / 3;
632
49
    }
633
49
    if (*(pfn->params.pole + a_offset + o) != double_stub)
634
24
        return true;
635
25
    return false;
636
49
}
637
638
/* Creates a tensor of Bezier coefficients by node interpolation. */
639
static inline int
640
make_interpolation_tensor(const gs_function_Sd_t *pfn, int *I, double *T,
641
                            int a_offset, int s_offset, int ii)
642
74
{
643
    /* Well, this function isn't obvious. Trying to explain what it does.
644
645
       Suppose we have a 4x4x4...x4 hypercube of nodes, and we want to build
646
       a multicubic interpolation function for the inner 2x2x2...x2 hypercube.
647
       We represent the multicubic function with a tensor of Besier poles,
648
       and the size of the tensor is 4x4x....x4. Note that the corners
649
       of the tensor are equal to the corners of the 2x2x...x2 hypercube.
650
651
       We organize the 'pole' array so that a tensor of a cell
652
       occupies the cell, and tensors for neighbour cells have a common hyperplane.
653
654
       For a 1-dimentional case let the nodes are n0, n1, n2, n3.
655
       It defines 3 cells n0...n1, n1...n2, n2...n3.
656
       For the 2nd cell n1...n2 let the tensor coefficients are q10, q11, q12, q13.
657
       We choose a cubic approximation, in which tangents at nodes n1, n2
658
       are parallel to (n2 - n0) and (n3 - n1) correspondingly.
659
       (Well, this doesn't give a the minimal curvity, but likely it is
660
       what Adobe implementations do, see the bug 687352,
661
       and we agree that it's some reasonable).
662
663
       Then we have :
664
665
       q11 = n0
666
       q12 = (n0/2 + 3*n1 - n2/2)/3;
667
       q11 = (n1/2 + 3*n2 - n3/2)/3;
668
       q13 = n2
669
670
       When the source node array have an insufficient nomber of nodes
671
       along a dimension to determine tangents a cell
672
       (this happens near the array boundaries),
673
       we simply duplicate ending nodes. This solution is done
674
       for the compatibility to the old code, and definitely
675
       there exists a better one. Likely Adobe does the same.
676
677
       For a 2-dimensional case we apply the 1-dimentional case through
678
       the first dimension, and then construct a surface by varying the
679
       second coordinate as a parameter. It gives a bicubic surface,
680
       and the result doesn't depend on the order of coordinates
681
       (I proved the latter with Matematica 3.0).
682
       Then we know that an interpolation by one coordinate and
683
       a differentiation by another coordinate are interchangeble operators.
684
       Due to that poles of the interpolated function are same as
685
       interpolated poles of the function (well, we didn't spend time
686
       for a strong proof, but this fact was confirmed with testing the
687
       implementation with POLE_CACHE_DEBUG).
688
689
       Then we apply the 2-dimentional considerations recursively
690
       to all dimensions. This is exactly what the function does.
691
692
     */
693
74
    int code;
694
695
74
    if (ii < 0) {
696
25
        if (POLE_CACHE_IGNORE || *(pfn->params.pole + a_offset) == double_stub) {
697
25
            code = load_vector(pfn, a_offset, s_offset);
698
25
            if (code < 0)
699
0
                return code;
700
25
        }
701
49
    } else {
702
49
        int Ii, ib, ie, i;
703
49
        int sa = pfn->params.array_step[ii];
704
49
        int ss = pfn->params.stream_step[ii];
705
706
49
        index_span(pfn, I, T, ii, &Ii, &ib, &ie);
707
49
        if (POLE_CACHE_IGNORE || !is_tensor_done(pfn, I, T, a_offset, ii)) {
708
50
            for (i = ib; i < ie; i++) {
709
25
                code = make_interpolation_tensor(pfn, I, T,
710
25
                                a_offset + i * sa, s_offset + i * ss, ii - 1);
711
25
                if (code < 0)
712
0
                    return code;
713
25
            }
714
25
            if (T[ii] != 0)
715
0
                interpolate_tensors(pfn, I, T, a_offset + ib * sa, sa, ie - ib - 1,
716
0
                                Ii - ib, ii - 1);
717
25
        }
718
49
    }
719
74
    return 0;
720
74
}
721
722
/* Creates a subarray of samples. */
723
static inline int
724
make_interpolation_nodes(const gs_function_Sd_t *pfn, double *T0, double *T1,
725
                            int *I, double *T,
726
                            int a_offset, int s_offset, int ii)
727
0
{
728
0
    int code;
729
730
0
    if (ii < 0) {
731
0
        if (POLE_CACHE_IGNORE || *(pfn->params.pole + a_offset) == double_stub) {
732
0
            code = load_vector(pfn, a_offset, s_offset);
733
0
            if (code < 0)
734
0
                return code;
735
0
        }
736
0
        if (pfn->params.Order == 3) {
737
0
            code = make_interpolation_tensor(pfn, I, T, 0, 0, pfn->params.m - 1);
738
0
            if (code < 0)
739
0
                return code;
740
0
        }
741
0
    } else {
742
0
        int i;
743
0
        int i0 = (int)floor(T0[ii]);
744
0
        int i1 = (int)ceil(T1[ii]);
745
0
        int sa = pfn->params.array_step[ii];
746
0
        int ss = pfn->params.stream_step[ii];
747
748
0
        if (i0 < 0 || i0 >= pfn->params.Size[ii])
749
0
            return_error(gs_error_unregistered); /* Must not happen. */
750
0
        if (i1 < 0 || i1 >= pfn->params.Size[ii])
751
0
            return_error(gs_error_unregistered); /* Must not happen. */
752
0
        I[ii] = i0;
753
0
        T[ii] = (i1 > i0 ? 1 : 0);
754
0
        for (i = i0; i <= i1; i++) {
755
0
            code = make_interpolation_nodes(pfn, T0, T1, I, T,
756
0
                            a_offset + i * sa, s_offset + i * ss, ii - 1);
757
0
            if (code < 0)
758
0
                return code;
759
0
        }
760
0
    }
761
0
    return 0;
762
0
}
763
764
static inline int
765
evaluate_from_tenzor(const gs_function_Sd_t *pfn, int *I, double *T, int offset, int ii, double *y)
766
98
{
767
98
    int s = pfn->params.array_step[ii], k, l, code;
768
769
98
    if (ii < 0) {
770
245
        for (k = 0; k < pfn->params.n; k++)
771
196
            y[k] = *(pfn->params.pole + offset + k);
772
49
    } else if (T[ii] == 0) {
773
49
        return evaluate_from_tenzor(pfn, I, T, offset + s * I[ii], ii - 1, y);
774
49
    } else {
775
0
        double t0 = T[ii], t1 = 1 - t0;
776
0
        double p[4][max_Sd_n];
777
778
0
        for (l = 0; l < 4; l++) {
779
0
            code = evaluate_from_tenzor(pfn, I, T, offset + s * I[ii] + l * (s / 3), ii - 1, p[l]);
780
0
            if (code < 0)
781
0
                return code;
782
0
        }
783
0
        for (k = 0; k < pfn->params.n; k++)
784
0
            y[k] = p[0][k] * t1 * t1 * t1 +
785
0
                   p[1][k] * t1 * t1 * t0 * 3 +
786
0
                   p[2][k] * t1 * t0 * t0 * 3 +
787
0
           p[3][k] * t0 * t0 * t0;
788
0
    }
789
49
    return 0;
790
98
}
791
792
/* Evaluate a Sampled function. */
793
/* A cubic interpolation with pole cache. */
794
/* Allows a fast check for extreme suspection with is_tensor_monotonic. */
795
static int
796
fn_Sd_evaluate_multicubic_cached(const gs_function_Sd_t *pfn, const float *in, float *out)
797
49
{
798
49
    double T[max_Sd_m], y[max_Sd_n];
799
49
    int I[max_Sd_m], k, code;
800
801
49
    decode_argument(pfn, in, T, I);
802
49
    code = make_interpolation_tensor(pfn, I, T, 0, 0, pfn->params.m - 1);
803
49
    if (code < 0)
804
0
        return code;
805
49
    evaluate_from_tenzor(pfn, I, T, 0, pfn->params.m - 1, y);
806
245
    for (k = 0; k < pfn->params.n; k++) {
807
196
        double yk = y[k];
808
809
196
        if (yk < pfn->params.Range[k * 2 + 0])
810
0
            yk = pfn->params.Range[k * 2 + 0];
811
196
        if (yk > pfn->params.Range[k * 2 + 1])
812
0
            yk = pfn->params.Range[k * 2 + 1];
813
196
        out[k] = yk;
814
196
    }
815
49
    return 0;
816
49
}
817
818
/* Evaluate a Sampled function. */
819
static int
820
fn_Sd_evaluate(const gs_function_t * pfn_common, const float *in, float *out)
821
95.4k
{
822
95.4k
    const gs_function_Sd_t *pfn = (const gs_function_Sd_t *)pfn_common;
823
95.4k
    int code;
824
825
95.4k
    if (pfn->params.Order == 3) {
826
49
        if (POLE_CACHE_GENERIC_1D || pfn->params.m > 1)
827
49
            code = fn_Sd_evaluate_multicubic_cached(pfn, in, out);
828
0
        else
829
0
            code = fn_Sd_evaluate_cubic_cached_1d(pfn, in, out);
830
# if POLE_CACHE_DEBUG
831
        {   float y[max_Sd_n];
832
            int k, code1;
833
834
            code1 = fn_Sd_evaluate_general(pfn_common, in, y);
835
            if (code != code1)
836
                return_error(gs_error_unregistered); /* Must not happen. */
837
            for (k = 0; k < pfn->params.n; k++) {
838
                if (any_abs(y[k] - out[k]) > 1e-6 * (pfn->params.Range[k * 2 + 1] - pfn->params.Range[k * 2 + 0]))
839
                    return_error(gs_error_unregistered); /* Must not happen. */
840
            }
841
        }
842
# endif
843
49
    } else
844
95.3k
        code = fn_Sd_evaluate_general(pfn_common, in, out);
845
95.4k
    return code;
846
95.4k
}
847
848
/* Map a function subdomain to the sample index subdomain. */
849
static inline int
850
get_scaled_range(const gs_function_Sd_t *const pfn,
851
                   const float *lower, const float *upper,
852
                   int i, float *pw0, float *pw1)
853
9.19k
{
854
9.19k
    float d0 = pfn->params.Domain[i * 2 + 0], d1 = pfn->params.Domain[i * 2 + 1];
855
9.19k
    float v0 = lower[i], v1 = upper[i];
856
9.19k
    float e0, e1, w0, w1, w;
857
9.19k
    const float small_noise = (float)1e-6;
858
859
9.19k
    if (v0 < d0 || v0 > d1)
860
2
        return_error(gs_error_rangecheck);
861
9.19k
    if (pfn->params.Encode)
862
1.66k
        e0 = pfn->params.Encode[i * 2 + 0], e1 = pfn->params.Encode[i * 2 + 1];
863
7.52k
    else
864
7.52k
        e0 = 0, e1 = (float)pfn->params.Size[i] - 1;
865
9.19k
    w0 = (v0 - d0) * (e1 - e0) / (d1 - d0) + e0;
866
9.19k
    if (w0 < 0)
867
0
        w0 = 0;
868
9.19k
    else if (w0 >= pfn->params.Size[i] - 1)
869
1.65k
        w0 = (float)pfn->params.Size[i] - 1;
870
9.19k
    w1 = (v1 - d0) * (e1 - e0) / (d1 - d0) + e0;
871
9.19k
    if (w1 < 0)
872
0
        w1 = 0;
873
9.19k
    else if (w1 >= pfn->params.Size[i] - 1)
874
4.07k
        w1 = (float)pfn->params.Size[i] - 1;
875
9.19k
    if (w0 > w1) {
876
108
        w = w0; w0 = w1; w1 = w;
877
108
    }
878
9.19k
    if (floor(w0 + 1) - w0 < small_noise * any_abs(e1 - e0))
879
0
        w0 = (floor(w0) + 1);
880
9.19k
    if (w1 - floor(w1) < small_noise * any_abs(e1 - e0))
881
5.83k
        w1 = floor(w1);
882
9.19k
    if (w0 > w1)
883
0
        w0 = w1;
884
9.19k
    *pw0 = w0;
885
9.19k
    *pw1 = w1;
886
9.19k
    return 0;
887
9.19k
}
888
889
/* Copy a tensor to a differently indexed pole array. */
890
static int
891
copy_poles(const gs_function_Sd_t *pfn, int *I, double *T0, double *T1, int a_offset,
892
                int ii, double *pole, int p_offset, int pole_step)
893
0
{
894
0
    int i, ei, sa, code;
895
0
    int order = pfn->params.Order;
896
897
0
    if (pole_step <= 0)
898
0
        return_error(gs_error_limitcheck); /* Too small buffer. */
899
0
    ei = (T0[ii] == T1[ii] ? 1 : order + 1);
900
0
    sa = pfn->params.array_step[ii];
901
0
    if (ii == 0) {
902
0
        for (i = 0; i < ei; i++)
903
0
            *(pole + p_offset + i * pole_step) =
904
0
                    *(pfn->params.pole + a_offset + I[ii] * sa + i * (sa / order));
905
0
    } else {
906
0
        for (i = 0; i < ei; i++) {
907
0
            code = copy_poles(pfn, I, T0, T1, a_offset + I[ii] * sa + i * (sa / order), ii - 1,
908
0
                            pole, p_offset + i * pole_step, pole_step / 4);
909
0
            if (code < 0)
910
0
                return code;
911
0
        }
912
0
    }
913
0
    return 0;
914
0
}
915
916
static inline void
917
subcurve(double *pole, int pole_step, double t0, double t1)
918
0
{
919
    /* Generated with subcurve.nb using Mathematica 3.0. */
920
0
    double q0 = pole[pole_step * 0];
921
0
    double q1 = pole[pole_step * 1];
922
0
    double q2 = pole[pole_step * 2];
923
0
    double q3 = pole[pole_step * 3];
924
0
    double t01 = t0 - 1, t11 = t1 - 1;
925
0
    double small = 1e-13;
926
927
0
#define Power2(a) (a) * (a)
928
0
#define Power3(a) (a) * (a) * (a)
929
0
    pole[pole_step * 0] = t0*(t0*(q3*t0 - 3*q2*t01) + 3*q1*Power2(t01)) - q0*Power3(t01);
930
0
    pole[pole_step * 1] = q1*t01*(-2*t0 - t1 + 3*t0*t1) + t0*(q2*t0 + 2*q2*t1 -
931
0
                            3*q2*t0*t1 + q3*t0*t1) - q0*t11*Power2(t01);
932
0
    pole[pole_step * 2] = t1*(2*q2*t0 + q2*t1 - 3*q2*t0*t1 + q3*t0*t1) +
933
0
                            q1*(-t0 - 2*t1 + 3*t0*t1)*t11 - q0*t01*Power2(t11);
934
0
    pole[pole_step * 3] = t1*(t1*(3*q2 - 3*q2*t1 + q3*t1) +
935
0
                            3*q1*Power2(t11)) - q0*Power3(t11);
936
0
#undef Power2
937
0
#undef Power3
938
0
    if (any_abs(pole[pole_step * 1] - pole[pole_step * 0]) < small)
939
0
        pole[pole_step * 1] = pole[pole_step * 0];
940
0
    if (any_abs(pole[pole_step * 2] - pole[pole_step * 3]) < small)
941
0
        pole[pole_step * 2] = pole[pole_step * 3];
942
0
}
943
944
static inline void
945
subline(double *pole, int pole_step, double t0, double t1)
946
0
{
947
0
    double q0 = pole[pole_step * 0];
948
0
    double q1 = pole[pole_step * 1];
949
950
0
    pole[pole_step * 0] = (1 - t0) * q0 + t0 * q1;
951
0
    pole[pole_step * 1] = (1 - t1) * q0 + t1 * q1;
952
0
}
953
954
static void
955
clamp_poles(double *T0, double *T1, int ii, int i, double * pole,
956
                int p_offset, int pole_step, int pole_step_i, int order)
957
0
{
958
0
    if (ii < 0) {
959
0
        if (order == 3)
960
0
            subcurve(pole + p_offset, pole_step_i, T0[i], T1[i]);
961
0
        else
962
0
            subline(pole + p_offset, pole_step_i, T0[i], T1[i]);
963
0
    } else if (i == ii) {
964
0
        clamp_poles(T0, T1, ii - 1, i, pole, p_offset, pole_step / 4, pole_step, order);
965
0
    } else {
966
0
        int j, ei = (T0[ii] == T1[ii] ? 1 : order + 1);
967
968
0
        for (j = 0; j < ei; j++)
969
0
            clamp_poles(T0, T1, ii - 1, i, pole, p_offset + j * pole_step,
970
0
                            pole_step / 4, pole_step_i, order);
971
0
    }
972
0
}
973
974
static inline int /* 3 - don't know, 2 - decreesing, 0 - constant, 1 - increasing. */
975
curve_monotonity(double *pole, int pole_step)
976
0
{
977
0
    double p0 = pole[pole_step * 0];
978
0
    double p1 = pole[pole_step * 1];
979
0
    double p2 = pole[pole_step * 2];
980
0
    double p3 = pole[pole_step * 3];
981
982
0
    if (p0 == p1 && any_abs(p1 - p2) < 1e-13 && p2 == p3)
983
0
        return 0;
984
0
    if (p0 <= p1 && p1 <= p2 && p2 <= p3)
985
0
        return 1;
986
0
    if (p0 >= p1 && p1 >= p2 && p2 >= p3)
987
0
        return 2;
988
    /* Maybe not monotonic.
989
       Don't want to solve quadratic equations, so return "don't know".
990
       This case should be rare.
991
     */
992
0
    return 3;
993
0
}
994
995
static inline int /* 2 - decreesing, 0 - constant, 1 - increasing. */
996
line_monotonity(double *pole, int pole_step)
997
0
{
998
0
    double p0 = pole[pole_step * 0];
999
0
    double p1 = pole[pole_step * 1];
1000
1001
0
    if (p1 - p0 > 1e-13)
1002
0
        return 1;
1003
0
    if (p0 - p1 > 1e-13)
1004
0
        return 2;
1005
0
    return 0;
1006
0
}
1007
1008
static int /* 3 bits per guide : 3 - non-monotonic or don't know,
1009
                    2 - decreesing, 0 - constant, 1 - increasing.
1010
                    The number of guides is order+1. */
1011
tensor_dimension_monotonity(const double *T0, const double *T1, int ii, int i0, double *pole,
1012
                int p_offset, int pole_step, int pole_step_i, int order)
1013
0
{
1014
0
    if (ii < 0) {
1015
0
        if (order == 3)
1016
0
            return curve_monotonity(pole + p_offset, pole_step_i);
1017
0
        else
1018
0
            return line_monotonity(pole + p_offset, pole_step_i);
1019
0
    } else if (i0 == ii) {
1020
        /* Delay the dimension till the end, and adjust pole_step. */
1021
0
        return tensor_dimension_monotonity(T0, T1, ii - 1, i0, pole, p_offset,
1022
0
                            pole_step / 4, pole_step, order);
1023
0
    } else {
1024
0
        int j, ei = (T0[ii] == T1[ii] ? 1 : order + 1), m = 0, mm;
1025
1026
0
        for (j = 0; j < ei; j++) {
1027
0
            mm = tensor_dimension_monotonity(T0, T1, ii - 1, i0, pole, p_offset + j * pole_step,
1028
0
                            pole_step/ 4, pole_step_i, order);
1029
0
            m |= mm << (j * 3);
1030
0
            if (mm == 3) {
1031
                /* If one guide is not monotonic, the dimension is not monotonic.
1032
                   Can return early. */
1033
0
                break;
1034
0
            }
1035
0
        }
1036
0
        return m;
1037
0
    }
1038
0
}
1039
1040
static inline int
1041
is_tensor_monotonic_by_dimension(const gs_function_Sd_t *pfn, int *I, double *T0, double *T1, int i0, int k,
1042
                    uint *mask /* 3 bits per guide : 3 - non-monotonic or don't know,
1043
                    2 - decreesing, 0 - constant, 1 - increasing.
1044
                    The number of guides is order+1. */)
1045
0
{
1046
0
    double pole[4*4*4]; /* For a while restricting with 3-in cubic functions.
1047
                 More arguments need a bigger buffer, but the rest of code is same. */
1048
0
    int i, code, ii = pfn->params.m - 1;
1049
0
    double TT0[3], TT1[3];
1050
1051
0
    *mask = 0;
1052
0
    if (ii >= 3) {
1053
         /* Unimplemented. We don't know practical cases,
1054
            because currently it is only called while decomposing a shading.  */
1055
0
        return_error(gs_error_limitcheck);
1056
0
    }
1057
0
    code = copy_poles(pfn, I, T0, T1, k, ii, pole, 0, count_of(pole) / 4);
1058
0
    if (code < 0)
1059
0
        return code;
1060
0
    for (i = ii; i >= 0; i--) {
1061
0
        TT0[i] = 0;
1062
0
        if (T0[i] != T1[i]) {
1063
0
            if (T0[i] != 0 || T1[i] != 1)
1064
0
                clamp_poles(T0, T1, ii, i, pole, 0, count_of(pole) / 4, -1, pfn->params.Order);
1065
0
            TT1[i] = 1;
1066
0
        } else
1067
0
            TT1[i] = 0;
1068
0
    }
1069
0
    *mask = tensor_dimension_monotonity(TT0, TT1, ii, i0, pole, 0,
1070
0
                        count_of(pole) / 4, 1, pfn->params.Order);
1071
0
    return 0;
1072
0
}
1073
1074
static int /* error code */
1075
is_lattice_monotonic_by_dimension(const gs_function_Sd_t *pfn, const double *T0, const double *T1,
1076
        int *I, double *S0, double *S1, int ii, int i0, int k,
1077
        uint *mask /* 3 bits per guide : 1 - non-monotonic or don't know, 0 - monotonic;
1078
                      The number of guides is order+1. */)
1079
0
{
1080
0
    if (ii == -1) {
1081
        /* fixme : could cache the cell monotonity against redundant evaluation. */
1082
0
        return is_tensor_monotonic_by_dimension(pfn, I, S0, S1, i0, k, mask);
1083
0
    } else {
1084
0
        int i1 = (ii > i0 ? ii : ii == 0 ? i0 : ii - 1); /* Delay the dimension i0 till the end of recursion. */
1085
0
        int j, code;
1086
0
        int bi = (int)floor(T0[i1]);
1087
0
        int ei = (int)floor(T1[i1]);
1088
0
        uint m, mm, m1 = 0x49249249 & ((1 << ((pfn->params.Order + 1) * 3)) - 1);
1089
1090
0
        if (floor(T1[i1]) == T1[i1])
1091
0
            ei --;
1092
0
        m = 0;
1093
0
        for (j = bi; j <= ei; j++) {
1094
            /* fixme : A better performance may be obtained with comparing central nodes with side ones. */
1095
0
            I[i1] = j;
1096
0
            S0[i1] = max(T0[i1] - j, 0);
1097
0
            S1[i1] = min(T1[i1] - j, 1);
1098
0
            code = is_lattice_monotonic_by_dimension(pfn, T0, T1, I, S0, S1, ii - 1, i0, k, &mm);
1099
0
            if (code < 0)
1100
0
                return code;
1101
0
            m |= mm;
1102
0
            if (m == m1) /* Don't return early - shadings need to know about all dimensions. */
1103
0
                break;
1104
0
        }
1105
0
        if (ii == 0) {
1106
            /* Detect non-monotonic guides. */
1107
0
            m = m & (m >> 1);
1108
0
        }
1109
0
        *mask = m;
1110
0
        return 0;
1111
0
    }
1112
0
}
1113
1114
static inline int /* error code */
1115
is_lattice_monotonic(const gs_function_Sd_t *pfn, const double *T0, const double *T1,
1116
         int *I, double *S0, double *S1,
1117
         int k, uint *mask /* 1 bit per dimension : 1 - non-monotonic or don't know,
1118
                      0 - monotonic. */)
1119
0
{
1120
0
    uint m, mm = 0;
1121
0
    int i, code;
1122
1123
0
    for (i = 0; i < pfn->params.m; i++) {
1124
0
        if (T0[i] != T1[i]) {
1125
0
            code = is_lattice_monotonic_by_dimension(pfn, T0, T1, I, S0, S1, pfn->params.m - 1, i, k, &m);
1126
0
            if (code < 0)
1127
0
                return code;
1128
0
            if (m)
1129
0
                mm |= 1 << i;
1130
0
        }
1131
0
    }
1132
0
    *mask = mm;
1133
0
    return 0;
1134
0
}
1135
1136
static int /* 3 bits per result : 3 - non-monotonic or don't know,
1137
               2 - decreesing, 0 - constant, 1 - increasing,
1138
               <0 - error. */
1139
fn_Sd_1arg_linear_monotonic_rec(const gs_function_Sd_t *const pfn, int i0, int i1,
1140
                                const double *V0, const double *V1)
1141
17.2M
{
1142
17.2M
    if (i1 - i0 <= 1) {
1143
8.61M
        int code = 0, i;
1144
1145
17.5M
        for (i = 0; i < pfn->params.n; i++) {
1146
8.98M
            if (V0[i] < V1[i])
1147
769k
                code |= 1 << (i * 3);
1148
8.21M
            else if (V0[i] > V1[i])
1149
382k
                code |= 2 << (i * 3);
1150
8.98M
        }
1151
8.61M
        return code;
1152
8.61M
    } else {
1153
8.60M
        double VV[MAX_FAST_COMPS];
1154
8.60M
        int ii = (i0 + i1) / 2, code, cod1;
1155
1156
8.60M
        code = load_vector_to(pfn, ii * pfn->params.n * pfn->params.BitsPerSample, VV);
1157
8.60M
        if (code < 0)
1158
0
            return code;
1159
8.60M
        if (code & (code >> 1))
1160
0
            return code; /* Not monotonic by some component of the result. */
1161
8.60M
        code = fn_Sd_1arg_linear_monotonic_rec(pfn, i0, ii, V0, VV);
1162
8.60M
        if (code < 0)
1163
0
            return code;
1164
8.60M
        cod1 = fn_Sd_1arg_linear_monotonic_rec(pfn, ii, i1, VV, V1);
1165
8.60M
        if (cod1 < 0)
1166
0
            return cod1;
1167
8.60M
        return code | cod1;
1168
8.60M
    }
1169
17.2M
}
1170
1171
static int
1172
fn_Sd_1arg_linear_monotonic(const gs_function_Sd_t *const pfn, double T0, double T1,
1173
                            uint *mask /* 1 - non-monotonic or don't know, 0 - monotonic. */)
1174
9.19k
{
1175
9.19k
    int i0 = (int)floor(T0);
1176
9.19k
    int i1 = (int)ceil(T1), code;
1177
9.19k
    double V0[MAX_FAST_COMPS], V1[MAX_FAST_COMPS];
1178
1179
9.19k
    if (i1 - i0 > 1) {
1180
5.78k
        code = load_vector_to(pfn, i0 * pfn->params.n * pfn->params.BitsPerSample, V0);
1181
5.78k
        if (code < 0)
1182
0
            return code;
1183
5.78k
        code = load_vector_to(pfn, i1 * pfn->params.n * pfn->params.BitsPerSample, V1);
1184
5.78k
        if (code < 0)
1185
0
            return code;
1186
5.78k
        code = fn_Sd_1arg_linear_monotonic_rec(pfn, i0, i1, V0, V1);
1187
5.78k
        if (code < 0)
1188
0
            return code;
1189
5.78k
        if (code & (code >> 1)) {
1190
3.10k
            *mask = 1;
1191
3.10k
            return 0;
1192
3.10k
        }
1193
5.78k
    }
1194
6.08k
    *mask = 0;
1195
6.08k
    return 1;
1196
9.19k
}
1197
1198
0
#define DEBUG_Sd_1arg 0
1199
1200
/* Test whether a Sampled function is monotonic. */
1201
static int /* 1 = monotonic, 0 = not or don't know, <0 = error. */
1202
fn_Sd_is_monotonic_aux(const gs_function_Sd_t *const pfn,
1203
                   const float *lower, const float *upper,
1204
                   uint *mask /* 1 bit per dimension : 1 - non-monotonic or don't know,
1205
                      0 - monotonic. */)
1206
9.19k
{
1207
9.19k
    int i, code, ii = pfn->params.m - 1;
1208
9.19k
    int I[4];
1209
9.19k
    double T0[count_of(I)], T1[count_of(I)];
1210
9.19k
    double S0[count_of(I)], S1[count_of(I)];
1211
9.19k
    uint m, mm, m1;
1212
#   if DEBUG_Sd_1arg
1213
    int code1, mask1;
1214
#   endif
1215
1216
9.19k
    if (ii >= count_of(T0)) {
1217
         /* Unimplemented. We don't know practical cases,
1218
            because currently it is only called while decomposing a shading.  */
1219
0
        return_error(gs_error_limitcheck);
1220
0
    }
1221
18.3k
    for (i = 0; i <= ii; i++) {
1222
9.19k
        float w0, w1;
1223
1224
9.19k
        code = get_scaled_range(pfn, lower, upper, i, &w0, &w1);
1225
9.19k
        if (code < 0)
1226
2
            return code;
1227
9.19k
        T0[i] = w0;
1228
9.19k
        T1[i] = w1;
1229
9.19k
    }
1230
9.19k
    if (pfn->params.m == 1 && pfn->params.Order == 1 && pfn->params.n <= MAX_FAST_COMPS) {
1231
9.19k
        code = fn_Sd_1arg_linear_monotonic(pfn, T0[0], T1[0], mask);
1232
9.19k
# if !DEBUG_Sd_1arg
1233
9.19k
            return code;
1234
# else
1235
            mask1 = *mask;
1236
            code1 = code;
1237
# endif
1238
9.19k
    }
1239
0
    m1 = (1 << pfn->params.m )- 1;
1240
0
    code = make_interpolation_nodes(pfn, T0, T1, I, S0, 0, 0, ii);
1241
0
    if (code < 0)
1242
0
        return code;
1243
0
    mm = 0;
1244
0
    for (i = 0; i < pfn->params.n; i++) {
1245
0
        code = is_lattice_monotonic(pfn, T0, T1, I, S0, S1, i, &m);
1246
0
        if (code < 0)
1247
0
            return code;
1248
0
        mm |= m;
1249
0
        if (mm == m1) /* Don't return early - shadings need to know about all dimensions. */
1250
0
            break;
1251
0
    }
1252
#   if DEBUG_Sd_1arg
1253
        if (mask1 != mm)
1254
            return_error(gs_error_unregistered);
1255
        if (code1 != !mm)
1256
            return_error(gs_error_unregistered);
1257
#   endif
1258
0
    *mask = mm;
1259
0
    return !mm;
1260
0
}
1261
1262
/* Test whether a Sampled function is monotonic. */
1263
/* 1 = monotonic, 0 = don't know, <0 = error. */
1264
static int
1265
fn_Sd_is_monotonic(const gs_function_t * pfn_common,
1266
                   const float *lower, const float *upper, uint *mask)
1267
9.19k
{
1268
9.19k
    const gs_function_Sd_t *const pfn =
1269
9.19k
        (const gs_function_Sd_t *)pfn_common;
1270
1271
9.19k
    return fn_Sd_is_monotonic_aux(pfn, lower, upper, mask);
1272
9.19k
}
1273
1274
/* Return Sampled function information. */
1275
static void
1276
fn_Sd_get_info(const gs_function_t *pfn_common, gs_function_info_t *pfi)
1277
0
{
1278
0
    const gs_function_Sd_t *const pfn =
1279
0
        (const gs_function_Sd_t *)pfn_common;
1280
0
    long size;
1281
0
    int i;
1282
1283
0
    gs_function_get_info_default(pfn_common, pfi);
1284
0
    pfi->DataSource = &pfn->params.DataSource;
1285
0
    for (i = 0, size = 1; i < pfn->params.m; ++i)
1286
0
        size *= pfn->params.Size[i];
1287
0
    pfi->data_size =
1288
0
        (size * pfn->params.n * pfn->params.BitsPerSample + 7) >> 3;
1289
0
}
1290
1291
/* Write Sampled function parameters on a parameter list. */
1292
static int
1293
fn_Sd_get_params(const gs_function_t *pfn_common, gs_param_list *plist)
1294
0
{
1295
0
    const gs_function_Sd_t *const pfn =
1296
0
        (const gs_function_Sd_t *)pfn_common;
1297
0
    int ecode = fn_common_get_params(pfn_common, plist);
1298
0
    int code;
1299
1300
0
    if (pfn->params.Order != 1) {
1301
0
        if ((code = param_write_int(plist, "Order", &pfn->params.Order)) < 0)
1302
0
            ecode = code;
1303
0
    }
1304
0
    if ((code = param_write_int(plist, "BitsPerSample",
1305
0
                                &pfn->params.BitsPerSample)) < 0)
1306
0
        ecode = code;
1307
0
    if (pfn->params.Encode) {
1308
0
        if ((code = param_write_float_values(plist, "Encode",
1309
0
                                             pfn->params.Encode,
1310
0
                                             2 * pfn->params.m, false)) < 0)
1311
0
            ecode = code;
1312
0
    }
1313
0
    if (pfn->params.Decode) {
1314
0
        if ((code = param_write_float_values(plist, "Decode",
1315
0
                                             pfn->params.Decode,
1316
0
                                             2 * pfn->params.n, false)) < 0)
1317
0
            ecode = code;
1318
0
    }
1319
0
    if (pfn->params.Size) {
1320
0
        if ((code = param_write_int_values(plist, "Size", pfn->params.Size,
1321
0
                                           pfn->params.m, false)) < 0)
1322
0
            ecode = code;
1323
0
    }
1324
0
    return ecode;
1325
0
}
1326
1327
/* Make a scaled copy of a Sampled function. */
1328
static int
1329
fn_Sd_make_scaled(const gs_function_Sd_t *pfn, gs_function_Sd_t **ppsfn,
1330
                  const gs_range_t *pranges, gs_memory_t *mem)
1331
0
{
1332
0
    gs_function_Sd_t *psfn =
1333
0
        gs_alloc_struct(mem, gs_function_Sd_t, &st_function_Sd,
1334
0
                        "fn_Sd_make_scaled");
1335
0
    int code;
1336
1337
0
    if (psfn == 0)
1338
0
        return_error(gs_error_VMerror);
1339
0
    psfn->params = pfn->params;
1340
0
    psfn->params.Encode = 0;    /* in case of failure */
1341
0
    psfn->params.Decode = 0;
1342
0
    psfn->params.Size =
1343
0
        fn_copy_values(pfn->params.Size, pfn->params.m, sizeof(int), mem);
1344
0
    if ((code = (psfn->params.Size == 0 ?
1345
0
                 gs_note_error(gs_error_VMerror) : 0)) < 0 ||
1346
0
        (code = fn_common_scale((gs_function_t *)psfn,
1347
0
                                (const gs_function_t *)pfn,
1348
0
                                pranges, mem)) < 0 ||
1349
0
        (code = fn_scale_pairs(&psfn->params.Encode, pfn->params.Encode,
1350
0
                               pfn->params.m, NULL, mem)) < 0 ||
1351
0
        (code = fn_scale_pairs(&psfn->params.Decode, pfn->params.Decode,
1352
0
                               pfn->params.n, pranges, mem)) < 0) {
1353
0
        gs_function_free((gs_function_t *)psfn, true, mem);
1354
0
    } else
1355
0
        *ppsfn = psfn;
1356
0
    return code;
1357
0
}
1358
1359
/* Free the parameters of a Sampled function. */
1360
void
1361
gs_function_Sd_free_params(gs_function_Sd_params_t * params, gs_memory_t * mem)
1362
1.81k
{
1363
1.81k
    gs_free_const_object(mem, params->Size, "Size");
1364
1.81k
    params->Size = NULL;
1365
1.81k
    gs_free_const_object(mem, params->Decode, "Decode");
1366
1.81k
    params->Decode = NULL;
1367
1.81k
    gs_free_const_object(mem, params->Encode, "Encode");
1368
1.81k
    params->Encode = NULL;
1369
1.81k
    fn_common_free_params((gs_function_params_t *) params, mem);
1370
1.81k
    if (params->DataSource.type == data_source_type_stream && params->DataSource.data.strm != NULL) {
1371
1.71k
        s_close_filters(&params->DataSource.data.strm, params->DataSource.data.strm->strm);
1372
1.71k
        params->DataSource.data.strm = NULL;
1373
1.71k
    }
1374
1.81k
    gs_free_object(mem, params->pole, "gs_function_Sd_free_params");
1375
1.81k
    params->pole = NULL;
1376
1.81k
    gs_free_object(mem, params->array_step, "gs_function_Sd_free_params");
1377
1.81k
    params->array_step = NULL;
1378
1.81k
    gs_free_object(mem, params->stream_step, "gs_function_Sd_free_params");
1379
1.81k
    params->stream_step = NULL;
1380
1.81k
}
1381
1382
/* aA helper for gs_function_Sd_serialize. */
1383
static int serialize_array(const float *a, int half_size, stream *s)
1384
0
{
1385
0
    uint n;
1386
0
    const float dummy[2] = {0, 0};
1387
0
    int i, code;
1388
1389
0
    if (a != NULL)
1390
0
        return sputs(s, (const byte *)a, sizeof(a[0]) * half_size * 2, &n);
1391
0
    for (i = 0; i < half_size; i++) {
1392
0
        code = sputs(s, (const byte *)dummy, sizeof(dummy), &n);
1393
0
        if (code < 0)
1394
0
            return code;
1395
0
    }
1396
0
    return 0;
1397
0
}
1398
1399
/* Serialize. */
1400
static int
1401
gs_function_Sd_serialize(const gs_function_t * pfn, stream *s)
1402
0
{
1403
0
    uint n;
1404
0
    const gs_function_Sd_params_t * p = (const gs_function_Sd_params_t *)&pfn->params;
1405
0
    gs_function_info_t info;
1406
0
    int code = fn_common_serialize(pfn, s);
1407
0
    ulong pos;
1408
0
    uint count;
1409
0
    byte buf[100];
1410
0
    const byte *ptr;
1411
1412
0
    if (code < 0)
1413
0
        return code;
1414
0
    code = sputs(s, (const byte *)&p->Order, sizeof(p->Order), &n);
1415
0
    if (code < 0)
1416
0
        return code;
1417
0
    code = sputs(s, (const byte *)&p->BitsPerSample, sizeof(p->BitsPerSample), &n);
1418
0
    if (code < 0)
1419
0
        return code;
1420
0
    code = serialize_array(p->Encode, p->m, s);
1421
0
    if (code < 0)
1422
0
        return code;
1423
0
    code = serialize_array(p->Decode, p->n, s);
1424
0
    if (code < 0)
1425
0
        return code;
1426
0
    gs_function_get_info(pfn, &info);
1427
0
    code = sputs(s, (const byte *)&info.data_size, sizeof(info.data_size), &n);
1428
0
    if (code < 0)
1429
0
        return code;
1430
0
    for (pos = 0; pos < info.data_size; pos += count) {
1431
0
        count = min(sizeof(buf), info.data_size - pos);
1432
0
        data_source_access_only(info.DataSource, pos, count, buf, &ptr);
1433
0
        code = sputs(s, ptr, count, &n);
1434
0
        if (code < 0)
1435
0
            return code;
1436
0
    }
1437
0
    return 0;
1438
0
}
1439
1440
/* Allocate and initialize a Sampled function. */
1441
int
1442
gs_function_Sd_init(gs_function_t ** ppfn,
1443
                  const gs_function_Sd_params_t * params, gs_memory_t * mem)
1444
1.84k
{
1445
1.84k
    static const gs_function_head_t function_Sd_head = {
1446
1.84k
        function_type_Sampled,
1447
1.84k
        {
1448
1.84k
            (fn_evaluate_proc_t) fn_Sd_evaluate,
1449
1.84k
            (fn_is_monotonic_proc_t) fn_Sd_is_monotonic,
1450
1.84k
            (fn_get_info_proc_t) fn_Sd_get_info,
1451
1.84k
            (fn_get_params_proc_t) fn_Sd_get_params,
1452
1.84k
            (fn_make_scaled_proc_t) fn_Sd_make_scaled,
1453
1.84k
            (fn_free_params_proc_t) gs_function_Sd_free_params,
1454
1.84k
            fn_common_free,
1455
1.84k
            (fn_serialize_proc_t) gs_function_Sd_serialize,
1456
1.84k
        }
1457
1.84k
    };
1458
1.84k
    int code;
1459
1.84k
    int i;
1460
1461
1.84k
    *ppfn = 0;      /* in case of error */
1462
1.84k
    code = fn_check_mnDR((const gs_function_params_t *)params,
1463
1.84k
                         params->m, params->n);
1464
1.84k
    if (code < 0)
1465
1
        return code;
1466
1.83k
    if (params->m > max_Sd_m)
1467
0
        return_error(gs_error_limitcheck);
1468
1.83k
    switch (params->Order) {
1469
0
        case 0:   /* use default */
1470
1.76k
        case 1:
1471
1.83k
        case 3:
1472
1.83k
            break;
1473
0
        default:
1474
0
            return_error(gs_error_rangecheck);
1475
1.83k
    }
1476
1.83k
    switch (params->BitsPerSample) {
1477
0
        case 1:
1478
0
        case 2:
1479
0
        case 4:
1480
1.70k
        case 8:
1481
1.70k
        case 12:
1482
1.78k
        case 16:
1483
1.78k
        case 24:
1484
1.78k
        case 32:
1485
1.78k
            break;
1486
53
        default:
1487
53
            return_error(gs_error_rangecheck);
1488
1.83k
    }
1489
3.58k
    for (i = 0; i < params->m; ++i)
1490
1.80k
        if (params->Size[i] <= 0)
1491
0
            return_error(gs_error_rangecheck);
1492
1.78k
    {
1493
1.78k
        gs_function_Sd_t *pfn =
1494
1.78k
            gs_alloc_struct(mem, gs_function_Sd_t, &st_function_Sd,
1495
1.78k
                            "gs_function_Sd_init");
1496
1.78k
        int bps, sa, ss, i, order, was;
1497
1498
1.78k
        if (pfn == 0)
1499
0
            return_error(gs_error_VMerror);
1500
1.78k
        pfn->params = *params;
1501
1.78k
        if (params->Order == 0)
1502
0
            pfn->params.Order = 1; /* default */
1503
1.78k
        pfn->params.pole = NULL;
1504
1.78k
        pfn->params.array_step = NULL;
1505
1.78k
        pfn->params.stream_step = NULL;
1506
1.78k
        pfn->head = function_Sd_head;
1507
1.78k
        pfn->params.array_size = 0;
1508
1.78k
        if (pfn->params.m == 1 && pfn->params.Order == 1 && pfn->params.n <= MAX_FAST_COMPS && !DEBUG_Sd_1arg) {
1509
            /* Won't use pole cache. Call fn_Sd_1arg_linear_monotonic instead. */
1510
1.70k
        } else {
1511
78
            pfn->params.array_step = (int *)gs_alloc_byte_array(mem,
1512
78
                                    max_Sd_m, sizeof(int), "gs_function_Sd_init");
1513
78
            pfn->params.stream_step = (int *)gs_alloc_byte_array(mem,
1514
78
                                    max_Sd_m, sizeof(int), "gs_function_Sd_init");
1515
78
            if (pfn->params.array_step == NULL || pfn->params.stream_step == NULL)
1516
0
                return_error(gs_error_VMerror);
1517
78
            bps = pfn->params.BitsPerSample;
1518
78
            sa = pfn->params.n;
1519
78
            ss = pfn->params.n * bps;
1520
78
            order = pfn->params.Order;
1521
172
            for (i = 0; i < pfn->params.m; i++) {
1522
94
                pfn->params.array_step[i] = sa * order;
1523
94
                was = sa;
1524
94
                sa = (pfn->params.Size[i] * order - (order - 1)) * sa;
1525
                /* If the calculation of sa went backwards then we overflowed! */
1526
94
                if (was > sa)
1527
0
                    return_error(gs_error_VMerror);
1528
94
                pfn->params.stream_step[i] = ss;
1529
94
                ss = pfn->params.Size[i] * ss;
1530
94
            }
1531
78
            pfn->params.pole = (double *)gs_alloc_byte_array(mem,
1532
78
                                    sa, sizeof(double), "gs_function_Sd_init");
1533
78
            if (pfn->params.pole == NULL)
1534
0
                return_error(gs_error_VMerror);
1535
482k
            for (i = 0; i < sa; i++)
1536
482k
                pfn->params.pole[i] = double_stub;
1537
78
            pfn->params.array_size = sa;
1538
78
        }
1539
1.78k
        *ppfn = (gs_function_t *) pfn;
1540
1.78k
    }
1541
0
    return 0;
1542
1.78k
}