aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/libs/libjpeg-turbo/jcdctmgr.c
blob: 7dae17a6e1494457638908fddb475689c9346f48 (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
/*
 * jcdctmgr.c
 *
 * This file was part of the Independent JPEG Group's software:
 * Copyright (C) 1994-1996, Thomas G. Lane.
 * libjpeg-turbo Modifications:
 * Copyright (C) 1999-2006, MIYASAKA Masaru.
 * Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
 * Copyright (C) 2011, 2014-2015, D. R. Commander.
 * For conditions of distribution and use, see the accompanying README.ijg
 * file.
 *
 * This file contains the forward-DCT management logic.
 * This code selects a particular DCT implementation to be used,
 * and it performs related housekeeping chores including coefficient
 * quantization.
 */

#define JPEG_INTERNALS
#include "jinclude.h"
#include "jpeglib.h"
#include "jdct.h"               /* Private declarations for DCT subsystem */
#include "jsimddct.h"


/* Private subobject for this module */

typedef void (*forward_DCT_method_ptr) (DCTELEM *data);
typedef void (*float_DCT_method_ptr) (FAST_FLOAT *data);

typedef void (*convsamp_method_ptr) (JSAMPARRAY sample_data,
                                     JDIMENSION start_col,
                                     DCTELEM *workspace);
typedef void (*float_convsamp_method_ptr) (JSAMPARRAY sample_data,
                                           JDIMENSION start_col,
                                           FAST_FLOAT *workspace);

typedef void (*quantize_method_ptr) (JCOEFPTR coef_block, DCTELEM *divisors,
                                     DCTELEM *workspace);
typedef void (*float_quantize_method_ptr) (JCOEFPTR coef_block,
                                           FAST_FLOAT *divisors,
                                           FAST_FLOAT *workspace);

METHODDEF(void) quantize(JCOEFPTR, DCTELEM *, DCTELEM *);

typedef struct {
  struct jpeg_forward_dct pub;  /* public fields */

  /* Pointer to the DCT routine actually in use */
  forward_DCT_method_ptr dct;
  convsamp_method_ptr convsamp;
  quantize_method_ptr quantize;

  /* The actual post-DCT divisors --- not identical to the quant table
   * entries, because of scaling (especially for an unnormalized DCT).
   * Each table is given in normal array order.
   */
  DCTELEM *divisors[NUM_QUANT_TBLS];

  /* work area for FDCT subroutine */
  DCTELEM *workspace;

#ifdef DCT_FLOAT_SUPPORTED
  /* Same as above for the floating-point case. */
  float_DCT_method_ptr float_dct;
  float_convsamp_method_ptr float_convsamp;
  float_quantize_method_ptr float_quantize;
  FAST_FLOAT *float_divisors[NUM_QUANT_TBLS];
  FAST_FLOAT *float_workspace;
#endif
} my_fdct_controller;

typedef my_fdct_controller *my_fdct_ptr;


#if BITS_IN_JSAMPLE == 8

/*
 * Find the highest bit in an integer through binary search.
 */

LOCAL(int)
flss(UINT16 val)
{
  int bit;

  bit = 16;

  if (!val)
    return 0;

  if (!(val & 0xff00)) {
    bit -= 8;
    val <<= 8;
  }
  if (!(val & 0xf000)) {
    bit -= 4;
    val <<= 4;
  }
  if (!(val & 0xc000)) {
    bit -= 2;
    val <<= 2;
  }
  if (!(val & 0x8000)) {
    bit -= 1;
    val <<= 1;
  }

  return bit;
}


/*
 * Compute values to do a division using reciprocal.
 *
 * This implementation is based on an algorithm described in
 *   "How to optimize for the Pentium family of microprocessors"
 *   (http://www.agner.org/assem/).
 * More information about the basic algorithm can be found in
 * the paper "Integer Division Using Reciprocals" by Robert Alverson.
 *
 * The basic idea is to replace x/d by x * d^-1. In order to store
 * d^-1 with enough precision we shift it left a few places. It turns
 * out that this algoright gives just enough precision, and also fits
 * into DCTELEM:
 *
 *   b = (the number of significant bits in divisor) - 1
 *   r = (word size) + b
 *   f = 2^r / divisor
 *
 * f will not be an integer for most cases, so we need to compensate
 * for the rounding error introduced:
 *
 *   no fractional part:
 *
 *       result = input >> r
 *
 *   fractional part of f < 0.5:
 *
 *       round f down to nearest integer
 *       result = ((input + 1) * f) >> r
 *
 *   fractional part of f > 0.5:
 *
 *       round f up to nearest integer
 *       result = (input * f) >> r
 *
 * This is the original algorithm that gives truncated results. But we
 * want properly rounded results, so we replace "input" with
 * "input + divisor/2".
 *
 * In order to allow SIMD implementations we also tweak the values to
 * allow the same calculation to be made at all times:
 *
 *   dctbl[0] = f rounded to nearest integer
 *   dctbl[1] = divisor / 2 (+ 1 if fractional part of f < 0.5)
 *   dctbl[2] = 1 << ((word size) * 2 - r)
 *   dctbl[3] = r - (word size)
 *
 * dctbl[2] is for stupid instruction sets where the shift operation
 * isn't member wise (e.g. MMX).
 *
 * The reason dctbl[2] and dctbl[3] reduce the shift with (word size)
 * is that most SIMD implementations have a "multiply and store top
 * half" operation.
 *
 * Lastly, we store each of the values in their own table instead
 * of in a consecutive manner, yet again in order to allow SIMD
 * routines.
 */

LOCAL(int)
compute_reciprocal(UINT16 divisor, DCTELEM *dtbl)
{
  UDCTELEM2 fq, fr;
  UDCTELEM c;
  int b, r;

  if (divisor == 1) {
    /* divisor == 1 means unquantized, so these reciprocal/correction/shift
     * values will cause the C quantization algorithm to act like the
     * identity function.  Since only the C quantization algorithm is used in
     * these cases, the scale value is irrelevant.
     */
    dtbl[DCTSIZE2 * 0] = (DCTELEM)1;                        /* reciprocal */
    dtbl[DCTSIZE2 * 1] = (DCTELEM)0;                        /* correction */
    dtbl[DCTSIZE2 * 2] = (DCTELEM)1;                        /* scale */
    dtbl[DCTSIZE2 * 3] = -(DCTELEM)(sizeof(DCTELEM) * 8);   /* shift */
    return 0;
  }

  b = flss(divisor) - 1;
  r  = sizeof(DCTELEM) * 8 + b;

  fq = ((UDCTELEM2)1 << r) / divisor;
  fr = ((UDCTELEM2)1 << r) % divisor;

  c = divisor / 2;                      /* for rounding */

  if (fr == 0) {                        /* divisor is power of two */
    /* fq will be one bit too large to fit in DCTELEM, so adjust */
    fq >>= 1;
    r--;
  } else if (fr <= (divisor / 2U)) {    /* fractional part is < 0.5 */
    c++;
  } else {                              /* fractional part is > 0.5 */
    fq++;
  }

  dtbl[DCTSIZE2 * 0] = (DCTELEM)fq;     /* reciprocal */
  dtbl[DCTSIZE2 * 1] = (DCTELEM)c;      /* correction + roundfactor */
#ifdef WITH_SIMD
  dtbl[DCTSIZE2 * 2] = (DCTELEM)(1 << (sizeof(DCTELEM) * 8 * 2 - r)); /* scale */
#else
  dtbl[DCTSIZE2 * 2] = 1;
#endif
  dtbl[DCTSIZE2 * 3] = (DCTELEM)r - sizeof(DCTELEM) * 8; /* shift */

  if (r <= 16) return 0;
  else return 1;
}

#endif


/*
 * Initialize for a processing pass.
 * Verify that all referenced Q-tables are present, and set up
 * the divisor table for each one.
 * In the current implementation, DCT of all components is done during
 * the first pass, even if only some components will be output in the
 * first scan.  Hence all components should be examined here.
 */

METHODDEF(void)
start_pass_fdctmgr(j_compress_ptr cinfo)
{
  my_fdct_ptr fdct = (my_fdct_ptr)cinfo->fdct;
  int ci, qtblno, i;
  jpeg_component_info *compptr;
  JQUANT_TBL *qtbl;
  DCTELEM *dtbl;

  for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components;
       ci++, compptr++) {
    qtblno = compptr->quant_tbl_no;
    /* Make sure specified quantization table is present */
    if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS ||
        cinfo->quant_tbl_ptrs[qtblno] == NULL)
      ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno);
    qtbl = cinfo->quant_tbl_ptrs[qtblno];
    /* Compute divisors for this quant table */
    /* We may do this more than once for same table, but it's not a big deal */
    switch (cinfo->dct_method) {
#ifdef DCT_ISLOW_SUPPORTED
    case JDCT_ISLOW:
      /* For LL&M IDCT method, divisors are equal to raw quantization
       * coefficients multiplied by 8 (to counteract scaling).
       */
      if (fdct->divisors[qtblno] == NULL) {
        fdct->divisors[qtblno] = (DCTELEM *)
          (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
                                      (DCTSIZE2 * 4) * sizeof(DCTELEM));
      }
      dtbl = fdct->divisors[qtblno];
      for (i = 0; i < DCTSIZE2; i++) {
#if BITS_IN_JSAMPLE == 8
        if (!compute_reciprocal(qtbl->quantval[i] << 3, &dtbl[i]) &&
            fdct->quantize == jsimd_quantize)
          fdct->quantize = quantize;
#else
        dtbl[i] = ((DCTELEM)qtbl->quantval[i]) << 3;
#endif
      }
      break;
#endif
#ifdef DCT_IFAST_SUPPORTED
    case JDCT_IFAST:
      {
        /* For AA&N IDCT method, divisors are equal to quantization
         * coefficients scaled by scalefactor[row]*scalefactor[col], where
         *   scalefactor[0] = 1
         *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
         * We apply a further scale factor of 8.
         */
#define CONST_BITS  14
        static const INT16 aanscales[DCTSIZE2] = {
          /* precomputed values scaled up by 14 bits */
          16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
          22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
          21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
          19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
          16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
          12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
           8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
           4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
        };
        SHIFT_TEMPS

        if (fdct->divisors[qtblno] == NULL) {
          fdct->divisors[qtblno] = (DCTELEM *)
            (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
                                        (DCTSIZE2 * 4) * sizeof(DCTELEM));
        }
        dtbl = fdct->divisors[qtblno];
        for (i = 0; i < DCTSIZE2; i++) {
#if BITS_IN_JSAMPLE == 8
          if (!compute_reciprocal(
                DESCALE(MULTIPLY16V16((JLONG)qtbl->quantval[i],
                                      (JLONG)aanscales[i]),
                        CONST_BITS - 3), &dtbl[i]) &&
              fdct->quantize == jsimd_quantize)
            fdct->quantize = quantize;
#else
          dtbl[i] = (DCTELEM)
            DESCALE(MULTIPLY16V16((JLONG)qtbl->quantval[i],
                                  (JLONG)aanscales[i]),
                    CONST_BITS - 3);
#endif
        }
      }
      break;
#endif
#ifdef DCT_FLOAT_SUPPORTED
    case JDCT_FLOAT:
      {
        /* For float AA&N IDCT method, divisors are equal to quantization
         * coefficients scaled by scalefactor[row]*scalefactor[col], where
         *   scalefactor[0] = 1
         *   scalefactor[k] = cos(k*PI/16) * sqrt(2)    for k=1..7
         * We apply a further scale factor of 8.
         * What's actually stored is 1/divisor so that the inner loop can
         * use a multiplication rather than a division.
         */
        FAST_FLOAT *fdtbl;
        int row, col;
        static const double aanscalefactor[DCTSIZE] = {
          1.0, 1.387039845, 1.306562965, 1.175875602,
          1.0, 0.785694958, 0.541196100, 0.275899379
        };

        if (fdct->float_divisors[qtblno] == NULL) {
          fdct->float_divisors[qtblno] = (FAST_FLOAT *)
            (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
                                        DCTSIZE2 * sizeof(FAST_FLOAT));
        }
        fdtbl = fdct->float_divisors[qtblno];
        i = 0;
        for (row = 0; row < DCTSIZE; row++) {
          for (col = 0; col < DCTSIZE; col++) {
            fdtbl[i] = (FAST_FLOAT)
              (1.0 / (((double)qtbl->quantval[i] *
                       aanscalefactor[row] * aanscalefactor[col] * 8.0)));
            i++;
          }
        }
      }
      break;
#endif
    default:
      ERREXIT(cinfo, JERR_NOT_COMPILED);
      break;
    }
  }
}


/*
 * Load data into workspace, applying unsigned->signed conversion.
 */

METHODDEF(void)
convsamp(JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace)
{
  register DCTELEM *workspaceptr;
  register JSAMPROW elemptr;
  register int elemr;

  workspaceptr = workspace;
  for (elemr = 0; elemr < DCTSIZE; elemr++) {
    elemptr = sample_data[elemr] + start_col;

#if DCTSIZE == 8                /* unroll the inner loop */
    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
    *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
#else
    {
      register int elemc;
      for (elemc = DCTSIZE; elemc > 0; elemc--)
        *workspaceptr++ = (*elemptr++) - CENTERJSAMPLE;
    }
#endif
  }
}


/*
 * Quantize/descale the coefficients, and store into coef_blocks[].
 */

METHODDEF(void)
quantize(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace)
{
  int i;
  DCTELEM temp;
  JCOEFPTR output_ptr = coef_block;

#if BITS_IN_JSAMPLE == 8

  UDCTELEM recip, corr;
  int shift;
  UDCTELEM2 product;

  for (i = 0; i < DCTSIZE2; i++) {
    temp = workspace[i];
    recip = divisors[i + DCTSIZE2 * 0];
    corr =  divisors[i + DCTSIZE2 * 1];
    shift = divisors[i + DCTSIZE2 * 3];

    if (temp < 0) {
      temp = -temp;
      product = (UDCTELEM2)(temp + corr) * recip;
      product >>= shift + sizeof(DCTELEM) * 8;
      temp = (DCTELEM)product;
      temp = -temp;
    } else {
      product = (UDCTELEM2)(temp + corr) * recip;
      product >>= shift + sizeof(DCTELEM) * 8;
      temp = (DCTELEM)product;
    }
    output_ptr[i] = (JCOEF)temp;
  }

#else

  register DCTELEM qval;

  for (i = 0; i < DCTSIZE2; i++) {
    qval = divisors[i];
    temp = workspace[i];
    /* Divide the coefficient value by qval, ensuring proper rounding.
     * Since C does not specify the direction of rounding for negative
     * quotients, we have to force the dividend positive for portability.
     *
     * In most files, at least half of the output values will be zero
     * (at default quantization settings, more like three-quarters...)
     * so we should ensure that this case is fast.  On many machines,
     * a comparison is enough cheaper than a divide to make a special test
     * a win.  Since both inputs will be nonnegative, we need only test
     * for a < b to discover whether a/b is 0.
     * If your machine's division is fast enough, define FAST_DIVIDE.
     */
#ifdef FAST_DIVIDE
#define DIVIDE_BY(a, b)  a /= b
#else
#define DIVIDE_BY(a, b)  if (a >= b) a /= b;  else a = 0
#endif
    if (temp < 0) {
      temp = -temp;
      temp += qval >> 1;        /* for rounding */
      DIVIDE_BY(temp, qval);
      temp = -temp;
    } else {
      temp += qval >> 1;        /* for rounding */
      DIVIDE_BY(temp, qval);
    }
    output_ptr[i] = (JCOEF)temp;
  }

#endif

}


/*
 * Perform forward DCT on one or more blocks of a component.
 *
 * The input samples are taken from the sample_data[] array starting at
 * position start_row/start_col, and moving to the right for any additional
 * blocks. The quantized coefficients are returned in coef_blocks[].
 */

METHODDEF(void)
forward_DCT(j_compress_ptr cinfo, jpeg_component_info *compptr,
            JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
            JDIMENSION start_row, JDIMENSION start_col, JDIMENSION num_blocks)
/* This version is used for integer DCT implementations. */
{
  /* This routine is heavily used, so it's worth coding it tightly. */
  my_fdct_ptr fdct = (my_fdct_ptr)cinfo->fdct;
  DCTELEM *divisors = fdct->divisors[compptr->quant_tbl_no];
  DCTELEM *workspace;
  JDIMENSION bi;

  /* Make sure the compiler doesn't look up these every pass */
  forward_DCT_method_ptr do_dct = fdct->dct;
  convsamp_method_ptr do_convsamp = fdct->convsamp;
  quantize_method_ptr do_quantize = fdct->quantize;
  workspace = fdct->workspace;

  sample_data += start_row;     /* fold in the vertical offset once */

  for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
    /* Load data into workspace, applying unsigned->signed conversion */
    (*do_convsamp) (sample_data, start_col, workspace);

    /* Perform the DCT */
    (*do_dct) (workspace);

    /* Quantize/descale the coefficients, and store into coef_blocks[] */
    (*do_quantize) (coef_blocks[bi], divisors, workspace);
  }
}


#ifdef DCT_FLOAT_SUPPORTED

METHODDEF(void)
convsamp_float(JSAMPARRAY sample_data, JDIMENSION start_col,
               FAST_FLOAT *workspace)
{
  register FAST_FLOAT *workspaceptr;
  register JSAMPROW elemptr;
  register int elemr;

  workspaceptr = workspace;
  for (elemr = 0; elemr < DCTSIZE; elemr++) {
    elemptr = sample_data[elemr] + start_col;
#if DCTSIZE == 8                /* unroll the inner loop */
    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
    *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
#else
    {
      register int elemc;
      for (elemc = DCTSIZE; elemc > 0; elemc--)
        *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - CENTERJSAMPLE);
    }
#endif
  }
}


METHODDEF(void)
quantize_float(JCOEFPTR coef_block, FAST_FLOAT *divisors,
               FAST_FLOAT *workspace)
{
  register FAST_FLOAT temp;
  register int i;
  register JCOEFPTR output_ptr = coef_block;

  for (i = 0; i < DCTSIZE2; i++) {
    /* Apply the quantization and scaling factor */
    temp = workspace[i] * divisors[i];

    /* Round to nearest integer.
     * Since C does not specify the direction of rounding for negative
     * quotients, we have to force the dividend positive for portability.
     * The maximum coefficient size is +-16K (for 12-bit data), so this
     * code should work for either 16-bit or 32-bit ints.
     */
    output_ptr[i] = (JCOEF)((int)(temp + (FAST_FLOAT)16384.5) - 16384);
  }
}


METHODDEF(void)
forward_DCT_float(j_compress_ptr cinfo, jpeg_component_info *compptr,
                  JSAMPARRAY sample_data, JBLOCKROW coef_blocks,
                  JDIMENSION start_row, JDIMENSION start_col,
                  JDIMENSION num_blocks)
/* This version is used for floating-point DCT implementations. */
{
  /* This routine is heavily used, so it's worth coding it tightly. */
  my_fdct_ptr fdct = (my_fdct_ptr)cinfo->fdct;
  FAST_FLOAT *divisors = fdct->float_divisors[compptr->quant_tbl_no];
  FAST_FLOAT *workspace;
  JDIMENSION bi;


  /* Make sure the compiler doesn't look up these every pass */
  float_DCT_method_ptr do_dct = fdct->float_dct;
  float_convsamp_method_ptr do_convsamp = fdct->float_convsamp;
  float_quantize_method_ptr do_quantize = fdct->float_quantize;
  workspace = fdct->float_workspace;

  sample_data += start_row;     /* fold in the vertical offset once */

  for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) {
    /* Load data into workspace, applying unsigned->signed conversion */
    (*do_convsamp) (sample_data, start_col, workspace);

    /* Perform the DCT */
    (*do_dct) (workspace);

    /* Quantize/descale the coefficients, and store into coef_blocks[] */
    (*do_quantize) (coef_blocks[bi], divisors, workspace);
  }
}

#endif /* DCT_FLOAT_SUPPORTED */


/*
 * Initialize FDCT manager.
 */

GLOBAL(void)
jinit_forward_dct(j_compress_ptr cinfo)
{
  my_fdct_ptr fdct;
  int i;

  fdct = (my_fdct_ptr)
    (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
                                sizeof(my_fdct_controller));
  cinfo->fdct = (struct jpeg_forward_dct *)fdct;
  fdct->pub.start_pass = start_pass_fdctmgr;

  /* First determine the DCT... */
  switch (cinfo->dct_method) {
#ifdef DCT_ISLOW_SUPPORTED
  case JDCT_ISLOW:
    fdct->pub.forward_DCT = forward_DCT;
    if (jsimd_can_fdct_islow())
      fdct->dct = jsimd_fdct_islow;
    else
      fdct->dct = jpeg_fdct_islow;
    break;
#endif
#ifdef DCT_IFAST_SUPPORTED
  case JDCT_IFAST:
    fdct->pub.forward_DCT = forward_DCT;
    if (jsimd_can_fdct_ifast())
      fdct->dct = jsimd_fdct_ifast;
    else
      fdct->dct = jpeg_fdct_ifast;
    break;
#endif
#ifdef DCT_FLOAT_SUPPORTED
  case JDCT_FLOAT:
    fdct->pub.forward_DCT = forward_DCT_float;
    if (jsimd_can_fdct_float())
      fdct->float_dct = jsimd_fdct_float;
    else
      fdct->float_dct = jpeg_fdct_float;
    break;
#endif
  default:
    ERREXIT(cinfo, JERR_NOT_COMPILED);
    break;
  }

  /* ...then the supporting stages. */
  switch (cinfo->dct_method) {
#ifdef DCT_ISLOW_SUPPORTED
  case JDCT_ISLOW:
#endif
#ifdef DCT_IFAST_SUPPORTED
  case JDCT_IFAST:
#endif
#if defined(DCT_ISLOW_SUPPORTED) || defined(DCT_IFAST_SUPPORTED)
    if (jsimd_can_convsamp())
      fdct->convsamp = jsimd_convsamp;
    else
      fdct->convsamp = convsamp;
    if (jsimd_can_quantize())
      fdct->quantize = jsimd_quantize;
    else
      fdct->quantize = quantize;
    break;
#endif
#ifdef DCT_FLOAT_SUPPORTED
  case JDCT_FLOAT:
    if (jsimd_can_convsamp_float())
      fdct->float_convsamp = jsimd_convsamp_float;
    else
      fdct->float_convsamp = convsamp_float;
    if (jsimd_can_quantize_float())
      fdct->float_quantize = jsimd_quantize_float;
    else
      fdct->float_quantize = quantize_float;
    break;
#endif
  default:
    ERREXIT(cinfo, JERR_NOT_COMPILED);
    break;
  }

  /* Allocate workspace memory */
#ifdef DCT_FLOAT_SUPPORTED
  if (cinfo->dct_method == JDCT_FLOAT)
    fdct->float_workspace = (FAST_FLOAT *)
      (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
                                  sizeof(FAST_FLOAT) * DCTSIZE2);
  else
#endif
    fdct->workspace = (DCTELEM *)
      (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE,
                                  sizeof(DCTELEM) * DCTSIZE2);

  /* Mark divisor tables unallocated */
  for (i = 0; i < NUM_QUANT_TBLS; i++) {
    fdct->divisors[i] = NULL;
#ifdef DCT_FLOAT_SUPPORTED
    fdct->float_divisors[i] = NULL;
#endif
  }
}