summaryrefslogtreecommitdiffstatshomepage
path: root/3rdparty/zlib/gzwrite.c
blob: c7b5651d70b994e20222a734c620f68e11e0dc84 (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
/* gzwrite.c -- zlib functions for writing gzip files
 * Copyright (C) 2004-2017 Mark Adler
 * For conditions of distribution and use, see copyright notice in zlib.h
 */

#include "gzguts.h"

/* Local functions */
local int gz_init OF((gz_statep));
local int gz_comp OF((gz_statep, int));
local int gz_zero OF((gz_statep, z_off64_t));
local z_size_t gz_write OF((gz_statep, voidpc, z_size_t));

/* Initialize state for writing a gzip file.  Mark initialization by setting
   state->size to non-zero.  Return -1 on a memory allocation failure, or 0 on
   success. */
local int gz_init(state)
    gz_statep state;
{
    int ret;
    z_streamp strm = &(state->strm);

    /* allocate input buffer (double size for gzprintf) */
    state->in = (unsigned char *)malloc(state->want << 1);
    if (state->in == NULL) {
        gz_error(state, Z_MEM_ERROR, "out of memory");
        return -1;
    }

    /* only need output buffer and deflate state if compressing */
    if (!state->direct) {
        /* allocate output buffer */
        state->out = (unsigned char *)malloc(state->want);
        if (state->out == NULL) {
            free(state->in);
            gz_error(state, Z_MEM_ERROR, "out of memory");
            return -1;
        }

        /* allocate deflate memory, set up for gzip compression */
        strm->zalloc = Z_NULL;
        strm->zfree = Z_NULL;
        strm->opaque = Z_NULL;
        ret = deflateInit2(strm, state->level, Z_DEFLATED,
                           MAX_WBITS + 16, DEF_MEM_LEVEL, state->strategy);
        if (ret != Z_OK) {
            free(state->out);
            free(state->in);
            gz_error(state, Z_MEM_ERROR, "out of memory");
            return -1;
        }
        strm->next_in = NULL;
    }

    /* mark state as initialized */
    state->size = state->want;

    /* initialize write buffer if compressing */
    if (!state->direct) {
        strm->avail_out = state->size;
        strm->next_out = state->out;
        state->x.next = strm->next_out;
    }
    return 0;
}

/* Compress whatever is at avail_in and next_in and write to the output file.
   Return -1 if there is an error writing to the output file or if gz_init()
   fails to allocate memory, otherwise 0.  flush is assumed to be a valid
   deflate() flush value.  If flush is Z_FINISH, then the deflate() state is
   reset to start a new gzip stream.  If gz->direct is true, then simply write
   to the output file without compressing, and ignore flush. */
local int gz_comp(state, flush)
    gz_statep state;
    int flush;
{
    int ret, writ;
    unsigned have, put, max = ((unsigned)-1 >> 2) + 1;
    z_streamp strm = &(state->strm);

    /* allocate memory if this is the first time through */
    if (state->size == 0 && gz_init(state) == -1)
        return -1;

    /* write directly if requested */
    if (state->direct) {
        while (strm->avail_in) {
            put = strm->avail_in > max ? max : strm->avail_in;
            writ = write(state->fd, strm->next_in, put);
            if (writ < 0) {
                gz_error(state, Z_ERRNO, zstrerror());
                return -1;
            }
            strm->avail_in -= (unsigned)writ;
            strm->next_in += writ;
        }
        return 0;
    }

    /* run deflate() on provided input until it produces no more output */
    ret = Z_OK;
    do {
        /* write out current buffer contents if full, or if flushing, but if
           doing Z_FINISH then don't write until we get to Z_STREAM_END */
        if (strm->avail_out == 0 || (flush != Z_NO_FLUSH &&
            (flush != Z_FINISH || ret == Z_STREAM_END))) {
            while (strm->next_out > state->x.next) {
                put = strm->next_out - state->x.next > (int)max ? max :
                      (unsigned)(strm->next_out - state->x.next);
                writ = write(state->fd, state->x.next, put);
                if (writ < 0) {
                    gz_error(state, Z_ERRNO, zstrerror());
                    return -1;
                }
                state->x.next += writ;
            }
            if (strm->avail_out == 0) {
                strm->avail_out = state->size;
                strm->next_out = state->out;
                state->x.next = state->out;
            }
        }

        /* compress */
        have = strm->avail_out;
        ret = deflate(strm, flush);
        if (ret == Z_STREAM_ERROR) {
            gz_error(state, Z_STREAM_ERROR,
                      "internal error: deflate stream corrupt");
            return -1;
        }
        have -= strm->avail_out;
    } while (have);

    /* if that completed a deflate stream, allow another to start */
    if (flush == Z_FINISH)
        deflateReset(strm);

    /* all done, no errors */
    return 0;
}

/* Compress len zeros to output.  Return -1 on a write error or memory
   allocation failure by gz_comp(), or 0 on success. */
local int gz_zero(state, len)
    gz_statep state;
    z_off64_t len;
{
    int first;
    unsigned n;
    z_streamp strm = &(state->strm);

    /* consume whatever's left in the input buffer */
    if (strm->avail_in && gz_comp(state, Z_NO_FLUSH) == -1)
        return -1;

    /* compress len zeros (len guaranteed > 0) */
    first = 1;
    while (len) {
        n = GT_OFF(state->size) || (z_off64_t)state->size > len ?
            (unsigned)len : state->size;
        if (first) {
            memset(state->in, 0, n);
            first = 0;
        }
        strm->avail_in = n;
        strm->next_in = state->in;
        state->x.pos += n;
        if (gz_comp(state, Z_NO_FLUSH) == -1)
            return -1;
        len -= n;
    }
    return 0;
}

/* Write len bytes from buf to file.  Return the number of bytes written.  If
   the returned value is less than len, then there was an error. */
local z_size_t gz_write(state, buf, len)
    gz_statep state;
    voidpc buf;
    z_size_t len;
{
    z_size_t put = len;

    /* if len is zero, avoid unnecessary operations */
    if (len == 0)
        return 0;

    /* allocate memory if this is the first time through */
    if (state->size == 0 && gz_init(state) == -1)
        return 0;

    /* check for seek request */
    if (state->seek) {
        state->seek = 0;
        if (gz_zero(state, state->skip) == -1)
            return 0;
    }

    /* for small len, copy to input buffer, otherwise compress directly */
    if (len < state->size) {
        /* copy to input buffer, compress when full */
        do {
            unsigned have, copy;

            if (state->strm.avail_in == 0)
                state->strm.next_in = state->in;
            have = (unsigned)((state->strm.next_in + state->strm.avail_in) -
                              state->in);
            copy = state->size - have;
            if (copy > len)
                copy = len;
            memcpy(state->in + have, buf, copy);
            state->strm.avail_in += copy;
            state->x.pos += copy;
            buf = (const char *)buf + copy;
            len -= copy;
            if (len && gz_comp(state, Z_NO_FLUSH) == -1)
                return 0;
        } while (len);
    }
    else {
        /* consume whatever's left in the input buffer */
        if (state->strm.avail_in && gz_comp(state, Z_NO_FLUSH) == -1)
            return 0;

        /* directly compress user buffer to file */
        state->strm.next_in = (z_const Bytef *)buf;
        do {
            unsigned n = (unsigned)-1;
            if (n > len)
                n = len;
            state->strm.avail_in = n;
            state->x.pos += n;
            if (gz_comp(state, Z_NO_FLUSH) == -1)
                return 0;
            len -= n;
        } while (len);
    }

    /* input was all buffered or compressed */
    return put;
}

/* -- see zlib.h -- */
int ZEXPORT gzwrite(file, buf, len)
    gzFile file;
    voidpc buf;
    unsigned len;
{
    gz_statep state;

    /* get internal structure */
    if (file == NULL)
        return 0;
    state = (gz_statep)file;

    /* check that we're writing and that there's no error */
    if (state->mode != GZ_WRITE || state->err != Z_OK)
        return 0;

    /* since an int is returned, make sure len fits in one, otherwise return
       with an error (this avoids a flaw in the interface) */
    if ((int)len < 0) {
        gz_error(state, Z_DATA_ERROR, "requested length does not fit in int");
        return 0;
    }

    /* write len bytes from buf (the return value will fit in an int) */
    return (int)gz_write(state, buf, len);
}

/* -- see zlib.h -- */
z_size_t ZEXPORT gzfwrite(buf, size, nitems, file)
    voidpc buf;
    z_size_t size;
    z_size_t nitems;
    gzFile file;
{
    z_size_t len;
    gz_statep state;

    /* get internal structure */
    if (file == NULL)
        return 0;
    state = (gz_statep)file;

    /* check that we're writing and that there's no error */
    if (state->mode != GZ_WRITE || state->err != Z_OK)
        return 0;

    /* compute bytes to read -- error on overflow */
    len = nitems * size;
    if (size && len / size != nitems) {
        gz_error(state, Z_STREAM_ERROR, "request does not fit in a size_t");
        return 0;
    }

    /* write len bytes to buf, return the number of full items written */
    return len ? gz_write(state, buf, len) / size : 0;
}

/* -- see zlib.h -- */
int ZEXPORT gzputc(file, c)
    gzFile file;
    int c;
{
    unsigned have;
    unsigned char buf[1];
    gz_statep state;
    z_streamp strm;

    /* get internal structure */
    if (file == NULL)
        return -1;
    state = (gz_statep)file;
    strm = &(state->strm);

    /* check that we're writing and that there's no error */
    if (state->mode != GZ_WRITE || state->err != Z_OK)
        return -1;

    /* check for seek request */
    if (state->seek) {
        state->seek = 0;
        if (gz_zero(state, state->skip) == -1)
            return -1;
    }

    /* try writing to input buffer for speed (state->size == 0 if buffer not
       initialized) */
    if (state->size) {
        if (strm->avail_in == 0)
            strm->next_in = state->in;
        have = (unsigned)((strm->next_in + strm->avail_in) - state->in);
        if (have < state->size) {
            state->in[have] = (unsigned char)c;
            strm->avail_in++;
            state->x.pos++;
            return c & 0xff;
        }
    }

    /* no room in buffer or not initialized, use gz_write() */
    buf[0] = (unsigned char)c;
    if (gz_write(state, buf, 1) != 1)
        return -1;
    return c & 0xff;
}

/* -- see zlib.h -- */
int ZEXPORT gzputs(file, str)
    gzFile file;
    const char *str;
{
    int ret;
    z_size_t len;
    gz_statep state;

    /* get internal structure */
    if (file == NULL)
        return -1;
    state = (gz_statep)file;

    /* check that we're writing and that there's no error */
    if (state->mode != GZ_WRITE || state->err != Z_OK)
        return -1;

    /* write string */
    len = strlen(str);
    ret = gz_write(state, str, len);
    return ret == 0 && len != 0 ? -1 : ret;
}

#if defined(STDC) || defined(Z_HAVE_STDARG_H)
#include <stdarg.h>

/* -- see zlib.h -- */
int ZEXPORTVA gzvprintf(gzFile file, const char *format, va_list va)
{
    int len;
    unsigned left;
    char *next;
    gz_statep state;
    z_streamp strm;

    /* get internal structure */
    if (file == NULL)
        return Z_STREAM_ERROR;
    state = (gz_statep)file;
    strm = &(state->strm);

    /* check that we're writing and that there's no error */
    if (state->mode != GZ_WRITE || state->err != Z_OK)
        return Z_STREAM_ERROR;

    /* make sure we have some buffer space */
    if (state->size == 0 && gz_init(state) == -1)
        return state->err;

    /* check for seek request */
    if (state->seek) {
        state->seek = 0;
        if (gz_zero(state, state->skip) == -1)
            return state->err;
    }

    /* do the printf() into the input buffer, put length in len -- the input
       buffer is double-sized just for this function, so there is guaranteed to
       be state->size bytes available after the current contents */
    if (strm->avail_in == 0)
        strm->next_in = state->in;
    next = (char *)(state->in + (strm->next_in - state->in) + strm->avail_in);
    next[state->size - 1] = 0;
#ifdef NO_vsnprintf
#  ifdef HAS_vsprintf_void
    (void)vsprintf(next, format, va);
    for (len = 0; len < state->size; len++)
        if (next[len] == 0) break;
#  else
    len = vsprintf(next, format, va);
#  endif
#else
#  ifdef HAS_vsnprintf_void
    (void)vsnprintf(next, state->size, format, va);
    len = strlen(next);
#  else
    len = vsnprintf(next, state->size, format, va);
#  endif
#endif

    /* check that printf() results fit in buffer */
    if (len == 0 || (unsigned)len >= state->size || next[state->size - 1] != 0)
        return 0;

    /* update buffer and position, compress first half if past that */
    strm->avail_in += (unsigned)len;
    state->x.pos += len;
    if (strm->avail_in >= state->size) {
        left = strm->avail_in - state->size;
        strm->avail_in = state->size;
        if (gz_comp(state, Z_NO_FLUSH) == -1)
            return state->err;
        memcpy(state->in, state->in + state->size, left);
        strm->next_in = state->in;
        strm->avail_in = left;
    }
    return len;
}

int ZEXPORTVA gzprintf(gzFile file, const char *format, ...)
{
    va_list va;
    int ret;

    va_start(va, format);
    ret = gzvprintf(file, format, va);
    va_end(va);
    return ret;
}

#else /* !STDC && !Z_HAVE_STDARG_H */

/* -- see zlib.h -- */
int ZEXPORTVA gzprintf (file, format, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10,
                       a11, a12, a13, a14, a15, a16, a17, a18, a19, a20)
    gzFile file;
    const char *format;
    int a1, a2, a3, a4, a5, a6, a7, a8, a9, a10,
        a11, a12, a13, a14, a15, a16, a17, a18, a19, a20;
{
    unsigned len, left;
    char *next;
    gz_statep state;
    z_streamp strm;

    /* get internal structure */
    if (file == NULL)
        return Z_STREAM_ERROR;
    state = (gz_statep)file;
    strm = &(state->strm);

    /* check that can really pass pointer in ints */
    if (sizeof(int) != sizeof(void *))
        return Z_STREAM_ERROR;

    /* check that we're writing and that there's no error */
    if (state->mode != GZ_WRITE || state->err != Z_OK)
        return Z_STREAM_ERROR;

    /* make sure we have some buffer space */
    if (state->size == 0 && gz_init(state) == -1)
        return state->error;

    /* check for seek request */
    if (state->seek) {
        state->seek = 0;
        if (gz_zero(state, state->skip) == -1)
            return state->error;
    }

    /* do the printf() into the input buffer, put length in len -- the input
       buffer is double-sized just for this function, so there is guaranteed to
       be state->size bytes available after the current contents */
    if (strm->avail_in == 0)
        strm->next_in = state->in;
    next = (char *)(strm->next_in + strm->avail_in);
    next[state->size - 1] = 0;
#ifdef NO_snprintf
#  ifdef HAS_sprintf_void
    sprintf(next, format, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10, a11, a12,
            a13, a14, a15, a16, a17, a18, a19, a20);
    for (len = 0; len < size; len++)
        if (next[len] == 0)
            break;
#  else
    len = sprintf(next, format, a1, a2, a3, a4, a5, a6, a7, a8, a9, a10, a11,
                  a12, a13, a14, a15, a16, a17, a18, a19, a20);
#  endif
#else
#  ifdef HAS_snprintf_void
    snprintf(next, state->size, format, a1, a2, a3, a4, a5, a6, a7, a8, a9,
             a10, a11, a12, a13, a14, a15, a16, a17, a18, a19, a20);
    len = strlen(next);
#  else
    len = snprintf(next, state->size, format, a1, a2, a3, a4, a5, a6, a7, a8,
                   a9, a10, a11, a12, a13, a14, a15, a16, a17, a18, a19, a20);
#  endif
#endif

    /* check that printf() results fit in buffer */
    if (len == 0 || len >= state->size || next[state->size - 1] != 0)
        return 0;

    /* update buffer and position, compress first half if past that */
    strm->avail_in += len;
    state->x.pos += len;
    if (strm->avail_in >= state->size) {
        left = strm->avail_in - state->size;
        strm->avail_in = state->size;
        if (gz_comp(state, Z_NO_FLUSH) == -1)
            return state->err;
        memcpy(state->in, state->in + state->size, left);
        strm->next_in = state->in;
        strm->avail_in = left;
    }
    return (int)len;
}

#endif

/* -- see zlib.h -- */
int ZEXPORT gzflush(file, flush)
    gzFile file;
    int flush;
{
    gz_statep state;

    /* get internal structure */
    if (file == NULL)
        return Z_STREAM_ERROR;
    state = (gz_statep)file;

    /* check that we're writing and that there's no error */
    if (state->mode != GZ_WRITE || state->err != Z_OK)
        return Z_STREAM_ERROR;

    /* check flush parameter */
    if (flush < 0 || flush > Z_FINISH)
        return Z_STREAM_ERROR;

    /* check for seek request */
    if (state->seek) {
        state->seek = 0;
        if (gz_zero(state, state->skip) == -1)
            return state->err;
    }

    /* compress remaining data with requested flush */
    (void)gz_comp(state, flush);
    return state->err;
}

/* -- see zlib.h -- */
int ZEXPORT gzsetparams(file, level, strategy)
    gzFile file;
    int level;
    int strategy;
{
    gz_statep state;
    z_streamp strm;

    /* get internal structure */
    if (file == NULL)
        return Z_STREAM_ERROR;
    state = (gz_statep)file;
    strm = &(state->strm);

    /* check that we're writing and that there's no error */
    if (state->mode != GZ_WRITE || state->err != Z_OK)
        return Z_STREAM_ERROR;

    /* if no change is requested, then do nothing */
    if (level == state->level && strategy == state->strategy)
        return Z_OK;

    /* check for seek request */
    if (state->seek) {
        state->seek = 0;
        if (gz_zero(state, state->skip) == -1)
            return state->err;
    }

    /* change compression parameters for subsequent input */
    if (state->size) {
        /* flush previous input with previous parameters before changing */
        if (strm->avail_in && gz_comp(state, Z_BLOCK) == -1)
            return state->err;
        deflateParams(strm, level, strategy);
    }
    state->level = level;
    state->strategy = strategy;
    return Z_OK;
}

/* -- see zlib.h -- */
int ZEXPORT gzclose_w(file)
    gzFile file;
{
    int ret = Z_OK;
    gz_statep state;

    /* get internal structure */
    if (file == NULL)
        return Z_STREAM_ERROR;
    state = (gz_statep)file;

    /* check that we're writing */
    if (state->mode != GZ_WRITE)
        return Z_STREAM_ERROR;

    /* check for seek request */
    if (state->seek) {
        state->seek = 0;
        if (gz_zero(state, state->skip) == -1)
            ret = state->err;
    }

    /* flush, free memory, and close file */
    if (gz_comp(state, Z_FINISH) == -1)
        ret = state->err;
    if (state->size) {
        if (!state->direct) {
            (void)deflateEnd(&(state->strm));
            free(state->out);
        }
        free(state->in);
    }
    gz_error(state, Z_OK, NULL);
    free(state->path);
    if (close(state->fd) == -1)
        ret = Z_ERRNO;
    free(state);
    return ret;
}
> 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 5113 5114 5115 5116 5117 5118 5119 5120 5121 5122 5123 5124 5125 5126 5127 5128 5129 5130 5131 5132 5133 5134 5135 5136 5137 5138 5139 5140 5141 5142 5143 5144 5145 5146 5147 5148 5149 5150 5151 5152 5153 5154 5155 5156 5157 5158 5159 5160 5161 5162 5163 5164 5165 5166 5167 5168 5169 5170 5171 5172 5173 5174 5175 5176 5177 5178 5179 5180 5181 5182 5183 5184 5185 5186 5187 5188 5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213 5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236 5237 5238 5239
/*
 * jidctint.c
 *
 * Copyright (C) 1991-1998, Thomas G. Lane.
 * Modification developed 2002-2015 by Guido Vollbeding.
 * This file is part of the Independent JPEG Group's software.
 * For conditions of distribution and use, see the accompanying README file.
 *
 * This file contains a slow-but-accurate integer implementation of the
 * inverse DCT (Discrete Cosine Transform).  In the IJG code, this routine
 * must also perform dequantization of the input coefficients.
 *
 * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT
 * on each row (or vice versa, but it's more convenient to emit a row at
 * a time).  Direct algorithms are also available, but they are much more
 * complex and seem not to be any faster when reduced to code.
 *
 * This implementation is based on an algorithm described in
 *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
 *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
 *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
 * The primary algorithm described there uses 11 multiplies and 29 adds.
 * We use their alternate method with 12 multiplies and 32 adds.
 * The advantage of this method is that no data path contains more than one
 * multiplication; this allows a very simple and accurate implementation in
 * scaled fixed-point arithmetic, with a minimal number of shifts.
 *
 * We also provide IDCT routines with various output sample block sizes for
 * direct resolution reduction or enlargement and for direct resolving the
 * common 2x1 and 1x2 subsampling cases without additional resampling: NxN
 * (N=1...16), 2NxN, and Nx2N (N=1...8) pixels for one 8x8 input DCT block.
 *
 * For N<8 we simply take the corresponding low-frequency coefficients of
 * the 8x8 input DCT block and apply an NxN point IDCT on the sub-block
 * to yield the downscaled outputs.
 * This can be seen as direct low-pass downsampling from the DCT domain
 * point of view rather than the usual spatial domain point of view,
 * yielding significant computational savings and results at least
 * as good as common bilinear (averaging) spatial downsampling.
 *
 * For N>8 we apply a partial NxN IDCT on the 8 input coefficients as
 * lower frequencies and higher frequencies assumed to be zero.
 * It turns out that the computational effort is similar to the 8x8 IDCT
 * regarding the output size.
 * Furthermore, the scaling and descaling is the same for all IDCT sizes.
 *
 * CAUTION: We rely on the FIX() macro except for the N=1,2,4,8 cases
 * since there would be too many additional constants to pre-calculate.
 */

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

#ifdef DCT_ISLOW_SUPPORTED


/*
 * This module is specialized to the case DCTSIZE = 8.
 */

#if DCTSIZE != 8
  Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */
#endif


/*
 * The poop on this scaling stuff is as follows:
 *
 * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
 * larger than the true IDCT outputs.  The final outputs are therefore
 * a factor of N larger than desired; since N=8 this can be cured by
 * a simple right shift at the end of the algorithm.  The advantage of
 * this arrangement is that we save two multiplications per 1-D IDCT,
 * because the y0 and y4 inputs need not be divided by sqrt(N).
 *
 * We have to do addition and subtraction of the integer inputs, which
 * is no problem, and multiplication by fractional constants, which is
 * a problem to do in integer arithmetic.  We multiply all the constants
 * by CONST_SCALE and convert them to integer constants (thus retaining
 * CONST_BITS bits of precision in the constants).  After doing a
 * multiplication we have to divide the product by CONST_SCALE, with proper
 * rounding, to produce the correct output.  This division can be done
 * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
 * as long as possible so that partial sums can be added together with
 * full fractional precision.
 *
 * The outputs of the first pass are scaled up by PASS1_BITS bits so that
 * they are represented to better-than-integral precision.  These outputs
 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
 * with the recommended scaling.  (To scale up 12-bit sample data further, an
 * intermediate INT32 array would be needed.)
 *
 * To avoid overflow of the 32-bit intermediate results in pass 2, we must
 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26.  Error analysis
 * shows that the values given below are the most effective.
 */

#if BITS_IN_JSAMPLE == 8
#define CONST_BITS  13
#define PASS1_BITS  2
#else
#define CONST_BITS  13
#define PASS1_BITS  1		/* lose a little precision to avoid overflow */
#endif

/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus
 * causing a lot of useless floating-point operations at run time.
 * To get around this we use the following pre-calculated constants.
 * If you change CONST_BITS you may want to add appropriate values.
 * (With a reasonable C compiler, you can just rely on the FIX() macro...)
 */

#if CONST_BITS == 13
#define FIX_0_298631336  ((INT32)  2446)	/* FIX(0.298631336) */
#define FIX_0_390180644  ((INT32)  3196)	/* FIX(0.390180644) */
#define FIX_0_541196100  ((INT32)  4433)	/* FIX(0.541196100) */
#define FIX_0_765366865  ((INT32)  6270)	/* FIX(0.765366865) */
#define FIX_0_899976223  ((INT32)  7373)	/* FIX(0.899976223) */
#define FIX_1_175875602  ((INT32)  9633)	/* FIX(1.175875602) */
#define FIX_1_501321110  ((INT32)  12299)	/* FIX(1.501321110) */
#define FIX_1_847759065  ((INT32)  15137)	/* FIX(1.847759065) */
#define FIX_1_961570560  ((INT32)  16069)	/* FIX(1.961570560) */
#define FIX_2_053119869  ((INT32)  16819)	/* FIX(2.053119869) */
#define FIX_2_562915447  ((INT32)  20995)	/* FIX(2.562915447) */
#define FIX_3_072711026  ((INT32)  25172)	/* FIX(3.072711026) */
#else
#define FIX_0_298631336  FIX(0.298631336)
#define FIX_0_390180644  FIX(0.390180644)
#define FIX_0_541196100  FIX(0.541196100)
#define FIX_0_765366865  FIX(0.765366865)
#define FIX_0_899976223  FIX(0.899976223)
#define FIX_1_175875602  FIX(1.175875602)
#define FIX_1_501321110  FIX(1.501321110)
#define FIX_1_847759065  FIX(1.847759065)
#define FIX_1_961570560  FIX(1.961570560)
#define FIX_2_053119869  FIX(2.053119869)
#define FIX_2_562915447  FIX(2.562915447)
#define FIX_3_072711026  FIX(3.072711026)
#endif


/* Multiply an INT32 variable by an INT32 constant to yield an INT32 result.
 * For 8-bit samples with the recommended scaling, all the variable
 * and constant values involved are no more than 16 bits wide, so a
 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply.
 * For 12-bit samples, a full 32-bit multiplication will be needed.
 */

#if BITS_IN_JSAMPLE == 8
#define MULTIPLY(var,const)  MULTIPLY16C16(var,const)
#else
#define MULTIPLY(var,const)  ((var) * (const))
#endif


/* Dequantize a coefficient by multiplying it by the multiplier-table
 * entry; produce an int result.  In this module, both inputs and result
 * are 16 bits or less, so either int or short multiply will work.
 */

#define DEQUANTIZE(coef,quantval)  (((ISLOW_MULT_TYPE) (coef)) * (quantval))


/*
 * Perform dequantization and inverse DCT on one block of coefficients.
 *
 * cK represents sqrt(2) * cos(K*pi/16).
 */

GLOBAL(void)
jpeg_idct_islow (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		 JCOEFPTR coef_block,
		 JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3;
  INT32 tmp10, tmp11, tmp12, tmp13;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[DCTSIZE2];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * Note results are scaled up by sqrt(8) compared to a true IDCT;
   * furthermore, we scale the results by 2**PASS1_BITS.
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = DCTSIZE; ctr > 0; ctr--) {
    /* Due to quantization, we will usually find that many of the input
     * coefficients are zero, especially the AC terms.  We can exploit this
     * by short-circuiting the IDCT calculation for any column in which all
     * the AC terms are zero.  In that case each output is equal to the
     * DC coefficient (with scale factor as needed).
     * With typical images and quantization tables, half or more of the
     * column DCT calculations can be simplified this way.
     */

    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
	inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
	inptr[DCTSIZE*7] == 0) {
      /* AC terms all zero */
      int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;

      wsptr[DCTSIZE*0] = dcval;
      wsptr[DCTSIZE*1] = dcval;
      wsptr[DCTSIZE*2] = dcval;
      wsptr[DCTSIZE*3] = dcval;
      wsptr[DCTSIZE*4] = dcval;
      wsptr[DCTSIZE*5] = dcval;
      wsptr[DCTSIZE*6] = dcval;
      wsptr[DCTSIZE*7] = dcval;

      inptr++;			/* advance pointers to next column */
      quantptr++;
      wsptr++;
      continue;
    }

    /* Even part: reverse the even part of the forward DCT.
     * The rotator is c(-6).
     */

    z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z2 <<= CONST_BITS;
    z3 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z2 += ONE << (CONST_BITS-PASS1_BITS-1);

    tmp0 = z2 + z3;
    tmp1 = z2 - z3;

    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);       /* c6 */
    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);     /* c2-c6 */
    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);     /* c2+c6 */

    tmp10 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;
    tmp11 = tmp1 + tmp3;
    tmp12 = tmp1 - tmp3;

    /* Odd part per figure 8; the matrix is unitary and hence its
     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
     */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
    tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);

    z2 = tmp0 + tmp2;
    z3 = tmp1 + tmp3;

    z1 = MULTIPLY(z2 + z3, FIX_1_175875602);       /*  c3 */
    z2 = MULTIPLY(z2, - FIX_1_961570560);          /* -c3-c5 */
    z3 = MULTIPLY(z3, - FIX_0_390180644);          /* -c3+c5 */
    z2 += z1;
    z3 += z1;

    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
    tmp0 = MULTIPLY(tmp0, FIX_0_298631336);        /* -c1+c3+c5-c7 */
    tmp3 = MULTIPLY(tmp3, FIX_1_501321110);        /*  c1+c3-c5-c7 */
    tmp0 += z1 + z2;
    tmp3 += z1 + z3;

    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
    tmp1 = MULTIPLY(tmp1, FIX_2_053119869);        /*  c1+c3-c5+c7 */
    tmp2 = MULTIPLY(tmp2, FIX_3_072711026);        /*  c1+c3+c5-c7 */
    tmp1 += z1 + z3;
    tmp2 += z1 + z2;

    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */

    wsptr[DCTSIZE*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);

    inptr++;			/* advance pointers to next column */
    quantptr++;
    wsptr++;
  }

  /* Pass 2: process rows from work array, store into output array.
   * Note that we must descale the results by a factor of 8 == 2**3,
   * and also undo the PASS1_BITS scaling.
   */

  wsptr = workspace;
  for (ctr = 0; ctr < DCTSIZE; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Add range center and fudge factor for final descale and range-limit. */
    z2 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));

    /* Rows of zeroes can be exploited in the same way as we did with columns.
     * However, the column calculation has created many nonzero AC terms, so
     * the simplification applies less often (typically 5% to 10% of the time).
     * On machines with very fast multiplication, it's possible that the
     * test takes more time than it's worth.  In that case this section
     * may be commented out.
     */

#ifndef NO_ZERO_ROW_TEST
    if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 &&
	wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) {
      /* AC terms all zero */
      JSAMPLE dcval = range_limit[(int) RIGHT_SHIFT(z2, PASS1_BITS+3)
				  & RANGE_MASK];

      outptr[0] = dcval;
      outptr[1] = dcval;
      outptr[2] = dcval;
      outptr[3] = dcval;
      outptr[4] = dcval;
      outptr[5] = dcval;
      outptr[6] = dcval;
      outptr[7] = dcval;

      wsptr += DCTSIZE;		/* advance pointer to next row */
      continue;
    }
#endif

    /* Even part: reverse the even part of the forward DCT.
     * The rotator is c(-6).
     */

    z3 = (INT32) wsptr[4];

    tmp0 = (z2 + z3) << CONST_BITS;
    tmp1 = (z2 - z3) << CONST_BITS;

    z2 = (INT32) wsptr[2];
    z3 = (INT32) wsptr[6];

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);       /* c6 */
    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);     /* c2-c6 */
    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);     /* c2+c6 */

    tmp10 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;
    tmp11 = tmp1 + tmp3;
    tmp12 = tmp1 - tmp3;

    /* Odd part per figure 8; the matrix is unitary and hence its
     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
     */

    tmp0 = (INT32) wsptr[7];
    tmp1 = (INT32) wsptr[5];
    tmp2 = (INT32) wsptr[3];
    tmp3 = (INT32) wsptr[1];

    z2 = tmp0 + tmp2;
    z3 = tmp1 + tmp3;

    z1 = MULTIPLY(z2 + z3, FIX_1_175875602);       /*  c3 */
    z2 = MULTIPLY(z2, - FIX_1_961570560);          /* -c3-c5 */
    z3 = MULTIPLY(z3, - FIX_0_390180644);          /* -c3+c5 */
    z2 += z1;
    z3 += z1;

    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
    tmp0 = MULTIPLY(tmp0, FIX_0_298631336);        /* -c1+c3+c5-c7 */
    tmp3 = MULTIPLY(tmp3, FIX_1_501321110);        /*  c1+c3-c5-c7 */
    tmp0 += z1 + z2;
    tmp3 += z1 + z3;

    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
    tmp1 = MULTIPLY(tmp1, FIX_2_053119869);        /*  c1+c3-c5+c7 */
    tmp2 = MULTIPLY(tmp2, FIX_3_072711026);        /*  c1+c3+c5-c7 */
    tmp1 += z1 + z3;
    tmp2 += z1 + z2;

    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += DCTSIZE;		/* advance pointer to next row */
  }
}

#ifdef IDCT_SCALING_SUPPORTED


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 7x7 output block.
 *
 * Optimized algorithm with 12 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/14).
 */

GLOBAL(void)
jpeg_idct_7x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12, tmp13;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[7*7];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp13 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp13 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp13 += ONE << (CONST_BITS-PASS1_BITS-1);

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734));     /* c4 */
    tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123));     /* c6 */
    tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
    tmp0 = z1 + z3;
    z2 -= tmp0;
    tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
    tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536));  /* c2-c4-c6 */
    tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249));  /* c2+c4+c6 */
    tmp13 += MULTIPLY(z2, FIX(1.414213562));         /* c0 */

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);

    tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347));      /* (c3+c1-c5)/2 */
    tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339));      /* (c3+c5-c1)/2 */
    tmp0 = tmp1 - tmp2;
    tmp1 += tmp2;
    tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276));    /* -c1 */
    tmp1 += tmp2;
    z2 = MULTIPLY(z1 + z3, FIX(0.613604268));        /* c5 */
    tmp0 += z2;
    tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693));     /* c3+c1-c5 */

    /* Final output stage */

    wsptr[7*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[7*6] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
    wsptr[7*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
    wsptr[7*5] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
    wsptr[7*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
    wsptr[7*4] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
    wsptr[7*3] = (int) RIGHT_SHIFT(tmp13, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 7 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 7; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp13 = (INT32) wsptr[0] +
	      ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	       (ONE << (PASS1_BITS+2)));
    tmp13 <<= CONST_BITS;

    z1 = (INT32) wsptr[2];
    z2 = (INT32) wsptr[4];
    z3 = (INT32) wsptr[6];

    tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734));     /* c4 */
    tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123));     /* c6 */
    tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
    tmp0 = z1 + z3;
    z2 -= tmp0;
    tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */
    tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536));  /* c2-c4-c6 */
    tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249));  /* c2+c4+c6 */
    tmp13 += MULTIPLY(z2, FIX(1.414213562));         /* c0 */

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];

    tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347));      /* (c3+c1-c5)/2 */
    tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339));      /* (c3+c5-c1)/2 */
    tmp0 = tmp1 - tmp2;
    tmp1 += tmp2;
    tmp2 = MULTIPLY(z2 + z3, - FIX(1.378756276));    /* -c1 */
    tmp1 += tmp2;
    z2 = MULTIPLY(z1 + z3, FIX(0.613604268));        /* c5 */
    tmp0 += z2;
    tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693));     /* c3+c1-c5 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 7;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a reduced-size 6x6 output block.
 *
 * Optimized algorithm with 3 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/12).
 */

GLOBAL(void)
jpeg_idct_6x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[6*6];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp0 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    tmp10 = MULTIPLY(tmp2, FIX(0.707106781));   /* c4 */
    tmp1 = tmp0 + tmp10;
    tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
    tmp10 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    tmp0 = MULTIPLY(tmp10, FIX(1.224744871));   /* c2 */
    tmp10 = tmp1 + tmp0;
    tmp12 = tmp1 - tmp0;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
    tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
    tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
    tmp1 = (z1 - z2 - z3) << PASS1_BITS;

    /* Final output stage */

    wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[6*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
    wsptr[6*1] = (int) (tmp11 + tmp1);
    wsptr[6*4] = (int) (tmp11 - tmp1);
    wsptr[6*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
    wsptr[6*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 6 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 6; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp0 = (INT32) wsptr[0] +
	     ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	      (ONE << (PASS1_BITS+2)));
    tmp0 <<= CONST_BITS;
    tmp2 = (INT32) wsptr[4];
    tmp10 = MULTIPLY(tmp2, FIX(0.707106781));   /* c4 */
    tmp1 = tmp0 + tmp10;
    tmp11 = tmp0 - tmp10 - tmp10;
    tmp10 = (INT32) wsptr[2];
    tmp0 = MULTIPLY(tmp10, FIX(1.224744871));   /* c2 */
    tmp10 = tmp1 + tmp0;
    tmp12 = tmp1 - tmp0;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
    tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
    tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
    tmp1 = (z1 - z2 - z3) << CONST_BITS;

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 6;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a reduced-size 5x5 output block.
 *
 * Optimized algorithm with 5 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/10).
 */

GLOBAL(void)
jpeg_idct_5x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp10, tmp11, tmp12;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[5*5];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp12 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp12 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp12 += ONE << (CONST_BITS-PASS1_BITS-1);
    tmp0 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    tmp1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
    z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
    z3 = tmp12 + z2;
    tmp10 = z3 + z1;
    tmp11 = z3 - z1;
    tmp12 -= z2 << 2;

    /* Odd part */

    z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);

    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));     /* c3 */
    tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148));   /* c1-c3 */
    tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899));   /* c1+c3 */

    /* Final output stage */

    wsptr[5*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[5*4] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
    wsptr[5*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
    wsptr[5*3] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
    wsptr[5*2] = (int) RIGHT_SHIFT(tmp12, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 5 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 5; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp12 = (INT32) wsptr[0] +
	      ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	       (ONE << (PASS1_BITS+2)));
    tmp12 <<= CONST_BITS;
    tmp0 = (INT32) wsptr[2];
    tmp1 = (INT32) wsptr[4];
    z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */
    z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */
    z3 = tmp12 + z2;
    tmp10 = z3 + z1;
    tmp11 = z3 - z1;
    tmp12 -= z2 << 2;

    /* Odd part */

    z2 = (INT32) wsptr[1];
    z3 = (INT32) wsptr[3];

    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));     /* c3 */
    tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148));   /* c1-c3 */
    tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899));   /* c1+c3 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 5;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a reduced-size 4x4 output block.
 *
 * Optimized algorithm with 3 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
 */

GLOBAL(void)
jpeg_idct_4x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp2, tmp10, tmp12;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[4*4];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 4; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    
    tmp10 = (tmp0 + tmp2) << PASS1_BITS;
    tmp12 = (tmp0 - tmp2) << PASS1_BITS;

    /* Odd part */
    /* Same rotation as in the even part of the 8x8 LL&M IDCT */

    z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);               /* c6 */
    /* Add fudge factor here for final descale. */
    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
    tmp0 = RIGHT_SHIFT(z1 + MULTIPLY(z2, FIX_0_765366865), /* c2-c6 */
		       CONST_BITS-PASS1_BITS);
    tmp2 = RIGHT_SHIFT(z1 - MULTIPLY(z3, FIX_1_847759065), /* c2+c6 */
		       CONST_BITS-PASS1_BITS);

    /* Final output stage */

    wsptr[4*0] = (int) (tmp10 + tmp0);
    wsptr[4*3] = (int) (tmp10 - tmp0);
    wsptr[4*1] = (int) (tmp12 + tmp2);
    wsptr[4*2] = (int) (tmp12 - tmp2);
  }

  /* Pass 2: process 4 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 4; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp0 = (INT32) wsptr[0] +
	     ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	      (ONE << (PASS1_BITS+2)));
    tmp2 = (INT32) wsptr[2];

    tmp10 = (tmp0 + tmp2) << CONST_BITS;
    tmp12 = (tmp0 - tmp2) << CONST_BITS;

    /* Odd part */
    /* Same rotation as in the even part of the 8x8 LL&M IDCT */

    z2 = (INT32) wsptr[1];
    z3 = (INT32) wsptr[3];

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);   /* c6 */
    tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
    tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 4;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a reduced-size 3x3 output block.
 *
 * Optimized algorithm with 2 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/6).
 */

GLOBAL(void)
jpeg_idct_3x3 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp2, tmp10, tmp12;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[3*3];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp0 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
    tmp10 = tmp0 + tmp12;
    tmp2 = tmp0 - tmp12 - tmp12;

    /* Odd part */

    tmp12 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */

    /* Final output stage */

    wsptr[3*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[3*2] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
    wsptr[3*1] = (int) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 3 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 3; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp0 = (INT32) wsptr[0] +
	     ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	      (ONE << (PASS1_BITS+2)));
    tmp0 <<= CONST_BITS;
    tmp2 = (INT32) wsptr[2];
    tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
    tmp10 = tmp0 + tmp12;
    tmp2 = tmp0 - tmp12 - tmp12;

    /* Odd part */

    tmp12 = (INT32) wsptr[1];
    tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 3;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a reduced-size 2x2 output block.
 *
 * Multiplication-less algorithm.
 */

GLOBAL(void)
jpeg_idct_2x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  DCTELEM tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
  ISLOW_MULT_TYPE * quantptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  ISHIFT_TEMPS

  /* Pass 1: process columns from input. */

  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;

  /* Column 0 */
  tmp4 = DEQUANTIZE(coef_block[DCTSIZE*0], quantptr[DCTSIZE*0]);
  tmp5 = DEQUANTIZE(coef_block[DCTSIZE*1], quantptr[DCTSIZE*1]);
  /* Add range center and fudge factor for final descale and range-limit. */
  tmp4 += (((DCTELEM) RANGE_CENTER) << 3) + (1 << 2);

  tmp0 = tmp4 + tmp5;
  tmp2 = tmp4 - tmp5;

  /* Column 1 */
  tmp4 = DEQUANTIZE(coef_block[DCTSIZE*0+1], quantptr[DCTSIZE*0+1]);
  tmp5 = DEQUANTIZE(coef_block[DCTSIZE*1+1], quantptr[DCTSIZE*1+1]);

  tmp1 = tmp4 + tmp5;
  tmp3 = tmp4 - tmp5;

  /* Pass 2: process 2 rows, store into output array. */

  /* Row 0 */
  outptr = output_buf[0] + output_col;

  outptr[0] = range_limit[(int) IRIGHT_SHIFT(tmp0 + tmp1, 3) & RANGE_MASK];
  outptr[1] = range_limit[(int) IRIGHT_SHIFT(tmp0 - tmp1, 3) & RANGE_MASK];

  /* Row 1 */
  outptr = output_buf[1] + output_col;

  outptr[0] = range_limit[(int) IRIGHT_SHIFT(tmp2 + tmp3, 3) & RANGE_MASK];
  outptr[1] = range_limit[(int) IRIGHT_SHIFT(tmp2 - tmp3, 3) & RANGE_MASK];
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a reduced-size 1x1 output block.
 *
 * We hardly need an inverse DCT routine for this: just take the
 * average pixel value, which is one-eighth of the DC coefficient.
 */

GLOBAL(void)
jpeg_idct_1x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  DCTELEM dcval;
  ISLOW_MULT_TYPE * quantptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  ISHIFT_TEMPS

  /* 1x1 is trivial: just take the DC coefficient divided by 8. */

  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;

  dcval = DEQUANTIZE(coef_block[0], quantptr[0]);
  /* Add range center and fudge factor for descale and range-limit. */
  dcval += (((DCTELEM) RANGE_CENTER) << 3) + (1 << 2);

  output_buf[0][output_col] =
    range_limit[(int) IRIGHT_SHIFT(dcval, 3) & RANGE_MASK];
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 9x9 output block.
 *
 * Optimized algorithm with 10 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/18).
 */

GLOBAL(void)
jpeg_idct_9x9 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13, tmp14;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*9];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp0 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    tmp3 = MULTIPLY(z3, FIX(0.707106781));      /* c6 */
    tmp1 = tmp0 + tmp3;
    tmp2 = tmp0 - tmp3 - tmp3;

    tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
    tmp11 = tmp2 + tmp0;
    tmp14 = tmp2 - tmp0 - tmp0;

    tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
    tmp2 = MULTIPLY(z1, FIX(1.083350441));      /* c4 */
    tmp3 = MULTIPLY(z2, FIX(0.245575608));      /* c8 */

    tmp10 = tmp1 + tmp0 - tmp3;
    tmp12 = tmp1 - tmp0 + tmp2;
    tmp13 = tmp1 - tmp2 + tmp3;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    z2 = MULTIPLY(z2, - FIX(1.224744871));           /* -c3 */

    tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955));      /* c5 */
    tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525));      /* c7 */
    tmp0 = tmp2 + tmp3 - z2;
    tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481));      /* c1 */
    tmp2 += z2 - tmp1;
    tmp3 += z2 + tmp1;
    tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */

    /* Final output stage */

    wsptr[8*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[8*8] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
    wsptr[8*1] = (int) RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS-PASS1_BITS);
    wsptr[8*7] = (int) RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS-PASS1_BITS);
    wsptr[8*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
    wsptr[8*6] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
    wsptr[8*3] = (int) RIGHT_SHIFT(tmp13 + tmp3, CONST_BITS-PASS1_BITS);
    wsptr[8*5] = (int) RIGHT_SHIFT(tmp13 - tmp3, CONST_BITS-PASS1_BITS);
    wsptr[8*4] = (int) RIGHT_SHIFT(tmp14, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 9 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 9; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp0 = (INT32) wsptr[0] +
	     ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	      (ONE << (PASS1_BITS+2)));
    tmp0 <<= CONST_BITS;

    z1 = (INT32) wsptr[2];
    z2 = (INT32) wsptr[4];
    z3 = (INT32) wsptr[6];

    tmp3 = MULTIPLY(z3, FIX(0.707106781));      /* c6 */
    tmp1 = tmp0 + tmp3;
    tmp2 = tmp0 - tmp3 - tmp3;

    tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */
    tmp11 = tmp2 + tmp0;
    tmp14 = tmp2 - tmp0 - tmp0;

    tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */
    tmp2 = MULTIPLY(z1, FIX(1.083350441));      /* c4 */
    tmp3 = MULTIPLY(z2, FIX(0.245575608));      /* c8 */

    tmp10 = tmp1 + tmp0 - tmp3;
    tmp12 = tmp1 - tmp0 + tmp2;
    tmp13 = tmp1 - tmp2 + tmp3;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z4 = (INT32) wsptr[7];

    z2 = MULTIPLY(z2, - FIX(1.224744871));           /* -c3 */

    tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955));      /* c5 */
    tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525));      /* c7 */
    tmp0 = tmp2 + tmp3 - z2;
    tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481));      /* c1 */
    tmp2 += z2 - tmp1;
    tmp3 += z2 + tmp1;
    tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp3,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp3,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp14,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 10x10 output block.
 *
 * Optimized algorithm with 12 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/20).
 */

GLOBAL(void)
jpeg_idct_10x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		 JCOEFPTR coef_block,
		 JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
  INT32 z1, z2, z3, z4, z5;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*10];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z3 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z3 += ONE << (CONST_BITS-PASS1_BITS-1);
    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z1 = MULTIPLY(z4, FIX(1.144122806));         /* c4 */
    z2 = MULTIPLY(z4, FIX(0.437016024));         /* c8 */
    tmp10 = z3 + z1;
    tmp11 = z3 - z2;

    tmp22 = RIGHT_SHIFT(z3 - ((z1 - z2) << 1),   /* c0 = (c4-c8)*2 */
			CONST_BITS-PASS1_BITS);

    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));    /* c6 */
    tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
    tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */

    tmp20 = tmp10 + tmp12;
    tmp24 = tmp10 - tmp12;
    tmp21 = tmp11 + tmp13;
    tmp23 = tmp11 - tmp13;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    tmp11 = z2 + z4;
    tmp13 = z2 - z4;

    tmp12 = MULTIPLY(tmp13, FIX(0.309016994));        /* (c3-c7)/2 */
    z5 = z3 << CONST_BITS;

    z2 = MULTIPLY(tmp11, FIX(0.951056516));           /* (c3+c7)/2 */
    z4 = z5 + tmp12;

    tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
    tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */

    z2 = MULTIPLY(tmp11, FIX(0.587785252));           /* (c1-c9)/2 */
    z4 = z5 - tmp12 - (tmp13 << (CONST_BITS - 1));

    tmp12 = (z1 - tmp13 - z3) << PASS1_BITS;

    tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
    tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */

    /* Final output stage */

    wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*9] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*8] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*2] = (int) (tmp22 + tmp12);
    wsptr[8*7] = (int) (tmp22 - tmp12);
    wsptr[8*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*6] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*5] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 10 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 10; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    z3 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));
    z3 <<= CONST_BITS;
    z4 = (INT32) wsptr[4];
    z1 = MULTIPLY(z4, FIX(1.144122806));         /* c4 */
    z2 = MULTIPLY(z4, FIX(0.437016024));         /* c8 */
    tmp10 = z3 + z1;
    tmp11 = z3 - z2;

    tmp22 = z3 - ((z1 - z2) << 1);               /* c0 = (c4-c8)*2 */

    z2 = (INT32) wsptr[2];
    z3 = (INT32) wsptr[6];

    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));    /* c6 */
    tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
    tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */

    tmp20 = tmp10 + tmp12;
    tmp24 = tmp10 - tmp12;
    tmp21 = tmp11 + tmp13;
    tmp23 = tmp11 - tmp13;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z3 <<= CONST_BITS;
    z4 = (INT32) wsptr[7];

    tmp11 = z2 + z4;
    tmp13 = z2 - z4;

    tmp12 = MULTIPLY(tmp13, FIX(0.309016994));        /* (c3-c7)/2 */

    z2 = MULTIPLY(tmp11, FIX(0.951056516));           /* (c3+c7)/2 */
    z4 = z3 + tmp12;

    tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
    tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */

    z2 = MULTIPLY(tmp11, FIX(0.587785252));           /* (c1-c9)/2 */
    z4 = z3 - tmp12 - (tmp13 << (CONST_BITS - 1));

    tmp12 = ((z1 - tmp13) << CONST_BITS) - z3;

    tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
    tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 11x11 output block.
 *
 * Optimized algorithm with 24 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/22).
 */

GLOBAL(void)
jpeg_idct_11x11 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		 JCOEFPTR coef_block,
		 JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*11];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp10 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp10 += ONE << (CONST_BITS-PASS1_BITS-1);

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132));     /* c2+c4 */
    tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045));     /* c2-c6 */
    z4 = z1 + z3;
    tmp24 = MULTIPLY(z4, - FIX(1.155664402));        /* -(c2-c10) */
    z4 -= z2;
    tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976));  /* c2 */
    tmp21 = tmp20 + tmp23 + tmp25 -
	    MULTIPLY(z2, FIX(1.821790775));          /* c2+c4+c10-c6 */
    tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
    tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
    tmp24 += tmp25;
    tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120));  /* c8+c10 */
    tmp24 += MULTIPLY(z2, FIX(1.944413522)) -        /* c2+c8 */
	     MULTIPLY(z1, FIX(1.390975730));         /* c4+c10 */
    tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562));  /* c0 */

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    tmp11 = z1 + z2;
    tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
    tmp11 = MULTIPLY(tmp11, FIX(0.887983902));           /* c3-c9 */
    tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295));         /* c5-c9 */
    tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
    tmp10 = tmp11 + tmp12 + tmp13 -
	    MULTIPLY(z1, FIX(0.923107866));              /* c7+c5+c3-c1-2*c9 */
    z1    = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
    tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588));        /* c1+c7+3*c9-c3 */
    tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623));        /* c3+c5-c7-c9 */
    z1    = MULTIPLY(z2 + z4, - FIX(1.798248910));       /* -(c1+c9) */
    tmp11 += z1;
    tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632));        /* c1+c5+c9-c7 */
    tmp14 += MULTIPLY(z2, - FIX(1.467221301)) +          /* -(c5+c9) */
	     MULTIPLY(z3, FIX(1.001388905)) -            /* c1-c9 */
	     MULTIPLY(z4, FIX(1.684843907));             /* c3+c9 */

    /* Final output stage */

    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*10] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 11 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 11; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp10 = (INT32) wsptr[0] +
	      ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	       (ONE << (PASS1_BITS+2)));
    tmp10 <<= CONST_BITS;

    z1 = (INT32) wsptr[2];
    z2 = (INT32) wsptr[4];
    z3 = (INT32) wsptr[6];

    tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132));     /* c2+c4 */
    tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045));     /* c2-c6 */
    z4 = z1 + z3;
    tmp24 = MULTIPLY(z4, - FIX(1.155664402));        /* -(c2-c10) */
    z4 -= z2;
    tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976));  /* c2 */
    tmp21 = tmp20 + tmp23 + tmp25 -
	    MULTIPLY(z2, FIX(1.821790775));          /* c2+c4+c10-c6 */
    tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */
    tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */
    tmp24 += tmp25;
    tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120));  /* c8+c10 */
    tmp24 += MULTIPLY(z2, FIX(1.944413522)) -        /* c2+c8 */
	     MULTIPLY(z1, FIX(1.390975730));         /* c4+c10 */
    tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562));  /* c0 */

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z4 = (INT32) wsptr[7];

    tmp11 = z1 + z2;
    tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */
    tmp11 = MULTIPLY(tmp11, FIX(0.887983902));           /* c3-c9 */
    tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295));         /* c5-c9 */
    tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */
    tmp10 = tmp11 + tmp12 + tmp13 -
	    MULTIPLY(z1, FIX(0.923107866));              /* c7+c5+c3-c1-2*c9 */
    z1    = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */
    tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588));        /* c1+c7+3*c9-c3 */
    tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623));        /* c3+c5-c7-c9 */
    z1    = MULTIPLY(z2 + z4, - FIX(1.798248910));       /* -(c1+c9) */
    tmp11 += z1;
    tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632));        /* c1+c5+c9-c7 */
    tmp14 += MULTIPLY(z2, - FIX(1.467221301)) +          /* -(c5+c9) */
	     MULTIPLY(z3, FIX(1.001388905)) -            /* c1-c9 */
	     MULTIPLY(z4, FIX(1.684843907));             /* c3+c9 */

    /* Final output stage */

    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 12x12 output block.
 *
 * Optimized algorithm with 15 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/24).
 */

GLOBAL(void)
jpeg_idct_12x12 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		 JCOEFPTR coef_block,
		 JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*12];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z3 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z3 += ONE << (CONST_BITS-PASS1_BITS-1);

    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */

    tmp10 = z3 + z4;
    tmp11 = z3 - z4;

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
    z1 <<= CONST_BITS;
    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
    z2 <<= CONST_BITS;

    tmp12 = z1 - z2;

    tmp21 = z3 + tmp12;
    tmp24 = z3 - tmp12;

    tmp12 = z4 + z2;

    tmp20 = tmp10 + tmp12;
    tmp25 = tmp10 - tmp12;

    tmp12 = z4 - z1 - z2;

    tmp22 = tmp11 + tmp12;
    tmp23 = tmp11 - tmp12;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    tmp11 = MULTIPLY(z2, FIX(1.306562965));                  /* c3 */
    tmp14 = MULTIPLY(z2, - FIX_0_541196100);                 /* -c9 */

    tmp10 = z1 + z3;
    tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669));          /* c7 */
    tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384));       /* c5-c7 */
    tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716));  /* c1-c5 */
    tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580));           /* -(c7+c11) */
    tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
    tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
    tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) -        /* c7-c11 */
	     MULTIPLY(z4, FIX(1.982889723));                 /* c5+c7 */

    z1 -= z4;
    z2 -= z3;
    z3 = MULTIPLY(z1 + z2, FIX_0_541196100);                 /* c9 */
    tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865);              /* c3-c9 */
    tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065);              /* c3+c9 */

    /* Final output stage */

    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*11] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*10] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 12 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 12; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    z3 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));
    z3 <<= CONST_BITS;

    z4 = (INT32) wsptr[4];
    z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */

    tmp10 = z3 + z4;
    tmp11 = z3 - z4;

    z1 = (INT32) wsptr[2];
    z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
    z1 <<= CONST_BITS;
    z2 = (INT32) wsptr[6];
    z2 <<= CONST_BITS;

    tmp12 = z1 - z2;

    tmp21 = z3 + tmp12;
    tmp24 = z3 - tmp12;

    tmp12 = z4 + z2;

    tmp20 = tmp10 + tmp12;
    tmp25 = tmp10 - tmp12;

    tmp12 = z4 - z1 - z2;

    tmp22 = tmp11 + tmp12;
    tmp23 = tmp11 - tmp12;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z4 = (INT32) wsptr[7];

    tmp11 = MULTIPLY(z2, FIX(1.306562965));                  /* c3 */
    tmp14 = MULTIPLY(z2, - FIX_0_541196100);                 /* -c9 */

    tmp10 = z1 + z3;
    tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669));          /* c7 */
    tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384));       /* c5-c7 */
    tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716));  /* c1-c5 */
    tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580));           /* -(c7+c11) */
    tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
    tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
    tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) -        /* c7-c11 */
	     MULTIPLY(z4, FIX(1.982889723));                 /* c5+c7 */

    z1 -= z4;
    z2 -= z3;
    z3 = MULTIPLY(z1 + z2, FIX_0_541196100);                 /* c9 */
    tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865);              /* c3-c9 */
    tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065);              /* c3+c9 */

    /* Final output stage */

    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 13x13 output block.
 *
 * Optimized algorithm with 29 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/26).
 */

GLOBAL(void)
jpeg_idct_13x13 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		 JCOEFPTR coef_block,
		 JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*13];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z1 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z1 += ONE << (CONST_BITS-PASS1_BITS-1);

    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    tmp10 = z3 + z4;
    tmp11 = z3 - z4;

    tmp12 = MULTIPLY(tmp10, FIX(1.155388986));                /* (c4+c6)/2 */
    tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1;           /* (c4-c6)/2 */

    tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13;   /* c2 */
    tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13;   /* c10 */

    tmp12 = MULTIPLY(tmp10, FIX(0.316450131));                /* (c8-c12)/2 */
    tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1;           /* (c8+c12)/2 */

    tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13;   /* c6 */
    tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */

    tmp12 = MULTIPLY(tmp10, FIX(0.435816023));                /* (c2-c10)/2 */
    tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1;           /* (c2+c10)/2 */

    tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
    tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */

    tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1;      /* c0 */

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651));     /* c3 */
    tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945));     /* c5 */
    tmp15 = z1 + z4;
    tmp13 = MULTIPLY(tmp15, FIX(0.937797057));       /* c7 */
    tmp10 = tmp11 + tmp12 + tmp13 -
	    MULTIPLY(z1, FIX(2.020082300));          /* c7+c5+c3-c1 */
    tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458));   /* -c11 */
    tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
    tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
    tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945));   /* -c5 */
    tmp11 += tmp14;
    tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
    tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813));   /* -c9 */
    tmp12 += tmp14;
    tmp13 += tmp14;
    tmp15 = MULTIPLY(tmp15, FIX(0.338443458));       /* c11 */
    tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
	    MULTIPLY(z2, FIX(0.466105296));          /* c1-c7 */
    z1    = MULTIPLY(z3 - z2, FIX(0.937797057));     /* c7 */
    tmp14 += z1;
    tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) -   /* c3-c7 */
	     MULTIPLY(z4, FIX(1.742345811));         /* c1+c11 */

    /* Final output stage */

    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*12] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*11] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*10] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp26, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 13 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 13; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    z1 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));
    z1 <<= CONST_BITS;

    z2 = (INT32) wsptr[2];
    z3 = (INT32) wsptr[4];
    z4 = (INT32) wsptr[6];

    tmp10 = z3 + z4;
    tmp11 = z3 - z4;

    tmp12 = MULTIPLY(tmp10, FIX(1.155388986));                /* (c4+c6)/2 */
    tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1;           /* (c4-c6)/2 */

    tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13;   /* c2 */
    tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13;   /* c10 */

    tmp12 = MULTIPLY(tmp10, FIX(0.316450131));                /* (c8-c12)/2 */
    tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1;           /* (c8+c12)/2 */

    tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13;   /* c6 */
    tmp25 = MULTIPLY(z2, - FIX(1.252223920)) + tmp12 + tmp13; /* c4 */

    tmp12 = MULTIPLY(tmp10, FIX(0.435816023));                /* (c2-c10)/2 */
    tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1;           /* (c2+c10)/2 */

    tmp23 = MULTIPLY(z2, - FIX(0.170464608)) - tmp12 - tmp13; /* c12 */
    tmp24 = MULTIPLY(z2, - FIX(0.803364869)) + tmp12 - tmp13; /* c8 */

    tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1;      /* c0 */

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z4 = (INT32) wsptr[7];

    tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651));     /* c3 */
    tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945));     /* c5 */
    tmp15 = z1 + z4;
    tmp13 = MULTIPLY(tmp15, FIX(0.937797057));       /* c7 */
    tmp10 = tmp11 + tmp12 + tmp13 -
	    MULTIPLY(z1, FIX(2.020082300));          /* c7+c5+c3-c1 */
    tmp14 = MULTIPLY(z2 + z3, - FIX(0.338443458));   /* -c11 */
    tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */
    tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */
    tmp14 = MULTIPLY(z2 + z4, - FIX(1.163874945));   /* -c5 */
    tmp11 += tmp14;
    tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */
    tmp14 = MULTIPLY(z3 + z4, - FIX(0.657217813));   /* -c9 */
    tmp12 += tmp14;
    tmp13 += tmp14;
    tmp15 = MULTIPLY(tmp15, FIX(0.338443458));       /* c11 */
    tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */
	    MULTIPLY(z2, FIX(0.466105296));          /* c1-c7 */
    z1    = MULTIPLY(z3 - z2, FIX(0.937797057));     /* c7 */
    tmp14 += z1;
    tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) -   /* c3-c7 */
	     MULTIPLY(z4, FIX(1.742345811));         /* c1+c11 */

    /* Final output stage */

    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 14x14 output block.
 *
 * Optimized algorithm with 20 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/28).
 */

GLOBAL(void)
jpeg_idct_14x14 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		 JCOEFPTR coef_block,
		 JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*14];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z1 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z2 = MULTIPLY(z4, FIX(1.274162392));         /* c4 */
    z3 = MULTIPLY(z4, FIX(0.314692123));         /* c12 */
    z4 = MULTIPLY(z4, FIX(0.881747734));         /* c8 */

    tmp10 = z1 + z2;
    tmp11 = z1 + z3;
    tmp12 = z1 - z4;

    tmp23 = RIGHT_SHIFT(z1 - ((z2 + z3 - z4) << 1), /* c0 = (c4+c12-c8)*2 */
			CONST_BITS-PASS1_BITS);

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    z3 = MULTIPLY(z1 + z2, FIX(1.105676686));    /* c6 */

    tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
    tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
    tmp15 = MULTIPLY(z1, FIX(0.613604268)) -     /* c10 */
	    MULTIPLY(z2, FIX(1.378756276));      /* c2 */

    tmp20 = tmp10 + tmp13;
    tmp26 = tmp10 - tmp13;
    tmp21 = tmp11 + tmp14;
    tmp25 = tmp11 - tmp14;
    tmp22 = tmp12 + tmp15;
    tmp24 = tmp12 - tmp15;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
    tmp13 = z4 << CONST_BITS;

    tmp14 = z1 + z3;
    tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607));           /* c3 */
    tmp12 = MULTIPLY(tmp14, FIX(1.197448846));             /* c5 */
    tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
    tmp14 = MULTIPLY(tmp14, FIX(0.752406978));             /* c9 */
    tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426));        /* c9+c11-c13 */
    z1    -= z2;
    tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13;        /* c11 */
    tmp16 += tmp15;
    z1    += z4;
    z4    = MULTIPLY(z2 + z3, - FIX(0.158341681)) - tmp13; /* -c13 */
    tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948));          /* c3-c9-c13 */
    tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773));          /* c3+c5-c13 */
    z4    = MULTIPLY(z3 - z2, FIX(1.405321284));           /* c1 */
    tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
    tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567));          /* c1+c11-c5 */

    tmp13 = (z1 - z3) << PASS1_BITS;

    /* Final output stage */

    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*13] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*12] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*11] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*3]  = (int) (tmp23 + tmp13);
    wsptr[8*10] = (int) (tmp23 - tmp13);
    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 14 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 14; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    z1 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));
    z1 <<= CONST_BITS;
    z4 = (INT32) wsptr[4];
    z2 = MULTIPLY(z4, FIX(1.274162392));         /* c4 */
    z3 = MULTIPLY(z4, FIX(0.314692123));         /* c12 */
    z4 = MULTIPLY(z4, FIX(0.881747734));         /* c8 */

    tmp10 = z1 + z2;
    tmp11 = z1 + z3;
    tmp12 = z1 - z4;

    tmp23 = z1 - ((z2 + z3 - z4) << 1);          /* c0 = (c4+c12-c8)*2 */

    z1 = (INT32) wsptr[2];
    z2 = (INT32) wsptr[6];

    z3 = MULTIPLY(z1 + z2, FIX(1.105676686));    /* c6 */

    tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
    tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
    tmp15 = MULTIPLY(z1, FIX(0.613604268)) -     /* c10 */
	    MULTIPLY(z2, FIX(1.378756276));      /* c2 */

    tmp20 = tmp10 + tmp13;
    tmp26 = tmp10 - tmp13;
    tmp21 = tmp11 + tmp14;
    tmp25 = tmp11 - tmp14;
    tmp22 = tmp12 + tmp15;
    tmp24 = tmp12 - tmp15;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z4 = (INT32) wsptr[7];
    z4 <<= CONST_BITS;

    tmp14 = z1 + z3;
    tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607));           /* c3 */
    tmp12 = MULTIPLY(tmp14, FIX(1.197448846));             /* c5 */
    tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
    tmp14 = MULTIPLY(tmp14, FIX(0.752406978));             /* c9 */
    tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426));        /* c9+c11-c13 */
    z1    -= z2;
    tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4;           /* c11 */
    tmp16 += tmp15;
    tmp13 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - z4;    /* -c13 */
    tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948));       /* c3-c9-c13 */
    tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773));       /* c3+c5-c13 */
    tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284));           /* c1 */
    tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
    tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567));       /* c1+c11-c5 */

    tmp13 = ((z1 - z3) << CONST_BITS) + z4;

    /* Final output stage */

    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 15x15 output block.
 *
 * Optimized algorithm with 22 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/30).
 */

GLOBAL(void)
jpeg_idct_15x15 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		 JCOEFPTR coef_block,
		 JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*15];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z1 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z1 += ONE << (CONST_BITS-PASS1_BITS-1);

    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
    tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */

    tmp12 = z1 - tmp10;
    tmp13 = z1 + tmp11;
    z1 -= (tmp11 - tmp10) << 1;             /* c0 = (c6-c12)*2 */

    z4 = z2 - z3;
    z3 += z2;
    tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
    tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
    z2 = MULTIPLY(z2, FIX(1.439773946));    /* c4+c14 */

    tmp20 = tmp13 + tmp10 + tmp11;
    tmp23 = tmp12 - tmp10 + tmp11 + z2;

    tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
    tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */

    tmp25 = tmp13 - tmp10 - tmp11;
    tmp26 = tmp12 + tmp10 - tmp11 - z2;

    tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
    tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */

    tmp21 = tmp12 + tmp10 + tmp11;
    tmp24 = tmp13 - tmp10 + tmp11;
    tmp11 += tmp11;
    tmp22 = z1 + tmp11;                     /* c10 = c6-c12 */
    tmp27 = z1 - tmp11 - tmp11;             /* c0 = (c6-c12)*2 */

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z3 = MULTIPLY(z4, FIX(1.224744871));                    /* c5 */
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    tmp13 = z2 - z4;
    tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876));         /* c9 */
    tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148));         /* c3-c9 */
    tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899));      /* c3+c9 */

    tmp13 = MULTIPLY(z2, - FIX(0.831253876));               /* -c9 */
    tmp15 = MULTIPLY(z2, - FIX(1.344997024));               /* -c3 */
    z2 = z1 - z4;
    tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353));            /* c1 */

    tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
    tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
    tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3;            /* c5 */
    z2 = MULTIPLY(z1 + z4, FIX(0.575212477));               /* c11 */
    tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3;      /* c7-c11 */
    tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3;      /* c11+c13 */

    /* Final output stage */

    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*14] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*13] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*12] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*11] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*10] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp27, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 15 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 15; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    z1 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));
    z1 <<= CONST_BITS;

    z2 = (INT32) wsptr[2];
    z3 = (INT32) wsptr[4];
    z4 = (INT32) wsptr[6];

    tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */
    tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */

    tmp12 = z1 - tmp10;
    tmp13 = z1 + tmp11;
    z1 -= (tmp11 - tmp10) << 1;             /* c0 = (c6-c12)*2 */

    z4 = z2 - z3;
    z3 += z2;
    tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */
    tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */
    z2 = MULTIPLY(z2, FIX(1.439773946));    /* c4+c14 */

    tmp20 = tmp13 + tmp10 + tmp11;
    tmp23 = tmp12 - tmp10 + tmp11 + z2;

    tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */
    tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */

    tmp25 = tmp13 - tmp10 - tmp11;
    tmp26 = tmp12 + tmp10 - tmp11 - z2;

    tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */
    tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */

    tmp21 = tmp12 + tmp10 + tmp11;
    tmp24 = tmp13 - tmp10 + tmp11;
    tmp11 += tmp11;
    tmp22 = z1 + tmp11;                     /* c10 = c6-c12 */
    tmp27 = z1 - tmp11 - tmp11;             /* c0 = (c6-c12)*2 */

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z4 = (INT32) wsptr[5];
    z3 = MULTIPLY(z4, FIX(1.224744871));                    /* c5 */
    z4 = (INT32) wsptr[7];

    tmp13 = z2 - z4;
    tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876));         /* c9 */
    tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148));         /* c3-c9 */
    tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899));      /* c3+c9 */

    tmp13 = MULTIPLY(z2, - FIX(0.831253876));               /* -c9 */
    tmp15 = MULTIPLY(z2, - FIX(1.344997024));               /* -c3 */
    z2 = z1 - z4;
    tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353));            /* c1 */

    tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */
    tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */
    tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3;            /* c5 */
    z2 = MULTIPLY(z1 + z4, FIX(0.575212477));               /* c11 */
    tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3;      /* c7-c11 */
    tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3;      /* c11+c13 */

    /* Final output stage */

    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp27,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 16x16 output block.
 *
 * Optimized algorithm with 28 multiplications in the 1-D kernel.
 * cK represents sqrt(2) * cos(K*pi/32).
 */

GLOBAL(void)
jpeg_idct_16x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		 JCOEFPTR coef_block,
		 JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*16];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp0 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp0 += 1 << (CONST_BITS-PASS1_BITS-1);

    z1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    tmp1 = MULTIPLY(z1, FIX(1.306562965));      /* c4[16] = c2[8] */
    tmp2 = MULTIPLY(z1, FIX_0_541196100);       /* c12[16] = c6[8] */

    tmp10 = tmp0 + tmp1;
    tmp11 = tmp0 - tmp1;
    tmp12 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
    z3 = z1 - z2;
    z4 = MULTIPLY(z3, FIX(0.275899379));        /* c14[16] = c7[8] */
    z3 = MULTIPLY(z3, FIX(1.387039845));        /* c2[16] = c1[8] */

    tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447);  /* (c6+c2)[16] = (c3+c1)[8] */
    tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223);  /* (c6-c14)[16] = (c3-c7)[8] */
    tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
    tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */

    tmp20 = tmp10 + tmp0;
    tmp27 = tmp10 - tmp0;
    tmp21 = tmp12 + tmp1;
    tmp26 = tmp12 - tmp1;
    tmp22 = tmp13 + tmp2;
    tmp25 = tmp13 - tmp2;
    tmp23 = tmp11 + tmp3;
    tmp24 = tmp11 - tmp3;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    tmp11 = z1 + z3;

    tmp1  = MULTIPLY(z1 + z2, FIX(1.353318001));   /* c3 */
    tmp2  = MULTIPLY(tmp11,   FIX(1.247225013));   /* c5 */
    tmp3  = MULTIPLY(z1 + z4, FIX(1.093201867));   /* c7 */
    tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586));   /* c9 */
    tmp11 = MULTIPLY(tmp11,   FIX(0.666655658));   /* c11 */
    tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528));   /* c13 */
    tmp0  = tmp1 + tmp2 + tmp3 -
	    MULTIPLY(z1, FIX(2.286341144));        /* c7+c5+c3-c1 */
    tmp13 = tmp10 + tmp11 + tmp12 -
	    MULTIPLY(z1, FIX(1.835730603));        /* c9+c11+c13-c15 */
    z1    = MULTIPLY(z2 + z3, FIX(0.138617169));   /* c15 */
    tmp1  += z1 + MULTIPLY(z2, FIX(0.071888074));  /* c9+c11-c3-c15 */
    tmp2  += z1 - MULTIPLY(z3, FIX(1.125726048));  /* c5+c7+c15-c3 */
    z1    = MULTIPLY(z3 - z2, FIX(1.407403738));   /* c1 */
    tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282));  /* c1+c11-c9-c13 */
    tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411));  /* c1+c5+c13-c7 */
    z2    += z4;
    z1    = MULTIPLY(z2, - FIX(0.666655658));      /* -c11 */
    tmp1  += z1;
    tmp3  += z1 + MULTIPLY(z4, FIX(1.065388962));  /* c3+c11+c15-c7 */
    z2    = MULTIPLY(z2, - FIX(1.247225013));      /* -c5 */
    tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809));  /* c1+c5+c9-c13 */
    tmp12 += z2;
    z2    = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
    tmp2  += z2;
    tmp3  += z2;
    z2    = MULTIPLY(z4 - z3, FIX(0.410524528));   /* c13 */
    tmp10 += z2;
    tmp11 += z2;

    /* Final output stage */

    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp0,  CONST_BITS-PASS1_BITS);
    wsptr[8*15] = (int) RIGHT_SHIFT(tmp20 - tmp0,  CONST_BITS-PASS1_BITS);
    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp1,  CONST_BITS-PASS1_BITS);
    wsptr[8*14] = (int) RIGHT_SHIFT(tmp21 - tmp1,  CONST_BITS-PASS1_BITS);
    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp2,  CONST_BITS-PASS1_BITS);
    wsptr[8*13] = (int) RIGHT_SHIFT(tmp22 - tmp2,  CONST_BITS-PASS1_BITS);
    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp3,  CONST_BITS-PASS1_BITS);
    wsptr[8*12] = (int) RIGHT_SHIFT(tmp23 - tmp3,  CONST_BITS-PASS1_BITS);
    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*11] = (int) RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*10] = (int) RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 16 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 16; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp0 = (INT32) wsptr[0] +
	     ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	      (ONE << (PASS1_BITS+2)));
    tmp0 <<= CONST_BITS;

    z1 = (INT32) wsptr[4];
    tmp1 = MULTIPLY(z1, FIX(1.306562965));      /* c4[16] = c2[8] */
    tmp2 = MULTIPLY(z1, FIX_0_541196100);       /* c12[16] = c6[8] */

    tmp10 = tmp0 + tmp1;
    tmp11 = tmp0 - tmp1;
    tmp12 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;

    z1 = (INT32) wsptr[2];
    z2 = (INT32) wsptr[6];
    z3 = z1 - z2;
    z4 = MULTIPLY(z3, FIX(0.275899379));        /* c14[16] = c7[8] */
    z3 = MULTIPLY(z3, FIX(1.387039845));        /* c2[16] = c1[8] */

    tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447);  /* (c6+c2)[16] = (c3+c1)[8] */
    tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223);  /* (c6-c14)[16] = (c3-c7)[8] */
    tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
    tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */

    tmp20 = tmp10 + tmp0;
    tmp27 = tmp10 - tmp0;
    tmp21 = tmp12 + tmp1;
    tmp26 = tmp12 - tmp1;
    tmp22 = tmp13 + tmp2;
    tmp25 = tmp13 - tmp2;
    tmp23 = tmp11 + tmp3;
    tmp24 = tmp11 - tmp3;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z4 = (INT32) wsptr[7];

    tmp11 = z1 + z3;

    tmp1  = MULTIPLY(z1 + z2, FIX(1.353318001));   /* c3 */
    tmp2  = MULTIPLY(tmp11,   FIX(1.247225013));   /* c5 */
    tmp3  = MULTIPLY(z1 + z4, FIX(1.093201867));   /* c7 */
    tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586));   /* c9 */
    tmp11 = MULTIPLY(tmp11,   FIX(0.666655658));   /* c11 */
    tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528));   /* c13 */
    tmp0  = tmp1 + tmp2 + tmp3 -
	    MULTIPLY(z1, FIX(2.286341144));        /* c7+c5+c3-c1 */
    tmp13 = tmp10 + tmp11 + tmp12 -
	    MULTIPLY(z1, FIX(1.835730603));        /* c9+c11+c13-c15 */
    z1    = MULTIPLY(z2 + z3, FIX(0.138617169));   /* c15 */
    tmp1  += z1 + MULTIPLY(z2, FIX(0.071888074));  /* c9+c11-c3-c15 */
    tmp2  += z1 - MULTIPLY(z3, FIX(1.125726048));  /* c5+c7+c15-c3 */
    z1    = MULTIPLY(z3 - z2, FIX(1.407403738));   /* c1 */
    tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282));  /* c1+c11-c9-c13 */
    tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411));  /* c1+c5+c13-c7 */
    z2    += z4;
    z1    = MULTIPLY(z2, - FIX(0.666655658));      /* -c11 */
    tmp1  += z1;
    tmp3  += z1 + MULTIPLY(z4, FIX(1.065388962));  /* c3+c11+c15-c7 */
    z2    = MULTIPLY(z2, - FIX(1.247225013));      /* -c5 */
    tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809));  /* c1+c5+c9-c13 */
    tmp12 += z2;
    z2    = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
    tmp2  += z2;
    tmp3  += z2;
    z2    = MULTIPLY(z4 - z3, FIX(0.410524528));   /* c13 */
    tmp10 += z2;
    tmp11 += z2;

    /* Final output stage */

    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp0,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[15] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp0,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp1,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp1,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp2,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp2,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp3,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp3,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp27 + tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp27 - tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 16x8 output block.
 *
 * 8-point IDCT in pass 1 (columns), 16-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_16x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		JCOEFPTR coef_block,
		JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*8];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * Note results are scaled up by sqrt(8) compared to a true IDCT;
   * furthermore, we scale the results by 2**PASS1_BITS.
   * 8-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = DCTSIZE; ctr > 0; ctr--) {
    /* Due to quantization, we will usually find that many of the input
     * coefficients are zero, especially the AC terms.  We can exploit this
     * by short-circuiting the IDCT calculation for any column in which all
     * the AC terms are zero.  In that case each output is equal to the
     * DC coefficient (with scale factor as needed).
     * With typical images and quantization tables, half or more of the
     * column DCT calculations can be simplified this way.
     */

    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
	inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
	inptr[DCTSIZE*7] == 0) {
      /* AC terms all zero */
      int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;

      wsptr[DCTSIZE*0] = dcval;
      wsptr[DCTSIZE*1] = dcval;
      wsptr[DCTSIZE*2] = dcval;
      wsptr[DCTSIZE*3] = dcval;
      wsptr[DCTSIZE*4] = dcval;
      wsptr[DCTSIZE*5] = dcval;
      wsptr[DCTSIZE*6] = dcval;
      wsptr[DCTSIZE*7] = dcval;

      inptr++;			/* advance pointers to next column */
      quantptr++;
      wsptr++;
      continue;
    }

    /* Even part: reverse the even part of the forward DCT.
     * The rotator is c(-6).
     */

    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);       /* c6 */
    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);     /* c2-c6 */
    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);     /* c2+c6 */

    z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z2 <<= CONST_BITS;
    z3 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z2 += ONE << (CONST_BITS-PASS1_BITS-1);

    tmp0 = z2 + z3;
    tmp1 = z2 - z3;

    tmp10 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;
    tmp11 = tmp1 + tmp3;
    tmp12 = tmp1 - tmp3;

    /* Odd part per figure 8; the matrix is unitary and hence its
     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
     */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
    tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);

    z2 = tmp0 + tmp2;
    z3 = tmp1 + tmp3;

    z1 = MULTIPLY(z2 + z3, FIX_1_175875602);       /*  c3 */
    z2 = MULTIPLY(z2, - FIX_1_961570560);          /* -c3-c5 */
    z3 = MULTIPLY(z3, - FIX_0_390180644);          /* -c3+c5 */
    z2 += z1;
    z3 += z1;

    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
    tmp0 = MULTIPLY(tmp0, FIX_0_298631336);        /* -c1+c3+c5-c7 */
    tmp3 = MULTIPLY(tmp3, FIX_1_501321110);        /*  c1+c3-c5-c7 */
    tmp0 += z1 + z2;
    tmp3 += z1 + z3;

    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
    tmp1 = MULTIPLY(tmp1, FIX_2_053119869);        /*  c1+c3-c5+c7 */
    tmp2 = MULTIPLY(tmp2, FIX_3_072711026);        /*  c1+c3+c5-c7 */
    tmp1 += z1 + z3;
    tmp2 += z1 + z2;

    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */

    wsptr[DCTSIZE*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[DCTSIZE*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);

    inptr++;			/* advance pointers to next column */
    quantptr++;
    wsptr++;
  }

  /* Pass 2: process 8 rows from work array, store into output array.
   * 16-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/32).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp0 = (INT32) wsptr[0] +
	     ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	      (ONE << (PASS1_BITS+2)));
    tmp0 <<= CONST_BITS;

    z1 = (INT32) wsptr[4];
    tmp1 = MULTIPLY(z1, FIX(1.306562965));      /* c4[16] = c2[8] */
    tmp2 = MULTIPLY(z1, FIX_0_541196100);       /* c12[16] = c6[8] */

    tmp10 = tmp0 + tmp1;
    tmp11 = tmp0 - tmp1;
    tmp12 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;

    z1 = (INT32) wsptr[2];
    z2 = (INT32) wsptr[6];
    z3 = z1 - z2;
    z4 = MULTIPLY(z3, FIX(0.275899379));        /* c14[16] = c7[8] */
    z3 = MULTIPLY(z3, FIX(1.387039845));        /* c2[16] = c1[8] */

    tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447);  /* (c6+c2)[16] = (c3+c1)[8] */
    tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223);  /* (c6-c14)[16] = (c3-c7)[8] */
    tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
    tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */

    tmp20 = tmp10 + tmp0;
    tmp27 = tmp10 - tmp0;
    tmp21 = tmp12 + tmp1;
    tmp26 = tmp12 - tmp1;
    tmp22 = tmp13 + tmp2;
    tmp25 = tmp13 - tmp2;
    tmp23 = tmp11 + tmp3;
    tmp24 = tmp11 - tmp3;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z4 = (INT32) wsptr[7];

    tmp11 = z1 + z3;

    tmp1  = MULTIPLY(z1 + z2, FIX(1.353318001));   /* c3 */
    tmp2  = MULTIPLY(tmp11,   FIX(1.247225013));   /* c5 */
    tmp3  = MULTIPLY(z1 + z4, FIX(1.093201867));   /* c7 */
    tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586));   /* c9 */
    tmp11 = MULTIPLY(tmp11,   FIX(0.666655658));   /* c11 */
    tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528));   /* c13 */
    tmp0  = tmp1 + tmp2 + tmp3 -
	    MULTIPLY(z1, FIX(2.286341144));        /* c7+c5+c3-c1 */
    tmp13 = tmp10 + tmp11 + tmp12 -
	    MULTIPLY(z1, FIX(1.835730603));        /* c9+c11+c13-c15 */
    z1    = MULTIPLY(z2 + z3, FIX(0.138617169));   /* c15 */
    tmp1  += z1 + MULTIPLY(z2, FIX(0.071888074));  /* c9+c11-c3-c15 */
    tmp2  += z1 - MULTIPLY(z3, FIX(1.125726048));  /* c5+c7+c15-c3 */
    z1    = MULTIPLY(z3 - z2, FIX(1.407403738));   /* c1 */
    tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282));  /* c1+c11-c9-c13 */
    tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411));  /* c1+c5+c13-c7 */
    z2    += z4;
    z1    = MULTIPLY(z2, - FIX(0.666655658));      /* -c11 */
    tmp1  += z1;
    tmp3  += z1 + MULTIPLY(z4, FIX(1.065388962));  /* c3+c11+c15-c7 */
    z2    = MULTIPLY(z2, - FIX(1.247225013));      /* -c5 */
    tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809));  /* c1+c5+c9-c13 */
    tmp12 += z2;
    z2    = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
    tmp2  += z2;
    tmp3  += z2;
    z2    = MULTIPLY(z4 - z3, FIX(0.410524528));   /* c13 */
    tmp10 += z2;
    tmp11 += z2;

    /* Final output stage */

    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp0,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[15] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp0,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp1,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[14] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp1,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp2,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp2,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp3,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp3,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp27 + tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp27 - tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 14x7 output block.
 *
 * 7-point IDCT in pass 1 (columns), 14-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_14x7 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		JCOEFPTR coef_block,
		JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*7];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 7-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/14).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp23 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp23 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp23 += ONE << (CONST_BITS-PASS1_BITS-1);

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    tmp20 = MULTIPLY(z2 - z3, FIX(0.881747734));       /* c4 */
    tmp22 = MULTIPLY(z1 - z2, FIX(0.314692123));       /* c6 */
    tmp21 = tmp20 + tmp22 + tmp23 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
    tmp10 = z1 + z3;
    z2 -= tmp10;
    tmp10 = MULTIPLY(tmp10, FIX(1.274162392)) + tmp23; /* c2 */
    tmp20 += tmp10 - MULTIPLY(z3, FIX(0.077722536));   /* c2-c4-c6 */
    tmp22 += tmp10 - MULTIPLY(z1, FIX(2.470602249));   /* c2+c4+c6 */
    tmp23 += MULTIPLY(z2, FIX(1.414213562));           /* c0 */

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);

    tmp11 = MULTIPLY(z1 + z2, FIX(0.935414347));       /* (c3+c1-c5)/2 */
    tmp12 = MULTIPLY(z1 - z2, FIX(0.170262339));       /* (c3+c5-c1)/2 */
    tmp10 = tmp11 - tmp12;
    tmp11 += tmp12;
    tmp12 = MULTIPLY(z2 + z3, - FIX(1.378756276));     /* -c1 */
    tmp11 += tmp12;
    z2 = MULTIPLY(z1 + z3, FIX(0.613604268));          /* c5 */
    tmp10 += z2;
    tmp12 += z2 + MULTIPLY(z3, FIX(1.870828693));      /* c3+c1-c5 */

    /* Final output stage */

    wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*6] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*5] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*4] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*3] = (int) RIGHT_SHIFT(tmp23, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 7 rows from work array, store into output array.
   * 14-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/28).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 7; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    z1 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));
    z1 <<= CONST_BITS;
    z4 = (INT32) wsptr[4];
    z2 = MULTIPLY(z4, FIX(1.274162392));         /* c4 */
    z3 = MULTIPLY(z4, FIX(0.314692123));         /* c12 */
    z4 = MULTIPLY(z4, FIX(0.881747734));         /* c8 */

    tmp10 = z1 + z2;
    tmp11 = z1 + z3;
    tmp12 = z1 - z4;

    tmp23 = z1 - ((z2 + z3 - z4) << 1);          /* c0 = (c4+c12-c8)*2 */

    z1 = (INT32) wsptr[2];
    z2 = (INT32) wsptr[6];

    z3 = MULTIPLY(z1 + z2, FIX(1.105676686));    /* c6 */

    tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
    tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
    tmp15 = MULTIPLY(z1, FIX(0.613604268)) -     /* c10 */
	    MULTIPLY(z2, FIX(1.378756276));      /* c2 */

    tmp20 = tmp10 + tmp13;
    tmp26 = tmp10 - tmp13;
    tmp21 = tmp11 + tmp14;
    tmp25 = tmp11 - tmp14;
    tmp22 = tmp12 + tmp15;
    tmp24 = tmp12 - tmp15;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z4 = (INT32) wsptr[7];
    z4 <<= CONST_BITS;

    tmp14 = z1 + z3;
    tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607));           /* c3 */
    tmp12 = MULTIPLY(tmp14, FIX(1.197448846));             /* c5 */
    tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
    tmp14 = MULTIPLY(tmp14, FIX(0.752406978));             /* c9 */
    tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426));        /* c9+c11-c13 */
    z1    -= z2;
    tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4;           /* c11 */
    tmp16 += tmp15;
    tmp13 = MULTIPLY(z2 + z3, - FIX(0.158341681)) - z4;    /* -c13 */
    tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948));       /* c3-c9-c13 */
    tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773));       /* c3+c5-c13 */
    tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284));           /* c1 */
    tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
    tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567));       /* c1+c11-c5 */

    tmp13 = ((z1 - z3) << CONST_BITS) + z4;

    /* Final output stage */

    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[13] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[12] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp26 + tmp16,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp26 - tmp16,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 12x6 output block.
 *
 * 6-point IDCT in pass 1 (columns), 12-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_12x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		JCOEFPTR coef_block,
		JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*6];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp10 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp10 += ONE << (CONST_BITS-PASS1_BITS-1);
    tmp12 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    tmp20 = MULTIPLY(tmp12, FIX(0.707106781));   /* c4 */
    tmp11 = tmp10 + tmp20;
    tmp21 = RIGHT_SHIFT(tmp10 - tmp20 - tmp20, CONST_BITS-PASS1_BITS);
    tmp20 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    tmp10 = MULTIPLY(tmp20, FIX(1.224744871));   /* c2 */
    tmp20 = tmp11 + tmp10;
    tmp22 = tmp11 - tmp10;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    tmp11 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
    tmp10 = tmp11 + ((z1 + z2) << CONST_BITS);
    tmp12 = tmp11 + ((z3 - z2) << CONST_BITS);
    tmp11 = (z1 - z2 - z3) << PASS1_BITS;

    /* Final output stage */

    wsptr[8*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*5] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*1] = (int) (tmp21 + tmp11);
    wsptr[8*4] = (int) (tmp21 - tmp11);
    wsptr[8*2] = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*3] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 6 rows from work array, store into output array.
   * 12-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/24).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 6; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    z3 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));
    z3 <<= CONST_BITS;

    z4 = (INT32) wsptr[4];
    z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */

    tmp10 = z3 + z4;
    tmp11 = z3 - z4;

    z1 = (INT32) wsptr[2];
    z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
    z1 <<= CONST_BITS;
    z2 = (INT32) wsptr[6];
    z2 <<= CONST_BITS;

    tmp12 = z1 - z2;

    tmp21 = z3 + tmp12;
    tmp24 = z3 - tmp12;

    tmp12 = z4 + z2;

    tmp20 = tmp10 + tmp12;
    tmp25 = tmp10 - tmp12;

    tmp12 = z4 - z1 - z2;

    tmp22 = tmp11 + tmp12;
    tmp23 = tmp11 - tmp12;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z4 = (INT32) wsptr[7];

    tmp11 = MULTIPLY(z2, FIX(1.306562965));                  /* c3 */
    tmp14 = MULTIPLY(z2, - FIX_0_541196100);                 /* -c9 */

    tmp10 = z1 + z3;
    tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669));          /* c7 */
    tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384));       /* c5-c7 */
    tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716));  /* c1-c5 */
    tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580));           /* -(c7+c11) */
    tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
    tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
    tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) -        /* c7-c11 */
	     MULTIPLY(z4, FIX(1.982889723));                 /* c5+c7 */

    z1 -= z4;
    z2 -= z3;
    z3 = MULTIPLY(z1 + z2, FIX_0_541196100);                 /* c9 */
    tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865);              /* c3-c9 */
    tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065);              /* c3+c9 */

    /* Final output stage */

    outptr[0]  = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[11] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[1]  = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[10] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[2]  = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[9]  = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[3]  = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[8]  = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[4]  = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[7]  = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[5]  = range_limit[(int) RIGHT_SHIFT(tmp25 + tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];
    outptr[6]  = range_limit[(int) RIGHT_SHIFT(tmp25 - tmp15,
					       CONST_BITS+PASS1_BITS+3)
			     & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 10x5 output block.
 *
 * 5-point IDCT in pass 1 (columns), 10-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_10x5 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		JCOEFPTR coef_block,
		JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*5];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 5-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/10).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp12 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp12 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp12 += ONE << (CONST_BITS-PASS1_BITS-1);
    tmp13 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    tmp14 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z1 = MULTIPLY(tmp13 + tmp14, FIX(0.790569415)); /* (c2+c4)/2 */
    z2 = MULTIPLY(tmp13 - tmp14, FIX(0.353553391)); /* (c2-c4)/2 */
    z3 = tmp12 + z2;
    tmp10 = z3 + z1;
    tmp11 = z3 - z1;
    tmp12 -= z2 << 2;

    /* Odd part */

    z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);

    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));       /* c3 */
    tmp13 = z1 + MULTIPLY(z2, FIX(0.513743148));    /* c1-c3 */
    tmp14 = z1 - MULTIPLY(z3, FIX(2.176250899));    /* c1+c3 */

    /* Final output stage */

    wsptr[8*0] = (int) RIGHT_SHIFT(tmp10 + tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*4] = (int) RIGHT_SHIFT(tmp10 - tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*1] = (int) RIGHT_SHIFT(tmp11 + tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*3] = (int) RIGHT_SHIFT(tmp11 - tmp14, CONST_BITS-PASS1_BITS);
    wsptr[8*2] = (int) RIGHT_SHIFT(tmp12, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 5 rows from work array, store into output array.
   * 10-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/20).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 5; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    z3 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));
    z3 <<= CONST_BITS;
    z4 = (INT32) wsptr[4];
    z1 = MULTIPLY(z4, FIX(1.144122806));         /* c4 */
    z2 = MULTIPLY(z4, FIX(0.437016024));         /* c8 */
    tmp10 = z3 + z1;
    tmp11 = z3 - z2;

    tmp22 = z3 - ((z1 - z2) << 1);               /* c0 = (c4-c8)*2 */

    z2 = (INT32) wsptr[2];
    z3 = (INT32) wsptr[6];

    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));    /* c6 */
    tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
    tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */

    tmp20 = tmp10 + tmp12;
    tmp24 = tmp10 - tmp12;
    tmp21 = tmp11 + tmp13;
    tmp23 = tmp11 - tmp13;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    z3 <<= CONST_BITS;
    z4 = (INT32) wsptr[7];

    tmp11 = z2 + z4;
    tmp13 = z2 - z4;

    tmp12 = MULTIPLY(tmp13, FIX(0.309016994));        /* (c3-c7)/2 */

    z2 = MULTIPLY(tmp11, FIX(0.951056516));           /* (c3+c7)/2 */
    z4 = z3 + tmp12;

    tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
    tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */

    z2 = MULTIPLY(tmp11, FIX(0.587785252));           /* (c1-c9)/2 */
    z4 = z3 - tmp12 - (tmp13 << (CONST_BITS - 1));

    tmp12 = ((z1 - tmp13) << CONST_BITS) - z3;

    tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
    tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[9] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[8] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23 + tmp13,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp23 - tmp13,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp24 + tmp14,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp24 - tmp14,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 8;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 8x4 output block.
 *
 * 4-point IDCT in pass 1 (columns), 8-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_8x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3;
  INT32 tmp10, tmp11, tmp12, tmp13;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*4];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 4-point IDCT kernel,
   * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);

    tmp10 = (tmp0 + tmp2) << PASS1_BITS;
    tmp12 = (tmp0 - tmp2) << PASS1_BITS;

    /* Odd part */
    /* Same rotation as in the even part of the 8x8 LL&M IDCT */

    z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);               /* c6 */
    /* Add fudge factor here for final descale. */
    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
    tmp0 = RIGHT_SHIFT(z1 + MULTIPLY(z2, FIX_0_765366865), /* c2-c6 */
		       CONST_BITS-PASS1_BITS);
    tmp2 = RIGHT_SHIFT(z1 - MULTIPLY(z3, FIX_1_847759065), /* c2+c6 */
		       CONST_BITS-PASS1_BITS);

    /* Final output stage */

    wsptr[8*0] = (int) (tmp10 + tmp0);
    wsptr[8*3] = (int) (tmp10 - tmp0);
    wsptr[8*1] = (int) (tmp12 + tmp2);
    wsptr[8*2] = (int) (tmp12 - tmp2);
  }

  /* Pass 2: process rows from work array, store into output array.
   * Note that we must descale the results by a factor of 8 == 2**3,
   * and also undo the PASS1_BITS scaling.
   * 8-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 4; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part: reverse the even part of the forward DCT.
     * The rotator is c(-6).
     */

    /* Add range center and fudge factor for final descale and range-limit. */
    z2 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));
    z3 = (INT32) wsptr[4];

    tmp0 = (z2 + z3) << CONST_BITS;
    tmp1 = (z2 - z3) << CONST_BITS;

    z2 = (INT32) wsptr[2];
    z3 = (INT32) wsptr[6];

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);       /* c6 */
    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);     /* c2-c6 */
    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);     /* c2+c6 */

    tmp10 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;
    tmp11 = tmp1 + tmp3;
    tmp12 = tmp1 - tmp3;

    /* Odd part per figure 8; the matrix is unitary and hence its
     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
     */

    tmp0 = (INT32) wsptr[7];
    tmp1 = (INT32) wsptr[5];
    tmp2 = (INT32) wsptr[3];
    tmp3 = (INT32) wsptr[1];

    z2 = tmp0 + tmp2;
    z3 = tmp1 + tmp3;

    z1 = MULTIPLY(z2 + z3, FIX_1_175875602);       /*  c3 */
    z2 = MULTIPLY(z2, - FIX_1_961570560);          /* -c3-c5 */
    z3 = MULTIPLY(z3, - FIX_0_390180644);          /* -c3+c5 */
    z2 += z1;
    z3 += z1;

    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
    tmp0 = MULTIPLY(tmp0, FIX_0_298631336);        /* -c1+c3+c5-c7 */
    tmp3 = MULTIPLY(tmp3, FIX_1_501321110);        /*  c1+c3-c5-c7 */
    tmp0 += z1 + z2;
    tmp3 += z1 + z3;

    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
    tmp1 = MULTIPLY(tmp1, FIX_2_053119869);        /*  c1+c3-c5+c7 */
    tmp2 = MULTIPLY(tmp2, FIX_3_072711026);        /*  c1+c3+c5-c7 */
    tmp1 += z1 + z3;
    tmp2 += z1 + z2;

    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += DCTSIZE;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a reduced-size 6x3 output block.
 *
 * 3-point IDCT in pass 1 (columns), 6-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_6x3 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[6*3];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 3-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/6).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp0 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
    tmp10 = tmp0 + tmp12;
    tmp2 = tmp0 - tmp12 - tmp12;

    /* Odd part */

    tmp12 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */

    /* Final output stage */

    wsptr[6*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[6*2] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
    wsptr[6*1] = (int) RIGHT_SHIFT(tmp2, CONST_BITS-PASS1_BITS);
  }
  
  /* Pass 2: process 3 rows from work array, store into output array.
   * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 3; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp0 = (INT32) wsptr[0] +
	     ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	      (ONE << (PASS1_BITS+2)));
    tmp0 <<= CONST_BITS;
    tmp2 = (INT32) wsptr[4];
    tmp10 = MULTIPLY(tmp2, FIX(0.707106781));   /* c4 */
    tmp1 = tmp0 + tmp10;
    tmp11 = tmp0 - tmp10 - tmp10;
    tmp10 = (INT32) wsptr[2];
    tmp0 = MULTIPLY(tmp10, FIX(1.224744871));   /* c2 */
    tmp10 = tmp1 + tmp0;
    tmp12 = tmp1 - tmp0;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
    tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
    tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
    tmp1 = (z1 - z2 - z3) << CONST_BITS;

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 6;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 4x2 output block.
 *
 * 2-point IDCT in pass 1 (columns), 4-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_4x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp2, tmp10, tmp12;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  INT32 * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  INT32 workspace[4*2];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array. */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 4; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp10 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);

    /* Odd part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);

    /* Final output stage */

    wsptr[4*0] = tmp10 + tmp0;
    wsptr[4*1] = tmp10 - tmp0;
  }

  /* Pass 2: process 2 rows from work array, store into output array.
   * 4-point IDCT kernel,
   * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 2; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp0 = wsptr[0] + ((((INT32) RANGE_CENTER) << 3) + (ONE << 2));
    tmp2 = wsptr[2];

    tmp10 = (tmp0 + tmp2) << CONST_BITS;
    tmp12 = (tmp0 - tmp2) << CONST_BITS;

    /* Odd part */
    /* Same rotation as in the even part of the 8x8 LL&M IDCT */

    z2 = wsptr[1];
    z3 = wsptr[3];

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);   /* c6 */
    tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
    tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
					      CONST_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
					      CONST_BITS+3)
			    & RANGE_MASK];

    wsptr += 4;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 2x1 output block.
 *
 * 1-point IDCT in pass 1 (columns), 2-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_2x1 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  DCTELEM tmp0, tmp1;
  ISLOW_MULT_TYPE * quantptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  ISHIFT_TEMPS

  /* Pass 1: empty. */

  /* Pass 2: process 1 row from input, store into output array. */

  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  outptr = output_buf[0] + output_col;

  /* Even part */

  tmp0 = DEQUANTIZE(coef_block[0], quantptr[0]);
  /* Add range center and fudge factor for final descale and range-limit. */
  tmp0 += (((DCTELEM) RANGE_CENTER) << 3) + (1 << 2);

  /* Odd part */

  tmp1 = DEQUANTIZE(coef_block[1], quantptr[1]);

  /* Final output stage */

  outptr[0] = range_limit[(int) IRIGHT_SHIFT(tmp0 + tmp1, 3) & RANGE_MASK];
  outptr[1] = range_limit[(int) IRIGHT_SHIFT(tmp0 - tmp1, 3) & RANGE_MASK];
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 8x16 output block.
 *
 * 16-point IDCT in pass 1 (columns), 8-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_8x16 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		JCOEFPTR coef_block,
		JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[8*16];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 16-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/32).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp0 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);

    z1 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    tmp1 = MULTIPLY(z1, FIX(1.306562965));      /* c4[16] = c2[8] */
    tmp2 = MULTIPLY(z1, FIX_0_541196100);       /* c12[16] = c6[8] */

    tmp10 = tmp0 + tmp1;
    tmp11 = tmp0 - tmp1;
    tmp12 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
    z3 = z1 - z2;
    z4 = MULTIPLY(z3, FIX(0.275899379));        /* c14[16] = c7[8] */
    z3 = MULTIPLY(z3, FIX(1.387039845));        /* c2[16] = c1[8] */

    tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447);  /* (c6+c2)[16] = (c3+c1)[8] */
    tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223);  /* (c6-c14)[16] = (c3-c7)[8] */
    tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */
    tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */

    tmp20 = tmp10 + tmp0;
    tmp27 = tmp10 - tmp0;
    tmp21 = tmp12 + tmp1;
    tmp26 = tmp12 - tmp1;
    tmp22 = tmp13 + tmp2;
    tmp25 = tmp13 - tmp2;
    tmp23 = tmp11 + tmp3;
    tmp24 = tmp11 - tmp3;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    tmp11 = z1 + z3;

    tmp1  = MULTIPLY(z1 + z2, FIX(1.353318001));   /* c3 */
    tmp2  = MULTIPLY(tmp11,   FIX(1.247225013));   /* c5 */
    tmp3  = MULTIPLY(z1 + z4, FIX(1.093201867));   /* c7 */
    tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586));   /* c9 */
    tmp11 = MULTIPLY(tmp11,   FIX(0.666655658));   /* c11 */
    tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528));   /* c13 */
    tmp0  = tmp1 + tmp2 + tmp3 -
	    MULTIPLY(z1, FIX(2.286341144));        /* c7+c5+c3-c1 */
    tmp13 = tmp10 + tmp11 + tmp12 -
	    MULTIPLY(z1, FIX(1.835730603));        /* c9+c11+c13-c15 */
    z1    = MULTIPLY(z2 + z3, FIX(0.138617169));   /* c15 */
    tmp1  += z1 + MULTIPLY(z2, FIX(0.071888074));  /* c9+c11-c3-c15 */
    tmp2  += z1 - MULTIPLY(z3, FIX(1.125726048));  /* c5+c7+c15-c3 */
    z1    = MULTIPLY(z3 - z2, FIX(1.407403738));   /* c1 */
    tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282));  /* c1+c11-c9-c13 */
    tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411));  /* c1+c5+c13-c7 */
    z2    += z4;
    z1    = MULTIPLY(z2, - FIX(0.666655658));      /* -c11 */
    tmp1  += z1;
    tmp3  += z1 + MULTIPLY(z4, FIX(1.065388962));  /* c3+c11+c15-c7 */
    z2    = MULTIPLY(z2, - FIX(1.247225013));      /* -c5 */
    tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809));  /* c1+c5+c9-c13 */
    tmp12 += z2;
    z2    = MULTIPLY(z3 + z4, - FIX(1.353318001)); /* -c3 */
    tmp2  += z2;
    tmp3  += z2;
    z2    = MULTIPLY(z4 - z3, FIX(0.410524528));   /* c13 */
    tmp10 += z2;
    tmp11 += z2;

    /* Final output stage */

    wsptr[8*0]  = (int) RIGHT_SHIFT(tmp20 + tmp0,  CONST_BITS-PASS1_BITS);
    wsptr[8*15] = (int) RIGHT_SHIFT(tmp20 - tmp0,  CONST_BITS-PASS1_BITS);
    wsptr[8*1]  = (int) RIGHT_SHIFT(tmp21 + tmp1,  CONST_BITS-PASS1_BITS);
    wsptr[8*14] = (int) RIGHT_SHIFT(tmp21 - tmp1,  CONST_BITS-PASS1_BITS);
    wsptr[8*2]  = (int) RIGHT_SHIFT(tmp22 + tmp2,  CONST_BITS-PASS1_BITS);
    wsptr[8*13] = (int) RIGHT_SHIFT(tmp22 - tmp2,  CONST_BITS-PASS1_BITS);
    wsptr[8*3]  = (int) RIGHT_SHIFT(tmp23 + tmp3,  CONST_BITS-PASS1_BITS);
    wsptr[8*12] = (int) RIGHT_SHIFT(tmp23 - tmp3,  CONST_BITS-PASS1_BITS);
    wsptr[8*4]  = (int) RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*11] = (int) RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[8*5]  = (int) RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*10] = (int) RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[8*6]  = (int) RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*9]  = (int) RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS-PASS1_BITS);
    wsptr[8*7]  = (int) RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS-PASS1_BITS);
    wsptr[8*8]  = (int) RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process rows from work array, store into output array.
   * Note that we must descale the results by a factor of 8 == 2**3,
   * and also undo the PASS1_BITS scaling.
   * 8-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 16; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part: reverse the even part of the forward DCT.
     * The rotator is c(-6).
     */

    /* Add range center and fudge factor for final descale and range-limit. */
    z2 = (INT32) wsptr[0] +
	   ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	    (ONE << (PASS1_BITS+2)));
    z3 = (INT32) wsptr[4];

    tmp0 = (z2 + z3) << CONST_BITS;
    tmp1 = (z2 - z3) << CONST_BITS;

    z2 = (INT32) wsptr[2];
    z3 = (INT32) wsptr[6];

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);       /* c6 */
    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);     /* c2-c6 */
    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);     /* c2+c6 */

    tmp10 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;
    tmp11 = tmp1 + tmp3;
    tmp12 = tmp1 - tmp3;

    /* Odd part per figure 8; the matrix is unitary and hence its
     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
     */

    tmp0 = (INT32) wsptr[7];
    tmp1 = (INT32) wsptr[5];
    tmp2 = (INT32) wsptr[3];
    tmp3 = (INT32) wsptr[1];

    z2 = tmp0 + tmp2;
    z3 = tmp1 + tmp3;

    z1 = MULTIPLY(z2 + z3, FIX_1_175875602);       /*  c3 */
    z2 = MULTIPLY(z2, - FIX_1_961570560);          /* -c3-c5 */
    z3 = MULTIPLY(z3, - FIX_0_390180644);          /* -c3+c5 */
    z2 += z1;
    z3 += z1;

    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
    tmp0 = MULTIPLY(tmp0, FIX_0_298631336);        /* -c1+c3+c5-c7 */
    tmp3 = MULTIPLY(tmp3, FIX_1_501321110);        /*  c1+c3-c5-c7 */
    tmp0 += z1 + z2;
    tmp3 += z1 + z3;

    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
    tmp1 = MULTIPLY(tmp1, FIX_2_053119869);        /*  c1+c3-c5+c7 */
    tmp2 = MULTIPLY(tmp2, FIX_3_072711026);        /*  c1+c3+c5-c7 */
    tmp1 += z1 + z3;
    tmp2 += z1 + z2;

    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp3,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[7] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp3,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp1,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp13 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp13 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += DCTSIZE;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 7x14 output block.
 *
 * 14-point IDCT in pass 1 (columns), 7-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_7x14 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		JCOEFPTR coef_block,
		JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[7*14];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 14-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/28).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z1 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z1 += ONE << (CONST_BITS-PASS1_BITS-1);
    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z2 = MULTIPLY(z4, FIX(1.274162392));         /* c4 */
    z3 = MULTIPLY(z4, FIX(0.314692123));         /* c12 */
    z4 = MULTIPLY(z4, FIX(0.881747734));         /* c8 */

    tmp10 = z1 + z2;
    tmp11 = z1 + z3;
    tmp12 = z1 - z4;

    tmp23 = RIGHT_SHIFT(z1 - ((z2 + z3 - z4) << 1), /* c0 = (c4+c12-c8)*2 */
			CONST_BITS-PASS1_BITS);

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    z3 = MULTIPLY(z1 + z2, FIX(1.105676686));    /* c6 */

    tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */
    tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */
    tmp15 = MULTIPLY(z1, FIX(0.613604268)) -     /* c10 */
	    MULTIPLY(z2, FIX(1.378756276));      /* c2 */

    tmp20 = tmp10 + tmp13;
    tmp26 = tmp10 - tmp13;
    tmp21 = tmp11 + tmp14;
    tmp25 = tmp11 - tmp14;
    tmp22 = tmp12 + tmp15;
    tmp24 = tmp12 - tmp15;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
    tmp13 = z4 << CONST_BITS;

    tmp14 = z1 + z3;
    tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607));           /* c3 */
    tmp12 = MULTIPLY(tmp14, FIX(1.197448846));             /* c5 */
    tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */
    tmp14 = MULTIPLY(tmp14, FIX(0.752406978));             /* c9 */
    tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426));        /* c9+c11-c13 */
    z1    -= z2;
    tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13;        /* c11 */
    tmp16 += tmp15;
    z1    += z4;
    z4    = MULTIPLY(z2 + z3, - FIX(0.158341681)) - tmp13; /* -c13 */
    tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948));          /* c3-c9-c13 */
    tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773));          /* c3+c5-c13 */
    z4    = MULTIPLY(z3 - z2, FIX(1.405321284));           /* c1 */
    tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */
    tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567));          /* c1+c11-c5 */

    tmp13 = (z1 - z3) << PASS1_BITS;

    /* Final output stage */

    wsptr[7*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[7*13] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[7*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[7*12] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[7*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[7*11] = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
    wsptr[7*3]  = (int) (tmp23 + tmp13);
    wsptr[7*10] = (int) (tmp23 - tmp13);
    wsptr[7*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
    wsptr[7*9]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
    wsptr[7*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
    wsptr[7*8]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
    wsptr[7*6]  = (int) RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS-PASS1_BITS);
    wsptr[7*7]  = (int) RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 14 rows from work array, store into output array.
   * 7-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/14).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 14; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp23 = (INT32) wsptr[0] +
	      ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	       (ONE << (PASS1_BITS+2)));
    tmp23 <<= CONST_BITS;

    z1 = (INT32) wsptr[2];
    z2 = (INT32) wsptr[4];
    z3 = (INT32) wsptr[6];

    tmp20 = MULTIPLY(z2 - z3, FIX(0.881747734));       /* c4 */
    tmp22 = MULTIPLY(z1 - z2, FIX(0.314692123));       /* c6 */
    tmp21 = tmp20 + tmp22 + tmp23 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */
    tmp10 = z1 + z3;
    z2 -= tmp10;
    tmp10 = MULTIPLY(tmp10, FIX(1.274162392)) + tmp23; /* c2 */
    tmp20 += tmp10 - MULTIPLY(z3, FIX(0.077722536));   /* c2-c4-c6 */
    tmp22 += tmp10 - MULTIPLY(z1, FIX(2.470602249));   /* c2+c4+c6 */
    tmp23 += MULTIPLY(z2, FIX(1.414213562));           /* c0 */

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];

    tmp11 = MULTIPLY(z1 + z2, FIX(0.935414347));       /* (c3+c1-c5)/2 */
    tmp12 = MULTIPLY(z1 - z2, FIX(0.170262339));       /* (c3+c5-c1)/2 */
    tmp10 = tmp11 - tmp12;
    tmp11 += tmp12;
    tmp12 = MULTIPLY(z2 + z3, - FIX(1.378756276));     /* -c1 */
    tmp11 += tmp12;
    z2 = MULTIPLY(z1 + z3, FIX(0.613604268));          /* c5 */
    tmp10 += z2;
    tmp12 += z2 + MULTIPLY(z3, FIX(1.870828693));      /* c3+c1-c5 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[6] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp23,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 7;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 6x12 output block.
 *
 * 12-point IDCT in pass 1 (columns), 6-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_6x12 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		JCOEFPTR coef_block,
		JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14, tmp15;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24, tmp25;
  INT32 z1, z2, z3, z4;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[6*12];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 12-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/24).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z3 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z3 += ONE << (CONST_BITS-PASS1_BITS-1);

    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */

    tmp10 = z3 + z4;
    tmp11 = z3 - z4;

    z1 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */
    z1 <<= CONST_BITS;
    z2 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);
    z2 <<= CONST_BITS;

    tmp12 = z1 - z2;

    tmp21 = z3 + tmp12;
    tmp24 = z3 - tmp12;

    tmp12 = z4 + z2;

    tmp20 = tmp10 + tmp12;
    tmp25 = tmp10 - tmp12;

    tmp12 = z4 - z1 - z2;

    tmp22 = tmp11 + tmp12;
    tmp23 = tmp11 - tmp12;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    tmp11 = MULTIPLY(z2, FIX(1.306562965));                  /* c3 */
    tmp14 = MULTIPLY(z2, - FIX_0_541196100);                 /* -c9 */

    tmp10 = z1 + z3;
    tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669));          /* c7 */
    tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384));       /* c5-c7 */
    tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716));  /* c1-c5 */
    tmp13 = MULTIPLY(z3 + z4, - FIX(1.045510580));           /* -(c7+c11) */
    tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */
    tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */
    tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) -        /* c7-c11 */
	     MULTIPLY(z4, FIX(1.982889723));                 /* c5+c7 */

    z1 -= z4;
    z2 -= z3;
    z3 = MULTIPLY(z1 + z2, FIX_0_541196100);                 /* c9 */
    tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865);              /* c3-c9 */
    tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065);              /* c3+c9 */

    /* Final output stage */

    wsptr[6*0]  = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[6*11] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[6*1]  = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[6*10] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[6*2]  = (int) RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS-PASS1_BITS);
    wsptr[6*9]  = (int) RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS-PASS1_BITS);
    wsptr[6*3]  = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
    wsptr[6*8]  = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
    wsptr[6*4]  = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
    wsptr[6*7]  = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
    wsptr[6*5]  = (int) RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS-PASS1_BITS);
    wsptr[6*6]  = (int) RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 12 rows from work array, store into output array.
   * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 12; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp10 = (INT32) wsptr[0] +
	      ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	       (ONE << (PASS1_BITS+2)));
    tmp10 <<= CONST_BITS;
    tmp12 = (INT32) wsptr[4];
    tmp20 = MULTIPLY(tmp12, FIX(0.707106781));   /* c4 */
    tmp11 = tmp10 + tmp20;
    tmp21 = tmp10 - tmp20 - tmp20;
    tmp20 = (INT32) wsptr[2];
    tmp10 = MULTIPLY(tmp20, FIX(1.224744871));   /* c2 */
    tmp20 = tmp11 + tmp10;
    tmp22 = tmp11 - tmp10;

    /* Odd part */

    z1 = (INT32) wsptr[1];
    z2 = (INT32) wsptr[3];
    z3 = (INT32) wsptr[5];
    tmp11 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
    tmp10 = tmp11 + ((z1 + z2) << CONST_BITS);
    tmp12 = tmp11 + ((z3 - z2) << CONST_BITS);
    tmp11 = (z1 - z2 - z3) << CONST_BITS;

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp20 + tmp10,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[5] = range_limit[(int) RIGHT_SHIFT(tmp20 - tmp10,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp21 + tmp11,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp21 - tmp11,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp22 + tmp12,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp22 - tmp12,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 6;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 5x10 output block.
 *
 * 10-point IDCT in pass 1 (columns), 5-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_5x10 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
		JCOEFPTR coef_block,
		JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp10, tmp11, tmp12, tmp13, tmp14;
  INT32 tmp20, tmp21, tmp22, tmp23, tmp24;
  INT32 z1, z2, z3, z4, z5;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[5*10];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 10-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/20).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    z3 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z3 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z3 += ONE << (CONST_BITS-PASS1_BITS-1);
    z4 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z1 = MULTIPLY(z4, FIX(1.144122806));         /* c4 */
    z2 = MULTIPLY(z4, FIX(0.437016024));         /* c8 */
    tmp10 = z3 + z1;
    tmp11 = z3 - z2;

    tmp22 = RIGHT_SHIFT(z3 - ((z1 - z2) << 1),   /* c0 = (c4-c8)*2 */
			CONST_BITS-PASS1_BITS);

    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));    /* c6 */
    tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */
    tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */

    tmp20 = tmp10 + tmp12;
    tmp24 = tmp10 - tmp12;
    tmp21 = tmp11 + tmp13;
    tmp23 = tmp11 - tmp13;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    z4 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);

    tmp11 = z2 + z4;
    tmp13 = z2 - z4;

    tmp12 = MULTIPLY(tmp13, FIX(0.309016994));        /* (c3-c7)/2 */
    z5 = z3 << CONST_BITS;

    z2 = MULTIPLY(tmp11, FIX(0.951056516));           /* (c3+c7)/2 */
    z4 = z5 + tmp12;

    tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */
    tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */

    z2 = MULTIPLY(tmp11, FIX(0.587785252));           /* (c1-c9)/2 */
    z4 = z5 - tmp12 - (tmp13 << (CONST_BITS - 1));

    tmp12 = (z1 - tmp13 - z3) << PASS1_BITS;

    tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */
    tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */

    /* Final output stage */

    wsptr[5*0] = (int) RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS-PASS1_BITS);
    wsptr[5*9] = (int) RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS-PASS1_BITS);
    wsptr[5*1] = (int) RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS-PASS1_BITS);
    wsptr[5*8] = (int) RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS-PASS1_BITS);
    wsptr[5*2] = (int) (tmp22 + tmp12);
    wsptr[5*7] = (int) (tmp22 - tmp12);
    wsptr[5*3] = (int) RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS-PASS1_BITS);
    wsptr[5*6] = (int) RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS-PASS1_BITS);
    wsptr[5*4] = (int) RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS-PASS1_BITS);
    wsptr[5*5] = (int) RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 10 rows from work array, store into output array.
   * 5-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/10).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 10; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp12 = (INT32) wsptr[0] +
	      ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	       (ONE << (PASS1_BITS+2)));
    tmp12 <<= CONST_BITS;
    tmp13 = (INT32) wsptr[2];
    tmp14 = (INT32) wsptr[4];
    z1 = MULTIPLY(tmp13 + tmp14, FIX(0.790569415)); /* (c2+c4)/2 */
    z2 = MULTIPLY(tmp13 - tmp14, FIX(0.353553391)); /* (c2-c4)/2 */
    z3 = tmp12 + z2;
    tmp10 = z3 + z1;
    tmp11 = z3 - z1;
    tmp12 -= z2 << 2;

    /* Odd part */

    z2 = (INT32) wsptr[1];
    z3 = (INT32) wsptr[3];

    z1 = MULTIPLY(z2 + z3, FIX(0.831253876));       /* c3 */
    tmp13 = z1 + MULTIPLY(z2, FIX(0.513743148));    /* c1-c3 */
    tmp14 = z1 - MULTIPLY(z3, FIX(2.176250899));    /* c1+c3 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp13,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[4] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp13,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp11 + tmp14,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp11 - tmp14,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 5;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 4x8 output block.
 *
 * 8-point IDCT in pass 1 (columns), 4-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_4x8 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp3;
  INT32 tmp10, tmp11, tmp12, tmp13;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[4*8];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * Note results are scaled up by sqrt(8) compared to a true IDCT;
   * furthermore, we scale the results by 2**PASS1_BITS.
   * 8-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/16).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 4; ctr > 0; ctr--) {
    /* Due to quantization, we will usually find that many of the input
     * coefficients are zero, especially the AC terms.  We can exploit this
     * by short-circuiting the IDCT calculation for any column in which all
     * the AC terms are zero.  In that case each output is equal to the
     * DC coefficient (with scale factor as needed).
     * With typical images and quantization tables, half or more of the
     * column DCT calculations can be simplified this way.
     */

    if (inptr[DCTSIZE*1] == 0 && inptr[DCTSIZE*2] == 0 &&
	inptr[DCTSIZE*3] == 0 && inptr[DCTSIZE*4] == 0 &&
	inptr[DCTSIZE*5] == 0 && inptr[DCTSIZE*6] == 0 &&
	inptr[DCTSIZE*7] == 0) {
      /* AC terms all zero */
      int dcval = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]) << PASS1_BITS;

      wsptr[4*0] = dcval;
      wsptr[4*1] = dcval;
      wsptr[4*2] = dcval;
      wsptr[4*3] = dcval;
      wsptr[4*4] = dcval;
      wsptr[4*5] = dcval;
      wsptr[4*6] = dcval;
      wsptr[4*7] = dcval;

      inptr++;			/* advance pointers to next column */
      quantptr++;
      wsptr++;
      continue;
    }

    /* Even part: reverse the even part of the forward DCT.
     * The rotator is c(-6).
     */

    z2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*6], quantptr[DCTSIZE*6]);

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);       /* c6 */
    tmp2 = z1 + MULTIPLY(z2, FIX_0_765366865);     /* c2-c6 */
    tmp3 = z1 - MULTIPLY(z3, FIX_1_847759065);     /* c2+c6 */

    z2 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    z2 <<= CONST_BITS;
    z3 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    z2 += ONE << (CONST_BITS-PASS1_BITS-1);

    tmp0 = z2 + z3;
    tmp1 = z2 - z3;

    tmp10 = tmp0 + tmp2;
    tmp13 = tmp0 - tmp2;
    tmp11 = tmp1 + tmp3;
    tmp12 = tmp1 - tmp3;

    /* Odd part per figure 8; the matrix is unitary and hence its
     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
     */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*7], quantptr[DCTSIZE*7]);
    tmp1 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    tmp3 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);

    z2 = tmp0 + tmp2;
    z3 = tmp1 + tmp3;

    z1 = MULTIPLY(z2 + z3, FIX_1_175875602);       /*  c3 */
    z2 = MULTIPLY(z2, - FIX_1_961570560);          /* -c3-c5 */
    z3 = MULTIPLY(z3, - FIX_0_390180644);          /* -c3+c5 */
    z2 += z1;
    z3 += z1;

    z1 = MULTIPLY(tmp0 + tmp3, - FIX_0_899976223); /* -c3+c7 */
    tmp0 = MULTIPLY(tmp0, FIX_0_298631336);        /* -c1+c3+c5-c7 */
    tmp3 = MULTIPLY(tmp3, FIX_1_501321110);        /*  c1+c3-c5-c7 */
    tmp0 += z1 + z2;
    tmp3 += z1 + z3;

    z1 = MULTIPLY(tmp1 + tmp2, - FIX_2_562915447); /* -c1-c3 */
    tmp1 = MULTIPLY(tmp1, FIX_2_053119869);        /*  c1+c3-c5+c7 */
    tmp2 = MULTIPLY(tmp2, FIX_3_072711026);        /*  c1+c3+c5-c7 */
    tmp1 += z1 + z3;
    tmp2 += z1 + z2;

    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */

    wsptr[4*0] = (int) RIGHT_SHIFT(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
    wsptr[4*7] = (int) RIGHT_SHIFT(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
    wsptr[4*1] = (int) RIGHT_SHIFT(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
    wsptr[4*6] = (int) RIGHT_SHIFT(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
    wsptr[4*2] = (int) RIGHT_SHIFT(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
    wsptr[4*5] = (int) RIGHT_SHIFT(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
    wsptr[4*3] = (int) RIGHT_SHIFT(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[4*4] = (int) RIGHT_SHIFT(tmp13 - tmp0, CONST_BITS-PASS1_BITS);

    inptr++;			/* advance pointers to next column */
    quantptr++;
    wsptr++;
  }

  /* Pass 2: process 8 rows from work array, store into output array.
   * 4-point IDCT kernel,
   * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 8; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp0 = (INT32) wsptr[0] +
	     ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	      (ONE << (PASS1_BITS+2)));
    tmp2 = (INT32) wsptr[2];

    tmp10 = (tmp0 + tmp2) << CONST_BITS;
    tmp12 = (tmp0 - tmp2) << CONST_BITS;

    /* Odd part */
    /* Same rotation as in the even part of the 8x8 LL&M IDCT */

    z2 = (INT32) wsptr[1];
    z3 = (INT32) wsptr[3];

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);   /* c6 */
    tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
    tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[3] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp12 + tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp12 - tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 4;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a reduced-size 3x6 output block.
 *
 * 6-point IDCT in pass 1 (columns), 3-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_3x6 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp1, tmp2, tmp10, tmp11, tmp12;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  int * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  int workspace[3*6];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 6-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/12).
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp0 <<= CONST_BITS;
    /* Add fudge factor here for final descale. */
    tmp0 += ONE << (CONST_BITS-PASS1_BITS-1);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*4], quantptr[DCTSIZE*4]);
    tmp10 = MULTIPLY(tmp2, FIX(0.707106781));   /* c4 */
    tmp1 = tmp0 + tmp10;
    tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS-PASS1_BITS);
    tmp10 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);
    tmp0 = MULTIPLY(tmp10, FIX(1.224744871));   /* c2 */
    tmp10 = tmp1 + tmp0;
    tmp12 = tmp1 - tmp0;

    /* Odd part */

    z1 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z2 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*5], quantptr[DCTSIZE*5]);
    tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */
    tmp0 = tmp1 + ((z1 + z2) << CONST_BITS);
    tmp2 = tmp1 + ((z3 - z2) << CONST_BITS);
    tmp1 = (z1 - z2 - z3) << PASS1_BITS;

    /* Final output stage */

    wsptr[3*0] = (int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS-PASS1_BITS);
    wsptr[3*5] = (int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS-PASS1_BITS);
    wsptr[3*1] = (int) (tmp11 + tmp1);
    wsptr[3*4] = (int) (tmp11 - tmp1);
    wsptr[3*2] = (int) RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS-PASS1_BITS);
    wsptr[3*3] = (int) RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS-PASS1_BITS);
  }

  /* Pass 2: process 6 rows from work array, store into output array.
   * 3-point IDCT kernel, cK represents sqrt(2) * cos(K*pi/6).
   */

  wsptr = workspace;
  for (ctr = 0; ctr < 6; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp0 = (INT32) wsptr[0] +
	     ((((INT32) RANGE_CENTER) << (PASS1_BITS+3)) +
	      (ONE << (PASS1_BITS+2)));
    tmp0 <<= CONST_BITS;
    tmp2 = (INT32) wsptr[2];
    tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */
    tmp10 = tmp0 + tmp12;
    tmp2 = tmp0 - tmp12 - tmp12;

    /* Odd part */

    tmp12 = (INT32) wsptr[1];
    tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[2] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp2,
					      CONST_BITS+PASS1_BITS+3)
			    & RANGE_MASK];

    wsptr += 3;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 2x4 output block.
 *
 * 4-point IDCT in pass 1 (columns), 2-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_2x4 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  INT32 tmp0, tmp2, tmp10, tmp12;
  INT32 z1, z2, z3;
  JCOEFPTR inptr;
  ISLOW_MULT_TYPE * quantptr;
  INT32 * wsptr;
  JSAMPROW outptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  int ctr;
  INT32 workspace[2*4];	/* buffers data between passes */
  SHIFT_TEMPS

  /* Pass 1: process columns from input, store into work array.
   * 4-point IDCT kernel,
   * cK represents sqrt(2) * cos(K*pi/16) [refers to 8-point IDCT].
   */

  inptr = coef_block;
  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;
  wsptr = workspace;
  for (ctr = 0; ctr < 2; ctr++, inptr++, quantptr++, wsptr++) {
    /* Even part */

    tmp0 = DEQUANTIZE(inptr[DCTSIZE*0], quantptr[DCTSIZE*0]);
    tmp2 = DEQUANTIZE(inptr[DCTSIZE*2], quantptr[DCTSIZE*2]);

    tmp10 = (tmp0 + tmp2) << CONST_BITS;
    tmp12 = (tmp0 - tmp2) << CONST_BITS;

    /* Odd part */
    /* Same rotation as in the even part of the 8x8 LL&M IDCT */

    z2 = DEQUANTIZE(inptr[DCTSIZE*1], quantptr[DCTSIZE*1]);
    z3 = DEQUANTIZE(inptr[DCTSIZE*3], quantptr[DCTSIZE*3]);

    z1 = MULTIPLY(z2 + z3, FIX_0_541196100);   /* c6 */
    tmp0 = z1 + MULTIPLY(z2, FIX_0_765366865); /* c2-c6 */
    tmp2 = z1 - MULTIPLY(z3, FIX_1_847759065); /* c2+c6 */

    /* Final output stage */

    wsptr[2*0] = tmp10 + tmp0;
    wsptr[2*3] = tmp10 - tmp0;
    wsptr[2*1] = tmp12 + tmp2;
    wsptr[2*2] = tmp12 - tmp2;
  }

  /* Pass 2: process 4 rows from work array, store into output array. */

  wsptr = workspace;
  for (ctr = 0; ctr < 4; ctr++) {
    outptr = output_buf[ctr] + output_col;

    /* Even part */

    /* Add range center and fudge factor for final descale and range-limit. */
    tmp10 = wsptr[0] +
	      ((((INT32) RANGE_CENTER) << (CONST_BITS+3)) +
	       (ONE << (CONST_BITS+2)));

    /* Odd part */

    tmp0 = wsptr[1];

    /* Final output stage */

    outptr[0] = range_limit[(int) RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS+3)
			    & RANGE_MASK];
    outptr[1] = range_limit[(int) RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS+3)
			    & RANGE_MASK];

    wsptr += 2;		/* advance pointer to next row */
  }
}


/*
 * Perform dequantization and inverse DCT on one block of coefficients,
 * producing a 1x2 output block.
 *
 * 2-point IDCT in pass 1 (columns), 1-point in pass 2 (rows).
 */

GLOBAL(void)
jpeg_idct_1x2 (j_decompress_ptr cinfo, jpeg_component_info * compptr,
	       JCOEFPTR coef_block,
	       JSAMPARRAY output_buf, JDIMENSION output_col)
{
  DCTELEM tmp0, tmp1;
  ISLOW_MULT_TYPE * quantptr;
  JSAMPLE *range_limit = IDCT_range_limit(cinfo);
  ISHIFT_TEMPS

  /* Process 1 column from input, store into output array. */

  quantptr = (ISLOW_MULT_TYPE *) compptr->dct_table;

  /* Even part */

  tmp0 = DEQUANTIZE(coef_block[DCTSIZE*0], quantptr[DCTSIZE*0]);
  /* Add range center and fudge factor for final descale and range-limit. */
  tmp0 += (((DCTELEM) RANGE_CENTER) << 3) + (1 << 2);

  /* Odd part */

  tmp1 = DEQUANTIZE(coef_block[DCTSIZE*1], quantptr[DCTSIZE*1]);

  /* Final output stage */

  output_buf[0][output_col] =
    range_limit[(int) IRIGHT_SHIFT(tmp0 + tmp1, 3) & RANGE_MASK];
  output_buf[1][output_col] =
    range_limit[(int) IRIGHT_SHIFT(tmp0 - tmp1, 3) & RANGE_MASK];
}

#endif /* IDCT_SCALING_SUPPORTED */
#endif /* DCT_ISLOW_SUPPORTED */