Statistics
| Branch: | Revision:

ffmpeg / libavcodec / jrevdct.c @ bb54f6ab

History | View | Annotate | Download (42.9 KB)

1
/*
2
 * jrevdct.c
3
 *
4
 * This file is part of the Independent JPEG Group's software.
5
 *
6
 * The authors make NO WARRANTY or representation, either express or implied,
7
 * with respect to this software, its quality, accuracy, merchantability, or
8
 * fitness for a particular purpose.  This software is provided "AS IS", and
9
 * you, its user, assume the entire risk as to its quality and accuracy.
10
 *
11
 * This software is copyright (C) 1991, 1992, Thomas G. Lane.
12
 * All Rights Reserved except as specified below.
13
 *
14
 * Permission is hereby granted to use, copy, modify, and distribute this
15
 * software (or portions thereof) for any purpose, without fee, subject to
16
 * these conditions:
17
 * (1) If any part of the source code for this software is distributed, then
18
 * this README file must be included, with this copyright and no-warranty
19
 * notice unaltered; and any additions, deletions, or changes to the original
20
 * files must be clearly indicated in accompanying documentation.
21
 * (2) If only executable code is distributed, then the accompanying
22
 * documentation must state that "this software is based in part on the work
23
 * of the Independent JPEG Group".
24
 * (3) Permission for use of this software is granted only if the user accepts
25
 * full responsibility for any undesirable consequences; the authors accept
26
 * NO LIABILITY for damages of any kind.
27
 *
28
 * These conditions apply to any software derived from or based on the IJG
29
 * code, not just to the unmodified library.  If you use our work, you ought
30
 * to acknowledge us.
31
 *
32
 * Permission is NOT granted for the use of any IJG author's name or company
33
 * name in advertising or publicity relating to this software or products
34
 * derived from it.  This software may be referred to only as "the Independent
35
 * JPEG Group's software".
36
 *
37
 * We specifically permit and encourage the use of this software as the basis
38
 * of commercial products, provided that all warranty or liability claims are
39
 * assumed by the product vendor.
40
 *
41
 * This file contains the basic inverse-DCT transformation subroutine.
42
 *
43
 * This implementation is based on an algorithm described in
44
 *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
45
 *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
46
 *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
47
 * The primary algorithm described there uses 11 multiplies and 29 adds.
48
 * We use their alternate method with 12 multiplies and 32 adds.
49
 * The advantage of this method is that no data path contains more than one
50
 * multiplication; this allows a very simple and accurate implementation in
51
 * scaled fixed-point arithmetic, with a minimal number of shifts.
52
 *
53
 * I've made lots of modifications to attempt to take advantage of the
54
 * sparse nature of the DCT matrices we're getting.  Although the logic
55
 * is cumbersome, it's straightforward and the resulting code is much
56
 * faster.
57
 *
58
 * A better way to do this would be to pass in the DCT block as a sparse
59
 * matrix, perhaps with the difference cases encoded.
60
 */
61

    
62
/**
63
 * @file jrevdct.c
64
 * Independent JPEG Group's LLM idct.
65
 */
66

    
67
#include "common.h"
68
#include "dsputil.h"
69

    
70
#define EIGHT_BIT_SAMPLES
71

    
72
#define DCTSIZE 8
73
#define DCTSIZE2 64
74

    
75
#define GLOBAL
76

    
77
#define RIGHT_SHIFT(x, n) ((x) >> (n))
78

    
79
typedef DCTELEM DCTBLOCK[DCTSIZE2];
80

    
81
#define CONST_BITS 13
82

    
83
/*
84
 * This routine is specialized to the case DCTSIZE = 8.
85
 */
86

    
87
#if DCTSIZE != 8
88
  Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
89
#endif
90

    
91

    
92
/*
93
 * A 2-D IDCT can be done by 1-D IDCT on each row followed by 1-D IDCT
94
 * on each column.  Direct algorithms are also available, but they are
95
 * much more complex and seem not to be any faster when reduced to code.
96
 *
97
 * The poop on this scaling stuff is as follows:
98
 *
99
 * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
100
 * larger than the true IDCT outputs.  The final outputs are therefore
101
 * a factor of N larger than desired; since N=8 this can be cured by
102
 * a simple right shift at the end of the algorithm.  The advantage of
103
 * this arrangement is that we save two multiplications per 1-D IDCT,
104
 * because the y0 and y4 inputs need not be divided by sqrt(N).
105
 *
106
 * We have to do addition and subtraction of the integer inputs, which
107
 * is no problem, and multiplication by fractional constants, which is
108
 * a problem to do in integer arithmetic.  We multiply all the constants
109
 * by CONST_SCALE and convert them to integer constants (thus retaining
110
 * CONST_BITS bits of precision in the constants).  After doing a
111
 * multiplication we have to divide the product by CONST_SCALE, with proper
112
 * rounding, to produce the correct output.  This division can be done
113
 * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
114
 * as long as possible so that partial sums can be added together with
115
 * full fractional precision.
116
 *
117
 * The outputs of the first pass are scaled up by PASS1_BITS bits so that
118
 * they are represented to better-than-integral precision.  These outputs
119
 * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
120
 * with the recommended scaling.  (To scale up 12-bit sample data further, an
121
 * intermediate int32 array would be needed.)
122
 *
123
 * To avoid overflow of the 32-bit intermediate results in pass 2, we must
124
 * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26.  Error analysis
125
 * shows that the values given below are the most effective.
126
 */
127

    
128
#ifdef EIGHT_BIT_SAMPLES
129
#define PASS1_BITS  2
130
#else
131
#define PASS1_BITS  1   /* lose a little precision to avoid overflow */
132
#endif
133

    
134
#define ONE         ((int32_t) 1)
135

    
136
#define CONST_SCALE (ONE << CONST_BITS)
137

    
138
/* Convert a positive real constant to an integer scaled by CONST_SCALE.
139
 * IMPORTANT: if your compiler doesn't do this arithmetic at compile time,
140
 * you will pay a significant penalty in run time.  In that case, figure
141
 * the correct integer constant values and insert them by hand.
142
 */
143

    
144
/* Actually FIX is no longer used, we precomputed them all */
145
#define FIX(x)  ((int32_t) ((x) * CONST_SCALE + 0.5))
146

    
147
/* Descale and correctly round an int32_t value that's scaled by N bits.
148
 * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
149
 * the fudge factor is correct for either sign of X.
150
 */
151

    
152
#define DESCALE(x,n)  RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
153

    
154
/* Multiply an int32_t variable by an int32_t constant to yield an int32_t result.
155
 * For 8-bit samples with the recommended scaling, all the variable
156
 * and constant values involved are no more than 16 bits wide, so a
157
 * 16x16->32 bit multiply can be used instead of a full 32x32 multiply;
158
 * this provides a useful speedup on many machines.
159
 * There is no way to specify a 16x16->32 multiply in portable C, but
160
 * some C compilers will do the right thing if you provide the correct
161
 * combination of casts.
162
 * NB: for 12-bit samples, a full 32-bit multiplication will be needed.
163
 */
164

    
165
#ifdef EIGHT_BIT_SAMPLES
166
#ifdef SHORTxSHORT_32           /* may work if 'int' is 32 bits */
167
#define MULTIPLY(var,const)  (((int16_t) (var)) * ((int16_t) (const)))
168
#endif
169
#ifdef SHORTxLCONST_32          /* known to work with Microsoft C 6.0 */
170
#define MULTIPLY(var,const)  (((int16_t) (var)) * ((int32_t) (const)))
171
#endif
172
#endif
173

    
174
#ifndef MULTIPLY                /* default definition */
175
#define MULTIPLY(var,const)  ((var) * (const))
176
#endif
177

    
178

    
179
/*
180
  Unlike our decoder where we approximate the FIXes, we need to use exact
181
ones here or successive P-frames will drift too much with Reference frame coding
182
*/
183
#define FIX_0_211164243 1730
184
#define FIX_0_275899380 2260
185
#define FIX_0_298631336 2446
186
#define FIX_0_390180644 3196
187
#define FIX_0_509795579 4176
188
#define FIX_0_541196100 4433
189
#define FIX_0_601344887 4926
190
#define FIX_0_765366865 6270
191
#define FIX_0_785694958 6436
192
#define FIX_0_899976223 7373
193
#define FIX_1_061594337 8697
194
#define FIX_1_111140466 9102
195
#define FIX_1_175875602 9633
196
#define FIX_1_306562965 10703
197
#define FIX_1_387039845 11363
198
#define FIX_1_451774981 11893
199
#define FIX_1_501321110 12299
200
#define FIX_1_662939225 13623
201
#define FIX_1_847759065 15137
202
#define FIX_1_961570560 16069
203
#define FIX_2_053119869 16819
204
#define FIX_2_172734803 17799
205
#define FIX_2_562915447 20995
206
#define FIX_3_072711026 25172
207

    
208
/*
209
 * Perform the inverse DCT on one block of coefficients.
210
 */
211

    
212
void j_rev_dct(DCTBLOCK data)
213
{
214
  int32_t tmp0, tmp1, tmp2, tmp3;
215
  int32_t tmp10, tmp11, tmp12, tmp13;
216
  int32_t z1, z2, z3, z4, z5;
217
  int32_t d0, d1, d2, d3, d4, d5, d6, d7;
218
  register DCTELEM *dataptr;
219
  int rowctr;
220

    
221
  /* Pass 1: process rows. */
222
  /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
223
  /* furthermore, we scale the results by 2**PASS1_BITS. */
224

    
225
  dataptr = data;
226

    
227
  for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
228
    /* Due to quantization, we will usually find that many of the input
229
     * coefficients are zero, especially the AC terms.  We can exploit this
230
     * by short-circuiting the IDCT calculation for any row in which all
231
     * the AC terms are zero.  In that case each output is equal to the
232
     * DC coefficient (with scale factor as needed).
233
     * With typical images and quantization tables, half or more of the
234
     * row DCT calculations can be simplified this way.
235
     */
236

    
237
    register int *idataptr = (int*)dataptr;
238

    
239
    /* WARNING: we do the same permutation as MMX idct to simplify the
240
       video core */
241
    d0 = dataptr[0];
242
    d2 = dataptr[1];
243
    d4 = dataptr[2];
244
    d6 = dataptr[3];
245
    d1 = dataptr[4];
246
    d3 = dataptr[5];
247
    d5 = dataptr[6];
248
    d7 = dataptr[7];
249

    
250
    if ((d1 | d2 | d3 | d4 | d5 | d6 | d7) == 0) {
251
      /* AC terms all zero */
252
      if (d0) {
253
          /* Compute a 32 bit value to assign. */
254
          DCTELEM dcval = (DCTELEM) (d0 << PASS1_BITS);
255
          register int v = (dcval & 0xffff) | ((dcval << 16) & 0xffff0000);
256

    
257
          idataptr[0] = v;
258
          idataptr[1] = v;
259
          idataptr[2] = v;
260
          idataptr[3] = v;
261
      }
262

    
263
      dataptr += DCTSIZE;       /* advance pointer to next row */
264
      continue;
265
    }
266

    
267
    /* Even part: reverse the even part of the forward DCT. */
268
    /* The rotator is sqrt(2)*c(-6). */
269
{
270
    if (d6) {
271
            if (d2) {
272
                    /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */
273
                    z1 = MULTIPLY(d2 + d6, FIX_0_541196100);
274
                    tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065);
275
                    tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865);
276

    
277
                    tmp0 = (d0 + d4) << CONST_BITS;
278
                    tmp1 = (d0 - d4) << CONST_BITS;
279

    
280
                    tmp10 = tmp0 + tmp3;
281
                    tmp13 = tmp0 - tmp3;
282
                    tmp11 = tmp1 + tmp2;
283
                    tmp12 = tmp1 - tmp2;
284
            } else {
285
                    /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */
286
                    tmp2 = MULTIPLY(-d6, FIX_1_306562965);
287
                    tmp3 = MULTIPLY(d6, FIX_0_541196100);
288

    
289
                    tmp0 = (d0 + d4) << CONST_BITS;
290
                    tmp1 = (d0 - d4) << CONST_BITS;
291

    
292
                    tmp10 = tmp0 + tmp3;
293
                    tmp13 = tmp0 - tmp3;
294
                    tmp11 = tmp1 + tmp2;
295
                    tmp12 = tmp1 - tmp2;
296
            }
297
    } else {
298
            if (d2) {
299
                    /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */
300
                    tmp2 = MULTIPLY(d2, FIX_0_541196100);
301
                    tmp3 = MULTIPLY(d2, FIX_1_306562965);
302

    
303
                    tmp0 = (d0 + d4) << CONST_BITS;
304
                    tmp1 = (d0 - d4) << CONST_BITS;
305

    
306
                    tmp10 = tmp0 + tmp3;
307
                    tmp13 = tmp0 - tmp3;
308
                    tmp11 = tmp1 + tmp2;
309
                    tmp12 = tmp1 - tmp2;
310
            } else {
311
                    /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */
312
                    tmp10 = tmp13 = (d0 + d4) << CONST_BITS;
313
                    tmp11 = tmp12 = (d0 - d4) << CONST_BITS;
314
            }
315
      }
316

    
317
    /* Odd part per figure 8; the matrix is unitary and hence its
318
     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
319
     */
320

    
321
    if (d7) {
322
        if (d5) {
323
            if (d3) {
324
                if (d1) {
325
                    /* d1 != 0, d3 != 0, d5 != 0, d7 != 0 */
326
                    z1 = d7 + d1;
327
                    z2 = d5 + d3;
328
                    z3 = d7 + d3;
329
                    z4 = d5 + d1;
330
                    z5 = MULTIPLY(z3 + z4, FIX_1_175875602);
331

    
332
                    tmp0 = MULTIPLY(d7, FIX_0_298631336);
333
                    tmp1 = MULTIPLY(d5, FIX_2_053119869);
334
                    tmp2 = MULTIPLY(d3, FIX_3_072711026);
335
                    tmp3 = MULTIPLY(d1, FIX_1_501321110);
336
                    z1 = MULTIPLY(-z1, FIX_0_899976223);
337
                    z2 = MULTIPLY(-z2, FIX_2_562915447);
338
                    z3 = MULTIPLY(-z3, FIX_1_961570560);
339
                    z4 = MULTIPLY(-z4, FIX_0_390180644);
340

    
341
                    z3 += z5;
342
                    z4 += z5;
343

    
344
                    tmp0 += z1 + z3;
345
                    tmp1 += z2 + z4;
346
                    tmp2 += z2 + z3;
347
                    tmp3 += z1 + z4;
348
                } else {
349
                    /* d1 == 0, d3 != 0, d5 != 0, d7 != 0 */
350
                    z2 = d5 + d3;
351
                    z3 = d7 + d3;
352
                    z5 = MULTIPLY(z3 + d5, FIX_1_175875602);
353

    
354
                    tmp0 = MULTIPLY(d7, FIX_0_298631336);
355
                    tmp1 = MULTIPLY(d5, FIX_2_053119869);
356
                    tmp2 = MULTIPLY(d3, FIX_3_072711026);
357
                    z1 = MULTIPLY(-d7, FIX_0_899976223);
358
                    z2 = MULTIPLY(-z2, FIX_2_562915447);
359
                    z3 = MULTIPLY(-z3, FIX_1_961570560);
360
                    z4 = MULTIPLY(-d5, FIX_0_390180644);
361

    
362
                    z3 += z5;
363
                    z4 += z5;
364

    
365
                    tmp0 += z1 + z3;
366
                    tmp1 += z2 + z4;
367
                    tmp2 += z2 + z3;
368
                    tmp3 = z1 + z4;
369
                }
370
            } else {
371
                if (d1) {
372
                    /* d1 != 0, d3 == 0, d5 != 0, d7 != 0 */
373
                    z1 = d7 + d1;
374
                    z4 = d5 + d1;
375
                    z5 = MULTIPLY(d7 + z4, FIX_1_175875602);
376

    
377
                    tmp0 = MULTIPLY(d7, FIX_0_298631336);
378
                    tmp1 = MULTIPLY(d5, FIX_2_053119869);
379
                    tmp3 = MULTIPLY(d1, FIX_1_501321110);
380
                    z1 = MULTIPLY(-z1, FIX_0_899976223);
381
                    z2 = MULTIPLY(-d5, FIX_2_562915447);
382
                    z3 = MULTIPLY(-d7, FIX_1_961570560);
383
                    z4 = MULTIPLY(-z4, FIX_0_390180644);
384

    
385
                    z3 += z5;
386
                    z4 += z5;
387

    
388
                    tmp0 += z1 + z3;
389
                    tmp1 += z2 + z4;
390
                    tmp2 = z2 + z3;
391
                    tmp3 += z1 + z4;
392
                } else {
393
                    /* d1 == 0, d3 == 0, d5 != 0, d7 != 0 */
394
                    tmp0 = MULTIPLY(-d7, FIX_0_601344887);
395
                    z1 = MULTIPLY(-d7, FIX_0_899976223);
396
                    z3 = MULTIPLY(-d7, FIX_1_961570560);
397
                    tmp1 = MULTIPLY(-d5, FIX_0_509795579);
398
                    z2 = MULTIPLY(-d5, FIX_2_562915447);
399
                    z4 = MULTIPLY(-d5, FIX_0_390180644);
400
                    z5 = MULTIPLY(d5 + d7, FIX_1_175875602);
401

    
402
                    z3 += z5;
403
                    z4 += z5;
404

    
405
                    tmp0 += z3;
406
                    tmp1 += z4;
407
                    tmp2 = z2 + z3;
408
                    tmp3 = z1 + z4;
409
                }
410
            }
411
        } else {
412
            if (d3) {
413
                if (d1) {
414
                    /* d1 != 0, d3 != 0, d5 == 0, d7 != 0 */
415
                    z1 = d7 + d1;
416
                    z3 = d7 + d3;
417
                    z5 = MULTIPLY(z3 + d1, FIX_1_175875602);
418

    
419
                    tmp0 = MULTIPLY(d7, FIX_0_298631336);
420
                    tmp2 = MULTIPLY(d3, FIX_3_072711026);
421
                    tmp3 = MULTIPLY(d1, FIX_1_501321110);
422
                    z1 = MULTIPLY(-z1, FIX_0_899976223);
423
                    z2 = MULTIPLY(-d3, FIX_2_562915447);
424
                    z3 = MULTIPLY(-z3, FIX_1_961570560);
425
                    z4 = MULTIPLY(-d1, FIX_0_390180644);
426

    
427
                    z3 += z5;
428
                    z4 += z5;
429

    
430
                    tmp0 += z1 + z3;
431
                    tmp1 = z2 + z4;
432
                    tmp2 += z2 + z3;
433
                    tmp3 += z1 + z4;
434
                } else {
435
                    /* d1 == 0, d3 != 0, d5 == 0, d7 != 0 */
436
                    z3 = d7 + d3;
437

    
438
                    tmp0 = MULTIPLY(-d7, FIX_0_601344887);
439
                    z1 = MULTIPLY(-d7, FIX_0_899976223);
440
                    tmp2 = MULTIPLY(d3, FIX_0_509795579);
441
                    z2 = MULTIPLY(-d3, FIX_2_562915447);
442
                    z5 = MULTIPLY(z3, FIX_1_175875602);
443
                    z3 = MULTIPLY(-z3, FIX_0_785694958);
444

    
445
                    tmp0 += z3;
446
                    tmp1 = z2 + z5;
447
                    tmp2 += z3;
448
                    tmp3 = z1 + z5;
449
                }
450
            } else {
451
                if (d1) {
452
                    /* d1 != 0, d3 == 0, d5 == 0, d7 != 0 */
453
                    z1 = d7 + d1;
454
                    z5 = MULTIPLY(z1, FIX_1_175875602);
455

    
456
                    z1 = MULTIPLY(z1, FIX_0_275899380);
457
                    z3 = MULTIPLY(-d7, FIX_1_961570560);
458
                    tmp0 = MULTIPLY(-d7, FIX_1_662939225);
459
                    z4 = MULTIPLY(-d1, FIX_0_390180644);
460
                    tmp3 = MULTIPLY(d1, FIX_1_111140466);
461

    
462
                    tmp0 += z1;
463
                    tmp1 = z4 + z5;
464
                    tmp2 = z3 + z5;
465
                    tmp3 += z1;
466
                } else {
467
                    /* d1 == 0, d3 == 0, d5 == 0, d7 != 0 */
468
                    tmp0 = MULTIPLY(-d7, FIX_1_387039845);
469
                    tmp1 = MULTIPLY(d7, FIX_1_175875602);
470
                    tmp2 = MULTIPLY(-d7, FIX_0_785694958);
471
                    tmp3 = MULTIPLY(d7, FIX_0_275899380);
472
                }
473
            }
474
        }
475
    } else {
476
        if (d5) {
477
            if (d3) {
478
                if (d1) {
479
                    /* d1 != 0, d3 != 0, d5 != 0, d7 == 0 */
480
                    z2 = d5 + d3;
481
                    z4 = d5 + d1;
482
                    z5 = MULTIPLY(d3 + z4, FIX_1_175875602);
483

    
484
                    tmp1 = MULTIPLY(d5, FIX_2_053119869);
485
                    tmp2 = MULTIPLY(d3, FIX_3_072711026);
486
                    tmp3 = MULTIPLY(d1, FIX_1_501321110);
487
                    z1 = MULTIPLY(-d1, FIX_0_899976223);
488
                    z2 = MULTIPLY(-z2, FIX_2_562915447);
489
                    z3 = MULTIPLY(-d3, FIX_1_961570560);
490
                    z4 = MULTIPLY(-z4, FIX_0_390180644);
491

    
492
                    z3 += z5;
493
                    z4 += z5;
494

    
495
                    tmp0 = z1 + z3;
496
                    tmp1 += z2 + z4;
497
                    tmp2 += z2 + z3;
498
                    tmp3 += z1 + z4;
499
                } else {
500
                    /* d1 == 0, d3 != 0, d5 != 0, d7 == 0 */
501
                    z2 = d5 + d3;
502

    
503
                    z5 = MULTIPLY(z2, FIX_1_175875602);
504
                    tmp1 = MULTIPLY(d5, FIX_1_662939225);
505
                    z4 = MULTIPLY(-d5, FIX_0_390180644);
506
                    z2 = MULTIPLY(-z2, FIX_1_387039845);
507
                    tmp2 = MULTIPLY(d3, FIX_1_111140466);
508
                    z3 = MULTIPLY(-d3, FIX_1_961570560);
509

    
510
                    tmp0 = z3 + z5;
511
                    tmp1 += z2;
512
                    tmp2 += z2;
513
                    tmp3 = z4 + z5;
514
                }
515
            } else {
516
                if (d1) {
517
                    /* d1 != 0, d3 == 0, d5 != 0, d7 == 0 */
518
                    z4 = d5 + d1;
519

    
520
                    z5 = MULTIPLY(z4, FIX_1_175875602);
521
                    z1 = MULTIPLY(-d1, FIX_0_899976223);
522
                    tmp3 = MULTIPLY(d1, FIX_0_601344887);
523
                    tmp1 = MULTIPLY(-d5, FIX_0_509795579);
524
                    z2 = MULTIPLY(-d5, FIX_2_562915447);
525
                    z4 = MULTIPLY(z4, FIX_0_785694958);
526

    
527
                    tmp0 = z1 + z5;
528
                    tmp1 += z4;
529
                    tmp2 = z2 + z5;
530
                    tmp3 += z4;
531
                } else {
532
                    /* d1 == 0, d3 == 0, d5 != 0, d7 == 0 */
533
                    tmp0 = MULTIPLY(d5, FIX_1_175875602);
534
                    tmp1 = MULTIPLY(d5, FIX_0_275899380);
535
                    tmp2 = MULTIPLY(-d5, FIX_1_387039845);
536
                    tmp3 = MULTIPLY(d5, FIX_0_785694958);
537
                }
538
            }
539
        } else {
540
            if (d3) {
541
                if (d1) {
542
                    /* d1 != 0, d3 != 0, d5 == 0, d7 == 0 */
543
                    z5 = d1 + d3;
544
                    tmp3 = MULTIPLY(d1, FIX_0_211164243);
545
                    tmp2 = MULTIPLY(-d3, FIX_1_451774981);
546
                    z1 = MULTIPLY(d1, FIX_1_061594337);
547
                    z2 = MULTIPLY(-d3, FIX_2_172734803);
548
                    z4 = MULTIPLY(z5, FIX_0_785694958);
549
                    z5 = MULTIPLY(z5, FIX_1_175875602);
550

    
551
                    tmp0 = z1 - z4;
552
                    tmp1 = z2 + z4;
553
                    tmp2 += z5;
554
                    tmp3 += z5;
555
                } else {
556
                    /* d1 == 0, d3 != 0, d5 == 0, d7 == 0 */
557
                    tmp0 = MULTIPLY(-d3, FIX_0_785694958);
558
                    tmp1 = MULTIPLY(-d3, FIX_1_387039845);
559
                    tmp2 = MULTIPLY(-d3, FIX_0_275899380);
560
                    tmp3 = MULTIPLY(d3, FIX_1_175875602);
561
                }
562
            } else {
563
                if (d1) {
564
                    /* d1 != 0, d3 == 0, d5 == 0, d7 == 0 */
565
                    tmp0 = MULTIPLY(d1, FIX_0_275899380);
566
                    tmp1 = MULTIPLY(d1, FIX_0_785694958);
567
                    tmp2 = MULTIPLY(d1, FIX_1_175875602);
568
                    tmp3 = MULTIPLY(d1, FIX_1_387039845);
569
                } else {
570
                    /* d1 == 0, d3 == 0, d5 == 0, d7 == 0 */
571
                    tmp0 = tmp1 = tmp2 = tmp3 = 0;
572
                }
573
            }
574
        }
575
    }
576
}
577
    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
578

    
579
    dataptr[0] = (DCTELEM) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
580
    dataptr[7] = (DCTELEM) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
581
    dataptr[1] = (DCTELEM) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
582
    dataptr[6] = (DCTELEM) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
583
    dataptr[2] = (DCTELEM) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
584
    dataptr[5] = (DCTELEM) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
585
    dataptr[3] = (DCTELEM) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
586
    dataptr[4] = (DCTELEM) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
587

    
588
    dataptr += DCTSIZE;         /* advance pointer to next row */
589
  }
590

    
591
  /* Pass 2: process columns. */
592
  /* Note that we must descale the results by a factor of 8 == 2**3, */
593
  /* and also undo the PASS1_BITS scaling. */
594

    
595
  dataptr = data;
596
  for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
597
    /* Columns of zeroes can be exploited in the same way as we did with rows.
598
     * However, the row calculation has created many nonzero AC terms, so the
599
     * simplification applies less often (typically 5% to 10% of the time).
600
     * On machines with very fast multiplication, it's possible that the
601
     * test takes more time than it's worth.  In that case this section
602
     * may be commented out.
603
     */
604

    
605
    d0 = dataptr[DCTSIZE*0];
606
    d1 = dataptr[DCTSIZE*1];
607
    d2 = dataptr[DCTSIZE*2];
608
    d3 = dataptr[DCTSIZE*3];
609
    d4 = dataptr[DCTSIZE*4];
610
    d5 = dataptr[DCTSIZE*5];
611
    d6 = dataptr[DCTSIZE*6];
612
    d7 = dataptr[DCTSIZE*7];
613

    
614
    /* Even part: reverse the even part of the forward DCT. */
615
    /* The rotator is sqrt(2)*c(-6). */
616
    if (d6) {
617
            if (d2) {
618
                    /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */
619
                    z1 = MULTIPLY(d2 + d6, FIX_0_541196100);
620
                    tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065);
621
                    tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865);
622

    
623
                    tmp0 = (d0 + d4) << CONST_BITS;
624
                    tmp1 = (d0 - d4) << CONST_BITS;
625

    
626
                    tmp10 = tmp0 + tmp3;
627
                    tmp13 = tmp0 - tmp3;
628
                    tmp11 = tmp1 + tmp2;
629
                    tmp12 = tmp1 - tmp2;
630
            } else {
631
                    /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */
632
                    tmp2 = MULTIPLY(-d6, FIX_1_306562965);
633
                    tmp3 = MULTIPLY(d6, FIX_0_541196100);
634

    
635
                    tmp0 = (d0 + d4) << CONST_BITS;
636
                    tmp1 = (d0 - d4) << CONST_BITS;
637

    
638
                    tmp10 = tmp0 + tmp3;
639
                    tmp13 = tmp0 - tmp3;
640
                    tmp11 = tmp1 + tmp2;
641
                    tmp12 = tmp1 - tmp2;
642
            }
643
    } else {
644
            if (d2) {
645
                    /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */
646
                    tmp2 = MULTIPLY(d2, FIX_0_541196100);
647
                    tmp3 = MULTIPLY(d2, FIX_1_306562965);
648

    
649
                    tmp0 = (d0 + d4) << CONST_BITS;
650
                    tmp1 = (d0 - d4) << CONST_BITS;
651

    
652
                    tmp10 = tmp0 + tmp3;
653
                    tmp13 = tmp0 - tmp3;
654
                    tmp11 = tmp1 + tmp2;
655
                    tmp12 = tmp1 - tmp2;
656
            } else {
657
                    /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */
658
                    tmp10 = tmp13 = (d0 + d4) << CONST_BITS;
659
                    tmp11 = tmp12 = (d0 - d4) << CONST_BITS;
660
            }
661
    }
662

    
663
    /* Odd part per figure 8; the matrix is unitary and hence its
664
     * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
665
     */
666
    if (d7) {
667
        if (d5) {
668
            if (d3) {
669
                if (d1) {
670
                    /* d1 != 0, d3 != 0, d5 != 0, d7 != 0 */
671
                    z1 = d7 + d1;
672
                    z2 = d5 + d3;
673
                    z3 = d7 + d3;
674
                    z4 = d5 + d1;
675
                    z5 = MULTIPLY(z3 + z4, FIX_1_175875602);
676

    
677
                    tmp0 = MULTIPLY(d7, FIX_0_298631336);
678
                    tmp1 = MULTIPLY(d5, FIX_2_053119869);
679
                    tmp2 = MULTIPLY(d3, FIX_3_072711026);
680
                    tmp3 = MULTIPLY(d1, FIX_1_501321110);
681
                    z1 = MULTIPLY(-z1, FIX_0_899976223);
682
                    z2 = MULTIPLY(-z2, FIX_2_562915447);
683
                    z3 = MULTIPLY(-z3, FIX_1_961570560);
684
                    z4 = MULTIPLY(-z4, FIX_0_390180644);
685

    
686
                    z3 += z5;
687
                    z4 += z5;
688

    
689
                    tmp0 += z1 + z3;
690
                    tmp1 += z2 + z4;
691
                    tmp2 += z2 + z3;
692
                    tmp3 += z1 + z4;
693
                } else {
694
                    /* d1 == 0, d3 != 0, d5 != 0, d7 != 0 */
695
                    z1 = d7;
696
                    z2 = d5 + d3;
697
                    z3 = d7 + d3;
698
                    z5 = MULTIPLY(z3 + d5, FIX_1_175875602);
699

    
700
                    tmp0 = MULTIPLY(d7, FIX_0_298631336);
701
                    tmp1 = MULTIPLY(d5, FIX_2_053119869);
702
                    tmp2 = MULTIPLY(d3, FIX_3_072711026);
703
                    z1 = MULTIPLY(-d7, FIX_0_899976223);
704
                    z2 = MULTIPLY(-z2, FIX_2_562915447);
705
                    z3 = MULTIPLY(-z3, FIX_1_961570560);
706
                    z4 = MULTIPLY(-d5, FIX_0_390180644);
707

    
708
                    z3 += z5;
709
                    z4 += z5;
710

    
711
                    tmp0 += z1 + z3;
712
                    tmp1 += z2 + z4;
713
                    tmp2 += z2 + z3;
714
                    tmp3 = z1 + z4;
715
                }
716
            } else {
717
                if (d1) {
718
                    /* d1 != 0, d3 == 0, d5 != 0, d7 != 0 */
719
                    z1 = d7 + d1;
720
                    z2 = d5;
721
                    z3 = d7;
722
                    z4 = d5 + d1;
723
                    z5 = MULTIPLY(z3 + z4, FIX_1_175875602);
724

    
725
                    tmp0 = MULTIPLY(d7, FIX_0_298631336);
726
                    tmp1 = MULTIPLY(d5, FIX_2_053119869);
727
                    tmp3 = MULTIPLY(d1, FIX_1_501321110);
728
                    z1 = MULTIPLY(-z1, FIX_0_899976223);
729
                    z2 = MULTIPLY(-d5, FIX_2_562915447);
730
                    z3 = MULTIPLY(-d7, FIX_1_961570560);
731
                    z4 = MULTIPLY(-z4, FIX_0_390180644);
732

    
733
                    z3 += z5;
734
                    z4 += z5;
735

    
736
                    tmp0 += z1 + z3;
737
                    tmp1 += z2 + z4;
738
                    tmp2 = z2 + z3;
739
                    tmp3 += z1 + z4;
740
                } else {
741
                    /* d1 == 0, d3 == 0, d5 != 0, d7 != 0 */
742
                    tmp0 = MULTIPLY(-d7, FIX_0_601344887);
743
                    z1 = MULTIPLY(-d7, FIX_0_899976223);
744
                    z3 = MULTIPLY(-d7, FIX_1_961570560);
745
                    tmp1 = MULTIPLY(-d5, FIX_0_509795579);
746
                    z2 = MULTIPLY(-d5, FIX_2_562915447);
747
                    z4 = MULTIPLY(-d5, FIX_0_390180644);
748
                    z5 = MULTIPLY(d5 + d7, FIX_1_175875602);
749

    
750
                    z3 += z5;
751
                    z4 += z5;
752

    
753
                    tmp0 += z3;
754
                    tmp1 += z4;
755
                    tmp2 = z2 + z3;
756
                    tmp3 = z1 + z4;
757
                }
758
            }
759
        } else {
760
            if (d3) {
761
                if (d1) {
762
                    /* d1 != 0, d3 != 0, d5 == 0, d7 != 0 */
763
                    z1 = d7 + d1;
764
                    z3 = d7 + d3;
765
                    z5 = MULTIPLY(z3 + d1, FIX_1_175875602);
766

    
767
                    tmp0 = MULTIPLY(d7, FIX_0_298631336);
768
                    tmp2 = MULTIPLY(d3, FIX_3_072711026);
769
                    tmp3 = MULTIPLY(d1, FIX_1_501321110);
770
                    z1 = MULTIPLY(-z1, FIX_0_899976223);
771
                    z2 = MULTIPLY(-d3, FIX_2_562915447);
772
                    z3 = MULTIPLY(-z3, FIX_1_961570560);
773
                    z4 = MULTIPLY(-d1, FIX_0_390180644);
774

    
775
                    z3 += z5;
776
                    z4 += z5;
777

    
778
                    tmp0 += z1 + z3;
779
                    tmp1 = z2 + z4;
780
                    tmp2 += z2 + z3;
781
                    tmp3 += z1 + z4;
782
                } else {
783
                    /* d1 == 0, d3 != 0, d5 == 0, d7 != 0 */
784
                    z3 = d7 + d3;
785

    
786
                    tmp0 = MULTIPLY(-d7, FIX_0_601344887);
787
                    z1 = MULTIPLY(-d7, FIX_0_899976223);
788
                    tmp2 = MULTIPLY(d3, FIX_0_509795579);
789
                    z2 = MULTIPLY(-d3, FIX_2_562915447);
790
                    z5 = MULTIPLY(z3, FIX_1_175875602);
791
                    z3 = MULTIPLY(-z3, FIX_0_785694958);
792

    
793
                    tmp0 += z3;
794
                    tmp1 = z2 + z5;
795
                    tmp2 += z3;
796
                    tmp3 = z1 + z5;
797
                }
798
            } else {
799
                if (d1) {
800
                    /* d1 != 0, d3 == 0, d5 == 0, d7 != 0 */
801
                    z1 = d7 + d1;
802
                    z5 = MULTIPLY(z1, FIX_1_175875602);
803

    
804
                    z1 = MULTIPLY(z1, FIX_0_275899380);
805
                    z3 = MULTIPLY(-d7, FIX_1_961570560);
806
                    tmp0 = MULTIPLY(-d7, FIX_1_662939225);
807
                    z4 = MULTIPLY(-d1, FIX_0_390180644);
808
                    tmp3 = MULTIPLY(d1, FIX_1_111140466);
809

    
810
                    tmp0 += z1;
811
                    tmp1 = z4 + z5;
812
                    tmp2 = z3 + z5;
813
                    tmp3 += z1;
814
                } else {
815
                    /* d1 == 0, d3 == 0, d5 == 0, d7 != 0 */
816
                    tmp0 = MULTIPLY(-d7, FIX_1_387039845);
817
                    tmp1 = MULTIPLY(d7, FIX_1_175875602);
818
                    tmp2 = MULTIPLY(-d7, FIX_0_785694958);
819
                    tmp3 = MULTIPLY(d7, FIX_0_275899380);
820
                }
821
            }
822
        }
823
    } else {
824
        if (d5) {
825
            if (d3) {
826
                if (d1) {
827
                    /* d1 != 0, d3 != 0, d5 != 0, d7 == 0 */
828
                    z2 = d5 + d3;
829
                    z4 = d5 + d1;
830
                    z5 = MULTIPLY(d3 + z4, FIX_1_175875602);
831

    
832
                    tmp1 = MULTIPLY(d5, FIX_2_053119869);
833
                    tmp2 = MULTIPLY(d3, FIX_3_072711026);
834
                    tmp3 = MULTIPLY(d1, FIX_1_501321110);
835
                    z1 = MULTIPLY(-d1, FIX_0_899976223);
836
                    z2 = MULTIPLY(-z2, FIX_2_562915447);
837
                    z3 = MULTIPLY(-d3, FIX_1_961570560);
838
                    z4 = MULTIPLY(-z4, FIX_0_390180644);
839

    
840
                    z3 += z5;
841
                    z4 += z5;
842

    
843
                    tmp0 = z1 + z3;
844
                    tmp1 += z2 + z4;
845
                    tmp2 += z2 + z3;
846
                    tmp3 += z1 + z4;
847
                } else {
848
                    /* d1 == 0, d3 != 0, d5 != 0, d7 == 0 */
849
                    z2 = d5 + d3;
850

    
851
                    z5 = MULTIPLY(z2, FIX_1_175875602);
852
                    tmp1 = MULTIPLY(d5, FIX_1_662939225);
853
                    z4 = MULTIPLY(-d5, FIX_0_390180644);
854
                    z2 = MULTIPLY(-z2, FIX_1_387039845);
855
                    tmp2 = MULTIPLY(d3, FIX_1_111140466);
856
                    z3 = MULTIPLY(-d3, FIX_1_961570560);
857

    
858
                    tmp0 = z3 + z5;
859
                    tmp1 += z2;
860
                    tmp2 += z2;
861
                    tmp3 = z4 + z5;
862
                }
863
            } else {
864
                if (d1) {
865
                    /* d1 != 0, d3 == 0, d5 != 0, d7 == 0 */
866
                    z4 = d5 + d1;
867

    
868
                    z5 = MULTIPLY(z4, FIX_1_175875602);
869
                    z1 = MULTIPLY(-d1, FIX_0_899976223);
870
                    tmp3 = MULTIPLY(d1, FIX_0_601344887);
871
                    tmp1 = MULTIPLY(-d5, FIX_0_509795579);
872
                    z2 = MULTIPLY(-d5, FIX_2_562915447);
873
                    z4 = MULTIPLY(z4, FIX_0_785694958);
874

    
875
                    tmp0 = z1 + z5;
876
                    tmp1 += z4;
877
                    tmp2 = z2 + z5;
878
                    tmp3 += z4;
879
                } else {
880
                    /* d1 == 0, d3 == 0, d5 != 0, d7 == 0 */
881
                    tmp0 = MULTIPLY(d5, FIX_1_175875602);
882
                    tmp1 = MULTIPLY(d5, FIX_0_275899380);
883
                    tmp2 = MULTIPLY(-d5, FIX_1_387039845);
884
                    tmp3 = MULTIPLY(d5, FIX_0_785694958);
885
                }
886
            }
887
        } else {
888
            if (d3) {
889
                if (d1) {
890
                    /* d1 != 0, d3 != 0, d5 == 0, d7 == 0 */
891
                    z5 = d1 + d3;
892
                    tmp3 = MULTIPLY(d1, FIX_0_211164243);
893
                    tmp2 = MULTIPLY(-d3, FIX_1_451774981);
894
                    z1 = MULTIPLY(d1, FIX_1_061594337);
895
                    z2 = MULTIPLY(-d3, FIX_2_172734803);
896
                    z4 = MULTIPLY(z5, FIX_0_785694958);
897
                    z5 = MULTIPLY(z5, FIX_1_175875602);
898

    
899
                    tmp0 = z1 - z4;
900
                    tmp1 = z2 + z4;
901
                    tmp2 += z5;
902
                    tmp3 += z5;
903
                } else {
904
                    /* d1 == 0, d3 != 0, d5 == 0, d7 == 0 */
905
                    tmp0 = MULTIPLY(-d3, FIX_0_785694958);
906
                    tmp1 = MULTIPLY(-d3, FIX_1_387039845);
907
                    tmp2 = MULTIPLY(-d3, FIX_0_275899380);
908
                    tmp3 = MULTIPLY(d3, FIX_1_175875602);
909
                }
910
            } else {
911
                if (d1) {
912
                    /* d1 != 0, d3 == 0, d5 == 0, d7 == 0 */
913
                    tmp0 = MULTIPLY(d1, FIX_0_275899380);
914
                    tmp1 = MULTIPLY(d1, FIX_0_785694958);
915
                    tmp2 = MULTIPLY(d1, FIX_1_175875602);
916
                    tmp3 = MULTIPLY(d1, FIX_1_387039845);
917
                } else {
918
                    /* d1 == 0, d3 == 0, d5 == 0, d7 == 0 */
919
                    tmp0 = tmp1 = tmp2 = tmp3 = 0;
920
                }
921
            }
922
        }
923
    }
924

    
925
    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
926

    
927
    dataptr[DCTSIZE*0] = (DCTELEM) DESCALE(tmp10 + tmp3,
928
                                           CONST_BITS+PASS1_BITS+3);
929
    dataptr[DCTSIZE*7] = (DCTELEM) DESCALE(tmp10 - tmp3,
930
                                           CONST_BITS+PASS1_BITS+3);
931
    dataptr[DCTSIZE*1] = (DCTELEM) DESCALE(tmp11 + tmp2,
932
                                           CONST_BITS+PASS1_BITS+3);
933
    dataptr[DCTSIZE*6] = (DCTELEM) DESCALE(tmp11 - tmp2,
934
                                           CONST_BITS+PASS1_BITS+3);
935
    dataptr[DCTSIZE*2] = (DCTELEM) DESCALE(tmp12 + tmp1,
936
                                           CONST_BITS+PASS1_BITS+3);
937
    dataptr[DCTSIZE*5] = (DCTELEM) DESCALE(tmp12 - tmp1,
938
                                           CONST_BITS+PASS1_BITS+3);
939
    dataptr[DCTSIZE*3] = (DCTELEM) DESCALE(tmp13 + tmp0,
940
                                           CONST_BITS+PASS1_BITS+3);
941
    dataptr[DCTSIZE*4] = (DCTELEM) DESCALE(tmp13 - tmp0,
942
                                           CONST_BITS+PASS1_BITS+3);
943

    
944
    dataptr++;                  /* advance pointer to next column */
945
  }
946
}
947

    
948
#undef DCTSIZE
949
#define DCTSIZE 4
950
#define DCTSTRIDE 8
951

    
952
void j_rev_dct4(DCTBLOCK data)
953
{
954
  int32_t tmp0, tmp1, tmp2, tmp3;
955
  int32_t tmp10, tmp11, tmp12, tmp13;
956
  int32_t z1;
957
  int32_t d0, d2, d4, d6;
958
  register DCTELEM *dataptr;
959
  int rowctr;
960

    
961
  /* Pass 1: process rows. */
962
  /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
963
  /* furthermore, we scale the results by 2**PASS1_BITS. */
964

    
965
  data[0] += 4;
966

    
967
  dataptr = data;
968

    
969
  for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
970
    /* Due to quantization, we will usually find that many of the input
971
     * coefficients are zero, especially the AC terms.  We can exploit this
972
     * by short-circuiting the IDCT calculation for any row in which all
973
     * the AC terms are zero.  In that case each output is equal to the
974
     * DC coefficient (with scale factor as needed).
975
     * With typical images and quantization tables, half or more of the
976
     * row DCT calculations can be simplified this way.
977
     */
978

    
979
    register int *idataptr = (int*)dataptr;
980

    
981
    d0 = dataptr[0];
982
    d2 = dataptr[1];
983
    d4 = dataptr[2];
984
    d6 = dataptr[3];
985

    
986
    if ((d2 | d4 | d6) == 0) {
987
      /* AC terms all zero */
988
      if (d0) {
989
          /* Compute a 32 bit value to assign. */
990
          DCTELEM dcval = (DCTELEM) (d0 << PASS1_BITS);
991
          register int v = (dcval & 0xffff) | ((dcval << 16) & 0xffff0000);
992

    
993
          idataptr[0] = v;
994
          idataptr[1] = v;
995
      }
996

    
997
      dataptr += DCTSTRIDE;     /* advance pointer to next row */
998
      continue;
999
    }
1000

    
1001
    /* Even part: reverse the even part of the forward DCT. */
1002
    /* The rotator is sqrt(2)*c(-6). */
1003
    if (d6) {
1004
            if (d2) {
1005
                    /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */
1006
                    z1 = MULTIPLY(d2 + d6, FIX_0_541196100);
1007
                    tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065);
1008
                    tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865);
1009

    
1010
                    tmp0 = (d0 + d4) << CONST_BITS;
1011
                    tmp1 = (d0 - d4) << CONST_BITS;
1012

    
1013
                    tmp10 = tmp0 + tmp3;
1014
                    tmp13 = tmp0 - tmp3;
1015
                    tmp11 = tmp1 + tmp2;
1016
                    tmp12 = tmp1 - tmp2;
1017
            } else {
1018
                    /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */
1019
                    tmp2 = MULTIPLY(-d6, FIX_1_306562965);
1020
                    tmp3 = MULTIPLY(d6, FIX_0_541196100);
1021

    
1022
                    tmp0 = (d0 + d4) << CONST_BITS;
1023
                    tmp1 = (d0 - d4) << CONST_BITS;
1024

    
1025
                    tmp10 = tmp0 + tmp3;
1026
                    tmp13 = tmp0 - tmp3;
1027
                    tmp11 = tmp1 + tmp2;
1028
                    tmp12 = tmp1 - tmp2;
1029
            }
1030
    } else {
1031
            if (d2) {
1032
                    /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */
1033
                    tmp2 = MULTIPLY(d2, FIX_0_541196100);
1034
                    tmp3 = MULTIPLY(d2, FIX_1_306562965);
1035

    
1036
                    tmp0 = (d0 + d4) << CONST_BITS;
1037
                    tmp1 = (d0 - d4) << CONST_BITS;
1038

    
1039
                    tmp10 = tmp0 + tmp3;
1040
                    tmp13 = tmp0 - tmp3;
1041
                    tmp11 = tmp1 + tmp2;
1042
                    tmp12 = tmp1 - tmp2;
1043
            } else {
1044
                    /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */
1045
                    tmp10 = tmp13 = (d0 + d4) << CONST_BITS;
1046
                    tmp11 = tmp12 = (d0 - d4) << CONST_BITS;
1047
            }
1048
      }
1049

    
1050
    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
1051

    
1052
    dataptr[0] = (DCTELEM) DESCALE(tmp10, CONST_BITS-PASS1_BITS);
1053
    dataptr[1] = (DCTELEM) DESCALE(tmp11, CONST_BITS-PASS1_BITS);
1054
    dataptr[2] = (DCTELEM) DESCALE(tmp12, CONST_BITS-PASS1_BITS);
1055
    dataptr[3] = (DCTELEM) DESCALE(tmp13, CONST_BITS-PASS1_BITS);
1056

    
1057
    dataptr += DCTSTRIDE;       /* advance pointer to next row */
1058
  }
1059

    
1060
  /* Pass 2: process columns. */
1061
  /* Note that we must descale the results by a factor of 8 == 2**3, */
1062
  /* and also undo the PASS1_BITS scaling. */
1063

    
1064
  dataptr = data;
1065
  for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
1066
    /* Columns of zeroes can be exploited in the same way as we did with rows.
1067
     * However, the row calculation has created many nonzero AC terms, so the
1068
     * simplification applies less often (typically 5% to 10% of the time).
1069
     * On machines with very fast multiplication, it's possible that the
1070
     * test takes more time than it's worth.  In that case this section
1071
     * may be commented out.
1072
     */
1073

    
1074
    d0 = dataptr[DCTSTRIDE*0];
1075
    d2 = dataptr[DCTSTRIDE*1];
1076
    d4 = dataptr[DCTSTRIDE*2];
1077
    d6 = dataptr[DCTSTRIDE*3];
1078

    
1079
    /* Even part: reverse the even part of the forward DCT. */
1080
    /* The rotator is sqrt(2)*c(-6). */
1081
    if (d6) {
1082
            if (d2) {
1083
                    /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */
1084
                    z1 = MULTIPLY(d2 + d6, FIX_0_541196100);
1085
                    tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065);
1086
                    tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865);
1087

    
1088
                    tmp0 = (d0 + d4) << CONST_BITS;
1089
                    tmp1 = (d0 - d4) << CONST_BITS;
1090

    
1091
                    tmp10 = tmp0 + tmp3;
1092
                    tmp13 = tmp0 - tmp3;
1093
                    tmp11 = tmp1 + tmp2;
1094
                    tmp12 = tmp1 - tmp2;
1095
            } else {
1096
                    /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */
1097
                    tmp2 = MULTIPLY(-d6, FIX_1_306562965);
1098
                    tmp3 = MULTIPLY(d6, FIX_0_541196100);
1099

    
1100
                    tmp0 = (d0 + d4) << CONST_BITS;
1101
                    tmp1 = (d0 - d4) << CONST_BITS;
1102

    
1103
                    tmp10 = tmp0 + tmp3;
1104
                    tmp13 = tmp0 - tmp3;
1105
                    tmp11 = tmp1 + tmp2;
1106
                    tmp12 = tmp1 - tmp2;
1107
            }
1108
    } else {
1109
            if (d2) {
1110
                    /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */
1111
                    tmp2 = MULTIPLY(d2, FIX_0_541196100);
1112
                    tmp3 = MULTIPLY(d2, FIX_1_306562965);
1113

    
1114
                    tmp0 = (d0 + d4) << CONST_BITS;
1115
                    tmp1 = (d0 - d4) << CONST_BITS;
1116

    
1117
                    tmp10 = tmp0 + tmp3;
1118
                    tmp13 = tmp0 - tmp3;
1119
                    tmp11 = tmp1 + tmp2;
1120
                    tmp12 = tmp1 - tmp2;
1121
            } else {
1122
                    /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */
1123
                    tmp10 = tmp13 = (d0 + d4) << CONST_BITS;
1124
                    tmp11 = tmp12 = (d0 - d4) << CONST_BITS;
1125
            }
1126
    }
1127

    
1128
    /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
1129

    
1130
    dataptr[DCTSTRIDE*0] = tmp10 >> (CONST_BITS+PASS1_BITS+3);
1131
    dataptr[DCTSTRIDE*1] = tmp11 >> (CONST_BITS+PASS1_BITS+3);
1132
    dataptr[DCTSTRIDE*2] = tmp12 >> (CONST_BITS+PASS1_BITS+3);
1133
    dataptr[DCTSTRIDE*3] = tmp13 >> (CONST_BITS+PASS1_BITS+3);
1134

    
1135
    dataptr++;                  /* advance pointer to next column */
1136
  }
1137
}
1138

    
1139
void j_rev_dct2(DCTBLOCK data){
1140
  int d00, d01, d10, d11;
1141

    
1142
  data[0] += 4;
1143
  d00 = data[0+0*DCTSTRIDE] + data[1+0*DCTSTRIDE];
1144
  d01 = data[0+0*DCTSTRIDE] - data[1+0*DCTSTRIDE];
1145
  d10 = data[0+1*DCTSTRIDE] + data[1+1*DCTSTRIDE];
1146
  d11 = data[0+1*DCTSTRIDE] - data[1+1*DCTSTRIDE];
1147

    
1148
  data[0+0*DCTSTRIDE]= (d00 + d10)>>3;
1149
  data[1+0*DCTSTRIDE]= (d01 + d11)>>3;
1150
  data[0+1*DCTSTRIDE]= (d00 - d10)>>3;
1151
  data[1+1*DCTSTRIDE]= (d01 - d11)>>3;
1152
}
1153

    
1154
void j_rev_dct1(DCTBLOCK data){
1155
  data[0] = (data[0] + 4)>>3;
1156
}
1157

    
1158
#undef FIX
1159
#undef CONST_BITS