• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * This file is part of the Independent JPEG Group's software.
3  *
4  * The authors make NO WARRANTY or representation, either express or implied,
5  * with respect to this software, its quality, accuracy, merchantability, or
6  * fitness for a particular purpose.  This software is provided "AS IS", and
7  * you, its user, assume the entire risk as to its quality and accuracy.
8  *
9  * This software is copyright (C) 1991, 1992, Thomas G. Lane.
10  * All Rights Reserved except as specified below.
11  *
12  * Permission is hereby granted to use, copy, modify, and distribute this
13  * software (or portions thereof) for any purpose, without fee, subject to
14  * these conditions:
15  * (1) If any part of the source code for this software is distributed, then
16  * this README file must be included, with this copyright and no-warranty
17  * notice unaltered; and any additions, deletions, or changes to the original
18  * files must be clearly indicated in accompanying documentation.
19  * (2) If only executable code is distributed, then the accompanying
20  * documentation must state that "this software is based in part on the work
21  * of the Independent JPEG Group".
22  * (3) Permission for use of this software is granted only if the user accepts
23  * full responsibility for any undesirable consequences; the authors accept
24  * NO LIABILITY for damages of any kind.
25  *
26  * These conditions apply to any software derived from or based on the IJG
27  * code, not just to the unmodified library.  If you use our work, you ought
28  * to acknowledge us.
29  *
30  * Permission is NOT granted for the use of any IJG author's name or company
31  * name in advertising or publicity relating to this software or products
32  * derived from it.  This software may be referred to only as "the Independent
33  * JPEG Group's software".
34  *
35  * We specifically permit and encourage the use of this software as the basis
36  * of commercial products, provided that all warranty or liability claims are
37  * assumed by the product vendor.
38  *
39  * This file contains the basic inverse-DCT transformation subroutine.
40  *
41  * This implementation is based on an algorithm described in
42  *   C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT
43  *   Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics,
44  *   Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991.
45  * The primary algorithm described there uses 11 multiplies and 29 adds.
46  * We use their alternate method with 12 multiplies and 32 adds.
47  * The advantage of this method is that no data path contains more than one
48  * multiplication; this allows a very simple and accurate implementation in
49  * scaled fixed-point arithmetic, with a minimal number of shifts.
50  *
51  * I've made lots of modifications to attempt to take advantage of the
52  * sparse nature of the DCT matrices we're getting.  Although the logic
53  * is cumbersome, it's straightforward and the resulting code is much
54  * faster.
55  *
56  * A better way to do this would be to pass in the DCT block as a sparse
57  * matrix, perhaps with the difference cases encoded.
58  */
59 
60 /**
61  * @file
62  * Independent JPEG Group's LLM idct.
63  */
64 
65 #include "libavutil/common.h"
66 #include "libavutil/intreadwrite.h"
67 
68 #include "dct.h"
69 #include "idctdsp.h"
70 
71 #define EIGHT_BIT_SAMPLES
72 
73 #define DCTSIZE 8
74 #define DCTSIZE2 64
75 
76 #define GLOBAL
77 
78 #define RIGHT_SHIFT(x, n) ((x) >> (n))
79 
80 typedef int16_t DCTBLOCK[DCTSIZE2];
81 
82 #define CONST_BITS 13
83 
84 /*
85  * This routine is specialized to the case DCTSIZE = 8.
86  */
87 
88 #if DCTSIZE != 8
89   Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */
90 #endif
91 
92 
93 /*
94  * A 2-D IDCT can be done by 1-D IDCT on each row followed by 1-D IDCT
95  * on each column.  Direct algorithms are also available, but they are
96  * much more complex and seem not to be any faster when reduced to code.
97  *
98  * The poop on this scaling stuff is as follows:
99  *
100  * Each 1-D IDCT step produces outputs which are a factor of sqrt(N)
101  * larger than the true IDCT outputs.  The final outputs are therefore
102  * a factor of N larger than desired; since N=8 this can be cured by
103  * a simple right shift at the end of the algorithm.  The advantage of
104  * this arrangement is that we save two multiplications per 1-D IDCT,
105  * because the y0 and y4 inputs need not be divided by sqrt(N).
106  *
107  * We have to do addition and subtraction of the integer inputs, which
108  * is no problem, and multiplication by fractional constants, which is
109  * a problem to do in integer arithmetic.  We multiply all the constants
110  * by CONST_SCALE and convert them to integer constants (thus retaining
111  * CONST_BITS bits of precision in the constants).  After doing a
112  * multiplication we have to divide the product by CONST_SCALE, with proper
113  * rounding, to produce the correct output.  This division can be done
114  * cheaply as a right shift of CONST_BITS bits.  We postpone shifting
115  * as long as possible so that partial sums can be added together with
116  * full fractional precision.
117  *
118  * The outputs of the first pass are scaled up by PASS1_BITS bits so that
119  * they are represented to better-than-integral precision.  These outputs
120  * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word
121  * with the recommended scaling.  (To scale up 12-bit sample data further, an
122  * intermediate int32 array would be needed.)
123  *
124  * To avoid overflow of the 32-bit intermediate results in pass 2, we must
125  * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26.  Error analysis
126  * shows that the values given below are the most effective.
127  */
128 
129 #ifdef EIGHT_BIT_SAMPLES
130 #define PASS1_BITS  2
131 #else
132 #define PASS1_BITS  1   /* lose a little precision to avoid overflow */
133 #endif
134 
135 #define ONE         ((int32_t) 1)
136 
137 #define CONST_SCALE (ONE << CONST_BITS)
138 
139 /* Convert a positive real constant to an integer scaled by CONST_SCALE.
140  * IMPORTANT: if your compiler doesn't do this arithmetic at compile time,
141  * you will pay a significant penalty in run time.  In that case, figure
142  * the correct integer constant values and insert them by hand.
143  */
144 
145 /* Actually FIX is no longer used, we precomputed them all */
146 #define FIX(x)  ((int32_t) ((x) * CONST_SCALE + 0.5))
147 
148 /* Descale and correctly round an int32_t value that's scaled by N bits.
149  * We assume RIGHT_SHIFT rounds towards minus infinity, so adding
150  * the fudge factor is correct for either sign of X.
151  */
152 
153 #define DESCALE(x,n)  RIGHT_SHIFT((x) + (ONE << ((n)-1)), n)
154 
155 /* Multiply an int32_t variable by an int32_t constant to yield an int32_t result.
156  * For 8-bit samples with the recommended scaling, all the variable
157  * and constant values involved are no more than 16 bits wide, so a
158  * 16x16->32 bit multiply can be used instead of a full 32x32 multiply;
159  * this provides a useful speedup on many machines.
160  * There is no way to specify a 16x16->32 multiply in portable C, but
161  * some C compilers will do the right thing if you provide the correct
162  * combination of casts.
163  * NB: for 12-bit samples, a full 32-bit multiplication will be needed.
164  */
165 
166 #ifdef EIGHT_BIT_SAMPLES
167 #ifdef SHORTxSHORT_32           /* may work if 'int' is 32 bits */
168 #define MULTIPLY(var,const)  (((int16_t) (var)) * ((int16_t) (const)))
169 #endif
170 #ifdef SHORTxLCONST_32          /* known to work with Microsoft C 6.0 */
171 #define MULTIPLY(var,const)  (((int16_t) (var)) * ((int32_t) (const)))
172 #endif
173 #endif
174 
175 #ifndef MULTIPLY                /* default definition */
176 #define MULTIPLY(var,const)  ((var) * (const))
177 #endif
178 
179 
180 /*
181   Unlike our decoder where we approximate the FIXes, we need to use exact
182 ones here or successive P-frames will drift too much with Reference frame coding
183 */
184 #define FIX_0_211164243 1730
185 #define FIX_0_275899380 2260
186 #define FIX_0_298631336 2446
187 #define FIX_0_390180644 3196
188 #define FIX_0_509795579 4176
189 #define FIX_0_541196100 4433
190 #define FIX_0_601344887 4926
191 #define FIX_0_765366865 6270
192 #define FIX_0_785694958 6436
193 #define FIX_0_899976223 7373
194 #define FIX_1_061594337 8697
195 #define FIX_1_111140466 9102
196 #define FIX_1_175875602 9633
197 #define FIX_1_306562965 10703
198 #define FIX_1_387039845 11363
199 #define FIX_1_451774981 11893
200 #define FIX_1_501321110 12299
201 #define FIX_1_662939225 13623
202 #define FIX_1_847759065 15137
203 #define FIX_1_961570560 16069
204 #define FIX_2_053119869 16819
205 #define FIX_2_172734803 17799
206 #define FIX_2_562915447 20995
207 #define FIX_3_072711026 25172
208 
209 /*
210  * Perform the inverse DCT on one block of coefficients.
211  */
212 
213 void ff_j_rev_dct(DCTBLOCK data)
214 {
215   int32_t tmp0, tmp1, tmp2, tmp3;
216   int32_t tmp10, tmp11, tmp12, tmp13;
217   int32_t z1, z2, z3, z4, z5;
218   int32_t d0, d1, d2, d3, d4, d5, d6, d7;
219   register int16_t *dataptr;
220   int rowctr;
221 
222   /* Pass 1: process rows. */
223   /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
224   /* furthermore, we scale the results by 2**PASS1_BITS. */
225 
226   dataptr = data;
227 
228   for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
229     /* Due to quantization, we will usually find that many of the input
230      * coefficients are zero, especially the AC terms.  We can exploit this
231      * by short-circuiting the IDCT calculation for any row in which all
232      * the AC terms are zero.  In that case each output is equal to the
233      * DC coefficient (with scale factor as needed).
234      * With typical images and quantization tables, half or more of the
235      * row DCT calculations can be simplified this way.
236      */
237 
238     register uint8_t *idataptr = (uint8_t*)dataptr;
239 
240     /* WARNING: we do the same permutation as MMX idct to simplify the
241        video core */
242     d0 = dataptr[0];
243     d2 = dataptr[1];
244     d4 = dataptr[2];
245     d6 = dataptr[3];
246     d1 = dataptr[4];
247     d3 = dataptr[5];
248     d5 = dataptr[6];
249     d7 = dataptr[7];
250 
251     if ((d1 | d2 | d3 | d4 | d5 | d6 | d7) == 0) {
252       /* AC terms all zero */
253       if (d0) {
254           /* Compute a 32 bit value to assign. */
255           int16_t dcval = (int16_t) (d0 * (1 << PASS1_BITS));
256           register int v = (dcval & 0xffff) | ((dcval * (1 << 16)) & 0xffff0000);
257 
258           AV_WN32A(&idataptr[ 0], v);
259           AV_WN32A(&idataptr[ 4], v);
260           AV_WN32A(&idataptr[ 8], v);
261           AV_WN32A(&idataptr[12], v);
262       }
263 
264       dataptr += DCTSIZE;       /* advance pointer to next row */
265       continue;
266     }
267 
268     /* Even part: reverse the even part of the forward DCT. */
269     /* The rotator is sqrt(2)*c(-6). */
270 {
271     if (d6) {
272             if (d2) {
273                     /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */
274                     z1 = MULTIPLY(d2 + d6, FIX_0_541196100);
275                     tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065);
276                     tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865);
277 
278                     tmp0 = (d0 + d4) * CONST_SCALE;
279                     tmp1 = (d0 - d4) * CONST_SCALE;
280 
281                     tmp10 = tmp0 + tmp3;
282                     tmp13 = tmp0 - tmp3;
283                     tmp11 = tmp1 + tmp2;
284                     tmp12 = tmp1 - tmp2;
285             } else {
286                     /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */
287                     tmp2 = MULTIPLY(-d6, FIX_1_306562965);
288                     tmp3 = MULTIPLY(d6, FIX_0_541196100);
289 
290                     tmp0 = (d0 + d4) * CONST_SCALE;
291                     tmp1 = (d0 - d4) * CONST_SCALE;
292 
293                     tmp10 = tmp0 + tmp3;
294                     tmp13 = tmp0 - tmp3;
295                     tmp11 = tmp1 + tmp2;
296                     tmp12 = tmp1 - tmp2;
297             }
298     } else {
299             if (d2) {
300                     /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */
301                     tmp2 = MULTIPLY(d2, FIX_0_541196100);
302                     tmp3 = MULTIPLY(d2, FIX_1_306562965);
303 
304                     tmp0 = (d0 + d4) * CONST_SCALE;
305                     tmp1 = (d0 - d4) * CONST_SCALE;
306 
307                     tmp10 = tmp0 + tmp3;
308                     tmp13 = tmp0 - tmp3;
309                     tmp11 = tmp1 + tmp2;
310                     tmp12 = tmp1 - tmp2;
311             } else {
312                     /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */
313                     tmp10 = tmp13 = (d0 + d4) * CONST_SCALE;
314                     tmp11 = tmp12 = (d0 - d4) * CONST_SCALE;
315             }
316       }
317 
318     /* Odd part per figure 8; the matrix is unitary and hence its
319      * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
320      */
321 
322     if (d7) {
323         if (d5) {
324             if (d3) {
325                 if (d1) {
326                     /* d1 != 0, d3 != 0, d5 != 0, d7 != 0 */
327                     z1 = d7 + d1;
328                     z2 = d5 + d3;
329                     z3 = d7 + d3;
330                     z4 = d5 + d1;
331                     z5 = MULTIPLY(z3 + z4, FIX_1_175875602);
332 
333                     tmp0 = MULTIPLY(d7, FIX_0_298631336);
334                     tmp1 = MULTIPLY(d5, FIX_2_053119869);
335                     tmp2 = MULTIPLY(d3, FIX_3_072711026);
336                     tmp3 = MULTIPLY(d1, FIX_1_501321110);
337                     z1 = MULTIPLY(-z1, FIX_0_899976223);
338                     z2 = MULTIPLY(-z2, FIX_2_562915447);
339                     z3 = MULTIPLY(-z3, FIX_1_961570560);
340                     z4 = MULTIPLY(-z4, FIX_0_390180644);
341 
342                     z3 += z5;
343                     z4 += z5;
344 
345                     tmp0 += z1 + z3;
346                     tmp1 += z2 + z4;
347                     tmp2 += z2 + z3;
348                     tmp3 += z1 + z4;
349                 } else {
350                     /* d1 == 0, d3 != 0, d5 != 0, d7 != 0 */
351                     z2 = d5 + d3;
352                     z3 = d7 + d3;
353                     z5 = MULTIPLY(z3 + d5, FIX_1_175875602);
354 
355                     tmp0 = MULTIPLY(d7, FIX_0_298631336);
356                     tmp1 = MULTIPLY(d5, FIX_2_053119869);
357                     tmp2 = MULTIPLY(d3, FIX_3_072711026);
358                     z1 = MULTIPLY(-d7, FIX_0_899976223);
359                     z2 = MULTIPLY(-z2, FIX_2_562915447);
360                     z3 = MULTIPLY(-z3, FIX_1_961570560);
361                     z4 = MULTIPLY(-d5, FIX_0_390180644);
362 
363                     z3 += z5;
364                     z4 += z5;
365 
366                     tmp0 += z1 + z3;
367                     tmp1 += z2 + z4;
368                     tmp2 += z2 + z3;
369                     tmp3 = z1 + z4;
370                 }
371             } else {
372                 if (d1) {
373                     /* d1 != 0, d3 == 0, d5 != 0, d7 != 0 */
374                     z1 = d7 + d1;
375                     z4 = d5 + d1;
376                     z5 = MULTIPLY(d7 + z4, FIX_1_175875602);
377 
378                     tmp0 = MULTIPLY(d7, FIX_0_298631336);
379                     tmp1 = MULTIPLY(d5, FIX_2_053119869);
380                     tmp3 = MULTIPLY(d1, FIX_1_501321110);
381                     z1 = MULTIPLY(-z1, FIX_0_899976223);
382                     z2 = MULTIPLY(-d5, FIX_2_562915447);
383                     z3 = MULTIPLY(-d7, FIX_1_961570560);
384                     z4 = MULTIPLY(-z4, FIX_0_390180644);
385 
386                     z3 += z5;
387                     z4 += z5;
388 
389                     tmp0 += z1 + z3;
390                     tmp1 += z2 + z4;
391                     tmp2 = z2 + z3;
392                     tmp3 += z1 + z4;
393                 } else {
394                     /* d1 == 0, d3 == 0, d5 != 0, d7 != 0 */
395                     tmp0 = MULTIPLY(-d7, FIX_0_601344887);
396                     z1 = MULTIPLY(-d7, FIX_0_899976223);
397                     z3 = MULTIPLY(-d7, FIX_1_961570560);
398                     tmp1 = MULTIPLY(-d5, FIX_0_509795579);
399                     z2 = MULTIPLY(-d5, FIX_2_562915447);
400                     z4 = MULTIPLY(-d5, FIX_0_390180644);
401                     z5 = MULTIPLY(d5 + d7, FIX_1_175875602);
402 
403                     z3 += z5;
404                     z4 += z5;
405 
406                     tmp0 += z3;
407                     tmp1 += z4;
408                     tmp2 = z2 + z3;
409                     tmp3 = z1 + z4;
410                 }
411             }
412         } else {
413             if (d3) {
414                 if (d1) {
415                     /* d1 != 0, d3 != 0, d5 == 0, d7 != 0 */
416                     z1 = d7 + d1;
417                     z3 = d7 + d3;
418                     z5 = MULTIPLY(z3 + d1, FIX_1_175875602);
419 
420                     tmp0 = MULTIPLY(d7, FIX_0_298631336);
421                     tmp2 = MULTIPLY(d3, FIX_3_072711026);
422                     tmp3 = MULTIPLY(d1, FIX_1_501321110);
423                     z1 = MULTIPLY(-z1, FIX_0_899976223);
424                     z2 = MULTIPLY(-d3, FIX_2_562915447);
425                     z3 = MULTIPLY(-z3, FIX_1_961570560);
426                     z4 = MULTIPLY(-d1, FIX_0_390180644);
427 
428                     z3 += z5;
429                     z4 += z5;
430 
431                     tmp0 += z1 + z3;
432                     tmp1 = z2 + z4;
433                     tmp2 += z2 + z3;
434                     tmp3 += z1 + z4;
435                 } else {
436                     /* d1 == 0, d3 != 0, d5 == 0, d7 != 0 */
437                     z3 = d7 + d3;
438 
439                     tmp0 = MULTIPLY(-d7, FIX_0_601344887);
440                     z1 = MULTIPLY(-d7, FIX_0_899976223);
441                     tmp2 = MULTIPLY(d3, FIX_0_509795579);
442                     z2 = MULTIPLY(-d3, FIX_2_562915447);
443                     z5 = MULTIPLY(z3, FIX_1_175875602);
444                     z3 = MULTIPLY(-z3, FIX_0_785694958);
445 
446                     tmp0 += z3;
447                     tmp1 = z2 + z5;
448                     tmp2 += z3;
449                     tmp3 = z1 + z5;
450                 }
451             } else {
452                 if (d1) {
453                     /* d1 != 0, d3 == 0, d5 == 0, d7 != 0 */
454                     z1 = d7 + d1;
455                     z5 = MULTIPLY(z1, FIX_1_175875602);
456 
457                     z1 = MULTIPLY(z1, FIX_0_275899380);
458                     z3 = MULTIPLY(-d7, FIX_1_961570560);
459                     tmp0 = MULTIPLY(-d7, FIX_1_662939225);
460                     z4 = MULTIPLY(-d1, FIX_0_390180644);
461                     tmp3 = MULTIPLY(d1, FIX_1_111140466);
462 
463                     tmp0 += z1;
464                     tmp1 = z4 + z5;
465                     tmp2 = z3 + z5;
466                     tmp3 += z1;
467                 } else {
468                     /* d1 == 0, d3 == 0, d5 == 0, d7 != 0 */
469                     tmp0 = MULTIPLY(-d7, FIX_1_387039845);
470                     tmp1 = MULTIPLY(d7, FIX_1_175875602);
471                     tmp2 = MULTIPLY(-d7, FIX_0_785694958);
472                     tmp3 = MULTIPLY(d7, FIX_0_275899380);
473                 }
474             }
475         }
476     } else {
477         if (d5) {
478             if (d3) {
479                 if (d1) {
480                     /* d1 != 0, d3 != 0, d5 != 0, d7 == 0 */
481                     z2 = d5 + d3;
482                     z4 = d5 + d1;
483                     z5 = MULTIPLY(d3 + z4, FIX_1_175875602);
484 
485                     tmp1 = MULTIPLY(d5, FIX_2_053119869);
486                     tmp2 = MULTIPLY(d3, FIX_3_072711026);
487                     tmp3 = MULTIPLY(d1, FIX_1_501321110);
488                     z1 = MULTIPLY(-d1, FIX_0_899976223);
489                     z2 = MULTIPLY(-z2, FIX_2_562915447);
490                     z3 = MULTIPLY(-d3, FIX_1_961570560);
491                     z4 = MULTIPLY(-z4, FIX_0_390180644);
492 
493                     z3 += z5;
494                     z4 += z5;
495 
496                     tmp0 = z1 + z3;
497                     tmp1 += z2 + z4;
498                     tmp2 += z2 + z3;
499                     tmp3 += z1 + z4;
500                 } else {
501                     /* d1 == 0, d3 != 0, d5 != 0, d7 == 0 */
502                     z2 = d5 + d3;
503 
504                     z5 = MULTIPLY(z2, FIX_1_175875602);
505                     tmp1 = MULTIPLY(d5, FIX_1_662939225);
506                     z4 = MULTIPLY(-d5, FIX_0_390180644);
507                     z2 = MULTIPLY(-z2, FIX_1_387039845);
508                     tmp2 = MULTIPLY(d3, FIX_1_111140466);
509                     z3 = MULTIPLY(-d3, FIX_1_961570560);
510 
511                     tmp0 = z3 + z5;
512                     tmp1 += z2;
513                     tmp2 += z2;
514                     tmp3 = z4 + z5;
515                 }
516             } else {
517                 if (d1) {
518                     /* d1 != 0, d3 == 0, d5 != 0, d7 == 0 */
519                     z4 = d5 + d1;
520 
521                     z5 = MULTIPLY(z4, FIX_1_175875602);
522                     z1 = MULTIPLY(-d1, FIX_0_899976223);
523                     tmp3 = MULTIPLY(d1, FIX_0_601344887);
524                     tmp1 = MULTIPLY(-d5, FIX_0_509795579);
525                     z2 = MULTIPLY(-d5, FIX_2_562915447);
526                     z4 = MULTIPLY(z4, FIX_0_785694958);
527 
528                     tmp0 = z1 + z5;
529                     tmp1 += z4;
530                     tmp2 = z2 + z5;
531                     tmp3 += z4;
532                 } else {
533                     /* d1 == 0, d3 == 0, d5 != 0, d7 == 0 */
534                     tmp0 = MULTIPLY(d5, FIX_1_175875602);
535                     tmp1 = MULTIPLY(d5, FIX_0_275899380);
536                     tmp2 = MULTIPLY(-d5, FIX_1_387039845);
537                     tmp3 = MULTIPLY(d5, FIX_0_785694958);
538                 }
539             }
540         } else {
541             if (d3) {
542                 if (d1) {
543                     /* d1 != 0, d3 != 0, d5 == 0, d7 == 0 */
544                     z5 = d1 + d3;
545                     tmp3 = MULTIPLY(d1, FIX_0_211164243);
546                     tmp2 = MULTIPLY(-d3, FIX_1_451774981);
547                     z1 = MULTIPLY(d1, FIX_1_061594337);
548                     z2 = MULTIPLY(-d3, FIX_2_172734803);
549                     z4 = MULTIPLY(z5, FIX_0_785694958);
550                     z5 = MULTIPLY(z5, FIX_1_175875602);
551 
552                     tmp0 = z1 - z4;
553                     tmp1 = z2 + z4;
554                     tmp2 += z5;
555                     tmp3 += z5;
556                 } else {
557                     /* d1 == 0, d3 != 0, d5 == 0, d7 == 0 */
558                     tmp0 = MULTIPLY(-d3, FIX_0_785694958);
559                     tmp1 = MULTIPLY(-d3, FIX_1_387039845);
560                     tmp2 = MULTIPLY(-d3, FIX_0_275899380);
561                     tmp3 = MULTIPLY(d3, FIX_1_175875602);
562                 }
563             } else {
564                 if (d1) {
565                     /* d1 != 0, d3 == 0, d5 == 0, d7 == 0 */
566                     tmp0 = MULTIPLY(d1, FIX_0_275899380);
567                     tmp1 = MULTIPLY(d1, FIX_0_785694958);
568                     tmp2 = MULTIPLY(d1, FIX_1_175875602);
569                     tmp3 = MULTIPLY(d1, FIX_1_387039845);
570                 } else {
571                     /* d1 == 0, d3 == 0, d5 == 0, d7 == 0 */
572                     tmp0 = tmp1 = tmp2 = tmp3 = 0;
573                 }
574             }
575         }
576     }
577 }
578     /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
579 
580     dataptr[0] = (int16_t) DESCALE(tmp10 + tmp3, CONST_BITS-PASS1_BITS);
581     dataptr[7] = (int16_t) DESCALE(tmp10 - tmp3, CONST_BITS-PASS1_BITS);
582     dataptr[1] = (int16_t) DESCALE(tmp11 + tmp2, CONST_BITS-PASS1_BITS);
583     dataptr[6] = (int16_t) DESCALE(tmp11 - tmp2, CONST_BITS-PASS1_BITS);
584     dataptr[2] = (int16_t) DESCALE(tmp12 + tmp1, CONST_BITS-PASS1_BITS);
585     dataptr[5] = (int16_t) DESCALE(tmp12 - tmp1, CONST_BITS-PASS1_BITS);
586     dataptr[3] = (int16_t) DESCALE(tmp13 + tmp0, CONST_BITS-PASS1_BITS);
587     dataptr[4] = (int16_t) DESCALE(tmp13 - tmp0, CONST_BITS-PASS1_BITS);
588 
589     dataptr += DCTSIZE;         /* advance pointer to next row */
590   }
591 
592   /* Pass 2: process columns. */
593   /* Note that we must descale the results by a factor of 8 == 2**3, */
594   /* and also undo the PASS1_BITS scaling. */
595 
596   dataptr = data;
597   for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
598     /* Columns of zeroes can be exploited in the same way as we did with rows.
599      * However, the row calculation has created many nonzero AC terms, so the
600      * simplification applies less often (typically 5% to 10% of the time).
601      * On machines with very fast multiplication, it's possible that the
602      * test takes more time than it's worth.  In that case this section
603      * may be commented out.
604      */
605 
606     d0 = dataptr[DCTSIZE*0];
607     d1 = dataptr[DCTSIZE*1];
608     d2 = dataptr[DCTSIZE*2];
609     d3 = dataptr[DCTSIZE*3];
610     d4 = dataptr[DCTSIZE*4];
611     d5 = dataptr[DCTSIZE*5];
612     d6 = dataptr[DCTSIZE*6];
613     d7 = dataptr[DCTSIZE*7];
614 
615     /* Even part: reverse the even part of the forward DCT. */
616     /* The rotator is sqrt(2)*c(-6). */
617     if (d6) {
618             if (d2) {
619                     /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */
620                     z1 = MULTIPLY(d2 + d6, FIX_0_541196100);
621                     tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065);
622                     tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865);
623 
624                     tmp0 = (d0 + d4) * CONST_SCALE;
625                     tmp1 = (d0 - d4) * CONST_SCALE;
626 
627                     tmp10 = tmp0 + tmp3;
628                     tmp13 = tmp0 - tmp3;
629                     tmp11 = tmp1 + tmp2;
630                     tmp12 = tmp1 - tmp2;
631             } else {
632                     /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */
633                     tmp2 = MULTIPLY(-d6, FIX_1_306562965);
634                     tmp3 = MULTIPLY(d6, FIX_0_541196100);
635 
636                     tmp0 = (d0 + d4) * CONST_SCALE;
637                     tmp1 = (d0 - d4) * CONST_SCALE;
638 
639                     tmp10 = tmp0 + tmp3;
640                     tmp13 = tmp0 - tmp3;
641                     tmp11 = tmp1 + tmp2;
642                     tmp12 = tmp1 - tmp2;
643             }
644     } else {
645             if (d2) {
646                     /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */
647                     tmp2 = MULTIPLY(d2, FIX_0_541196100);
648                     tmp3 = MULTIPLY(d2, FIX_1_306562965);
649 
650                     tmp0 = (d0 + d4) * CONST_SCALE;
651                     tmp1 = (d0 - d4) * CONST_SCALE;
652 
653                     tmp10 = tmp0 + tmp3;
654                     tmp13 = tmp0 - tmp3;
655                     tmp11 = tmp1 + tmp2;
656                     tmp12 = tmp1 - tmp2;
657             } else {
658                     /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */
659                     tmp10 = tmp13 = (d0 + d4) * CONST_SCALE;
660                     tmp11 = tmp12 = (d0 - d4) * CONST_SCALE;
661             }
662     }
663 
664     /* Odd part per figure 8; the matrix is unitary and hence its
665      * transpose is its inverse.  i0..i3 are y7,y5,y3,y1 respectively.
666      */
667     if (d7) {
668         if (d5) {
669             if (d3) {
670                 if (d1) {
671                     /* d1 != 0, d3 != 0, d5 != 0, d7 != 0 */
672                     z1 = d7 + d1;
673                     z2 = d5 + d3;
674                     z3 = d7 + d3;
675                     z4 = d5 + d1;
676                     z5 = MULTIPLY(z3 + z4, FIX_1_175875602);
677 
678                     tmp0 = MULTIPLY(d7, FIX_0_298631336);
679                     tmp1 = MULTIPLY(d5, FIX_2_053119869);
680                     tmp2 = MULTIPLY(d3, FIX_3_072711026);
681                     tmp3 = MULTIPLY(d1, FIX_1_501321110);
682                     z1 = MULTIPLY(-z1, FIX_0_899976223);
683                     z2 = MULTIPLY(-z2, FIX_2_562915447);
684                     z3 = MULTIPLY(-z3, FIX_1_961570560);
685                     z4 = MULTIPLY(-z4, FIX_0_390180644);
686 
687                     z3 += z5;
688                     z4 += z5;
689 
690                     tmp0 += z1 + z3;
691                     tmp1 += z2 + z4;
692                     tmp2 += z2 + z3;
693                     tmp3 += z1 + z4;
694                 } else {
695                     /* d1 == 0, d3 != 0, d5 != 0, d7 != 0 */
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                     z3 = d7;
721                     z4 = d5 + d1;
722                     z5 = MULTIPLY(z3 + z4, FIX_1_175875602);
723 
724                     tmp0 = MULTIPLY(d7, FIX_0_298631336);
725                     tmp1 = MULTIPLY(d5, FIX_2_053119869);
726                     tmp3 = MULTIPLY(d1, FIX_1_501321110);
727                     z1 = MULTIPLY(-z1, FIX_0_899976223);
728                     z2 = MULTIPLY(-d5, FIX_2_562915447);
729                     z3 = MULTIPLY(-d7, FIX_1_961570560);
730                     z4 = MULTIPLY(-z4, FIX_0_390180644);
731 
732                     z3 += z5;
733                     z4 += z5;
734 
735                     tmp0 += z1 + z3;
736                     tmp1 += z2 + z4;
737                     tmp2 = z2 + z3;
738                     tmp3 += z1 + z4;
739                 } else {
740                     /* d1 == 0, d3 == 0, d5 != 0, d7 != 0 */
741                     tmp0 = MULTIPLY(-d7, FIX_0_601344887);
742                     z1 = MULTIPLY(-d7, FIX_0_899976223);
743                     z3 = MULTIPLY(-d7, FIX_1_961570560);
744                     tmp1 = MULTIPLY(-d5, FIX_0_509795579);
745                     z2 = MULTIPLY(-d5, FIX_2_562915447);
746                     z4 = MULTIPLY(-d5, FIX_0_390180644);
747                     z5 = MULTIPLY(d5 + d7, FIX_1_175875602);
748 
749                     z3 += z5;
750                     z4 += z5;
751 
752                     tmp0 += z3;
753                     tmp1 += z4;
754                     tmp2 = z2 + z3;
755                     tmp3 = z1 + z4;
756                 }
757             }
758         } else {
759             if (d3) {
760                 if (d1) {
761                     /* d1 != 0, d3 != 0, d5 == 0, d7 != 0 */
762                     z1 = d7 + d1;
763                     z3 = d7 + d3;
764                     z5 = MULTIPLY(z3 + d1, FIX_1_175875602);
765 
766                     tmp0 = MULTIPLY(d7, FIX_0_298631336);
767                     tmp2 = MULTIPLY(d3, FIX_3_072711026);
768                     tmp3 = MULTIPLY(d1, FIX_1_501321110);
769                     z1 = MULTIPLY(-z1, FIX_0_899976223);
770                     z2 = MULTIPLY(-d3, FIX_2_562915447);
771                     z3 = MULTIPLY(-z3, FIX_1_961570560);
772                     z4 = MULTIPLY(-d1, FIX_0_390180644);
773 
774                     z3 += z5;
775                     z4 += z5;
776 
777                     tmp0 += z1 + z3;
778                     tmp1 = z2 + z4;
779                     tmp2 += z2 + z3;
780                     tmp3 += z1 + z4;
781                 } else {
782                     /* d1 == 0, d3 != 0, d5 == 0, d7 != 0 */
783                     z3 = d7 + d3;
784 
785                     tmp0 = MULTIPLY(-d7, FIX_0_601344887);
786                     z1 = MULTIPLY(-d7, FIX_0_899976223);
787                     tmp2 = MULTIPLY(d3, FIX_0_509795579);
788                     z2 = MULTIPLY(-d3, FIX_2_562915447);
789                     z5 = MULTIPLY(z3, FIX_1_175875602);
790                     z3 = MULTIPLY(-z3, FIX_0_785694958);
791 
792                     tmp0 += z3;
793                     tmp1 = z2 + z5;
794                     tmp2 += z3;
795                     tmp3 = z1 + z5;
796                 }
797             } else {
798                 if (d1) {
799                     /* d1 != 0, d3 == 0, d5 == 0, d7 != 0 */
800                     z1 = d7 + d1;
801                     z5 = MULTIPLY(z1, FIX_1_175875602);
802 
803                     z1 = MULTIPLY(z1, FIX_0_275899380);
804                     z3 = MULTIPLY(-d7, FIX_1_961570560);
805                     tmp0 = MULTIPLY(-d7, FIX_1_662939225);
806                     z4 = MULTIPLY(-d1, FIX_0_390180644);
807                     tmp3 = MULTIPLY(d1, FIX_1_111140466);
808 
809                     tmp0 += z1;
810                     tmp1 = z4 + z5;
811                     tmp2 = z3 + z5;
812                     tmp3 += z1;
813                 } else {
814                     /* d1 == 0, d3 == 0, d5 == 0, d7 != 0 */
815                     tmp0 = MULTIPLY(-d7, FIX_1_387039845);
816                     tmp1 = MULTIPLY(d7, FIX_1_175875602);
817                     tmp2 = MULTIPLY(-d7, FIX_0_785694958);
818                     tmp3 = MULTIPLY(d7, FIX_0_275899380);
819                 }
820             }
821         }
822     } else {
823         if (d5) {
824             if (d3) {
825                 if (d1) {
826                     /* d1 != 0, d3 != 0, d5 != 0, d7 == 0 */
827                     z2 = d5 + d3;
828                     z4 = d5 + d1;
829                     z5 = MULTIPLY(d3 + z4, FIX_1_175875602);
830 
831                     tmp1 = MULTIPLY(d5, FIX_2_053119869);
832                     tmp2 = MULTIPLY(d3, FIX_3_072711026);
833                     tmp3 = MULTIPLY(d1, FIX_1_501321110);
834                     z1 = MULTIPLY(-d1, FIX_0_899976223);
835                     z2 = MULTIPLY(-z2, FIX_2_562915447);
836                     z3 = MULTIPLY(-d3, FIX_1_961570560);
837                     z4 = MULTIPLY(-z4, FIX_0_390180644);
838 
839                     z3 += z5;
840                     z4 += z5;
841 
842                     tmp0 = z1 + z3;
843                     tmp1 += z2 + z4;
844                     tmp2 += z2 + z3;
845                     tmp3 += z1 + z4;
846                 } else {
847                     /* d1 == 0, d3 != 0, d5 != 0, d7 == 0 */
848                     z2 = d5 + d3;
849 
850                     z5 = MULTIPLY(z2, FIX_1_175875602);
851                     tmp1 = MULTIPLY(d5, FIX_1_662939225);
852                     z4 = MULTIPLY(-d5, FIX_0_390180644);
853                     z2 = MULTIPLY(-z2, FIX_1_387039845);
854                     tmp2 = MULTIPLY(d3, FIX_1_111140466);
855                     z3 = MULTIPLY(-d3, FIX_1_961570560);
856 
857                     tmp0 = z3 + z5;
858                     tmp1 += z2;
859                     tmp2 += z2;
860                     tmp3 = z4 + z5;
861                 }
862             } else {
863                 if (d1) {
864                     /* d1 != 0, d3 == 0, d5 != 0, d7 == 0 */
865                     z4 = d5 + d1;
866 
867                     z5 = MULTIPLY(z4, FIX_1_175875602);
868                     z1 = MULTIPLY(-d1, FIX_0_899976223);
869                     tmp3 = MULTIPLY(d1, FIX_0_601344887);
870                     tmp1 = MULTIPLY(-d5, FIX_0_509795579);
871                     z2 = MULTIPLY(-d5, FIX_2_562915447);
872                     z4 = MULTIPLY(z4, FIX_0_785694958);
873 
874                     tmp0 = z1 + z5;
875                     tmp1 += z4;
876                     tmp2 = z2 + z5;
877                     tmp3 += z4;
878                 } else {
879                     /* d1 == 0, d3 == 0, d5 != 0, d7 == 0 */
880                     tmp0 = MULTIPLY(d5, FIX_1_175875602);
881                     tmp1 = MULTIPLY(d5, FIX_0_275899380);
882                     tmp2 = MULTIPLY(-d5, FIX_1_387039845);
883                     tmp3 = MULTIPLY(d5, FIX_0_785694958);
884                 }
885             }
886         } else {
887             if (d3) {
888                 if (d1) {
889                     /* d1 != 0, d3 != 0, d5 == 0, d7 == 0 */
890                     z5 = d1 + d3;
891                     tmp3 = MULTIPLY(d1, FIX_0_211164243);
892                     tmp2 = MULTIPLY(-d3, FIX_1_451774981);
893                     z1 = MULTIPLY(d1, FIX_1_061594337);
894                     z2 = MULTIPLY(-d3, FIX_2_172734803);
895                     z4 = MULTIPLY(z5, FIX_0_785694958);
896                     z5 = MULTIPLY(z5, FIX_1_175875602);
897 
898                     tmp0 = z1 - z4;
899                     tmp1 = z2 + z4;
900                     tmp2 += z5;
901                     tmp3 += z5;
902                 } else {
903                     /* d1 == 0, d3 != 0, d5 == 0, d7 == 0 */
904                     tmp0 = MULTIPLY(-d3, FIX_0_785694958);
905                     tmp1 = MULTIPLY(-d3, FIX_1_387039845);
906                     tmp2 = MULTIPLY(-d3, FIX_0_275899380);
907                     tmp3 = MULTIPLY(d3, FIX_1_175875602);
908                 }
909             } else {
910                 if (d1) {
911                     /* d1 != 0, d3 == 0, d5 == 0, d7 == 0 */
912                     tmp0 = MULTIPLY(d1, FIX_0_275899380);
913                     tmp1 = MULTIPLY(d1, FIX_0_785694958);
914                     tmp2 = MULTIPLY(d1, FIX_1_175875602);
915                     tmp3 = MULTIPLY(d1, FIX_1_387039845);
916                 } else {
917                     /* d1 == 0, d3 == 0, d5 == 0, d7 == 0 */
918                     tmp0 = tmp1 = tmp2 = tmp3 = 0;
919                 }
920             }
921         }
922     }
923 
924     /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
925 
926     dataptr[DCTSIZE*0] = (int16_t) DESCALE(tmp10 + tmp3,
927                                            CONST_BITS+PASS1_BITS+3);
928     dataptr[DCTSIZE*7] = (int16_t) DESCALE(tmp10 - tmp3,
929                                            CONST_BITS+PASS1_BITS+3);
930     dataptr[DCTSIZE*1] = (int16_t) DESCALE(tmp11 + tmp2,
931                                            CONST_BITS+PASS1_BITS+3);
932     dataptr[DCTSIZE*6] = (int16_t) DESCALE(tmp11 - tmp2,
933                                            CONST_BITS+PASS1_BITS+3);
934     dataptr[DCTSIZE*2] = (int16_t) DESCALE(tmp12 + tmp1,
935                                            CONST_BITS+PASS1_BITS+3);
936     dataptr[DCTSIZE*5] = (int16_t) DESCALE(tmp12 - tmp1,
937                                            CONST_BITS+PASS1_BITS+3);
938     dataptr[DCTSIZE*3] = (int16_t) DESCALE(tmp13 + tmp0,
939                                            CONST_BITS+PASS1_BITS+3);
940     dataptr[DCTSIZE*4] = (int16_t) DESCALE(tmp13 - tmp0,
941                                            CONST_BITS+PASS1_BITS+3);
942 
943     dataptr++;                  /* advance pointer to next column */
944   }
945 }
946 
947 #undef DCTSIZE
948 #define DCTSIZE 4
949 #define DCTSTRIDE 8
950 
ff_j_rev_dct4(DCTBLOCK data)951 void ff_j_rev_dct4(DCTBLOCK data)
952 {
953   int32_t tmp0, tmp1, tmp2, tmp3;
954   int32_t tmp10, tmp11, tmp12, tmp13;
955   int32_t z1;
956   int32_t d0, d2, d4, d6;
957   register int16_t *dataptr;
958   int rowctr;
959 
960   /* Pass 1: process rows. */
961   /* Note results are scaled up by sqrt(8) compared to a true IDCT; */
962   /* furthermore, we scale the results by 2**PASS1_BITS. */
963 
964   data[0] += 4;
965 
966   dataptr = data;
967 
968   for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
969     /* Due to quantization, we will usually find that many of the input
970      * coefficients are zero, especially the AC terms.  We can exploit this
971      * by short-circuiting the IDCT calculation for any row in which all
972      * the AC terms are zero.  In that case each output is equal to the
973      * DC coefficient (with scale factor as needed).
974      * With typical images and quantization tables, half or more of the
975      * row DCT calculations can be simplified this way.
976      */
977 
978     register uint8_t *idataptr = (uint8_t*)dataptr;
979 
980     d0 = dataptr[0];
981     d2 = dataptr[1];
982     d4 = dataptr[2];
983     d6 = dataptr[3];
984 
985     if ((d2 | d4 | d6) == 0) {
986       /* AC terms all zero */
987       if (d0) {
988           /* Compute a 32 bit value to assign. */
989           int16_t dcval = (int16_t) (d0 << PASS1_BITS);
990           register int v = (dcval & 0xffff) | ((dcval << 16) & 0xffff0000);
991 
992           AV_WN32A(&idataptr[0], v);
993           AV_WN32A(&idataptr[4], v);
994       }
995 
996       dataptr += DCTSTRIDE;     /* advance pointer to next row */
997       continue;
998     }
999 
1000     /* Even part: reverse the even part of the forward DCT. */
1001     /* The rotator is sqrt(2)*c(-6). */
1002     if (d6) {
1003             if (d2) {
1004                     /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */
1005                     z1 = MULTIPLY(d2 + d6, FIX_0_541196100);
1006                     tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065);
1007                     tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865);
1008 
1009                     tmp0 = (d0 + d4) << CONST_BITS;
1010                     tmp1 = (d0 - d4) << CONST_BITS;
1011 
1012                     tmp10 = tmp0 + tmp3;
1013                     tmp13 = tmp0 - tmp3;
1014                     tmp11 = tmp1 + tmp2;
1015                     tmp12 = tmp1 - tmp2;
1016             } else {
1017                     /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */
1018                     tmp2 = MULTIPLY(-d6, FIX_1_306562965);
1019                     tmp3 = MULTIPLY(d6, FIX_0_541196100);
1020 
1021                     tmp0 = (d0 + d4) << CONST_BITS;
1022                     tmp1 = (d0 - d4) << CONST_BITS;
1023 
1024                     tmp10 = tmp0 + tmp3;
1025                     tmp13 = tmp0 - tmp3;
1026                     tmp11 = tmp1 + tmp2;
1027                     tmp12 = tmp1 - tmp2;
1028             }
1029     } else {
1030             if (d2) {
1031                     /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */
1032                     tmp2 = MULTIPLY(d2, FIX_0_541196100);
1033                     tmp3 = MULTIPLY(d2, FIX_1_306562965);
1034 
1035                     tmp0 = (d0 + d4) << CONST_BITS;
1036                     tmp1 = (d0 - d4) << CONST_BITS;
1037 
1038                     tmp10 = tmp0 + tmp3;
1039                     tmp13 = tmp0 - tmp3;
1040                     tmp11 = tmp1 + tmp2;
1041                     tmp12 = tmp1 - tmp2;
1042             } else {
1043                     /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */
1044                     tmp10 = tmp13 = (d0 + d4) << CONST_BITS;
1045                     tmp11 = tmp12 = (d0 - d4) << CONST_BITS;
1046             }
1047       }
1048 
1049     /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
1050 
1051     dataptr[0] = (int16_t) DESCALE(tmp10, CONST_BITS-PASS1_BITS);
1052     dataptr[1] = (int16_t) DESCALE(tmp11, CONST_BITS-PASS1_BITS);
1053     dataptr[2] = (int16_t) DESCALE(tmp12, CONST_BITS-PASS1_BITS);
1054     dataptr[3] = (int16_t) DESCALE(tmp13, CONST_BITS-PASS1_BITS);
1055 
1056     dataptr += DCTSTRIDE;       /* advance pointer to next row */
1057   }
1058 
1059   /* Pass 2: process columns. */
1060   /* Note that we must descale the results by a factor of 8 == 2**3, */
1061   /* and also undo the PASS1_BITS scaling. */
1062 
1063   dataptr = data;
1064   for (rowctr = DCTSIZE-1; rowctr >= 0; rowctr--) {
1065     /* Columns of zeroes can be exploited in the same way as we did with rows.
1066      * However, the row calculation has created many nonzero AC terms, so the
1067      * simplification applies less often (typically 5% to 10% of the time).
1068      * On machines with very fast multiplication, it's possible that the
1069      * test takes more time than it's worth.  In that case this section
1070      * may be commented out.
1071      */
1072 
1073     d0 = dataptr[DCTSTRIDE*0];
1074     d2 = dataptr[DCTSTRIDE*1];
1075     d4 = dataptr[DCTSTRIDE*2];
1076     d6 = dataptr[DCTSTRIDE*3];
1077 
1078     /* Even part: reverse the even part of the forward DCT. */
1079     /* The rotator is sqrt(2)*c(-6). */
1080     if (d6) {
1081             if (d2) {
1082                     /* d0 != 0, d2 != 0, d4 != 0, d6 != 0 */
1083                     z1 = MULTIPLY(d2 + d6, FIX_0_541196100);
1084                     tmp2 = z1 + MULTIPLY(-d6, FIX_1_847759065);
1085                     tmp3 = z1 + MULTIPLY(d2, FIX_0_765366865);
1086 
1087                     tmp0 = (d0 + d4) << CONST_BITS;
1088                     tmp1 = (d0 - d4) << CONST_BITS;
1089 
1090                     tmp10 = tmp0 + tmp3;
1091                     tmp13 = tmp0 - tmp3;
1092                     tmp11 = tmp1 + tmp2;
1093                     tmp12 = tmp1 - tmp2;
1094             } else {
1095                     /* d0 != 0, d2 == 0, d4 != 0, d6 != 0 */
1096                     tmp2 = MULTIPLY(-d6, FIX_1_306562965);
1097                     tmp3 = MULTIPLY(d6, FIX_0_541196100);
1098 
1099                     tmp0 = (d0 + d4) << CONST_BITS;
1100                     tmp1 = (d0 - d4) << CONST_BITS;
1101 
1102                     tmp10 = tmp0 + tmp3;
1103                     tmp13 = tmp0 - tmp3;
1104                     tmp11 = tmp1 + tmp2;
1105                     tmp12 = tmp1 - tmp2;
1106             }
1107     } else {
1108             if (d2) {
1109                     /* d0 != 0, d2 != 0, d4 != 0, d6 == 0 */
1110                     tmp2 = MULTIPLY(d2, FIX_0_541196100);
1111                     tmp3 = MULTIPLY(d2, FIX_1_306562965);
1112 
1113                     tmp0 = (d0 + d4) << CONST_BITS;
1114                     tmp1 = (d0 - d4) << CONST_BITS;
1115 
1116                     tmp10 = tmp0 + tmp3;
1117                     tmp13 = tmp0 - tmp3;
1118                     tmp11 = tmp1 + tmp2;
1119                     tmp12 = tmp1 - tmp2;
1120             } else {
1121                     /* d0 != 0, d2 == 0, d4 != 0, d6 == 0 */
1122                     tmp10 = tmp13 = (d0 + d4) << CONST_BITS;
1123                     tmp11 = tmp12 = (d0 - d4) << CONST_BITS;
1124             }
1125     }
1126 
1127     /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */
1128 
1129     dataptr[DCTSTRIDE*0] = tmp10 >> (CONST_BITS+PASS1_BITS+3);
1130     dataptr[DCTSTRIDE*1] = tmp11 >> (CONST_BITS+PASS1_BITS+3);
1131     dataptr[DCTSTRIDE*2] = tmp12 >> (CONST_BITS+PASS1_BITS+3);
1132     dataptr[DCTSTRIDE*3] = tmp13 >> (CONST_BITS+PASS1_BITS+3);
1133 
1134     dataptr++;                  /* advance pointer to next column */
1135   }
1136 }
1137 
ff_j_rev_dct2(DCTBLOCK data)1138 void ff_j_rev_dct2(DCTBLOCK data){
1139   int d00, d01, d10, d11;
1140 
1141   data[0] += 4;
1142   d00 = data[0+0*DCTSTRIDE] + data[1+0*DCTSTRIDE];
1143   d01 = data[0+0*DCTSTRIDE] - data[1+0*DCTSTRIDE];
1144   d10 = data[0+1*DCTSTRIDE] + data[1+1*DCTSTRIDE];
1145   d11 = data[0+1*DCTSTRIDE] - data[1+1*DCTSTRIDE];
1146 
1147   data[0+0*DCTSTRIDE]= (d00 + d10)>>3;
1148   data[1+0*DCTSTRIDE]= (d01 + d11)>>3;
1149   data[0+1*DCTSTRIDE]= (d00 - d10)>>3;
1150   data[1+1*DCTSTRIDE]= (d01 - d11)>>3;
1151 }
1152 
ff_j_rev_dct1(DCTBLOCK data)1153 void ff_j_rev_dct1(DCTBLOCK data){
1154   data[0] = (data[0] + 4)>>3;
1155 }
1156 
1157 #undef FIX
1158 #undef CONST_BITS
1159 
ff_jref_idct_put(uint8_t * dest,ptrdiff_t line_size,int16_t * block)1160 void ff_jref_idct_put(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
1161 {
1162     ff_j_rev_dct(block);
1163     ff_put_pixels_clamped_c(block, dest, line_size);
1164 }
1165 
ff_jref_idct_add(uint8_t * dest,ptrdiff_t line_size,int16_t * block)1166 void ff_jref_idct_add(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
1167 {
1168     ff_j_rev_dct(block);
1169     ff_add_pixels_clamped_c(block, dest, line_size);
1170 }
1171