Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
[ffmpeg.git] / libavcodec / simple_idct.c
1 /*
2     Copyright (C) 2001 Michael Niedermayer (michaelni@gmx.at)
3
4     This program is free software; you can redistribute it and/or modify
5     it under the terms of the GNU General Public License as published by
6     the Free Software Foundation; either version 2 of the License, or
7     (at your option) any later version.
8
9     This program is distributed in the hope that it will be useful,
10     but WITHOUT ANY WARRANTY; without even the implied warranty of
11     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12     GNU General Public License for more details.
13
14     You should have received a copy of the GNU General Public License
15     along with this program; if not, write to the Free Software
16     Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
17 */
18
19 /*
20   based upon some outcommented c code from mpeg2dec (idct_mmx.c written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>)
21 */
22
23 #include <inttypes.h>
24
25 #include "simple_idct.h"
26 #include "../config.h"
27
28 #if 0
29 #define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */
30 #define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */
31 #define W3 2408 /* 2048*sqrt (2)*cos (3*pi/16) */
32 #define W4 2048 /* 2048*sqrt (2)*cos (4*pi/16) */
33 #define W5 1609 /* 2048*sqrt (2)*cos (5*pi/16) */
34 #define W6 1108 /* 2048*sqrt (2)*cos (6*pi/16) */
35 #define W7 565  /* 2048*sqrt (2)*cos (7*pi/16) */
36 #define ROW_SHIFT 8
37 #define COL_SHIFT 17
38 #else
39 #define W1  22725  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
40 #define W2  21407  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
41 #define W3  19266  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
42 #define W4  16384  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
43 #define W5  12873  //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
44 #define W6  8867   //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
45 #define W7  4520   //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
46 #define ROW_SHIFT 11
47 #define COL_SHIFT 20 // 6
48 #endif
49
50 /* 8x8 Matrix used to do a trivial (slow) 8 point IDCT */
51 static int coeff[64]={
52         W4, W4, W4, W4, W4, W4, W4, W4,
53         W1, W3, W5, W7,-W7,-W5,-W3,-W1,
54         W2, W6,-W6,-W2,-W2,-W6, W6, W2,
55         W3,-W7,-W1,-W5, W5, W1, W7,-W3,
56         W4,-W4,-W4, W4, W4,-W4,-W4, W4,
57         W5,-W1, W7, W3,-W3,-W7, W1,-W5,
58         W6,-W2, W2,-W6,-W6, W2,-W2, W6,
59         W7,-W5, W3,-W1, W1,-W3, W5,-W7
60 };
61
62 static int inline idctRowCondZ (int16_t * row)
63 {
64         int a0, a1, a2, a3, b0, b1, b2, b3;
65
66         if( !( ((uint32_t*)row)[0]|((uint32_t*)row)[1] |((uint32_t*)row)[2] |((uint32_t*)row)[3])) {
67 /*              row[0] = row[1] = row[2] = row[3] = row[4] =
68                         row[5] = row[6] = row[7] = 0;*/
69                 return 0;
70         }
71
72         if(!( ((uint32_t*)row)[2] |((uint32_t*)row)[3] )){
73                 a0 = W4*row[0] + W2*row[2] + (1<<(ROW_SHIFT-1));
74                 a1 = W4*row[0] + W6*row[2] + (1<<(ROW_SHIFT-1));
75                 a2 = W4*row[0] - W6*row[2] + (1<<(ROW_SHIFT-1));
76                 a3 = W4*row[0] - W2*row[2] + (1<<(ROW_SHIFT-1));
77
78                 b0 = W1*row[1] + W3*row[3];
79                 b1 = W3*row[1] - W7*row[3];
80                 b2 = W5*row[1] - W1*row[3];
81                 b3 = W7*row[1] - W5*row[3];
82         }else{
83                 a0 = W4*row[0] + W2*row[2] + W4*row[4] + W6*row[6] + (1<<(ROW_SHIFT-1));
84                 a1 = W4*row[0] + W6*row[2] - W4*row[4] - W2*row[6] + (1<<(ROW_SHIFT-1));
85                 a2 = W4*row[0] - W6*row[2] - W4*row[4] + W2*row[6] + (1<<(ROW_SHIFT-1));
86                 a3 = W4*row[0] - W2*row[2] + W4*row[4] - W6*row[6] + (1<<(ROW_SHIFT-1));
87
88                 b0 = W1*row[1] + W3*row[3] + W5*row[5] + W7*row[7];
89                 b1 = W3*row[1] - W7*row[3] - W1*row[5] - W5*row[7];
90                 b2 = W5*row[1] - W1*row[3] + W7*row[5] + W3*row[7];
91                 b3 = W7*row[1] - W5*row[3] + W3*row[5] - W1*row[7];
92         }
93
94         row[0] = (a0 + b0) >> ROW_SHIFT;
95         row[1] = (a1 + b1) >> ROW_SHIFT;
96         row[2] = (a2 + b2) >> ROW_SHIFT;
97         row[3] = (a3 + b3) >> ROW_SHIFT;
98         row[4] = (a3 - b3) >> ROW_SHIFT;
99         row[5] = (a2 - b2) >> ROW_SHIFT;
100         row[6] = (a1 - b1) >> ROW_SHIFT;
101         row[7] = (a0 - b0) >> ROW_SHIFT;
102         
103         return 1;
104 }
105
106 #ifdef ARCH_ALPHA
107 static int inline idctRowCondDC(int16_t *row)
108 {
109         int_fast32_t a0, a1, a2, a3, b0, b1, b2, b3;
110         uint64_t *lrow = (uint64_t *) row;
111
112         if (lrow[1] == 0) {
113                 if (lrow[0] == 0)
114                         return 0;
115                 if ((lrow[0] & ~0xffffULL) == 0) {
116                         uint64_t v;
117
118                         a0 = W4 * row[0];
119                         a0 += 1 << (ROW_SHIFT - 1);
120                         a0 >>= ROW_SHIFT;
121                         v = (uint16_t) a0;
122                         v += v << 16;
123                         v += v << 32;
124                         lrow[0] = v;
125                         lrow[1] = v;
126
127                         return 1;
128                 }
129         }
130
131         a0 = W4 * row[0];
132         a1 = W4 * row[0];
133         a2 = W4 * row[0];
134         a3 = W4 * row[0];
135
136         if (row[2]) {
137                 a0 += W2 * row[2];
138                 a1 += W6 * row[2];
139                 a2 -= W6 * row[2];
140                 a3 -= W2 * row[2];
141         }
142
143         if (row[4]) {
144                 a0 += W4 * row[4];
145                 a1 -= W4 * row[4];
146                 a2 -= W4 * row[4];
147                 a3 += W4 * row[4];
148         }
149
150         if (row[6]) {
151                 a0 += W6 * row[6];
152                 a1 -= W2 * row[6];
153                 a2 += W2 * row[6];
154                 a3 -= W6 * row[6];
155         }
156
157         a0 += 1 << (ROW_SHIFT - 1);
158         a1 += 1 << (ROW_SHIFT - 1);
159         a2 += 1 << (ROW_SHIFT - 1);
160         a3 += 1 << (ROW_SHIFT - 1);
161
162         if (row[1]) {
163                 b0 = W1 * row[1];
164                 b1 = W3 * row[1];
165                 b2 = W5 * row[1];
166                 b3 = W7 * row[1];
167         } else {
168                 b0 = 0;
169                 b1 = 0;
170                 b2 = 0;
171                 b3 = 0;
172         }
173
174         if (row[3]) {
175                 b0 += W3 * row[3];
176                 b1 -= W7 * row[3];
177                 b2 -= W1 * row[3];
178                 b3 -= W5 * row[3];
179         }
180
181         if (row[5]) {
182                 b0 += W5 * row[5];
183                 b1 -= W1 * row[5];
184                 b2 += W7 * row[5];
185                 b3 += W3 * row[5];
186         }
187
188         if (row[7]) {
189                 b0 += W7 * row[7];
190                 b1 -= W5 * row[7];
191                 b2 += W3 * row[7];
192                 b3 -= W1 * row[7];
193         }
194
195         row[0] = (a0 + b0) >> ROW_SHIFT;
196         row[1] = (a1 + b1) >> ROW_SHIFT;
197         row[2] = (a2 + b2) >> ROW_SHIFT;
198         row[3] = (a3 + b3) >> ROW_SHIFT;
199         row[4] = (a3 - b3) >> ROW_SHIFT;
200         row[5] = (a2 - b2) >> ROW_SHIFT;
201         row[6] = (a1 - b1) >> ROW_SHIFT;
202         row[7] = (a0 - b0) >> ROW_SHIFT;
203
204         return 1;
205 }
206 #else  /* not ARCH_ALPHA */
207 static int inline idctRowCondDC (int16_t * row)
208 {
209         int a0, a1, a2, a3, b0, b1, b2, b3;
210
211         if( !( ((uint32_t*)row)[1] |((uint32_t*)row)[2] |((uint32_t*)row)[3]| row[1])) {
212 //              row[0] = row[1] = row[2] = row[3] = row[4] = row[5] = row[6] = row[7] = row[0]<<3;
213                 uint16_t temp= row[0]<<3;
214                 ((uint32_t*)row)[0]=((uint32_t*)row)[1]=
215                 ((uint32_t*)row)[2]=((uint32_t*)row)[3]= temp + (temp<<16);
216                 return 0;
217         }
218
219         if(!( ((uint32_t*)row)[2] |((uint32_t*)row)[3] )){
220                 a0 = W4*row[0] + W2*row[2] + (1<<(ROW_SHIFT-1));
221                 a1 = W4*row[0] + W6*row[2] + (1<<(ROW_SHIFT-1));
222                 a2 = W4*row[0] - W6*row[2] + (1<<(ROW_SHIFT-1));
223                 a3 = W4*row[0] - W2*row[2] + (1<<(ROW_SHIFT-1));
224
225                 b0 = W1*row[1] + W3*row[3];
226                 b1 = W3*row[1] - W7*row[3];
227                 b2 = W5*row[1] - W1*row[3];
228                 b3 = W7*row[1] - W5*row[3];
229         }else{
230                 a0 = W4*row[0] + W2*row[2] + W4*row[4] + W6*row[6] + (1<<(ROW_SHIFT-1));
231                 a1 = W4*row[0] + W6*row[2] - W4*row[4] - W2*row[6] + (1<<(ROW_SHIFT-1));
232                 a2 = W4*row[0] - W6*row[2] - W4*row[4] + W2*row[6] + (1<<(ROW_SHIFT-1));
233                 a3 = W4*row[0] - W2*row[2] + W4*row[4] - W6*row[6] + (1<<(ROW_SHIFT-1));
234
235                 b0 = W1*row[1] + W3*row[3] + W5*row[5] + W7*row[7];
236                 b1 = W3*row[1] - W7*row[3] - W1*row[5] - W5*row[7];
237                 b2 = W5*row[1] - W1*row[3] + W7*row[5] + W3*row[7];
238                 b3 = W7*row[1] - W5*row[3] + W3*row[5] - W1*row[7];
239         }
240
241         row[0] = (a0 + b0) >> ROW_SHIFT;
242         row[7] = (a0 - b0) >> ROW_SHIFT;
243         row[1] = (a1 + b1) >> ROW_SHIFT;
244         row[6] = (a1 - b1) >> ROW_SHIFT;
245         row[2] = (a2 + b2) >> ROW_SHIFT;
246         row[5] = (a2 - b2) >> ROW_SHIFT;
247         row[3] = (a3 + b3) >> ROW_SHIFT;
248         row[4] = (a3 - b3) >> ROW_SHIFT;
249         
250         return 1;
251 }
252 #endif /* not ARCH_ALPHA */
253
254 static void inline idctCol (int16_t * col)
255 {
256
257 /*
258         if( !(col[8*1] | col[8*2] |col[8*3] |col[8*4] |col[8*5] |col[8*6] | col[8*7])) {
259                 col[8*0] = col[8*1] = col[8*2] = col[8*3] = col[8*4] =
260                         col[8*5] = col[8*6] = col[8*7] = col[8*0]<<3;
261                 return;
262         }*/
263
264         int a0, a1, a2, a3, b0, b1, b2, b3;
265         col[0] += (1<<(COL_SHIFT-1))/W4;
266         a0 = W4*col[8*0] + W2*col[8*2] + W4*col[8*4] + W6*col[8*6];
267         a1 = W4*col[8*0] + W6*col[8*2] - W4*col[8*4] - W2*col[8*6];
268         a2 = W4*col[8*0] - W6*col[8*2] - W4*col[8*4] + W2*col[8*6];
269         a3 = W4*col[8*0] - W2*col[8*2] + W4*col[8*4] - W6*col[8*6];
270
271         b0 = W1*col[8*1] + W3*col[8*3] + W5*col[8*5] + W7*col[8*7];
272         b1 = W3*col[8*1] - W7*col[8*3] - W1*col[8*5] - W5*col[8*7];
273         b2 = W5*col[8*1] - W1*col[8*3] + W7*col[8*5] + W3*col[8*7];
274         b3 = W7*col[8*1] - W5*col[8*3] + W3*col[8*5] - W1*col[8*7];
275
276         col[8*0] = (a0 + b0) >> COL_SHIFT;
277         col[8*7] = (a0 - b0) >> COL_SHIFT;
278         col[8*1] = (a1 + b1) >> COL_SHIFT;
279         col[8*6] = (a1 - b1) >> COL_SHIFT;
280         col[8*2] = (a2 + b2) >> COL_SHIFT;
281         col[8*5] = (a2 - b2) >> COL_SHIFT;
282         col[8*3] = (a3 + b3) >> COL_SHIFT;
283         col[8*4] = (a3 - b3) >> COL_SHIFT;
284 }
285
286 static void inline idctSparseCol (int16_t * col)
287 {
288         int a0, a1, a2, a3, b0, b1, b2, b3;
289         col[0] += (1<<(COL_SHIFT-1))/W4;
290         a0 = W4*col[8*0];
291         a1 = W4*col[8*0];
292         a2 = W4*col[8*0];
293         a3 = W4*col[8*0];
294
295         if(col[8*2]){
296                 a0 +=  + W2*col[8*2];
297                 a1 +=  + W6*col[8*2];
298                 a2 +=  - W6*col[8*2];
299                 a3 +=  - W2*col[8*2];
300         }
301
302         if(col[8*4]){
303                 a0 += + W4*col[8*4];
304                 a1 += - W4*col[8*4];
305                 a2 += - W4*col[8*4];
306                 a3 += + W4*col[8*4];
307         }
308
309         if(col[8*6]){
310                 a0 += + W6*col[8*6];
311                 a1 += - W2*col[8*6];
312                 a2 += + W2*col[8*6];
313                 a3 += - W6*col[8*6];
314         }
315
316         if(col[8*1]){
317                 b0 = W1*col[8*1];
318                 b1 = W3*col[8*1];
319                 b2 = W5*col[8*1];
320                 b3 = W7*col[8*1];
321         }else{
322                 b0 = 
323                 b1 = 
324                 b2 = 
325                 b3 = 0;
326         }
327
328         if(col[8*3]){
329                 b0 += + W3*col[8*3];
330                 b1 += - W7*col[8*3];
331                 b2 += - W1*col[8*3];
332                 b3 += - W5*col[8*3];
333         }
334
335         if(col[8*5]){
336                 b0 += + W5*col[8*5];
337                 b1 += - W1*col[8*5];
338                 b2 += + W7*col[8*5];
339                 b3 += + W3*col[8*5];
340         }
341
342         if(col[8*7]){
343                 b0 += + W7*col[8*7];
344                 b1 += - W5*col[8*7];
345                 b2 += + W3*col[8*7];
346                 b3 += - W1*col[8*7];
347         }
348
349 #ifndef ARCH_ALPHA
350         if(!(b0|b1|b2|b3)){
351                 col[8*0] = (a0) >> COL_SHIFT;
352                 col[8*7] = (a0) >> COL_SHIFT;
353                 col[8*1] = (a1) >> COL_SHIFT;
354                 col[8*6] = (a1) >> COL_SHIFT;
355                 col[8*2] = (a2) >> COL_SHIFT;
356                 col[8*5] = (a2) >> COL_SHIFT;
357                 col[8*3] = (a3) >> COL_SHIFT;
358                 col[8*4] = (a3) >> COL_SHIFT;
359         }else{
360 #endif
361                 col[8*0] = (a0 + b0) >> COL_SHIFT;
362                 col[8*7] = (a0 - b0) >> COL_SHIFT;
363                 col[8*1] = (a1 + b1) >> COL_SHIFT;
364                 col[8*6] = (a1 - b1) >> COL_SHIFT;
365                 col[8*2] = (a2 + b2) >> COL_SHIFT;
366                 col[8*5] = (a2 - b2) >> COL_SHIFT;
367                 col[8*3] = (a3 + b3) >> COL_SHIFT;
368                 col[8*4] = (a3 - b3) >> COL_SHIFT;
369 #ifndef ARCH_ALPHA
370         }
371 #endif
372 }
373
374 static void inline idctSparse2Col (int16_t * col)
375 {
376         int a0, a1, a2, a3, b0, b1, b2, b3;
377         col[0] += (1<<(COL_SHIFT-1))/W4;
378         a0 = W4*col[8*0];
379         a1 = W4*col[8*0];
380         a2 = W4*col[8*0];
381         a3 = W4*col[8*0];
382
383         if(col[8*2]){
384                 a0 +=  + W2*col[8*2];
385                 a1 +=  + W6*col[8*2];
386                 a2 +=  - W6*col[8*2];
387                 a3 +=  - W2*col[8*2];
388         }
389
390         if(col[8*4]){
391                 a0 += + W4*col[8*4];
392                 a1 += - W4*col[8*4];
393                 a2 += - W4*col[8*4];
394                 a3 += + W4*col[8*4];
395         }
396
397         if(col[8*6]){
398                 a0 += + W6*col[8*6];
399                 a1 += - W2*col[8*6];
400                 a2 += + W2*col[8*6];
401                 a3 += - W6*col[8*6];
402         }
403
404         if(col[8*1] || 1){
405                 b0 = W1*col[8*1];
406                 b1 = W3*col[8*1];
407                 b2 = W5*col[8*1];
408                 b3 = W7*col[8*1];
409         }else{
410                 b0 = 
411                 b1 = 
412                 b2 = 
413                 b3 = 0;
414         }
415
416         if(col[8*3]){
417                 b0 += + W3*col[8*3];
418                 b1 += - W7*col[8*3];
419                 b2 += - W1*col[8*3];
420                 b3 += - W5*col[8*3];
421         }
422
423         if(col[8*5]){
424                 b0 += + W5*col[8*5];
425                 b1 += - W1*col[8*5];
426                 b2 += + W7*col[8*5];
427                 b3 += + W3*col[8*5];
428         }
429
430         if(col[8*7]){
431                 b0 += + W7*col[8*7];
432                 b1 += - W5*col[8*7];
433                 b2 += + W3*col[8*7];
434                 b3 += - W1*col[8*7];
435         }
436
437         col[8*0] = (a0 + b0) >> COL_SHIFT;
438         col[8*7] = (a0 - b0) >> COL_SHIFT;
439         col[8*1] = (a1 + b1) >> COL_SHIFT;
440         col[8*6] = (a1 - b1) >> COL_SHIFT;
441         col[8*2] = (a2 + b2) >> COL_SHIFT;
442         col[8*5] = (a2 - b2) >> COL_SHIFT;
443         col[8*3] = (a3 + b3) >> COL_SHIFT;
444         col[8*4] = (a3 - b3) >> COL_SHIFT;
445 }
446
447 #ifdef ARCH_ALPHA
448 /* If all rows but the first one are zero after row transformation,
449    all rows will be identical after column transformation.  */
450 static inline void idctCol2(int16_t *col)
451 {
452         int i;
453         uint64_t l, r;
454         uint64_t *lcol = (uint64_t *) col;
455
456         for (i = 0; i < 8; ++i) {
457                 int a0 = col[0] + (1 << (COL_SHIFT - 1)) / W4;
458
459                 a0 *= W4;
460                 col[0] = a0 >> COL_SHIFT;
461                 ++col;
462         }
463
464         l = lcol[0];
465         r = lcol[1];
466         lcol[ 2] = l; lcol[ 3] = r;
467         lcol[ 4] = l; lcol[ 5] = r;
468         lcol[ 6] = l; lcol[ 7] = r;
469         lcol[ 8] = l; lcol[ 9] = r;
470         lcol[10] = l; lcol[11] = r;
471         lcol[12] = l; lcol[13] = r;
472         lcol[14] = l; lcol[15] = r;
473 }
474 #endif
475
476 void simple_idct (short *block)
477 {
478
479         int i;
480         
481 #if 0
482         int nonZero[8];
483         int buffer[64];
484         int nNonZero=0;
485         
486         idctRowCondDC(block);
487         
488         for(i=1; i<8; i++)
489         {
490                 nonZero[nNonZero]=i;
491                 nNonZero+= idctRowCondZ(block + i*8);
492         }
493         
494         if(nNonZero==0)
495         {
496                 for(i=0; i<8; i++)
497                 {
498                         block[i   ]=
499                         block[i+8 ]=
500                         block[i+16]=
501                         block[i+24]=
502                         block[i+32]=
503                         block[i+40]=
504                         block[i+48]=
505                         block[i+56]= (W4*block[i] + (1<<(COL_SHIFT-1))) >> COL_SHIFT;
506                 }       
507         }
508         else if(nNonZero==1)
509         {
510                 int index= nonZero[0]*8;
511                 for(i=0; i<8; i++)
512                 {
513                         int bias= W4*block[i] + (1<<(COL_SHIFT-1));
514                         int c= block[i + index];
515                         block[i   ]= (c*coeff[index  ] + bias) >> COL_SHIFT;
516                         block[i+8 ]= (c*coeff[index+1] + bias) >> COL_SHIFT;
517                         block[i+16]= (c*coeff[index+2] + bias) >> COL_SHIFT;
518                         block[i+24]= (c*coeff[index+3] + bias) >> COL_SHIFT;
519                         block[i+32]= (c*coeff[index+4] + bias) >> COL_SHIFT;
520                         block[i+40]= (c*coeff[index+5] + bias) >> COL_SHIFT;
521                         block[i+48]= (c*coeff[index+6] + bias) >> COL_SHIFT;
522                         block[i+56]= (c*coeff[index+7] + bias) >> COL_SHIFT;
523                 }       
524         }
525 /*      else if(nNonZero==2)
526         {
527                 int index1= nonZero[0]*8;
528                 int index2= nonZero[1]*8;
529                 for(i=0; i<8; i++)
530                 {
531                         int bias= W4*block[i] + (1<<(COL_SHIFT-1));
532                         int c1= block[i + index1];
533                         int c2= block[i + index2];
534                         block[i   ]= (c1*coeff[index1  ] + c2*coeff[index2  ] + bias) >> COL_SHIFT;
535                         block[i+8 ]= (c1*coeff[index1+1] + c2*coeff[index2+1] + bias) >> COL_SHIFT;
536                         block[i+16]= (c1*coeff[index1+2] + c2*coeff[index2+2] + bias) >> COL_SHIFT;
537                         block[i+24]= (c1*coeff[index1+3] + c2*coeff[index2+3] + bias) >> COL_SHIFT;
538                         block[i+32]= (c1*coeff[index1+4] + c2*coeff[index2+4] + bias) >> COL_SHIFT;
539                         block[i+40]= (c1*coeff[index1+5] + c2*coeff[index2+5] + bias) >> COL_SHIFT;
540                         block[i+48]= (c1*coeff[index1+6] + c2*coeff[index2+6] + bias) >> COL_SHIFT;
541                         block[i+56]= (c1*coeff[index1+7] + c2*coeff[index2+7] + bias) >> COL_SHIFT;
542                 }       
543         }*/
544         else
545         {
546                 for(i=0; i<8; i++)
547                         idctSparse2Col(block + i);
548         }
549 #elif defined(ARCH_ALPHA)
550         int shortcut = 1;
551
552         for (i = 0; i < 8; i++) {
553                 int anynonzero = idctRowCondDC(block + 8 * i);
554                 if (i > 0 && anynonzero)
555                         shortcut = 0;
556         }
557
558         if (shortcut) {
559                 idctCol2(block);
560         } else {
561                 for (i = 0; i < 8; i++)
562                         idctSparseCol(block + i);
563         }
564 #else
565         for(i=0; i<8; i++)
566                 idctRowCondDC(block + i*8);
567         
568         for(i=0; i<8; i++)
569                 idctSparseCol(block + i);
570 #endif
571 }