comparison src/wma/libffwma/simple_idct.c @ 878:aa044d71838f trunk

[svn] Goodbye ffmpeg, welcome back WMA.
author chainsaw
date Mon, 19 Mar 2007 13:15:52 -0700
parents
children
comparison
equal deleted inserted replaced
877:ecf3cb97dba1 878:aa044d71838f
1 /*
2 * Simple IDCT
3 *
4 * Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21 /**
22 * @file simple_idct.c
23 * simpleidct in C.
24 */
25
26 /*
27 based upon some outcommented c code from mpeg2dec (idct_mmx.c
28 written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>)
29 */
30 #include "avcodec.h"
31 #include "dsputil.h"
32 #include "simple_idct.h"
33
34 #if 0
35 #define W1 2841 /* 2048*sqrt (2)*cos (1*pi/16) */
36 #define W2 2676 /* 2048*sqrt (2)*cos (2*pi/16) */
37 #define W3 2408 /* 2048*sqrt (2)*cos (3*pi/16) */
38 #define W4 2048 /* 2048*sqrt (2)*cos (4*pi/16) */
39 #define W5 1609 /* 2048*sqrt (2)*cos (5*pi/16) */
40 #define W6 1108 /* 2048*sqrt (2)*cos (6*pi/16) */
41 #define W7 565 /* 2048*sqrt (2)*cos (7*pi/16) */
42 #define ROW_SHIFT 8
43 #define COL_SHIFT 17
44 #else
45 #define W1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
46 #define W2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
47 #define W3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
48 #define W4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
49 #define W5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
50 #define W6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
51 #define W7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
52 #define ROW_SHIFT 11
53 #define COL_SHIFT 20 // 6
54 #endif
55
56 #if defined(ARCH_POWERPC_405)
57
58 /* signed 16x16 -> 32 multiply add accumulate */
59 #define MAC16(rt, ra, rb) \
60 asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
61
62 /* signed 16x16 -> 32 multiply */
63 #define MUL16(rt, ra, rb) \
64 asm ("mullhw %0, %1, %2" : "=r" (rt) : "r" (ra), "r" (rb));
65
66 #else
67
68 /* signed 16x16 -> 32 multiply add accumulate */
69 #define MAC16(rt, ra, rb) rt += (ra) * (rb)
70
71 /* signed 16x16 -> 32 multiply */
72 #define MUL16(rt, ra, rb) rt = (ra) * (rb)
73
74 #endif
75
76 static inline void idctRowCondDC (DCTELEM * row)
77 {
78 int a0, a1, a2, a3, b0, b1, b2, b3;
79 #ifdef FAST_64BIT
80 uint64_t temp;
81 #else
82 uint32_t temp;
83 #endif
84
85 #ifdef FAST_64BIT
86 #ifdef WORDS_BIGENDIAN
87 #define ROW0_MASK 0xffff000000000000LL
88 #else
89 #define ROW0_MASK 0xffffLL
90 #endif
91 if(sizeof(DCTELEM)==2){
92 if ( ((((uint64_t *)row)[0] & ~ROW0_MASK) |
93 ((uint64_t *)row)[1]) == 0) {
94 temp = (row[0] << 3) & 0xffff;
95 temp += temp << 16;
96 temp += temp << 32;
97 ((uint64_t *)row)[0] = temp;
98 ((uint64_t *)row)[1] = temp;
99 return;
100 }
101 }else{
102 if (!(row[1]|row[2]|row[3]|row[4]|row[5]|row[6]|row[7])) {
103 row[0]=row[1]=row[2]=row[3]=row[4]=row[5]=row[6]=row[7]= row[0] << 3;
104 return;
105 }
106 }
107 #else
108 if(sizeof(DCTELEM)==2){
109 if (!(((uint32_t*)row)[1] |
110 ((uint32_t*)row)[2] |
111 ((uint32_t*)row)[3] |
112 row[1])) {
113 temp = (row[0] << 3) & 0xffff;
114 temp += temp << 16;
115 ((uint32_t*)row)[0]=((uint32_t*)row)[1] =
116 ((uint32_t*)row)[2]=((uint32_t*)row)[3] = temp;
117 return;
118 }
119 }else{
120 if (!(row[1]|row[2]|row[3]|row[4]|row[5]|row[6]|row[7])) {
121 row[0]=row[1]=row[2]=row[3]=row[4]=row[5]=row[6]=row[7]= row[0] << 3;
122 return;
123 }
124 }
125 #endif
126
127 a0 = (W4 * row[0]) + (1 << (ROW_SHIFT - 1));
128 a1 = a0;
129 a2 = a0;
130 a3 = a0;
131
132 /* no need to optimize : gcc does it */
133 a0 += W2 * row[2];
134 a1 += W6 * row[2];
135 a2 -= W6 * row[2];
136 a3 -= W2 * row[2];
137
138 MUL16(b0, W1, row[1]);
139 MAC16(b0, W3, row[3]);
140 MUL16(b1, W3, row[1]);
141 MAC16(b1, -W7, row[3]);
142 MUL16(b2, W5, row[1]);
143 MAC16(b2, -W1, row[3]);
144 MUL16(b3, W7, row[1]);
145 MAC16(b3, -W5, row[3]);
146
147 #ifdef FAST_64BIT
148 temp = ((uint64_t*)row)[1];
149 #else
150 temp = ((uint32_t*)row)[2] | ((uint32_t*)row)[3];
151 #endif
152 if (temp != 0) {
153 a0 += W4*row[4] + W6*row[6];
154 a1 += - W4*row[4] - W2*row[6];
155 a2 += - W4*row[4] + W2*row[6];
156 a3 += W4*row[4] - W6*row[6];
157
158 MAC16(b0, W5, row[5]);
159 MAC16(b0, W7, row[7]);
160
161 MAC16(b1, -W1, row[5]);
162 MAC16(b1, -W5, row[7]);
163
164 MAC16(b2, W7, row[5]);
165 MAC16(b2, W3, row[7]);
166
167 MAC16(b3, W3, row[5]);
168 MAC16(b3, -W1, row[7]);
169 }
170
171 row[0] = (a0 + b0) >> ROW_SHIFT;
172 row[7] = (a0 - b0) >> ROW_SHIFT;
173 row[1] = (a1 + b1) >> ROW_SHIFT;
174 row[6] = (a1 - b1) >> ROW_SHIFT;
175 row[2] = (a2 + b2) >> ROW_SHIFT;
176 row[5] = (a2 - b2) >> ROW_SHIFT;
177 row[3] = (a3 + b3) >> ROW_SHIFT;
178 row[4] = (a3 - b3) >> ROW_SHIFT;
179 }
180
181 static inline void idctSparseColPut (uint8_t *dest, int line_size,
182 DCTELEM * col)
183 {
184 int a0, a1, a2, a3, b0, b1, b2, b3;
185 uint8_t *cm = cropTbl + MAX_NEG_CROP;
186
187 /* XXX: I did that only to give same values as previous code */
188 a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4));
189 a1 = a0;
190 a2 = a0;
191 a3 = a0;
192
193 a0 += + W2*col[8*2];
194 a1 += + W6*col[8*2];
195 a2 += - W6*col[8*2];
196 a3 += - W2*col[8*2];
197
198 MUL16(b0, W1, col[8*1]);
199 MUL16(b1, W3, col[8*1]);
200 MUL16(b2, W5, col[8*1]);
201 MUL16(b3, W7, col[8*1]);
202
203 MAC16(b0, + W3, col[8*3]);
204 MAC16(b1, - W7, col[8*3]);
205 MAC16(b2, - W1, col[8*3]);
206 MAC16(b3, - W5, col[8*3]);
207
208 if(col[8*4]){
209 a0 += + W4*col[8*4];
210 a1 += - W4*col[8*4];
211 a2 += - W4*col[8*4];
212 a3 += + W4*col[8*4];
213 }
214
215 if (col[8*5]) {
216 MAC16(b0, + W5, col[8*5]);
217 MAC16(b1, - W1, col[8*5]);
218 MAC16(b2, + W7, col[8*5]);
219 MAC16(b3, + W3, col[8*5]);
220 }
221
222 if(col[8*6]){
223 a0 += + W6*col[8*6];
224 a1 += - W2*col[8*6];
225 a2 += + W2*col[8*6];
226 a3 += - W6*col[8*6];
227 }
228
229 if (col[8*7]) {
230 MAC16(b0, + W7, col[8*7]);
231 MAC16(b1, - W5, col[8*7]);
232 MAC16(b2, + W3, col[8*7]);
233 MAC16(b3, - W1, col[8*7]);
234 }
235
236 dest[0] = cm[(a0 + b0) >> COL_SHIFT];
237 dest += line_size;
238 dest[0] = cm[(a1 + b1) >> COL_SHIFT];
239 dest += line_size;
240 dest[0] = cm[(a2 + b2) >> COL_SHIFT];
241 dest += line_size;
242 dest[0] = cm[(a3 + b3) >> COL_SHIFT];
243 dest += line_size;
244 dest[0] = cm[(a3 - b3) >> COL_SHIFT];
245 dest += line_size;
246 dest[0] = cm[(a2 - b2) >> COL_SHIFT];
247 dest += line_size;
248 dest[0] = cm[(a1 - b1) >> COL_SHIFT];
249 dest += line_size;
250 dest[0] = cm[(a0 - b0) >> COL_SHIFT];
251 }
252
253 static inline void idctSparseColAdd (uint8_t *dest, int line_size,
254 DCTELEM * col)
255 {
256 int a0, a1, a2, a3, b0, b1, b2, b3;
257 uint8_t *cm = cropTbl + MAX_NEG_CROP;
258
259 /* XXX: I did that only to give same values as previous code */
260 a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4));
261 a1 = a0;
262 a2 = a0;
263 a3 = a0;
264
265 a0 += + W2*col[8*2];
266 a1 += + W6*col[8*2];
267 a2 += - W6*col[8*2];
268 a3 += - W2*col[8*2];
269
270 MUL16(b0, W1, col[8*1]);
271 MUL16(b1, W3, col[8*1]);
272 MUL16(b2, W5, col[8*1]);
273 MUL16(b3, W7, col[8*1]);
274
275 MAC16(b0, + W3, col[8*3]);
276 MAC16(b1, - W7, col[8*3]);
277 MAC16(b2, - W1, col[8*3]);
278 MAC16(b3, - W5, col[8*3]);
279
280 if(col[8*4]){
281 a0 += + W4*col[8*4];
282 a1 += - W4*col[8*4];
283 a2 += - W4*col[8*4];
284 a3 += + W4*col[8*4];
285 }
286
287 if (col[8*5]) {
288 MAC16(b0, + W5, col[8*5]);
289 MAC16(b1, - W1, col[8*5]);
290 MAC16(b2, + W7, col[8*5]);
291 MAC16(b3, + W3, col[8*5]);
292 }
293
294 if(col[8*6]){
295 a0 += + W6*col[8*6];
296 a1 += - W2*col[8*6];
297 a2 += + W2*col[8*6];
298 a3 += - W6*col[8*6];
299 }
300
301 if (col[8*7]) {
302 MAC16(b0, + W7, col[8*7]);
303 MAC16(b1, - W5, col[8*7]);
304 MAC16(b2, + W3, col[8*7]);
305 MAC16(b3, - W1, col[8*7]);
306 }
307
308 dest[0] = cm[dest[0] + ((a0 + b0) >> COL_SHIFT)];
309 dest += line_size;
310 dest[0] = cm[dest[0] + ((a1 + b1) >> COL_SHIFT)];
311 dest += line_size;
312 dest[0] = cm[dest[0] + ((a2 + b2) >> COL_SHIFT)];
313 dest += line_size;
314 dest[0] = cm[dest[0] + ((a3 + b3) >> COL_SHIFT)];
315 dest += line_size;
316 dest[0] = cm[dest[0] + ((a3 - b3) >> COL_SHIFT)];
317 dest += line_size;
318 dest[0] = cm[dest[0] + ((a2 - b2) >> COL_SHIFT)];
319 dest += line_size;
320 dest[0] = cm[dest[0] + ((a1 - b1) >> COL_SHIFT)];
321 dest += line_size;
322 dest[0] = cm[dest[0] + ((a0 - b0) >> COL_SHIFT)];
323 }
324
325 static inline void idctSparseCol (DCTELEM * col)
326 {
327 int a0, a1, a2, a3, b0, b1, b2, b3;
328
329 /* XXX: I did that only to give same values as previous code */
330 a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4));
331 a1 = a0;
332 a2 = a0;
333 a3 = a0;
334
335 a0 += + W2*col[8*2];
336 a1 += + W6*col[8*2];
337 a2 += - W6*col[8*2];
338 a3 += - W2*col[8*2];
339
340 MUL16(b0, W1, col[8*1]);
341 MUL16(b1, W3, col[8*1]);
342 MUL16(b2, W5, col[8*1]);
343 MUL16(b3, W7, col[8*1]);
344
345 MAC16(b0, + W3, col[8*3]);
346 MAC16(b1, - W7, col[8*3]);
347 MAC16(b2, - W1, col[8*3]);
348 MAC16(b3, - W5, col[8*3]);
349
350 if(col[8*4]){
351 a0 += + W4*col[8*4];
352 a1 += - W4*col[8*4];
353 a2 += - W4*col[8*4];
354 a3 += + W4*col[8*4];
355 }
356
357 if (col[8*5]) {
358 MAC16(b0, + W5, col[8*5]);
359 MAC16(b1, - W1, col[8*5]);
360 MAC16(b2, + W7, col[8*5]);
361 MAC16(b3, + W3, col[8*5]);
362 }
363
364 if(col[8*6]){
365 a0 += + W6*col[8*6];
366 a1 += - W2*col[8*6];
367 a2 += + W2*col[8*6];
368 a3 += - W6*col[8*6];
369 }
370
371 if (col[8*7]) {
372 MAC16(b0, + W7, col[8*7]);
373 MAC16(b1, - W5, col[8*7]);
374 MAC16(b2, + W3, col[8*7]);
375 MAC16(b3, - W1, col[8*7]);
376 }
377
378 col[0 ] = ((a0 + b0) >> COL_SHIFT);
379 col[8 ] = ((a1 + b1) >> COL_SHIFT);
380 col[16] = ((a2 + b2) >> COL_SHIFT);
381 col[24] = ((a3 + b3) >> COL_SHIFT);
382 col[32] = ((a3 - b3) >> COL_SHIFT);
383 col[40] = ((a2 - b2) >> COL_SHIFT);
384 col[48] = ((a1 - b1) >> COL_SHIFT);
385 col[56] = ((a0 - b0) >> COL_SHIFT);
386 }
387
388 void simple_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
389 {
390 int i;
391 for(i=0; i<8; i++)
392 idctRowCondDC(block + i*8);
393
394 for(i=0; i<8; i++)
395 idctSparseColPut(dest + i, line_size, block + i);
396 }
397
398 void simple_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
399 {
400 int i;
401 for(i=0; i<8; i++)
402 idctRowCondDC(block + i*8);
403
404 for(i=0; i<8; i++)
405 idctSparseColAdd(dest + i, line_size, block + i);
406 }
407
408 void simple_idct(DCTELEM *block)
409 {
410 int i;
411 for(i=0; i<8; i++)
412 idctRowCondDC(block + i*8);
413
414 for(i=0; i<8; i++)
415 idctSparseCol(block + i);
416 }
417
418 /* 2x4x8 idct */
419
420 #define CN_SHIFT 12
421 #define C_FIX(x) ((int)((x) * (1 << CN_SHIFT) + 0.5))
422 #define C1 C_FIX(0.6532814824)
423 #define C2 C_FIX(0.2705980501)
424
425 /* row idct is multiple by 16 * sqrt(2.0), col idct4 is normalized,
426 and the butterfly must be multiplied by 0.5 * sqrt(2.0) */
427 #define C_SHIFT (4+1+12)
428
429 static inline void idct4col(uint8_t *dest, int line_size, const DCTELEM *col)
430 {
431 int c0, c1, c2, c3, a0, a1, a2, a3;
432 const uint8_t *cm = cropTbl + MAX_NEG_CROP;
433
434 a0 = col[8*0];
435 a1 = col[8*2];
436 a2 = col[8*4];
437 a3 = col[8*6];
438 c0 = ((a0 + a2) << (CN_SHIFT - 1)) + (1 << (C_SHIFT - 1));
439 c2 = ((a0 - a2) << (CN_SHIFT - 1)) + (1 << (C_SHIFT - 1));
440 c1 = a1 * C1 + a3 * C2;
441 c3 = a1 * C2 - a3 * C1;
442 dest[0] = cm[(c0 + c1) >> C_SHIFT];
443 dest += line_size;
444 dest[0] = cm[(c2 + c3) >> C_SHIFT];
445 dest += line_size;
446 dest[0] = cm[(c2 - c3) >> C_SHIFT];
447 dest += line_size;
448 dest[0] = cm[(c0 - c1) >> C_SHIFT];
449 }
450
451 #define BF(k) \
452 {\
453 int a0, a1;\
454 a0 = ptr[k];\
455 a1 = ptr[8 + k];\
456 ptr[k] = a0 + a1;\
457 ptr[8 + k] = a0 - a1;\
458 }
459
460 /* only used by DV codec. The input must be interlaced. 128 is added
461 to the pixels before clamping to avoid systematic error
462 (1024*sqrt(2)) offset would be needed otherwise. */
463 /* XXX: I think a 1.0/sqrt(2) normalization should be needed to
464 compensate the extra butterfly stage - I don't have the full DV
465 specification */
466 void simple_idct248_put(uint8_t *dest, int line_size, DCTELEM *block)
467 {
468 int i;
469 DCTELEM *ptr;
470
471 /* butterfly */
472 ptr = block;
473 for(i=0;i<4;i++) {
474 BF(0);
475 BF(1);
476 BF(2);
477 BF(3);
478 BF(4);
479 BF(5);
480 BF(6);
481 BF(7);
482 ptr += 2 * 8;
483 }
484
485 /* IDCT8 on each line */
486 for(i=0; i<8; i++) {
487 idctRowCondDC(block + i*8);
488 }
489
490 /* IDCT4 and store */
491 for(i=0;i<8;i++) {
492 idct4col(dest + i, 2 * line_size, block + i);
493 idct4col(dest + line_size + i, 2 * line_size, block + 8 + i);
494 }
495 }
496
497 /* 8x4 & 4x8 WMV2 IDCT */
498 #undef CN_SHIFT
499 #undef C_SHIFT
500 #undef C_FIX
501 #undef C1
502 #undef C2
503 #define CN_SHIFT 12
504 #define C_FIX(x) ((int)((x) * 1.414213562 * (1 << CN_SHIFT) + 0.5))
505 #define C1 C_FIX(0.6532814824)
506 #define C2 C_FIX(0.2705980501)
507 #define C3 C_FIX(0.5)
508 #define C_SHIFT (4+1+12)
509 static inline void idct4col_add(uint8_t *dest, int line_size, const DCTELEM *col)
510 {
511 int c0, c1, c2, c3, a0, a1, a2, a3;
512 const uint8_t *cm = cropTbl + MAX_NEG_CROP;
513
514 a0 = col[8*0];
515 a1 = col[8*1];
516 a2 = col[8*2];
517 a3 = col[8*3];
518 c0 = (a0 + a2)*C3 + (1 << (C_SHIFT - 1));
519 c2 = (a0 - a2)*C3 + (1 << (C_SHIFT - 1));
520 c1 = a1 * C1 + a3 * C2;
521 c3 = a1 * C2 - a3 * C1;
522 dest[0] = cm[dest[0] + ((c0 + c1) >> C_SHIFT)];
523 dest += line_size;
524 dest[0] = cm[dest[0] + ((c2 + c3) >> C_SHIFT)];
525 dest += line_size;
526 dest[0] = cm[dest[0] + ((c2 - c3) >> C_SHIFT)];
527 dest += line_size;
528 dest[0] = cm[dest[0] + ((c0 - c1) >> C_SHIFT)];
529 }
530
531 #define RN_SHIFT 15
532 #define R_FIX(x) ((int)((x) * 1.414213562 * (1 << RN_SHIFT) + 0.5))
533 #define R1 R_FIX(0.6532814824)
534 #define R2 R_FIX(0.2705980501)
535 #define R3 R_FIX(0.5)
536 #define R_SHIFT 11
537 static inline void idct4row(DCTELEM *row)
538 {
539 int c0, c1, c2, c3, a0, a1, a2, a3;
540 //const uint8_t *cm = cropTbl + MAX_NEG_CROP;
541
542 a0 = row[0];
543 a1 = row[1];
544 a2 = row[2];
545 a3 = row[3];
546 c0 = (a0 + a2)*R3 + (1 << (R_SHIFT - 1));
547 c2 = (a0 - a2)*R3 + (1 << (R_SHIFT - 1));
548 c1 = a1 * R1 + a3 * R2;
549 c3 = a1 * R2 - a3 * R1;
550 row[0]= (c0 + c1) >> R_SHIFT;
551 row[1]= (c2 + c3) >> R_SHIFT;
552 row[2]= (c2 - c3) >> R_SHIFT;
553 row[3]= (c0 - c1) >> R_SHIFT;
554 }
555
556 void simple_idct84_add(uint8_t *dest, int line_size, DCTELEM *block)
557 {
558 int i;
559
560 /* IDCT8 on each line */
561 for(i=0; i<4; i++) {
562 idctRowCondDC(block + i*8);
563 }
564
565 /* IDCT4 and store */
566 for(i=0;i<8;i++) {
567 idct4col_add(dest + i, line_size, block + i);
568 }
569 }
570
571 void simple_idct48_add(uint8_t *dest, int line_size, DCTELEM *block)
572 {
573 int i;
574
575 /* IDCT4 on each line */
576 for(i=0; i<8; i++) {
577 idct4row(block + i*8);
578 }
579
580 /* IDCT8 and store */
581 for(i=0; i<4; i++){
582 idctSparseColAdd(dest + i, line_size, block + i);
583 }
584 }
585