4 * Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of Libav.
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
29 based upon some outcommented c code from mpeg2dec (idct_mmx.c
30 written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>)
35 #include "simple_idct.h"
37 #define W1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
38 #define W2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
39 #define W3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
40 #define W4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
41 #define W5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
42 #define W6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
43 #define W7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
45 #define COL_SHIFT 20 // 6
47 static inline void idctRowCondDC (DCTELEM * row)
49 int a0, a1, a2, a3, b0, b1, b2, b3;
58 #define ROW0_MASK 0xffff000000000000LL
60 #define ROW0_MASK 0xffffLL
62 if(sizeof(DCTELEM)==2){
63 if ( ((((uint64_t *)row)[0] & ~ROW0_MASK) |
64 ((uint64_t *)row)[1]) == 0) {
65 temp = (row[0] << 3) & 0xffff;
68 ((uint64_t *)row)[0] = temp;
69 ((uint64_t *)row)[1] = temp;
73 if (!(row[1]|row[2]|row[3]|row[4]|row[5]|row[6]|row[7])) {
74 row[0]=row[1]=row[2]=row[3]=row[4]=row[5]=row[6]=row[7]= row[0] << 3;
79 if(sizeof(DCTELEM)==2){
80 if (!(((uint32_t*)row)[1] |
84 temp = (row[0] << 3) & 0xffff;
86 ((uint32_t*)row)[0]=((uint32_t*)row)[1] =
87 ((uint32_t*)row)[2]=((uint32_t*)row)[3] = temp;
91 if (!(row[1]|row[2]|row[3]|row[4]|row[5]|row[6]|row[7])) {
92 row[0]=row[1]=row[2]=row[3]=row[4]=row[5]=row[6]=row[7]= row[0] << 3;
98 a0 = (W4 * row[0]) + (1 << (ROW_SHIFT - 1));
103 /* no need to optimize : gcc does it */
109 b0 = MUL16(W1, row[1]);
110 MAC16(b0, W3, row[3]);
111 b1 = MUL16(W3, row[1]);
112 MAC16(b1, -W7, row[3]);
113 b2 = MUL16(W5, row[1]);
114 MAC16(b2, -W1, row[3]);
115 b3 = MUL16(W7, row[1]);
116 MAC16(b3, -W5, row[3]);
119 temp = ((uint64_t*)row)[1];
121 temp = ((uint32_t*)row)[2] | ((uint32_t*)row)[3];
124 a0 += W4*row[4] + W6*row[6];
125 a1 += - W4*row[4] - W2*row[6];
126 a2 += - W4*row[4] + W2*row[6];
127 a3 += W4*row[4] - W6*row[6];
129 MAC16(b0, W5, row[5]);
130 MAC16(b0, W7, row[7]);
132 MAC16(b1, -W1, row[5]);
133 MAC16(b1, -W5, row[7]);
135 MAC16(b2, W7, row[5]);
136 MAC16(b2, W3, row[7]);
138 MAC16(b3, W3, row[5]);
139 MAC16(b3, -W1, row[7]);
142 row[0] = (a0 + b0) >> ROW_SHIFT;
143 row[7] = (a0 - b0) >> ROW_SHIFT;
144 row[1] = (a1 + b1) >> ROW_SHIFT;
145 row[6] = (a1 - b1) >> ROW_SHIFT;
146 row[2] = (a2 + b2) >> ROW_SHIFT;
147 row[5] = (a2 - b2) >> ROW_SHIFT;
148 row[3] = (a3 + b3) >> ROW_SHIFT;
149 row[4] = (a3 - b3) >> ROW_SHIFT;
152 static inline void idctSparseColPut (uint8_t *dest, int line_size,
155 int a0, a1, a2, a3, b0, b1, b2, b3;
156 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
158 /* XXX: I did that only to give same values as previous code */
159 a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4));
169 b0 = MUL16(W1, col[8*1]);
170 b1 = MUL16(W3, col[8*1]);
171 b2 = MUL16(W5, col[8*1]);
172 b3 = MUL16(W7, col[8*1]);
174 MAC16(b0, + W3, col[8*3]);
175 MAC16(b1, - W7, col[8*3]);
176 MAC16(b2, - W1, col[8*3]);
177 MAC16(b3, - W5, col[8*3]);
187 MAC16(b0, + W5, col[8*5]);
188 MAC16(b1, - W1, col[8*5]);
189 MAC16(b2, + W7, col[8*5]);
190 MAC16(b3, + W3, col[8*5]);
201 MAC16(b0, + W7, col[8*7]);
202 MAC16(b1, - W5, col[8*7]);
203 MAC16(b2, + W3, col[8*7]);
204 MAC16(b3, - W1, col[8*7]);
207 dest[0] = cm[(a0 + b0) >> COL_SHIFT];
209 dest[0] = cm[(a1 + b1) >> COL_SHIFT];
211 dest[0] = cm[(a2 + b2) >> COL_SHIFT];
213 dest[0] = cm[(a3 + b3) >> COL_SHIFT];
215 dest[0] = cm[(a3 - b3) >> COL_SHIFT];
217 dest[0] = cm[(a2 - b2) >> COL_SHIFT];
219 dest[0] = cm[(a1 - b1) >> COL_SHIFT];
221 dest[0] = cm[(a0 - b0) >> COL_SHIFT];
224 static inline void idctSparseColAdd (uint8_t *dest, int line_size,
227 int a0, a1, a2, a3, b0, b1, b2, b3;
228 uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
230 /* XXX: I did that only to give same values as previous code */
231 a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4));
241 b0 = MUL16(W1, col[8*1]);
242 b1 = MUL16(W3, col[8*1]);
243 b2 = MUL16(W5, col[8*1]);
244 b3 = MUL16(W7, col[8*1]);
246 MAC16(b0, + W3, col[8*3]);
247 MAC16(b1, - W7, col[8*3]);
248 MAC16(b2, - W1, col[8*3]);
249 MAC16(b3, - W5, col[8*3]);
259 MAC16(b0, + W5, col[8*5]);
260 MAC16(b1, - W1, col[8*5]);
261 MAC16(b2, + W7, col[8*5]);
262 MAC16(b3, + W3, col[8*5]);
273 MAC16(b0, + W7, col[8*7]);
274 MAC16(b1, - W5, col[8*7]);
275 MAC16(b2, + W3, col[8*7]);
276 MAC16(b3, - W1, col[8*7]);
279 dest[0] = cm[dest[0] + ((a0 + b0) >> COL_SHIFT)];
281 dest[0] = cm[dest[0] + ((a1 + b1) >> COL_SHIFT)];
283 dest[0] = cm[dest[0] + ((a2 + b2) >> COL_SHIFT)];
285 dest[0] = cm[dest[0] + ((a3 + b3) >> COL_SHIFT)];
287 dest[0] = cm[dest[0] + ((a3 - b3) >> COL_SHIFT)];
289 dest[0] = cm[dest[0] + ((a2 - b2) >> COL_SHIFT)];
291 dest[0] = cm[dest[0] + ((a1 - b1) >> COL_SHIFT)];
293 dest[0] = cm[dest[0] + ((a0 - b0) >> COL_SHIFT)];
296 static inline void idctSparseCol (DCTELEM * col)
298 int a0, a1, a2, a3, b0, b1, b2, b3;
300 /* XXX: I did that only to give same values as previous code */
301 a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4));
311 b0 = MUL16(W1, col[8*1]);
312 b1 = MUL16(W3, col[8*1]);
313 b2 = MUL16(W5, col[8*1]);
314 b3 = MUL16(W7, col[8*1]);
316 MAC16(b0, + W3, col[8*3]);
317 MAC16(b1, - W7, col[8*3]);
318 MAC16(b2, - W1, col[8*3]);
319 MAC16(b3, - W5, col[8*3]);
329 MAC16(b0, + W5, col[8*5]);
330 MAC16(b1, - W1, col[8*5]);
331 MAC16(b2, + W7, col[8*5]);
332 MAC16(b3, + W3, col[8*5]);
343 MAC16(b0, + W7, col[8*7]);
344 MAC16(b1, - W5, col[8*7]);
345 MAC16(b2, + W3, col[8*7]);
346 MAC16(b3, - W1, col[8*7]);
349 col[0 ] = ((a0 + b0) >> COL_SHIFT);
350 col[8 ] = ((a1 + b1) >> COL_SHIFT);
351 col[16] = ((a2 + b2) >> COL_SHIFT);
352 col[24] = ((a3 + b3) >> COL_SHIFT);
353 col[32] = ((a3 - b3) >> COL_SHIFT);
354 col[40] = ((a2 - b2) >> COL_SHIFT);
355 col[48] = ((a1 - b1) >> COL_SHIFT);
356 col[56] = ((a0 - b0) >> COL_SHIFT);
359 void ff_simple_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
363 idctRowCondDC(block + i*8);
366 idctSparseColPut(dest + i, line_size, block + i);
369 void ff_simple_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
373 idctRowCondDC(block + i*8);
376 idctSparseColAdd(dest + i, line_size, block + i);
379 void ff_simple_idct(DCTELEM *block)
383 idctRowCondDC(block + i*8);
386 idctSparseCol(block + i);
392 #define C_FIX(x) ((int)((x) * (1 << CN_SHIFT) + 0.5))
393 #define C1 C_FIX(0.6532814824)
394 #define C2 C_FIX(0.2705980501)
396 /* row idct is multiple by 16 * sqrt(2.0), col idct4 is normalized,
397 and the butterfly must be multiplied by 0.5 * sqrt(2.0) */
398 #define C_SHIFT (4+1+12)
400 static inline void idct4col_put(uint8_t *dest, int line_size, const DCTELEM *col)
402 int c0, c1, c2, c3, a0, a1, a2, a3;
403 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
409 c0 = ((a0 + a2) << (CN_SHIFT - 1)) + (1 << (C_SHIFT - 1));
410 c2 = ((a0 - a2) << (CN_SHIFT - 1)) + (1 << (C_SHIFT - 1));
411 c1 = a1 * C1 + a3 * C2;
412 c3 = a1 * C2 - a3 * C1;
413 dest[0] = cm[(c0 + c1) >> C_SHIFT];
415 dest[0] = cm[(c2 + c3) >> C_SHIFT];
417 dest[0] = cm[(c2 - c3) >> C_SHIFT];
419 dest[0] = cm[(c0 - c1) >> C_SHIFT];
428 ptr[8 + k] = a0 - a1;\
431 /* only used by DV codec. The input must be interlaced. 128 is added
432 to the pixels before clamping to avoid systematic error
433 (1024*sqrt(2)) offset would be needed otherwise. */
434 /* XXX: I think a 1.0/sqrt(2) normalization should be needed to
435 compensate the extra butterfly stage - I don't have the full DV
437 void ff_simple_idct248_put(uint8_t *dest, int line_size, DCTELEM *block)
456 /* IDCT8 on each line */
458 idctRowCondDC(block + i*8);
461 /* IDCT4 and store */
463 idct4col_put(dest + i, 2 * line_size, block + i);
464 idct4col_put(dest + line_size + i, 2 * line_size, block + 8 + i);
468 /* 8x4 & 4x8 WMV2 IDCT */
475 #define C_FIX(x) ((int)((x) * 1.414213562 * (1 << CN_SHIFT) + 0.5))
476 #define C1 C_FIX(0.6532814824)
477 #define C2 C_FIX(0.2705980501)
478 #define C3 C_FIX(0.5)
479 #define C_SHIFT (4+1+12)
480 static inline void idct4col_add(uint8_t *dest, int line_size, const DCTELEM *col)
482 int c0, c1, c2, c3, a0, a1, a2, a3;
483 const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
489 c0 = (a0 + a2)*C3 + (1 << (C_SHIFT - 1));
490 c2 = (a0 - a2)*C3 + (1 << (C_SHIFT - 1));
491 c1 = a1 * C1 + a3 * C2;
492 c3 = a1 * C2 - a3 * C1;
493 dest[0] = cm[dest[0] + ((c0 + c1) >> C_SHIFT)];
495 dest[0] = cm[dest[0] + ((c2 + c3) >> C_SHIFT)];
497 dest[0] = cm[dest[0] + ((c2 - c3) >> C_SHIFT)];
499 dest[0] = cm[dest[0] + ((c0 - c1) >> C_SHIFT)];
503 #define R_FIX(x) ((int)((x) * 1.414213562 * (1 << RN_SHIFT) + 0.5))
504 #define R1 R_FIX(0.6532814824)
505 #define R2 R_FIX(0.2705980501)
506 #define R3 R_FIX(0.5)
508 static inline void idct4row(DCTELEM *row)
510 int c0, c1, c2, c3, a0, a1, a2, a3;
511 //const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
517 c0 = (a0 + a2)*R3 + (1 << (R_SHIFT - 1));
518 c2 = (a0 - a2)*R3 + (1 << (R_SHIFT - 1));
519 c1 = a1 * R1 + a3 * R2;
520 c3 = a1 * R2 - a3 * R1;
521 row[0]= (c0 + c1) >> R_SHIFT;
522 row[1]= (c2 + c3) >> R_SHIFT;
523 row[2]= (c2 - c3) >> R_SHIFT;
524 row[3]= (c0 - c1) >> R_SHIFT;
527 void ff_simple_idct84_add(uint8_t *dest, int line_size, DCTELEM *block)
531 /* IDCT8 on each line */
533 idctRowCondDC(block + i*8);
536 /* IDCT4 and store */
538 idct4col_add(dest + i, line_size, block + i);
542 void ff_simple_idct48_add(uint8_t *dest, int line_size, DCTELEM *block)
546 /* IDCT4 on each line */
548 idct4row(block + i*8);
551 /* IDCT8 and store */
553 idctSparseColAdd(dest + i, line_size, block + i);
557 void ff_simple_idct44_add(uint8_t *dest, int line_size, DCTELEM *block)
561 /* IDCT4 on each line */
563 idct4row(block + i*8);
566 /* IDCT4 and store */
568 idct4col_add(dest + i, line_size, block + i);