4 * Copyright (c) 2001 Michael Niedermayer <michaelni@gmx.at>
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.
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.
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., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 based upon some outcommented c code from mpeg2dec (idct_mmx.c
22 written by Aaron Holtzman <aholtzma@ess.engr.uvic.ca>)
26 #include "simple_idct.h"
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) */
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 16383 //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
47 #define COL_SHIFT 20 // 6
54 #if defined(ARCH_POWERPC_405)
56 /* signed 16x16 -> 32 multiply add accumulate */
57 #define MAC16(rt, ra, rb) \
58 asm ("maclhw %0, %2, %3" : "=r" (rt) : "0" (rt), "r" (ra), "r" (rb));
60 /* signed 16x16 -> 32 multiply */
61 #define MUL16(rt, ra, rb) \
62 asm ("mullhw %0, %1, %2" : "=r" (rt) : "r" (ra), "r" (rb));
66 /* signed 16x16 -> 32 multiply add accumulate */
67 #define MAC16(rt, ra, rb) rt += (ra) * (rb)
69 /* signed 16x16 -> 32 multiply */
70 #define MUL16(rt, ra, rb) rt = (ra) * (rb)
75 /* 0: all entries 0, 1: only first entry nonzero, 2: otherwise */
76 static inline int idctRowCondDC(int16_t *row)
78 int_fast32_t a0, a1, a2, a3, b0, b1, b2, b3;
79 uint64_t *lrow = (uint64_t *) row;
84 if ((lrow[0] & ~0xffffULL) == 0) {
88 a0 += 1 << (ROW_SHIFT - 1);
100 a0 = (W4 * row[0]) + (1 << (ROW_SHIFT - 1));
159 row[0] = (a0 + b0) >> ROW_SHIFT;
160 row[1] = (a1 + b1) >> ROW_SHIFT;
161 row[2] = (a2 + b2) >> ROW_SHIFT;
162 row[3] = (a3 + b3) >> ROW_SHIFT;
163 row[4] = (a3 - b3) >> ROW_SHIFT;
164 row[5] = (a2 - b2) >> ROW_SHIFT;
165 row[6] = (a1 - b1) >> ROW_SHIFT;
166 row[7] = (a0 - b0) >> ROW_SHIFT;
171 inline static void idctSparseCol(int16_t *col)
173 int a0, a1, a2, a3, b0, b1, b2, b3;
175 col[0] += (1 << (COL_SHIFT - 1)) / W4;
177 a0 = W4 * col[8 * 0];
178 a1 = W4 * col[8 * 0];
179 a2 = W4 * col[8 * 0];
180 a3 = W4 * col[8 * 0];
183 a0 += W2 * col[8 * 2];
184 a1 += W6 * col[8 * 2];
185 a2 -= W6 * col[8 * 2];
186 a3 -= W2 * col[8 * 2];
190 a0 += W4 * col[8 * 4];
191 a1 -= W4 * col[8 * 4];
192 a2 -= W4 * col[8 * 4];
193 a3 += W4 * col[8 * 4];
197 a0 += W6 * col[8 * 6];
198 a1 -= W2 * col[8 * 6];
199 a2 += W2 * col[8 * 6];
200 a3 -= W6 * col[8 * 6];
204 b0 = W1 * col[8 * 1];
205 b1 = W3 * col[8 * 1];
206 b2 = W5 * col[8 * 1];
207 b3 = W7 * col[8 * 1];
209 b0 = b1 = b2 = b3 = 0;
213 b0 += W3 * col[8 * 3];
214 b1 -= W7 * col[8 * 3];
215 b2 -= W1 * col[8 * 3];
216 b3 -= W5 * col[8 * 3];
220 b0 += W5 * col[8 * 5];
221 b1 -= W1 * col[8 * 5];
222 b2 += W7 * col[8 * 5];
223 b3 += W3 * col[8 * 5];
227 b0 += W7 * col[8 * 7];
228 b1 -= W5 * col[8 * 7];
229 b2 += W3 * col[8 * 7];
230 b3 -= W1 * col[8 * 7];
233 col[8 * 0] = (a0 + b0) >> COL_SHIFT;
234 col[8 * 7] = (a0 - b0) >> COL_SHIFT;
235 col[8 * 1] = (a1 + b1) >> COL_SHIFT;
236 col[8 * 6] = (a1 - b1) >> COL_SHIFT;
237 col[8 * 2] = (a2 + b2) >> COL_SHIFT;
238 col[8 * 5] = (a2 - b2) >> COL_SHIFT;
239 col[8 * 3] = (a3 + b3) >> COL_SHIFT;
240 col[8 * 4] = (a3 - b3) >> COL_SHIFT;
243 #else /* not ARCH_ALPHA */
245 static inline void idctRowCondDC (int16_t * row)
247 int a0, a1, a2, a3, b0, b1, b2, b3;
255 #ifdef WORDS_BIGENDIAN
256 #define ROW0_MASK 0xffff000000000000LL
258 #define ROW0_MASK 0xffffLL
260 if ( ((((uint64_t *)row)[0] & ~ROW0_MASK) |
261 ((uint64_t *)row)[1]) == 0) {
262 temp = (row[0] << 3) & 0xffff;
265 ((uint64_t *)row)[0] = temp;
266 ((uint64_t *)row)[1] = temp;
270 if (!(((uint32_t*)row)[1] |
271 ((uint32_t*)row)[2] |
272 ((uint32_t*)row)[3] |
274 temp = (row[0] << 3) & 0xffff;
276 ((uint32_t*)row)[0]=((uint32_t*)row)[1] =
277 ((uint32_t*)row)[2]=((uint32_t*)row)[3] = temp;
282 a0 = (W4 * row[0]) + (1 << (ROW_SHIFT - 1));
287 /* no need to optimize : gcc does it */
293 MUL16(b0, W1, row[1]);
294 MAC16(b0, W3, row[3]);
295 MUL16(b1, W3, row[1]);
296 MAC16(b1, -W7, row[3]);
297 MUL16(b2, W5, row[1]);
298 MAC16(b2, -W1, row[3]);
299 MUL16(b3, W7, row[1]);
300 MAC16(b3, -W5, row[3]);
303 temp = ((uint64_t*)row)[1];
305 temp = ((uint32_t*)row)[2] | ((uint32_t*)row)[3];
308 a0 += W4*row[4] + W6*row[6];
309 a1 += - W4*row[4] - W2*row[6];
310 a2 += - W4*row[4] + W2*row[6];
311 a3 += W4*row[4] - W6*row[6];
313 MAC16(b0, W5, row[5]);
314 MAC16(b0, W7, row[7]);
316 MAC16(b1, -W1, row[5]);
317 MAC16(b1, -W5, row[7]);
319 MAC16(b2, W7, row[5]);
320 MAC16(b2, W3, row[7]);
322 MAC16(b3, W3, row[5]);
323 MAC16(b3, -W1, row[7]);
326 row[0] = (a0 + b0) >> ROW_SHIFT;
327 row[7] = (a0 - b0) >> ROW_SHIFT;
328 row[1] = (a1 + b1) >> ROW_SHIFT;
329 row[6] = (a1 - b1) >> ROW_SHIFT;
330 row[2] = (a2 + b2) >> ROW_SHIFT;
331 row[5] = (a2 - b2) >> ROW_SHIFT;
332 row[3] = (a3 + b3) >> ROW_SHIFT;
333 row[4] = (a3 - b3) >> ROW_SHIFT;
335 #endif /* not ARCH_ALPHA */
337 static inline void idctSparseColPut (UINT8 *dest, int line_size,
340 int a0, a1, a2, a3, b0, b1, b2, b3;
341 UINT8 *cm = cropTbl + MAX_NEG_CROP;
343 /* XXX: I did that only to give same values as previous code */
344 a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4));
354 MUL16(b0, W1, col[8*1]);
355 MUL16(b1, W3, col[8*1]);
356 MUL16(b2, W5, col[8*1]);
357 MUL16(b3, W7, col[8*1]);
359 MAC16(b0, + W3, col[8*3]);
360 MAC16(b1, - W7, col[8*3]);
361 MAC16(b2, - W1, col[8*3]);
362 MAC16(b3, - W5, col[8*3]);
372 MAC16(b0, + W5, col[8*5]);
373 MAC16(b1, - W1, col[8*5]);
374 MAC16(b2, + W7, col[8*5]);
375 MAC16(b3, + W3, col[8*5]);
386 MAC16(b0, + W7, col[8*7]);
387 MAC16(b1, - W5, col[8*7]);
388 MAC16(b2, + W3, col[8*7]);
389 MAC16(b3, - W1, col[8*7]);
392 dest[0] = cm[(a0 + b0) >> COL_SHIFT];
394 dest[0] = cm[(a1 + b1) >> COL_SHIFT];
396 dest[0] = cm[(a2 + b2) >> COL_SHIFT];
398 dest[0] = cm[(a3 + b3) >> COL_SHIFT];
400 dest[0] = cm[(a3 - b3) >> COL_SHIFT];
402 dest[0] = cm[(a2 - b2) >> COL_SHIFT];
404 dest[0] = cm[(a1 - b1) >> COL_SHIFT];
406 dest[0] = cm[(a0 - b0) >> COL_SHIFT];
409 static inline void idctSparseColAdd (UINT8 *dest, int line_size,
412 int a0, a1, a2, a3, b0, b1, b2, b3;
413 UINT8 *cm = cropTbl + MAX_NEG_CROP;
415 /* XXX: I did that only to give same values as previous code */
416 a0 = W4 * (col[8*0] + ((1<<(COL_SHIFT-1))/W4));
426 MUL16(b0, W1, col[8*1]);
427 MUL16(b1, W3, col[8*1]);
428 MUL16(b2, W5, col[8*1]);
429 MUL16(b3, W7, col[8*1]);
431 MAC16(b0, + W3, col[8*3]);
432 MAC16(b1, - W7, col[8*3]);
433 MAC16(b2, - W1, col[8*3]);
434 MAC16(b3, - W5, col[8*3]);
444 MAC16(b0, + W5, col[8*5]);
445 MAC16(b1, - W1, col[8*5]);
446 MAC16(b2, + W7, col[8*5]);
447 MAC16(b3, + W3, col[8*5]);
458 MAC16(b0, + W7, col[8*7]);
459 MAC16(b1, - W5, col[8*7]);
460 MAC16(b2, + W3, col[8*7]);
461 MAC16(b3, - W1, col[8*7]);
464 dest[0] = cm[dest[0] + ((a0 + b0) >> COL_SHIFT)];
466 dest[0] = cm[dest[0] + ((a1 + b1) >> COL_SHIFT)];
468 dest[0] = cm[dest[0] + ((a2 + b2) >> COL_SHIFT)];
470 dest[0] = cm[dest[0] + ((a3 + b3) >> COL_SHIFT)];
472 dest[0] = cm[dest[0] + ((a3 - b3) >> COL_SHIFT)];
474 dest[0] = cm[dest[0] + ((a2 - b2) >> COL_SHIFT)];
476 dest[0] = cm[dest[0] + ((a1 - b1) >> COL_SHIFT)];
478 dest[0] = cm[dest[0] + ((a0 - b0) >> COL_SHIFT)];
482 /* If all rows but the first one are zero after row transformation,
483 all rows will be identical after column transformation. */
484 static inline void idctCol2(int16_t *col)
488 uint64_t *lcol = (uint64_t *) col;
490 for (i = 0; i < 8; ++i) {
491 int a0 = col[0] + (1 << (COL_SHIFT - 1)) / W4;
494 col[0] = a0 >> COL_SHIFT;
500 lcol[ 2] = l; lcol[ 3] = r;
501 lcol[ 4] = l; lcol[ 5] = r;
502 lcol[ 6] = l; lcol[ 7] = r;
503 lcol[ 8] = l; lcol[ 9] = r;
504 lcol[10] = l; lcol[11] = r;
505 lcol[12] = l; lcol[13] = r;
506 lcol[14] = l; lcol[15] = r;
509 void simple_idct (short *block)
513 int rowsZero = 1; /* all rows except row 0 zero */
514 int rowsConstant = 1; /* all rows consist of a constant value */
516 for (i = 0; i < 8; i++) {
517 int sparseness = idctRowCondDC(block + 8 * i);
519 if (i > 0 && sparseness > 0)
527 } else if (rowsConstant) {
528 uint64_t *lblock = (uint64_t *) block;
530 idctSparseCol(block);
531 for (i = 0; i < 8; i++) {
532 uint64_t v = (uint16_t) block[i * 8];
541 for (i = 0; i < 8; i++)
542 idctSparseCol(block + i);
546 /* XXX: suppress this mess */
547 void simple_idct_put(UINT8 *dest, int line_size, DCTELEM *block)
550 put_pixels_clamped(block, dest, line_size);
553 void simple_idct_add(UINT8 *dest, int line_size, DCTELEM *block)
556 add_pixels_clamped(block, dest, line_size);
561 void simple_idct_put(UINT8 *dest, int line_size, INT16 *block)
565 idctRowCondDC(block + i*8);
568 idctSparseColPut(dest + i, line_size, block + i);
571 void simple_idct_add(UINT8 *dest, int line_size, INT16 *block)
575 idctRowCondDC(block + i*8);
578 idctSparseColAdd(dest + i, line_size, block + i);