1 /*****************************************************************************
2 * dct.c: h264 encoder library
3 *****************************************************************************
4 * Copyright (C) 2003 Laurent Aimar
5 * $Id: dct.c,v 1.1 2004/06/03 19:27:06 fenrir Exp $
7 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
22 *****************************************************************************/
36 # include "i386/dct.h"
43 static inline int clip_uint8( int a )
52 * XXX For all dct dc : input could be equal to output so ...
55 static void dct2x2dc( int16_t d[2][2] )
59 tmp[0][0] = d[0][0] + d[0][1];
60 tmp[1][0] = d[0][0] - d[0][1];
61 tmp[0][1] = d[1][0] + d[1][1];
62 tmp[1][1] = d[1][0] - d[1][1];
64 d[0][0] = tmp[0][0] + tmp[0][1];
65 d[0][1] = tmp[1][0] + tmp[1][1];
66 d[1][0] = tmp[0][0] - tmp[0][1];
67 d[1][1] = tmp[1][0] - tmp[1][1];
70 static void dct4x4dc( int16_t d[4][4] )
77 for( i = 0; i < 4; i++ )
79 s01 = d[i][0] + d[i][1];
80 d01 = d[i][0] - d[i][1];
81 s23 = d[i][2] + d[i][3];
82 d23 = d[i][2] - d[i][3];
84 tmp[0][i] = s01 + s23;
85 tmp[1][i] = s01 - s23;
86 tmp[2][i] = d01 - d23;
87 tmp[3][i] = d01 + d23;
90 for( i = 0; i < 4; i++ )
92 s01 = tmp[i][0] + tmp[i][1];
93 d01 = tmp[i][0] - tmp[i][1];
94 s23 = tmp[i][2] + tmp[i][3];
95 d23 = tmp[i][2] - tmp[i][3];
97 d[0][i] = ( s01 + s23 + 1 ) >> 1;
98 d[1][i] = ( s01 - s23 + 1 ) >> 1;
99 d[2][i] = ( d01 - d23 + 1 ) >> 1;
100 d[3][i] = ( d01 + d23 + 1 ) >> 1;
104 static void idct4x4dc( int16_t d[4][4] )
111 for( i = 0; i < 4; i++ )
113 s01 = d[0][i] + d[1][i];
114 d01 = d[0][i] - d[1][i];
115 s23 = d[2][i] + d[3][i];
116 d23 = d[2][i] - d[3][i];
118 tmp[0][i] = s01 + s23;
119 tmp[1][i] = s01 - s23;
120 tmp[2][i] = d01 - d23;
121 tmp[3][i] = d01 + d23;
124 for( i = 0; i < 4; i++ )
126 s01 = tmp[i][0] + tmp[i][1];
127 d01 = tmp[i][0] - tmp[i][1];
128 s23 = tmp[i][2] + tmp[i][3];
129 d23 = tmp[i][2] - tmp[i][3];
138 static inline void pixel_sub_wxh( int16_t *diff, int i_size,
139 uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
142 for( y = 0; y < i_size; y++ )
144 for( x = 0; x < i_size; x++ )
146 diff[x + y*i_size] = pix1[x] - pix2[x];
153 static void sub4x4_dct( int16_t dct[4][4], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
159 pixel_sub_wxh( (int16_t*)d, 4, pix1, i_pix1, pix2, i_pix2 );
161 for( i = 0; i < 4; i++ )
163 const int s03 = d[i][0] + d[i][3];
164 const int s12 = d[i][1] + d[i][2];
165 const int d03 = d[i][0] - d[i][3];
166 const int d12 = d[i][1] - d[i][2];
168 tmp[0][i] = s03 + s12;
169 tmp[1][i] = 2*d03 + d12;
170 tmp[2][i] = s03 - s12;
171 tmp[3][i] = d03 - 2*d12;
174 for( i = 0; i < 4; i++ )
176 const int s03 = tmp[i][0] + tmp[i][3];
177 const int s12 = tmp[i][1] + tmp[i][2];
178 const int d03 = tmp[i][0] - tmp[i][3];
179 const int d12 = tmp[i][1] - tmp[i][2];
181 dct[0][i] = s03 + s12;
182 dct[1][i] = 2*d03 + d12;
183 dct[2][i] = s03 - s12;
184 dct[3][i] = d03 - 2*d12;
188 static void sub8x8_dct( int16_t dct[4][4][4], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
190 sub4x4_dct( dct[0], &pix1[0], i_pix1, &pix2[0], i_pix2 );
191 sub4x4_dct( dct[1], &pix1[4], i_pix1, &pix2[4], i_pix2 );
192 sub4x4_dct( dct[2], &pix1[4*i_pix1+0], i_pix1, &pix2[4*i_pix2+0], i_pix2 );
193 sub4x4_dct( dct[3], &pix1[4*i_pix1+4], i_pix1, &pix2[4*i_pix2+4], i_pix2 );
196 static void sub16x16_dct( int16_t dct[16][4][4], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
198 sub8x8_dct( &dct[ 0], pix1, i_pix1, pix2, i_pix2 );
199 sub8x8_dct( &dct[ 4], &pix1[8], i_pix1, &pix2[8], i_pix2 );
200 sub8x8_dct( &dct[ 8], &pix1[8*i_pix1], i_pix1, &pix2[8*i_pix2], i_pix2 );
201 sub8x8_dct( &dct[12], &pix1[8*i_pix1+8], i_pix1, &pix2[8*i_pix2+8], i_pix2 );
205 static void add4x4_idct( uint8_t *p_dst, int i_dst, int16_t dct[4][4] )
212 for( i = 0; i < 4; i++ )
214 const int s02 = dct[i][0] + dct[i][2];
215 const int d02 = dct[i][0] - dct[i][2];
216 const int s13 = dct[i][1] + (dct[i][3]>>1);
217 const int d13 = (dct[i][1]>>1) - dct[i][3];
219 tmp[i][0] = s02 + s13;
220 tmp[i][1] = d02 + d13;
221 tmp[i][2] = d02 - d13;
222 tmp[i][3] = s02 - s13;
225 for( i = 0; i < 4; i++ )
227 const int s02 = tmp[0][i] + tmp[2][i];
228 const int d02 = tmp[0][i] - tmp[2][i];
229 const int s13 = tmp[1][i] + (tmp[3][i]>>1);
230 const int d13 = (tmp[1][i]>>1) - tmp[3][i];
232 d[0][i] = ( s02 + s13 + 32 ) >> 6;
233 d[1][i] = ( d02 + d13 + 32 ) >> 6;
234 d[2][i] = ( d02 - d13 + 32 ) >> 6;
235 d[3][i] = ( s02 - s13 + 32 ) >> 6;
239 for( y = 0; y < 4; y++ )
241 for( x = 0; x < 4; x++ )
243 p_dst[x] = clip_uint8( p_dst[x] + d[y][x] );
249 static void add8x8_idct( uint8_t *p_dst, int i_dst, int16_t dct[4][4][4] )
251 add4x4_idct( p_dst, i_dst, dct[0] );
252 add4x4_idct( &p_dst[4], i_dst, dct[1] );
253 add4x4_idct( &p_dst[4*i_dst+0], i_dst, dct[2] );
254 add4x4_idct( &p_dst[4*i_dst+4], i_dst, dct[3] );
257 static void add16x16_idct( uint8_t *p_dst, int i_dst, int16_t dct[16][4][4] )
259 add8x8_idct( &p_dst[0], i_dst, &dct[0] );
260 add8x8_idct( &p_dst[8], i_dst, &dct[4] );
261 add8x8_idct( &p_dst[8*i_dst], i_dst, &dct[8] );
262 add8x8_idct( &p_dst[8*i_dst+8], i_dst, &dct[12] );
265 /****************************************************************************
267 ****************************************************************************/
270 const int s07 = SRC(0) + SRC(7);\
271 const int s16 = SRC(1) + SRC(6);\
272 const int s25 = SRC(2) + SRC(5);\
273 const int s34 = SRC(3) + SRC(4);\
274 const int a0 = s07 + s34;\
275 const int a1 = s16 + s25;\
276 const int a2 = s07 - s34;\
277 const int a3 = s16 - s25;\
278 const int d07 = SRC(0) - SRC(7);\
279 const int d16 = SRC(1) - SRC(6);\
280 const int d25 = SRC(2) - SRC(5);\
281 const int d34 = SRC(3) - SRC(4);\
282 const int a4 = d16 + d25 + (d07 + (d07>>1));\
283 const int a5 = d07 - d34 - (d25 + (d25>>1));\
284 const int a6 = d07 + d34 - (d16 + (d16>>1));\
285 const int a7 = d16 - d25 + (d34 + (d34>>1));\
287 SRC(1) = a4 + (a7>>2);\
288 SRC(2) = a2 + (a3>>1);\
289 SRC(3) = a5 + (a6>>2);\
291 SRC(5) = a6 - (a5>>2);\
292 SRC(6) = (a2>>1) - a3 ;\
293 SRC(7) = (a4>>2) - a7 ;\
296 static void sub8x8_dct8( int16_t dct[8][8], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
300 pixel_sub_wxh( (int16_t*)dct, 8, pix1, i_pix1, pix2, i_pix2 );
302 #define SRC(x) dct[i][x]
303 for( i = 0; i < 8; i++ )
307 #define SRC(x) dct[x][i]
308 for( i = 0; i < 8; i++ )
313 static void sub16x16_dct8( int16_t dct[4][8][8], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
315 sub8x8_dct8( dct[0], pix1, i_pix1, pix2, i_pix2 );
316 sub8x8_dct8( dct[1], &pix1[8], i_pix1, &pix2[8], i_pix2 );
317 sub8x8_dct8( dct[2], &pix1[8*i_pix1], i_pix1, &pix2[8*i_pix2], i_pix2 );
318 sub8x8_dct8( dct[3], &pix1[8*i_pix1+8], i_pix1, &pix2[8*i_pix2+8], i_pix2 );
322 const int a0 = SRC(0) + SRC(4);\
323 const int a2 = SRC(0) - SRC(4);\
324 const int a4 = (SRC(2)>>1) - SRC(6);\
325 const int a6 = (SRC(6)>>1) + SRC(2);\
326 const int b0 = a0 + a6;\
327 const int b2 = a2 + a4;\
328 const int b4 = a2 - a4;\
329 const int b6 = a0 - a6;\
330 const int a1 = -SRC(3) + SRC(5) - SRC(7) - (SRC(7)>>1);\
331 const int a3 = SRC(1) + SRC(7) - SRC(3) - (SRC(3)>>1);\
332 const int a5 = -SRC(1) + SRC(7) + SRC(5) + (SRC(5)>>1);\
333 const int a7 = SRC(3) + SRC(5) + SRC(1) + (SRC(1)>>1);\
334 const int b1 = (a7>>2) + a1;\
335 const int b3 = a3 + (a5>>2);\
336 const int b5 = (a3>>2) - a5;\
337 const int b7 = a7 - (a1>>2);\
348 static void add8x8_idct8( uint8_t *dst, int i_dst, int16_t dct[8][8] )
352 dct[0][0] += 32; // rounding for the >>6 at the end
354 #define SRC(x) dct[i][x]
355 #define DST(x,rhs) dct[i][x] = (rhs)
356 for( i = 0; i < 8; i++ )
361 #define SRC(x) dct[x][i]
362 #define DST(x,rhs) dst[i + x*i_dst] = clip_uint8( dst[i + x*i_dst] + ((rhs) >> 6) );
363 for( i = 0; i < 8; i++ )
369 static void add16x16_idct8( uint8_t *dst, int i_dst, int16_t dct[4][8][8] )
371 add8x8_idct8( &dst[0], i_dst, dct[0] );
372 add8x8_idct8( &dst[8], i_dst, dct[1] );
373 add8x8_idct8( &dst[8*i_dst], i_dst, dct[2] );
374 add8x8_idct8( &dst[8*i_dst+8], i_dst, dct[3] );
378 /****************************************************************************
380 ****************************************************************************/
381 void x264_dct_init( int cpu, x264_dct_function_t *dctf )
383 dctf->sub4x4_dct = sub4x4_dct;
384 dctf->add4x4_idct = add4x4_idct;
386 dctf->sub8x8_dct = sub8x8_dct;
387 dctf->add8x8_idct = add8x8_idct;
389 dctf->sub16x16_dct = sub16x16_dct;
390 dctf->add16x16_idct = add16x16_idct;
392 dctf->sub8x8_dct8 = sub8x8_dct8;
393 dctf->add8x8_idct8 = add8x8_idct8;
395 dctf->sub16x16_dct8 = sub16x16_dct8;
396 dctf->add16x16_idct8 = add16x16_idct8;
398 dctf->dct4x4dc = dct4x4dc;
399 dctf->idct4x4dc = idct4x4dc;
401 dctf->dct2x2dc = dct2x2dc;
402 dctf->idct2x2dc = dct2x2dc;
405 if( cpu&X264_CPU_MMXEXT )
407 dctf->sub4x4_dct = x264_sub4x4_dct_mmxext;
408 dctf->sub8x8_dct = x264_sub8x8_dct_mmxext;
409 dctf->sub16x16_dct = x264_sub16x16_dct_mmxext;
411 dctf->add4x4_idct = x264_add4x4_idct_mmxext;
412 dctf->add8x8_idct = x264_add8x8_idct_mmxext;
413 dctf->add16x16_idct = x264_add16x16_idct_mmxext;
415 dctf->sub8x8_dct8 = x264_sub8x8_dct8_mmxext;
416 dctf->sub16x16_dct8 = x264_sub16x16_dct8_mmxext;
418 dctf->add8x8_idct8 = x264_add8x8_idct8_mmxext;
419 dctf->add16x16_idct8= x264_add16x16_idct8_mmxext;
421 dctf->dct4x4dc = x264_dct4x4dc_mmxext;
422 dctf->idct4x4dc = x264_idct4x4dc_mmxext;
426 if( cpu&X264_CPU_ALTIVEC )
428 dctf->sub4x4_dct = x264_sub4x4_dct_altivec;
429 dctf->sub8x8_dct = x264_sub8x8_dct_altivec;
430 dctf->sub16x16_dct = x264_sub16x16_dct_altivec;