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 *****************************************************************************/
26 # include "i386/dct.h"
32 int x264_dct4_weight2_zigzag[2][16];
33 int x264_dct8_weight2_zigzag[2][64];
35 static inline int clip_uint8( int a )
44 * XXX For all dct dc : input could be equal to output so ...
47 static void dct2x2dc( int16_t d[2][2] )
51 tmp[0][0] = d[0][0] + d[0][1];
52 tmp[1][0] = d[0][0] - d[0][1];
53 tmp[0][1] = d[1][0] + d[1][1];
54 tmp[1][1] = d[1][0] - d[1][1];
56 d[0][0] = tmp[0][0] + tmp[0][1];
57 d[1][0] = tmp[1][0] + tmp[1][1];
58 d[0][1] = tmp[0][0] - tmp[0][1];
59 d[1][1] = tmp[1][0] - tmp[1][1];
62 static void dct4x4dc( int16_t d[4][4] )
69 for( i = 0; i < 4; i++ )
71 s01 = d[i][0] + d[i][1];
72 d01 = d[i][0] - d[i][1];
73 s23 = d[i][2] + d[i][3];
74 d23 = d[i][2] - d[i][3];
76 tmp[0][i] = s01 + s23;
77 tmp[1][i] = s01 - s23;
78 tmp[2][i] = d01 - d23;
79 tmp[3][i] = d01 + d23;
82 for( i = 0; i < 4; i++ )
84 s01 = tmp[i][0] + tmp[i][1];
85 d01 = tmp[i][0] - tmp[i][1];
86 s23 = tmp[i][2] + tmp[i][3];
87 d23 = tmp[i][2] - tmp[i][3];
89 d[i][0] = ( s01 + s23 + 1 ) >> 1;
90 d[i][1] = ( s01 - s23 + 1 ) >> 1;
91 d[i][2] = ( d01 - d23 + 1 ) >> 1;
92 d[i][3] = ( d01 + d23 + 1 ) >> 1;
96 static void idct4x4dc( int16_t d[4][4] )
103 for( i = 0; i < 4; i++ )
105 s01 = d[i][0] + d[i][1];
106 d01 = d[i][0] - d[i][1];
107 s23 = d[i][2] + d[i][3];
108 d23 = d[i][2] - d[i][3];
110 tmp[0][i] = s01 + s23;
111 tmp[1][i] = s01 - s23;
112 tmp[2][i] = d01 - d23;
113 tmp[3][i] = d01 + d23;
116 for( i = 0; i < 4; i++ )
118 s01 = tmp[i][0] + tmp[i][1];
119 d01 = tmp[i][0] - tmp[i][1];
120 s23 = tmp[i][2] + tmp[i][3];
121 d23 = tmp[i][2] - tmp[i][3];
130 static inline void pixel_sub_wxh( int16_t *diff, int i_size,
131 uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
134 for( y = 0; y < i_size; y++ )
136 for( x = 0; x < i_size; x++ )
138 diff[x + y*i_size] = pix1[x] - pix2[x];
145 static void sub4x4_dct( int16_t dct[4][4], uint8_t *pix1, uint8_t *pix2 )
151 pixel_sub_wxh( (int16_t*)d, 4, pix1, FENC_STRIDE, pix2, FDEC_STRIDE );
153 for( i = 0; i < 4; i++ )
155 const int s03 = d[i][0] + d[i][3];
156 const int s12 = d[i][1] + d[i][2];
157 const int d03 = d[i][0] - d[i][3];
158 const int d12 = d[i][1] - d[i][2];
160 tmp[0][i] = s03 + s12;
161 tmp[1][i] = 2*d03 + d12;
162 tmp[2][i] = s03 - s12;
163 tmp[3][i] = d03 - 2*d12;
166 for( i = 0; i < 4; i++ )
168 const int s03 = tmp[i][0] + tmp[i][3];
169 const int s12 = tmp[i][1] + tmp[i][2];
170 const int d03 = tmp[i][0] - tmp[i][3];
171 const int d12 = tmp[i][1] - tmp[i][2];
173 dct[i][0] = s03 + s12;
174 dct[i][1] = 2*d03 + d12;
175 dct[i][2] = s03 - s12;
176 dct[i][3] = d03 - 2*d12;
180 static void sub8x8_dct( int16_t dct[4][4][4], uint8_t *pix1, uint8_t *pix2 )
182 sub4x4_dct( dct[0], &pix1[0], &pix2[0] );
183 sub4x4_dct( dct[1], &pix1[4], &pix2[4] );
184 sub4x4_dct( dct[2], &pix1[4*FENC_STRIDE+0], &pix2[4*FDEC_STRIDE+0] );
185 sub4x4_dct( dct[3], &pix1[4*FENC_STRIDE+4], &pix2[4*FDEC_STRIDE+4] );
188 static void sub16x16_dct( int16_t dct[16][4][4], uint8_t *pix1, uint8_t *pix2 )
190 sub8x8_dct( &dct[ 0], &pix1[0], &pix2[0] );
191 sub8x8_dct( &dct[ 4], &pix1[8], &pix2[8] );
192 sub8x8_dct( &dct[ 8], &pix1[8*FENC_STRIDE+0], &pix2[8*FDEC_STRIDE+0] );
193 sub8x8_dct( &dct[12], &pix1[8*FENC_STRIDE+8], &pix2[8*FDEC_STRIDE+8] );
197 static void add4x4_idct( uint8_t *p_dst, int16_t dct[4][4] )
204 for( i = 0; i < 4; i++ )
206 const int s02 = dct[0][i] + dct[2][i];
207 const int d02 = dct[0][i] - dct[2][i];
208 const int s13 = dct[1][i] + (dct[3][i]>>1);
209 const int d13 = (dct[1][i]>>1) - dct[3][i];
211 tmp[i][0] = s02 + s13;
212 tmp[i][1] = d02 + d13;
213 tmp[i][2] = d02 - d13;
214 tmp[i][3] = s02 - s13;
217 for( i = 0; i < 4; i++ )
219 const int s02 = tmp[0][i] + tmp[2][i];
220 const int d02 = tmp[0][i] - tmp[2][i];
221 const int s13 = tmp[1][i] + (tmp[3][i]>>1);
222 const int d13 = (tmp[1][i]>>1) - tmp[3][i];
224 d[0][i] = ( s02 + s13 + 32 ) >> 6;
225 d[1][i] = ( d02 + d13 + 32 ) >> 6;
226 d[2][i] = ( d02 - d13 + 32 ) >> 6;
227 d[3][i] = ( s02 - s13 + 32 ) >> 6;
231 for( y = 0; y < 4; y++ )
233 for( x = 0; x < 4; x++ )
235 p_dst[x] = clip_uint8( p_dst[x] + d[y][x] );
237 p_dst += FDEC_STRIDE;
241 static void add8x8_idct( uint8_t *p_dst, int16_t dct[4][4][4] )
243 add4x4_idct( &p_dst[0], dct[0] );
244 add4x4_idct( &p_dst[4], dct[1] );
245 add4x4_idct( &p_dst[4*FDEC_STRIDE+0], dct[2] );
246 add4x4_idct( &p_dst[4*FDEC_STRIDE+4], dct[3] );
249 static void add16x16_idct( uint8_t *p_dst, int16_t dct[16][4][4] )
251 add8x8_idct( &p_dst[0], &dct[0] );
252 add8x8_idct( &p_dst[8], &dct[4] );
253 add8x8_idct( &p_dst[8*FDEC_STRIDE+0], &dct[8] );
254 add8x8_idct( &p_dst[8*FDEC_STRIDE+8], &dct[12] );
257 /****************************************************************************
259 ****************************************************************************/
262 const int s07 = SRC(0) + SRC(7);\
263 const int s16 = SRC(1) + SRC(6);\
264 const int s25 = SRC(2) + SRC(5);\
265 const int s34 = SRC(3) + SRC(4);\
266 const int a0 = s07 + s34;\
267 const int a1 = s16 + s25;\
268 const int a2 = s07 - s34;\
269 const int a3 = s16 - s25;\
270 const int d07 = SRC(0) - SRC(7);\
271 const int d16 = SRC(1) - SRC(6);\
272 const int d25 = SRC(2) - SRC(5);\
273 const int d34 = SRC(3) - SRC(4);\
274 const int a4 = d16 + d25 + (d07 + (d07>>1));\
275 const int a5 = d07 - d34 - (d25 + (d25>>1));\
276 const int a6 = d07 + d34 - (d16 + (d16>>1));\
277 const int a7 = d16 - d25 + (d34 + (d34>>1));\
279 DST(1) = a4 + (a7>>2);\
280 DST(2) = a2 + (a3>>1);\
281 DST(3) = a5 + (a6>>2);\
283 DST(5) = a6 - (a5>>2);\
284 DST(6) = (a2>>1) - a3 ;\
285 DST(7) = (a4>>2) - a7 ;\
288 static void sub8x8_dct8( int16_t dct[8][8], uint8_t *pix1, uint8_t *pix2 )
293 pixel_sub_wxh( (int16_t*)tmp, 8, pix1, FENC_STRIDE, pix2, FDEC_STRIDE );
295 #define SRC(x) tmp[x][i]
296 #define DST(x) tmp[x][i]
297 for( i = 0; i < 8; i++ )
302 #define SRC(x) tmp[i][x]
303 #define DST(x) dct[x][i]
304 for( i = 0; i < 8; i++ )
310 static void sub16x16_dct8( int16_t dct[4][8][8], uint8_t *pix1, uint8_t *pix2 )
312 sub8x8_dct8( dct[0], &pix1[0], &pix2[0] );
313 sub8x8_dct8( dct[1], &pix1[8], &pix2[8] );
314 sub8x8_dct8( dct[2], &pix1[8*FENC_STRIDE+0], &pix2[8*FDEC_STRIDE+0] );
315 sub8x8_dct8( dct[3], &pix1[8*FENC_STRIDE+8], &pix2[8*FDEC_STRIDE+8] );
319 const int a0 = SRC(0) + SRC(4);\
320 const int a2 = SRC(0) - SRC(4);\
321 const int a4 = (SRC(2)>>1) - SRC(6);\
322 const int a6 = (SRC(6)>>1) + SRC(2);\
323 const int b0 = a0 + a6;\
324 const int b2 = a2 + a4;\
325 const int b4 = a2 - a4;\
326 const int b6 = a0 - a6;\
327 const int a1 = -SRC(3) + SRC(5) - SRC(7) - (SRC(7)>>1);\
328 const int a3 = SRC(1) + SRC(7) - SRC(3) - (SRC(3)>>1);\
329 const int a5 = -SRC(1) + SRC(7) + SRC(5) + (SRC(5)>>1);\
330 const int a7 = SRC(3) + SRC(5) + SRC(1) + (SRC(1)>>1);\
331 const int b1 = (a7>>2) + a1;\
332 const int b3 = a3 + (a5>>2);\
333 const int b5 = (a3>>2) - a5;\
334 const int b7 = a7 - (a1>>2);\
345 static void add8x8_idct8( uint8_t *dst, int16_t dct[8][8] )
349 dct[0][0] += 32; // rounding for the >>6 at the end
351 #define SRC(x) dct[x][i]
352 #define DST(x,rhs) dct[x][i] = (rhs)
353 for( i = 0; i < 8; i++ )
358 #define SRC(x) dct[i][x]
359 #define DST(x,rhs) dst[i + x*FDEC_STRIDE] = clip_uint8( dst[i + x*FDEC_STRIDE] + ((rhs) >> 6) );
360 for( i = 0; i < 8; i++ )
366 static void add16x16_idct8( uint8_t *dst, int16_t dct[4][8][8] )
368 add8x8_idct8( &dst[0], dct[0] );
369 add8x8_idct8( &dst[8], dct[1] );
370 add8x8_idct8( &dst[8*FDEC_STRIDE+0], dct[2] );
371 add8x8_idct8( &dst[8*FDEC_STRIDE+8], dct[3] );
375 /****************************************************************************
377 ****************************************************************************/
378 void x264_dct_init( int cpu, x264_dct_function_t *dctf )
380 dctf->sub4x4_dct = sub4x4_dct;
381 dctf->add4x4_idct = add4x4_idct;
383 dctf->sub8x8_dct = sub8x8_dct;
384 dctf->add8x8_idct = add8x8_idct;
386 dctf->sub16x16_dct = sub16x16_dct;
387 dctf->add16x16_idct = add16x16_idct;
389 dctf->sub8x8_dct8 = sub8x8_dct8;
390 dctf->add8x8_idct8 = add8x8_idct8;
392 dctf->sub16x16_dct8 = sub16x16_dct8;
393 dctf->add16x16_idct8 = add16x16_idct8;
395 dctf->dct4x4dc = dct4x4dc;
396 dctf->idct4x4dc = idct4x4dc;
398 dctf->dct2x2dc = dct2x2dc;
399 dctf->idct2x2dc = dct2x2dc;
402 if( cpu&X264_CPU_MMX )
404 dctf->sub4x4_dct = x264_sub4x4_dct_mmx;
405 dctf->sub8x8_dct = x264_sub8x8_dct_mmx;
406 dctf->sub16x16_dct = x264_sub16x16_dct_mmx;
408 dctf->add4x4_idct = x264_add4x4_idct_mmx;
409 dctf->add8x8_idct = x264_add8x8_idct_mmx;
410 dctf->add16x16_idct = x264_add16x16_idct_mmx;
412 dctf->dct4x4dc = x264_dct4x4dc_mmx;
413 dctf->idct4x4dc = x264_idct4x4dc_mmx;
416 dctf->sub8x8_dct8 = x264_sub8x8_dct8_mmx;
417 dctf->sub16x16_dct8 = x264_sub16x16_dct8_mmx;
419 dctf->add8x8_idct8 = x264_add8x8_idct8_mmx;
420 dctf->add16x16_idct8= x264_add16x16_idct8_mmx;
425 #if defined(HAVE_SSE2) && defined(ARCH_X86_64)
426 if( cpu&X264_CPU_SSE2 )
428 dctf->sub8x8_dct8 = x264_sub8x8_dct8_sse2;
429 dctf->sub16x16_dct8 = x264_sub16x16_dct8_sse2;
431 dctf->add8x8_idct8 = x264_add8x8_idct8_sse2;
432 dctf->add16x16_idct8= x264_add16x16_idct8_sse2;
437 if( cpu&X264_CPU_ALTIVEC )
439 dctf->sub4x4_dct = x264_sub4x4_dct_altivec;
440 dctf->sub8x8_dct = x264_sub8x8_dct_altivec;
441 dctf->sub16x16_dct = x264_sub16x16_dct_altivec;
446 void x264_dct_init_weights( void )
451 for( i=0; i<16; i++ )
452 x264_dct4_weight2_zigzag[j][i] = x264_dct4_weight2_tab[ x264_zigzag_scan4[j][i] ];
453 for( i=0; i<64; i++ )
454 x264_dct8_weight2_zigzag[j][i] = x264_dct8_weight2_tab[ x264_zigzag_scan8[j][i] ];