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"
33 static inline int clip_uint8( int a )
42 * XXX For all dct dc : input could be equal to output so ...
45 static void dct2x2dc( int16_t d[2][2] )
49 tmp[0][0] = d[0][0] + d[0][1];
50 tmp[1][0] = d[0][0] - d[0][1];
51 tmp[0][1] = d[1][0] + d[1][1];
52 tmp[1][1] = d[1][0] - d[1][1];
54 d[0][0] = tmp[0][0] + tmp[0][1];
55 d[1][0] = tmp[1][0] + tmp[1][1];
56 d[0][1] = tmp[0][0] - tmp[0][1];
57 d[1][1] = tmp[1][0] - tmp[1][1];
60 static void dct4x4dc( int16_t d[4][4] )
67 for( i = 0; i < 4; i++ )
69 s01 = d[i][0] + d[i][1];
70 d01 = d[i][0] - d[i][1];
71 s23 = d[i][2] + d[i][3];
72 d23 = d[i][2] - d[i][3];
74 tmp[0][i] = s01 + s23;
75 tmp[1][i] = s01 - s23;
76 tmp[2][i] = d01 - d23;
77 tmp[3][i] = d01 + d23;
80 for( i = 0; i < 4; i++ )
82 s01 = tmp[i][0] + tmp[i][1];
83 d01 = tmp[i][0] - tmp[i][1];
84 s23 = tmp[i][2] + tmp[i][3];
85 d23 = tmp[i][2] - tmp[i][3];
87 d[i][0] = ( s01 + s23 + 1 ) >> 1;
88 d[i][1] = ( s01 - s23 + 1 ) >> 1;
89 d[i][2] = ( d01 - d23 + 1 ) >> 1;
90 d[i][3] = ( d01 + d23 + 1 ) >> 1;
94 static void idct4x4dc( int16_t d[4][4] )
101 for( i = 0; i < 4; i++ )
103 s01 = d[i][0] + d[i][1];
104 d01 = d[i][0] - d[i][1];
105 s23 = d[i][2] + d[i][3];
106 d23 = d[i][2] - d[i][3];
108 tmp[0][i] = s01 + s23;
109 tmp[1][i] = s01 - s23;
110 tmp[2][i] = d01 - d23;
111 tmp[3][i] = d01 + d23;
114 for( i = 0; i < 4; i++ )
116 s01 = tmp[i][0] + tmp[i][1];
117 d01 = tmp[i][0] - tmp[i][1];
118 s23 = tmp[i][2] + tmp[i][3];
119 d23 = tmp[i][2] - tmp[i][3];
128 static inline void pixel_sub_wxh( int16_t *diff, int i_size,
129 uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
132 for( y = 0; y < i_size; y++ )
134 for( x = 0; x < i_size; x++ )
136 diff[x + y*i_size] = pix1[x] - pix2[x];
143 static void sub4x4_dct( int16_t dct[4][4], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
149 pixel_sub_wxh( (int16_t*)d, 4, pix1, i_pix1, pix2, i_pix2 );
151 for( i = 0; i < 4; i++ )
153 const int s03 = d[i][0] + d[i][3];
154 const int s12 = d[i][1] + d[i][2];
155 const int d03 = d[i][0] - d[i][3];
156 const int d12 = d[i][1] - d[i][2];
158 tmp[0][i] = s03 + s12;
159 tmp[1][i] = 2*d03 + d12;
160 tmp[2][i] = s03 - s12;
161 tmp[3][i] = d03 - 2*d12;
164 for( i = 0; i < 4; i++ )
166 const int s03 = tmp[i][0] + tmp[i][3];
167 const int s12 = tmp[i][1] + tmp[i][2];
168 const int d03 = tmp[i][0] - tmp[i][3];
169 const int d12 = tmp[i][1] - tmp[i][2];
171 dct[i][0] = s03 + s12;
172 dct[i][1] = 2*d03 + d12;
173 dct[i][2] = s03 - s12;
174 dct[i][3] = d03 - 2*d12;
178 static void sub8x8_dct( int16_t dct[4][4][4], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
180 sub4x4_dct( dct[0], &pix1[0], i_pix1, &pix2[0], i_pix2 );
181 sub4x4_dct( dct[1], &pix1[4], i_pix1, &pix2[4], i_pix2 );
182 sub4x4_dct( dct[2], &pix1[4*i_pix1+0], i_pix1, &pix2[4*i_pix2+0], i_pix2 );
183 sub4x4_dct( dct[3], &pix1[4*i_pix1+4], i_pix1, &pix2[4*i_pix2+4], i_pix2 );
186 static void sub16x16_dct( int16_t dct[16][4][4], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
188 sub8x8_dct( &dct[ 0], pix1, i_pix1, pix2, i_pix2 );
189 sub8x8_dct( &dct[ 4], &pix1[8], i_pix1, &pix2[8], i_pix2 );
190 sub8x8_dct( &dct[ 8], &pix1[8*i_pix1], i_pix1, &pix2[8*i_pix2], i_pix2 );
191 sub8x8_dct( &dct[12], &pix1[8*i_pix1+8], i_pix1, &pix2[8*i_pix2+8], i_pix2 );
195 static void add4x4_idct( uint8_t *p_dst, int i_dst, int16_t dct[4][4] )
202 for( i = 0; i < 4; i++ )
204 const int s02 = dct[0][i] + dct[2][i];
205 const int d02 = dct[0][i] - dct[2][i];
206 const int s13 = dct[1][i] + (dct[3][i]>>1);
207 const int d13 = (dct[1][i]>>1) - dct[3][i];
209 tmp[i][0] = s02 + s13;
210 tmp[i][1] = d02 + d13;
211 tmp[i][2] = d02 - d13;
212 tmp[i][3] = s02 - s13;
215 for( i = 0; i < 4; i++ )
217 const int s02 = tmp[0][i] + tmp[2][i];
218 const int d02 = tmp[0][i] - tmp[2][i];
219 const int s13 = tmp[1][i] + (tmp[3][i]>>1);
220 const int d13 = (tmp[1][i]>>1) - tmp[3][i];
222 d[0][i] = ( s02 + s13 + 32 ) >> 6;
223 d[1][i] = ( d02 + d13 + 32 ) >> 6;
224 d[2][i] = ( d02 - d13 + 32 ) >> 6;
225 d[3][i] = ( s02 - s13 + 32 ) >> 6;
229 for( y = 0; y < 4; y++ )
231 for( x = 0; x < 4; x++ )
233 p_dst[x] = clip_uint8( p_dst[x] + d[y][x] );
239 static void add8x8_idct( uint8_t *p_dst, int i_dst, int16_t dct[4][4][4] )
241 add4x4_idct( p_dst, i_dst, dct[0] );
242 add4x4_idct( &p_dst[4], i_dst, dct[1] );
243 add4x4_idct( &p_dst[4*i_dst+0], i_dst, dct[2] );
244 add4x4_idct( &p_dst[4*i_dst+4], i_dst, dct[3] );
247 static void add16x16_idct( uint8_t *p_dst, int i_dst, int16_t dct[16][4][4] )
249 add8x8_idct( &p_dst[0], i_dst, &dct[0] );
250 add8x8_idct( &p_dst[8], i_dst, &dct[4] );
251 add8x8_idct( &p_dst[8*i_dst], i_dst, &dct[8] );
252 add8x8_idct( &p_dst[8*i_dst+8], i_dst, &dct[12] );
255 /****************************************************************************
257 ****************************************************************************/
260 const int s07 = SRC(0) + SRC(7);\
261 const int s16 = SRC(1) + SRC(6);\
262 const int s25 = SRC(2) + SRC(5);\
263 const int s34 = SRC(3) + SRC(4);\
264 const int a0 = s07 + s34;\
265 const int a1 = s16 + s25;\
266 const int a2 = s07 - s34;\
267 const int a3 = s16 - s25;\
268 const int d07 = SRC(0) - SRC(7);\
269 const int d16 = SRC(1) - SRC(6);\
270 const int d25 = SRC(2) - SRC(5);\
271 const int d34 = SRC(3) - SRC(4);\
272 const int a4 = d16 + d25 + (d07 + (d07>>1));\
273 const int a5 = d07 - d34 - (d25 + (d25>>1));\
274 const int a6 = d07 + d34 - (d16 + (d16>>1));\
275 const int a7 = d16 - d25 + (d34 + (d34>>1));\
277 DST(1) = a4 + (a7>>2);\
278 DST(2) = a2 + (a3>>1);\
279 DST(3) = a5 + (a6>>2);\
281 DST(5) = a6 - (a5>>2);\
282 DST(6) = (a2>>1) - a3 ;\
283 DST(7) = (a4>>2) - a7 ;\
286 static void sub8x8_dct8( int16_t dct[8][8], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
291 pixel_sub_wxh( (int16_t*)tmp, 8, pix1, i_pix1, pix2, i_pix2 );
293 #define SRC(x) tmp[x][i]
294 #define DST(x) tmp[x][i]
295 for( i = 0; i < 8; i++ )
300 #define SRC(x) tmp[i][x]
301 #define DST(x) dct[x][i]
302 for( i = 0; i < 8; i++ )
308 static void sub16x16_dct8( int16_t dct[4][8][8], uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
310 sub8x8_dct8( dct[0], pix1, i_pix1, pix2, i_pix2 );
311 sub8x8_dct8( dct[1], &pix1[8], i_pix1, &pix2[8], i_pix2 );
312 sub8x8_dct8( dct[2], &pix1[8*i_pix1], i_pix1, &pix2[8*i_pix2], i_pix2 );
313 sub8x8_dct8( dct[3], &pix1[8*i_pix1+8], i_pix1, &pix2[8*i_pix2+8], i_pix2 );
317 const int a0 = SRC(0) + SRC(4);\
318 const int a2 = SRC(0) - SRC(4);\
319 const int a4 = (SRC(2)>>1) - SRC(6);\
320 const int a6 = (SRC(6)>>1) + SRC(2);\
321 const int b0 = a0 + a6;\
322 const int b2 = a2 + a4;\
323 const int b4 = a2 - a4;\
324 const int b6 = a0 - a6;\
325 const int a1 = -SRC(3) + SRC(5) - SRC(7) - (SRC(7)>>1);\
326 const int a3 = SRC(1) + SRC(7) - SRC(3) - (SRC(3)>>1);\
327 const int a5 = -SRC(1) + SRC(7) + SRC(5) + (SRC(5)>>1);\
328 const int a7 = SRC(3) + SRC(5) + SRC(1) + (SRC(1)>>1);\
329 const int b1 = (a7>>2) + a1;\
330 const int b3 = a3 + (a5>>2);\
331 const int b5 = (a3>>2) - a5;\
332 const int b7 = a7 - (a1>>2);\
343 static void add8x8_idct8( uint8_t *dst, int i_dst, int16_t dct[8][8] )
347 dct[0][0] += 32; // rounding for the >>6 at the end
349 #define SRC(x) dct[x][i]
350 #define DST(x,rhs) dct[x][i] = (rhs)
351 for( i = 0; i < 8; i++ )
356 #define SRC(x) dct[i][x]
357 #define DST(x,rhs) dst[i + x*i_dst] = clip_uint8( dst[i + x*i_dst] + ((rhs) >> 6) );
358 for( i = 0; i < 8; i++ )
364 static void add16x16_idct8( uint8_t *dst, int i_dst, int16_t dct[4][8][8] )
366 add8x8_idct8( &dst[0], i_dst, dct[0] );
367 add8x8_idct8( &dst[8], i_dst, dct[1] );
368 add8x8_idct8( &dst[8*i_dst], i_dst, dct[2] );
369 add8x8_idct8( &dst[8*i_dst+8], i_dst, dct[3] );
373 /****************************************************************************
375 ****************************************************************************/
376 void x264_dct_init( int cpu, x264_dct_function_t *dctf )
378 dctf->sub4x4_dct = sub4x4_dct;
379 dctf->add4x4_idct = add4x4_idct;
381 dctf->sub8x8_dct = sub8x8_dct;
382 dctf->add8x8_idct = add8x8_idct;
384 dctf->sub16x16_dct = sub16x16_dct;
385 dctf->add16x16_idct = add16x16_idct;
387 dctf->sub8x8_dct8 = sub8x8_dct8;
388 dctf->add8x8_idct8 = add8x8_idct8;
390 dctf->sub16x16_dct8 = sub16x16_dct8;
391 dctf->add16x16_idct8 = add16x16_idct8;
393 dctf->dct4x4dc = dct4x4dc;
394 dctf->idct4x4dc = idct4x4dc;
396 dctf->dct2x2dc = dct2x2dc;
397 dctf->idct2x2dc = dct2x2dc;
400 if( cpu&X264_CPU_MMX )
402 dctf->sub4x4_dct = x264_sub4x4_dct_mmx;
403 dctf->sub8x8_dct = x264_sub8x8_dct_mmx;
404 dctf->sub16x16_dct = x264_sub16x16_dct_mmx;
406 dctf->add4x4_idct = x264_add4x4_idct_mmx;
407 dctf->add8x8_idct = x264_add8x8_idct_mmx;
408 dctf->add16x16_idct = x264_add16x16_idct_mmx;
410 dctf->dct4x4dc = x264_dct4x4dc_mmx;
411 dctf->idct4x4dc = x264_idct4x4dc_mmx;
414 dctf->sub8x8_dct8 = x264_sub8x8_dct8_mmx;
415 dctf->sub16x16_dct8 = x264_sub16x16_dct8_mmx;
417 dctf->add8x8_idct8 = x264_add8x8_idct8_mmx;
418 dctf->add16x16_idct8= x264_add16x16_idct8_mmx;
423 #if defined(HAVE_SSE2) && defined(ARCH_X86_64)
424 if( cpu&X264_CPU_SSE2 )
426 dctf->sub8x8_dct8 = x264_sub8x8_dct8_sse2;
427 dctf->sub16x16_dct8 = x264_sub16x16_dct8_sse2;
429 dctf->add8x8_idct8 = x264_add8x8_idct8_sse2;
430 dctf->add16x16_idct8= x264_add16x16_idct8_sse2;
434 /* FIXME altivec dct is not transposed yet
436 if( cpu&X264_CPU_ALTIVEC )
438 dctf->sub4x4_dct = x264_sub4x4_dct_altivec;
439 dctf->sub8x8_dct = x264_sub8x8_dct_altivec;
440 dctf->sub16x16_dct = x264_sub16x16_dct_altivec;