drop support for pre-SSE3 assemblers
[x262.git] / common / pixel.c
1 /*****************************************************************************
2  * pixel.c: h264 encoder
3  *****************************************************************************
4  * Copyright (C) 2003 Laurent Aimar
5  * $Id: pixel.c,v 1.1 2004/06/03 19:27:07 fenrir Exp $
6  *
7  * Authors: Laurent Aimar <fenrir@via.ecp.fr>
8  *
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.
13  *
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.
18  *
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  *****************************************************************************/
23
24 #include "common.h"
25
26 #ifdef HAVE_MMX
27 #   include "x86/pixel.h"
28 #endif
29 #ifdef ARCH_PPC
30 #   include "ppc/pixel.h"
31 #endif
32 #ifdef ARCH_UltraSparc
33 #   include "sparc/pixel.h"
34 #endif
35
36
37 /****************************************************************************
38  * pixel_sad_WxH
39  ****************************************************************************/
40 #define PIXEL_SAD_C( name, lx, ly ) \
41 static int name( uint8_t *pix1, int i_stride_pix1,  \
42                  uint8_t *pix2, int i_stride_pix2 ) \
43 {                                                   \
44     int i_sum = 0;                                  \
45     int x, y;                                       \
46     for( y = 0; y < ly; y++ )                       \
47     {                                               \
48         for( x = 0; x < lx; x++ )                   \
49         {                                           \
50             i_sum += abs( pix1[x] - pix2[x] );      \
51         }                                           \
52         pix1 += i_stride_pix1;                      \
53         pix2 += i_stride_pix2;                      \
54     }                                               \
55     return i_sum;                                   \
56 }
57
58
59 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
60 PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
61 PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
62 PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
63 PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
64 PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
65 PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
66
67
68 /****************************************************************************
69  * pixel_ssd_WxH
70  ****************************************************************************/
71 #define PIXEL_SSD_C( name, lx, ly ) \
72 static int name( uint8_t *pix1, int i_stride_pix1,  \
73                  uint8_t *pix2, int i_stride_pix2 ) \
74 {                                                   \
75     int i_sum = 0;                                  \
76     int x, y;                                       \
77     for( y = 0; y < ly; y++ )                       \
78     {                                               \
79         for( x = 0; x < lx; x++ )                   \
80         {                                           \
81             int d = pix1[x] - pix2[x];              \
82             i_sum += d*d;                           \
83         }                                           \
84         pix1 += i_stride_pix1;                      \
85         pix2 += i_stride_pix2;                      \
86     }                                               \
87     return i_sum;                                   \
88 }
89
90 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
91 PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
92 PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
93 PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
94 PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
95 PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
96 PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
97
98 int64_t x264_pixel_ssd_wxh( x264_pixel_function_t *pf, uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2, int i_width, int i_height )
99 {
100     int64_t i_ssd = 0;
101     int x, y;
102     int align = !(((long)pix1 | (long)pix2 | i_pix1 | i_pix2) & 15);
103
104 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
105                                           pix2 + y*i_pix2 + x, i_pix2 );
106     for( y = 0; y < i_height-15; y += 16 )
107     {
108         x = 0;
109         if( align )
110             for( ; x < i_width-15; x += 16 )
111                 SSD(PIXEL_16x16);
112         for( ; x < i_width-7; x += 8 )
113             SSD(PIXEL_8x16);
114     }
115     if( y < i_height-7 )
116         for( x = 0; x < i_width-7; x += 8 )
117             SSD(PIXEL_8x8);
118 #undef SSD
119
120 #define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
121     if( i_width % 8 != 0 )
122     {
123         for( y = 0; y < (i_height & ~7); y++ )
124             for( x = i_width & ~7; x < i_width; x++ )
125                 SSD1;
126     }
127     if( i_height % 8 != 0 )
128     {
129         for( y = i_height & ~7; y < i_height; y++ )
130             for( x = 0; x < i_width; x++ )
131                 SSD1;
132     }
133 #undef SSD1
134
135     return i_ssd;
136 }
137
138
139 static inline void pixel_sub_wxh( int16_t *diff, int i_size,
140                                   uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
141 {
142     int y, x;
143     for( y = 0; y < i_size; y++ )
144     {
145         for( x = 0; x < i_size; x++ )
146         {
147             diff[x + y*i_size] = pix1[x] - pix2[x];
148         }
149         pix1 += i_pix1;
150         pix2 += i_pix2;
151     }
152 }
153
154
155 /****************************************************************************
156  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
157  ****************************************************************************/
158 static int pixel_satd_wxh( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2, int i_width, int i_height )
159 {
160     int16_t tmp[4][4];
161     int16_t diff[4][4];
162     int x, y;
163     int i_satd = 0;
164
165     for( y = 0; y < i_height; y += 4 )
166     {
167         for( x = 0; x < i_width; x += 4 )
168         {
169             int d;
170
171             pixel_sub_wxh( (int16_t*)diff, 4, &pix1[x], i_pix1, &pix2[x], i_pix2 );
172
173             for( d = 0; d < 4; d++ )
174             {
175                 int s01, s23;
176                 int d01, d23;
177
178                 s01 = diff[d][0] + diff[d][1]; s23 = diff[d][2] + diff[d][3];
179                 d01 = diff[d][0] - diff[d][1]; d23 = diff[d][2] - diff[d][3];
180
181                 tmp[d][0] = s01 + s23;
182                 tmp[d][1] = s01 - s23;
183                 tmp[d][2] = d01 - d23;
184                 tmp[d][3] = d01 + d23;
185             }
186             for( d = 0; d < 4; d++ )
187             {
188                 int s01, s23;
189                 int d01, d23;
190
191                 s01 = tmp[0][d] + tmp[1][d]; s23 = tmp[2][d] + tmp[3][d];
192                 d01 = tmp[0][d] - tmp[1][d]; d23 = tmp[2][d] - tmp[3][d];
193
194                 i_satd += abs( s01 + s23 ) + abs( s01 - s23 ) + abs( d01 - d23 ) + abs( d01 + d23 );
195             }
196
197         }
198         pix1 += 4 * i_pix1;
199         pix2 += 4 * i_pix2;
200     }
201
202     return i_satd / 2;
203 }
204 #define PIXEL_SATD_C( name, width, height ) \
205 static int name( uint8_t *pix1, int i_stride_pix1, \
206                  uint8_t *pix2, int i_stride_pix2 ) \
207 { \
208     return pixel_satd_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ); \
209 }
210 PIXEL_SATD_C( x264_pixel_satd_16x16, 16, 16 )
211 PIXEL_SATD_C( x264_pixel_satd_16x8,  16, 8 )
212 PIXEL_SATD_C( x264_pixel_satd_8x16,  8, 16 )
213 PIXEL_SATD_C( x264_pixel_satd_8x8,   8, 8 )
214 PIXEL_SATD_C( x264_pixel_satd_8x4,   8, 4 )
215 PIXEL_SATD_C( x264_pixel_satd_4x8,   4, 8 )
216 PIXEL_SATD_C( x264_pixel_satd_4x4,   4, 4 )
217
218
219 /****************************************************************************
220  * pixel_sa8d_WxH: sum of 8x8 Hadamard transformed differences
221  ****************************************************************************/
222 #define SA8D_1D {\
223     const int a0 = SRC(0) + SRC(4);\
224     const int a4 = SRC(0) - SRC(4);\
225     const int a1 = SRC(1) + SRC(5);\
226     const int a5 = SRC(1) - SRC(5);\
227     const int a2 = SRC(2) + SRC(6);\
228     const int a6 = SRC(2) - SRC(6);\
229     const int a3 = SRC(3) + SRC(7);\
230     const int a7 = SRC(3) - SRC(7);\
231     const int b0 = a0 + a2;\
232     const int b2 = a0 - a2;\
233     const int b1 = a1 + a3;\
234     const int b3 = a1 - a3;\
235     const int b4 = a4 + a6;\
236     const int b6 = a4 - a6;\
237     const int b5 = a5 + a7;\
238     const int b7 = a5 - a7;\
239     DST(0, b0 + b1);\
240     DST(1, b0 - b1);\
241     DST(2, b2 + b3);\
242     DST(3, b2 - b3);\
243     DST(4, b4 + b5);\
244     DST(5, b4 - b5);\
245     DST(6, b6 + b7);\
246     DST(7, b6 - b7);\
247 }
248
249 static inline int pixel_sa8d_wxh( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2,
250                                   int i_width, int i_height )
251 {
252     int16_t diff[8][8];
253     int i_satd = 0;
254     int x, y;
255
256     for( y = 0; y < i_height; y += 8 )
257     {
258         for( x = 0; x < i_width; x += 8 )
259         {
260             int i;
261             pixel_sub_wxh( (int16_t*)diff, 8, pix1+x, i_pix1, pix2+x, i_pix2 );
262
263 #define SRC(x)     diff[i][x]
264 #define DST(x,rhs) diff[i][x] = (rhs)
265             for( i = 0; i < 8; i++ )
266                 SA8D_1D
267 #undef SRC
268 #undef DST
269
270 #define SRC(x)     diff[x][i]
271 #define DST(x,rhs) i_satd += abs(rhs)
272             for( i = 0; i < 8; i++ )
273                 SA8D_1D
274 #undef SRC
275 #undef DST
276         }
277         pix1 += 8 * i_pix1;
278         pix2 += 8 * i_pix2;
279     }
280
281     return i_satd;
282 }
283
284 #define PIXEL_SA8D_C( width, height ) \
285 static int x264_pixel_sa8d_##width##x##height( uint8_t *pix1, int i_stride_pix1, \
286                                                uint8_t *pix2, int i_stride_pix2 ) \
287 { \
288     return ( pixel_sa8d_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ) + 2 ) >> 2; \
289 }
290 PIXEL_SA8D_C( 16, 16 )
291 PIXEL_SA8D_C( 16, 8 )
292 PIXEL_SA8D_C( 8, 16 )
293 PIXEL_SA8D_C( 8, 8 )
294
295 /****************************************************************************
296  * pixel_sad_x4
297  ****************************************************************************/
298 #define SAD_X( size ) \
299 static void x264_pixel_sad_x3_##size( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, int i_stride, int scores[3] )\
300 {\
301     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
302     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
303     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
304 }\
305 static void x264_pixel_sad_x4_##size( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, uint8_t *pix3, int i_stride, int scores[4] )\
306 {\
307     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
308     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
309     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
310     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
311 }
312
313 SAD_X( 16x16 )
314 SAD_X( 16x8 )
315 SAD_X( 8x16 )
316 SAD_X( 8x8 )
317 SAD_X( 8x4 )
318 SAD_X( 4x8 )
319 SAD_X( 4x4 )
320
321 #ifdef ARCH_UltraSparc
322 SAD_X( 16x16_vis )
323 SAD_X( 16x8_vis )
324 SAD_X( 8x16_vis )
325 SAD_X( 8x8_vis )
326 #endif
327
328 /****************************************************************************
329  * pixel_satd_x4
330  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
331  ****************************************************************************/
332
333 #define SATD_X( size, cpu ) \
334 static void x264_pixel_satd_x3_##size##cpu( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, int i_stride, int scores[3] )\
335 {\
336     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
337     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
338     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
339 }\
340 static void x264_pixel_satd_x4_##size##cpu( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, uint8_t *pix3, int i_stride, int scores[4] )\
341 {\
342     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
343     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
344     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
345     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
346 }
347 #define SATD_X_DECL5( cpu )\
348 SATD_X( 16x16, cpu )\
349 SATD_X( 16x8, cpu )\
350 SATD_X( 8x16, cpu )\
351 SATD_X( 8x8, cpu )\
352 SATD_X( 8x4, cpu )
353 #define SATD_X_DECL7( cpu )\
354 SATD_X_DECL5( cpu )\
355 SATD_X( 4x8, cpu )\
356 SATD_X( 4x4, cpu )
357
358 SATD_X_DECL7()
359 #ifdef HAVE_MMX
360 SATD_X_DECL7( _mmxext )
361 SATD_X_DECL5( _sse2 )
362 SATD_X_DECL7( _ssse3 )
363 #endif
364
365 /****************************************************************************
366  * structural similarity metric
367  ****************************************************************************/
368 static void ssim_4x4x2_core( const uint8_t *pix1, int stride1,
369                              const uint8_t *pix2, int stride2,
370                              int sums[2][4])
371 {
372     int x, y, z;
373     for(z=0; z<2; z++)
374     {
375         uint32_t s1=0, s2=0, ss=0, s12=0;
376         for(y=0; y<4; y++)
377             for(x=0; x<4; x++)
378             {
379                 int a = pix1[x+y*stride1];
380                 int b = pix2[x+y*stride2];
381                 s1  += a;
382                 s2  += b;
383                 ss  += a*a;
384                 ss  += b*b;
385                 s12 += a*b;
386             }
387         sums[z][0] = s1;
388         sums[z][1] = s2;
389         sums[z][2] = ss;
390         sums[z][3] = s12;
391         pix1 += 4;
392         pix2 += 4;
393     }
394 }
395
396 static float ssim_end1( int s1, int s2, int ss, int s12 )
397 {
398     static const int ssim_c1 = (int)(.01*.01*255*255*64 + .5);
399     static const int ssim_c2 = (int)(.03*.03*255*255*64*63 + .5);
400     int vars = ss*64 - s1*s1 - s2*s2;
401     int covar = s12*64 - s1*s2;
402     return (float)(2*s1*s2 + ssim_c1) * (float)(2*covar + ssim_c2)\
403            / ((float)(s1*s1 + s2*s2 + ssim_c1) * (float)(vars + ssim_c2));
404 }
405
406 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
407 {
408     int i;
409     float ssim = 0.0;
410     for( i = 0; i < width; i++ )
411         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
412                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
413                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
414                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
415     return ssim;
416 }
417
418 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
419                            uint8_t *pix1, int stride1,
420                            uint8_t *pix2, int stride2,
421                            int width, int height )
422 {
423     int x, y, z;
424     float ssim = 0.0;
425     int (*sum0)[4] = x264_malloc(4 * (width/4+3) * sizeof(int));
426     int (*sum1)[4] = x264_malloc(4 * (width/4+3) * sizeof(int));
427     width >>= 2;
428     height >>= 2;
429     z = 0;
430     for( y = 1; y < height; y++ )
431     {
432         for( ; z <= y; z++ )
433         {
434             XCHG( void*, sum0, sum1 );
435             for( x = 0; x < width; x+=2 )
436                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
437         }
438         for( x = 0; x < width-1; x += 4 )
439             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
440     }
441     x264_free(sum0);
442     x264_free(sum1);
443     return ssim / ((width-1) * (height-1));
444 }
445
446
447 /****************************************************************************
448  * successive elimination
449  ****************************************************************************/
450 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
451                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
452 {
453     int nmv=0, i;
454     for( i=0; i<width; i++, sums++ )
455     {
456         int ads = abs( enc_dc[0] - sums[0] )
457                 + abs( enc_dc[1] - sums[8] )
458                 + abs( enc_dc[2] - sums[delta] )
459                 + abs( enc_dc[3] - sums[delta+8] )
460                 + cost_mvx[i];
461         if( ads < thresh )
462             mvs[nmv++] = i;
463     }
464     return nmv;
465 }
466
467 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
468                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
469 {
470     int nmv=0, i;
471     for( i=0; i<width; i++, sums++ )
472     {
473         int ads = abs( enc_dc[0] - sums[0] )
474                 + abs( enc_dc[1] - sums[delta] )
475                 + cost_mvx[i];
476         if( ads < thresh )
477             mvs[nmv++] = i;
478     }
479     return nmv;
480 }
481
482 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
483                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
484 {
485     int nmv=0, i;
486     for( i=0; i<width; i++, sums++ )
487     {
488         int ads = abs( enc_dc[0] - sums[0] )
489                 + cost_mvx[i];
490         if( ads < thresh )
491             mvs[nmv++] = i;
492     }
493     return nmv;
494 }
495
496
497 /****************************************************************************
498  * x264_pixel_init:
499  ****************************************************************************/
500 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
501 {
502     memset( pixf, 0, sizeof(*pixf) );
503
504 #define INIT2( name, cpu ) \
505     pixf->name[PIXEL_16x16] = x264_pixel_##name##_16x16##cpu;\
506     pixf->name[PIXEL_16x8]  = x264_pixel_##name##_16x8##cpu;
507 #define INIT4( name, cpu ) \
508     INIT2( name, cpu ) \
509     pixf->name[PIXEL_8x16]  = x264_pixel_##name##_8x16##cpu;\
510     pixf->name[PIXEL_8x8]   = x264_pixel_##name##_8x8##cpu;
511 #define INIT5( name, cpu ) \
512     INIT4( name, cpu ) \
513     pixf->name[PIXEL_8x4]   = x264_pixel_##name##_8x4##cpu;
514 #define INIT7( name, cpu ) \
515     INIT5( name, cpu ) \
516     pixf->name[PIXEL_4x8]   = x264_pixel_##name##_4x8##cpu;\
517     pixf->name[PIXEL_4x4]   = x264_pixel_##name##_4x4##cpu;
518
519 #define INIT_ADS( cpu ) \
520     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
521     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
522     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
523
524     INIT7( sad, );
525     INIT7( sad_x3, );
526     INIT7( sad_x4, );
527     INIT7( ssd, );
528     INIT7( satd, );
529     INIT7( satd_x3, );
530     INIT7( satd_x4, );
531     INIT4( sa8d, );
532     INIT_ADS( );
533
534     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
535     pixf->ssim_end4 = ssim_end4;
536
537 #ifdef HAVE_MMX
538     if( cpu&X264_CPU_MMX )
539     {
540         INIT7( ssd, _mmx );
541     }
542
543     if( cpu&X264_CPU_MMXEXT )
544     {
545         INIT7( sad, _mmxext );
546         INIT7( sad_x3, _mmxext );
547         INIT7( sad_x4, _mmxext );
548         INIT7( satd, _mmxext );
549         INIT7( satd_x3, _mmxext );
550         INIT7( satd_x4, _mmxext );
551         INIT_ADS( _mmxext );
552
553 #ifdef ARCH_X86
554         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmxext;
555         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmxext;
556         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmxext;
557         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_mmxext;
558
559         if( cpu&X264_CPU_CACHELINE_SPLIT )
560         {
561             if( cpu&X264_CPU_CACHELINE_32 )
562             {
563                 INIT5( sad, _cache32_mmxext );
564                 INIT4( sad_x3, _cache32_mmxext );
565                 INIT4( sad_x4, _cache32_mmxext );
566             }
567             else
568             {
569                 INIT5( sad, _cache64_mmxext );
570                 INIT4( sad_x3, _cache64_mmxext );
571                 INIT4( sad_x4, _cache64_mmxext );
572             }
573         }
574 #else
575         if( cpu&X264_CPU_CACHELINE_SPLIT )
576         {
577             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmxext;
578             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmxext;
579             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmxext;
580             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmxext;
581             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmxext;
582             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmxext;
583             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmxext;
584         }
585 #endif
586         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
587         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
588         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
589     }
590
591     // disable on AMD processors since it is slower
592     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_3DNOW) )
593     {
594         INIT2( sad, _sse2 );
595         INIT2( sad_x3, _sse2 );
596         INIT2( sad_x4, _sse2 );
597         INIT5( satd, _sse2 );
598         INIT5( satd_x3, _sse2 );
599         INIT5( satd_x4, _sse2 );
600         INIT_ADS( _sse2 );
601
602 #ifdef ARCH_X86
603         if( cpu&X264_CPU_CACHELINE_SPLIT )
604         {
605             INIT2( sad, _cache64_sse2 );
606             INIT2( sad_x3, _cache64_sse2 );
607             INIT2( sad_x4, _cache64_sse2 );
608         }
609 #endif
610     }
611     // these are faster on both Intel and AMD
612     if( cpu&X264_CPU_SSE2 )
613     {
614         INIT5( ssd, _sse2 );
615         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
616         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
617         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
618         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
619 #ifdef ARCH_X86_64
620         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
621 #endif
622     }
623
624     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_SPLIT) )
625     {
626         INIT2( sad, _sse3 );
627         INIT2( sad_x3, _sse3 );
628         INIT2( sad_x4, _sse3 );
629     }
630
631     if( cpu&X264_CPU_SSSE3 )
632     {
633         INIT7( satd, _ssse3 );
634         INIT7( satd_x3, _ssse3 );
635         INIT7( satd_x4, _ssse3 );
636         INIT_ADS( _ssse3 );
637         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
638         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
639         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
640         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
641         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_ssse3;
642 #ifdef ARCH_X86_64
643         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_ssse3;
644 #endif
645         if( cpu&X264_CPU_CACHELINE_SPLIT )
646         {
647             INIT2( sad, _cache64_ssse3 );
648             INIT2( sad_x3, _cache64_ssse3 );
649             INIT2( sad_x4, _cache64_ssse3 );
650         }
651     }
652 #endif //HAVE_MMX
653
654 #ifdef ARCH_PPC
655     if( cpu&X264_CPU_ALTIVEC )
656     {
657         x264_pixel_altivec_init( pixf );
658     }
659 #endif
660 #ifdef ARCH_UltraSparc
661     INIT4( sad, _vis );
662     INIT4( sad_x3, _vis );
663     INIT4( sad_x4, _vis );
664 #endif
665
666     pixf->ads[PIXEL_8x16] =
667     pixf->ads[PIXEL_8x4] =
668     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
669     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
670 }
671