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 $
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 *****************************************************************************/
30 # include "i386/pixel.h"
33 # include "ppc/pixel.h"
35 #ifdef ARCH_UltraSparc
36 # include "sparc/pixel.h"
40 /****************************************************************************
42 ****************************************************************************/
43 #define PIXEL_SAD_C( name, lx, ly ) \
44 static int name( uint8_t *pix1, int i_stride_pix1, \
45 uint8_t *pix2, int i_stride_pix2 ) \
49 for( y = 0; y < ly; y++ ) \
51 for( x = 0; x < lx; x++ ) \
53 i_sum += abs( pix1[x] - pix2[x] ); \
55 pix1 += i_stride_pix1; \
56 pix2 += i_stride_pix2; \
62 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
63 PIXEL_SAD_C( x264_pixel_sad_16x8, 16, 8 )
64 PIXEL_SAD_C( x264_pixel_sad_8x16, 8, 16 )
65 PIXEL_SAD_C( x264_pixel_sad_8x8, 8, 8 )
66 PIXEL_SAD_C( x264_pixel_sad_8x4, 8, 4 )
67 PIXEL_SAD_C( x264_pixel_sad_4x8, 4, 8 )
68 PIXEL_SAD_C( x264_pixel_sad_4x4, 4, 4 )
71 /****************************************************************************
73 ****************************************************************************/
74 #define PIXEL_SSD_C( name, lx, ly ) \
75 static int name( uint8_t *pix1, int i_stride_pix1, \
76 uint8_t *pix2, int i_stride_pix2 ) \
80 for( y = 0; y < ly; y++ ) \
82 for( x = 0; x < lx; x++ ) \
84 int d = pix1[x] - pix2[x]; \
87 pix1 += i_stride_pix1; \
88 pix2 += i_stride_pix2; \
93 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
94 PIXEL_SSD_C( x264_pixel_ssd_16x8, 16, 8 )
95 PIXEL_SSD_C( x264_pixel_ssd_8x16, 8, 16 )
96 PIXEL_SSD_C( x264_pixel_ssd_8x8, 8, 8 )
97 PIXEL_SSD_C( x264_pixel_ssd_8x4, 8, 4 )
98 PIXEL_SSD_C( x264_pixel_ssd_4x8, 4, 8 )
99 PIXEL_SSD_C( x264_pixel_ssd_4x4, 4, 4 )
101 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 )
106 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
107 pix2 + y*i_pix2 + x, i_pix2 );
108 for( y = 0; y < i_height-15; y += 16 )
110 for( x = 0; x < i_width-15; x += 16 )
116 for( x = 0; x < i_width-7; x += 8 )
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 )
123 for( y = 0; y < (i_height & ~7); y++ )
124 for( x = i_width & ~7; x < i_width; x++ )
127 if( i_height % 8 != 0 )
129 for( y = i_height & ~7; y < i_height; y++ )
130 for( x = 0; x < i_width; x++ )
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 )
143 for( y = 0; y < i_size; y++ )
145 for( x = 0; x < i_size; x++ )
147 diff[x + y*i_size] = pix1[x] - pix2[x];
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 )
165 for( y = 0; y < i_height; y += 4 )
167 for( x = 0; x < i_width; x += 4 )
171 pixel_sub_wxh( (int16_t*)diff, 4, &pix1[x], i_pix1, &pix2[x], i_pix2 );
173 for( d = 0; d < 4; d++ )
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];
181 tmp[d][0] = s01 + s23;
182 tmp[d][1] = s01 - s23;
183 tmp[d][2] = d01 - d23;
184 tmp[d][3] = d01 + d23;
186 for( d = 0; d < 4; d++ )
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];
194 i_satd += abs( s01 + s23 ) + abs( s01 - s23 ) + abs( d01 - d23 ) + abs( d01 + d23 );
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 ) \
208 return pixel_satd_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ); \
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 )
219 /****************************************************************************
220 * pixel_sa8d_WxH: sum of 8x8 Hadamard transformed differences
221 ****************************************************************************/
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;\
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 )
256 for( y = 0; y < i_height; y += 8 )
258 for( x = 0; x < i_width; x += 8 )
261 pixel_sub_wxh( (int16_t*)diff, 8, pix1+x, i_pix1, pix2+x, i_pix2 );
263 #define SRC(x) diff[i][x]
264 #define DST(x,rhs) diff[i][x] = (rhs)
265 for( i = 0; i < 8; i++ )
270 #define SRC(x) diff[x][i]
271 #define DST(x,rhs) i_satd += abs(rhs)
272 for( i = 0; i < 8; i++ )
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 ) \
288 return ( pixel_sa8d_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ) + 2 ) >> 2; \
290 PIXEL_SA8D_C( 16, 16 )
291 PIXEL_SA8D_C( 16, 8 )
292 PIXEL_SA8D_C( 8, 16 )
295 /****************************************************************************
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] )\
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 );\
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] )\
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 );\
321 #ifdef ARCH_UltraSparc
328 /****************************************************************************
329 * structural similarity metric
330 ****************************************************************************/
331 static void ssim_4x4x2_core( const uint8_t *pix1, int stride1,
332 const uint8_t *pix2, int stride2,
338 uint32_t s1=0, s2=0, ss=0, s12=0;
342 int a = pix1[x+y*stride1];
343 int b = pix2[x+y*stride2];
359 static float ssim_end1( int s1, int s2, int ss, int s12 )
361 static const int ssim_c1 = (int)(.01*.01*255*255*64 + .5);
362 static const int ssim_c2 = (int)(.03*.03*255*255*64*63 + .5);
363 int vars = ss*64 - s1*s1 - s2*s2;
364 int covar = s12*64 - s1*s2;
365 return (float)(2*s1*s2 + ssim_c1) * (float)(2*covar + ssim_c2)\
366 / ((float)(s1*s1 + s2*s2 + ssim_c1) * (float)(vars + ssim_c2));
369 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
373 for( i = 0; i < width; i++ )
374 ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
375 sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
376 sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
377 sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
381 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
382 uint8_t *pix1, int stride1,
383 uint8_t *pix2, int stride2,
384 int width, int height )
388 int (*sum0)[4] = x264_alloca(4 * (width/4+3) * sizeof(int));
389 int (*sum1)[4] = x264_alloca(4 * (width/4+3) * sizeof(int));
393 for( y = 1; y < height; y++ )
397 XCHG( void*, sum0, sum1 );
398 for( x = 0; x < width; x+=2 )
399 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
401 for( x = 0; x < width-1; x += 4 )
402 ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
404 return ssim / ((width-1) * (height-1));
408 /****************************************************************************
409 * successive elimination
410 ****************************************************************************/
411 static void pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
412 uint16_t *res, int width )
415 for( i=0; i<width; i++, sums++ )
416 res[i] = abs( enc_dc[0] - sums[0] )
417 + abs( enc_dc[1] - sums[8] )
418 + abs( enc_dc[2] - sums[delta] )
419 + abs( enc_dc[3] - sums[delta+8] );
422 static void pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
423 uint16_t *res, int width )
426 for( i=0; i<width; i++, sums++ )
427 res[i] = abs( enc_dc[0] - sums[0] )
428 + abs( enc_dc[1] - sums[delta] );
431 static void pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
432 uint16_t *res, int width )
435 for( i=0; i<width; i++, sums++ )
436 res[i] = abs( enc_dc[0] - sums[0] );
440 /****************************************************************************
442 ****************************************************************************/
443 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
445 memset( pixf, 0, sizeof(*pixf) );
447 #define INIT( name, cpu ) \
448 pixf->name[PIXEL_16x16] = x264_pixel_##name##_16x16##cpu;\
449 pixf->name[PIXEL_16x8] = x264_pixel_##name##_16x8##cpu;\
450 pixf->name[PIXEL_8x16] = x264_pixel_##name##_8x16##cpu;\
451 pixf->name[PIXEL_8x8] = x264_pixel_##name##_8x8##cpu;\
452 pixf->name[PIXEL_8x4] = x264_pixel_##name##_8x4##cpu;\
453 pixf->name[PIXEL_4x8] = x264_pixel_##name##_4x8##cpu;\
454 pixf->name[PIXEL_4x4] = x264_pixel_##name##_4x4##cpu;
462 pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16;
463 pixf->sa8d[PIXEL_16x8] = x264_pixel_sa8d_16x8;
464 pixf->sa8d[PIXEL_8x16] = x264_pixel_sa8d_8x16;
465 pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8;
466 pixf->ssim_4x4x2_core = ssim_4x4x2_core;
467 pixf->ssim_end4 = ssim_end4;
469 pixf->ads[PIXEL_16x16] = pixel_ads4;
470 pixf->ads[PIXEL_16x8] = pixel_ads2;
471 pixf->ads[PIXEL_8x8] = pixel_ads1;
474 if( cpu&X264_CPU_MMX )
479 if( cpu&X264_CPU_MMXEXT )
481 INIT( sad, _mmxext );
482 INIT( sad_x3, _mmxext );
483 INIT( sad_x4, _mmxext );
484 INIT( satd, _mmxext );
486 pixf->sad_pde[PIXEL_16x16] = x264_pixel_sad_pde_16x16_mmxext;
487 pixf->sad_pde[PIXEL_16x8 ] = x264_pixel_sad_pde_16x8_mmxext;
488 pixf->sad_pde[PIXEL_8x16 ] = x264_pixel_sad_pde_8x16_mmxext;
490 pixf->ads[PIXEL_16x16] = x264_pixel_ads4_mmxext;
491 pixf->ads[PIXEL_16x8 ] = x264_pixel_ads2_mmxext;
492 pixf->ads[PIXEL_8x8 ] = x264_pixel_ads1_mmxext;
495 pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmxext;
496 pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8_mmxext;
497 pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmxext;
498 pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_mmxext;
500 pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
501 pixf->intra_satd_x3_8x8c = x264_intra_satd_x3_8x8c_mmxext;
502 pixf->intra_satd_x3_4x4 = x264_intra_satd_x3_4x4_mmxext;
505 // disable on AMD processors since it is slower
506 if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_3DNOW) )
508 pixf->sad[PIXEL_16x16] = x264_pixel_sad_16x16_sse2;
509 pixf->sad[PIXEL_16x8 ] = x264_pixel_sad_16x8_sse2;
511 pixf->satd[PIXEL_16x16]= x264_pixel_satd_16x16_sse2;
512 pixf->satd[PIXEL_16x8] = x264_pixel_satd_16x8_sse2;
513 pixf->satd[PIXEL_8x16] = x264_pixel_satd_8x16_sse2;
514 pixf->satd[PIXEL_8x8] = x264_pixel_satd_8x8_sse2;
515 pixf->satd[PIXEL_8x4] = x264_pixel_satd_8x4_sse2;
518 pixf->sad_x3[PIXEL_16x16] = x264_pixel_sad_x3_16x16_sse2;
519 pixf->sad_x3[PIXEL_16x8 ] = x264_pixel_sad_x3_16x8_sse2;
521 pixf->sad_x4[PIXEL_16x16] = x264_pixel_sad_x4_16x16_sse2;
522 pixf->sad_x4[PIXEL_16x8 ] = x264_pixel_sad_x4_16x8_sse2;
525 // these are faster on both Intel and AMD
526 if( cpu&X264_CPU_SSE2 )
528 pixf->ssd[PIXEL_16x16] = x264_pixel_ssd_16x16_sse2;
529 pixf->ssd[PIXEL_16x8] = x264_pixel_ssd_16x8_sse2;
530 pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_sse2;
531 pixf->ssim_end4 = x264_pixel_ssim_end4_sse2;
534 pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
535 pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8_sse2;
536 pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
540 if( cpu&X264_CPU_SSSE3 )
543 pixf->satd[PIXEL_16x16]= x264_pixel_satd_16x16_ssse3;
544 pixf->satd[PIXEL_16x8] = x264_pixel_satd_16x8_ssse3;
545 pixf->satd[PIXEL_8x16] = x264_pixel_satd_8x16_ssse3;
546 pixf->satd[PIXEL_8x8] = x264_pixel_satd_8x8_ssse3;
547 pixf->satd[PIXEL_8x4] = x264_pixel_satd_8x4_ssse3;
549 pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
550 pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8_ssse3;
557 if( cpu&X264_CPU_ALTIVEC )
559 x264_pixel_altivec_init( pixf );
562 #ifdef ARCH_UltraSparc
563 pixf->sad[PIXEL_8x8] = x264_pixel_sad_8x8_vis;
564 pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_vis;
565 pixf->sad[PIXEL_16x8] = x264_pixel_sad_16x8_vis;
566 pixf->sad[PIXEL_16x16] = x264_pixel_sad_16x16_vis;
568 pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_vis;
569 pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_vis;
570 pixf->sad_x3[PIXEL_16x8] = x264_pixel_sad_x3_16x8_vis;
571 pixf->sad_x3[PIXEL_16x16] = x264_pixel_sad_x3_16x16_vis;
573 pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_vis;
574 pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_vis;
575 pixf->sad_x4[PIXEL_16x8] = x264_pixel_sad_x4_16x8_vis;
576 pixf->sad_x4[PIXEL_16x16] = x264_pixel_sad_x4_16x16_vis;
579 pixf->ads[PIXEL_8x16] =
580 pixf->ads[PIXEL_8x4] =
581 pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
582 pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];