* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
*****************************************************************************/
-#include <string.h>
-
#include "common.h"
-#include "clip1.h"
-#ifdef HAVE_MMXEXT
-# include "i386/pixel.h"
+#ifdef HAVE_MMX
+# include "x86/pixel.h"
#endif
#ifdef ARCH_PPC
# include "ppc/pixel.h"
{
int64_t i_ssd = 0;
int x, y;
+ int align = !(((long)pix1 | (long)pix2 | i_pix1 | i_pix2) & 15);
#define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
pix2 + y*i_pix2 + x, i_pix2 );
for( y = 0; y < i_height-15; y += 16 )
{
- for( x = 0; x < i_width-15; x += 16 )
- SSD(PIXEL_16x16);
- if( x < i_width-7 )
+ x = 0;
+ if( align )
+ for( ; x < i_width-15; x += 16 )
+ SSD(PIXEL_16x16);
+ for( ; x < i_width-7; x += 8 )
SSD(PIXEL_8x16);
}
if( y < i_height-7 )
PIXEL_SA8D_C( 8, 16 )
PIXEL_SA8D_C( 8, 8 )
+/****************************************************************************
+ * pixel_sad_x4
+ ****************************************************************************/
#define SAD_X( size ) \
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] )\
{\
SAD_X( 8x8_vis )
#endif
+/****************************************************************************
+ * pixel_satd_x4
+ * no faster than single satd, but needed for satd to be a drop-in replacement for sad
+ ****************************************************************************/
+
+#define SATD_X( size, cpu ) \
+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] )\
+{\
+ scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
+ scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
+ scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
+}\
+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] )\
+{\
+ scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
+ scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
+ scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
+ scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
+}
+#define SATD_X_DECL5( cpu )\
+SATD_X( 16x16, cpu )\
+SATD_X( 16x8, cpu )\
+SATD_X( 8x16, cpu )\
+SATD_X( 8x8, cpu )\
+SATD_X( 8x4, cpu )
+#define SATD_X_DECL7( cpu )\
+SATD_X_DECL5( cpu )\
+SATD_X( 4x8, cpu )\
+SATD_X( 4x4, cpu )
+
+SATD_X_DECL7()
+#ifdef HAVE_MMX
+SATD_X_DECL7( _mmxext )
+SATD_X_DECL5( _sse2 )
+#ifdef HAVE_SSE3
+SATD_X_DECL7( _ssse3 )
+#endif
+#endif
+
+/****************************************************************************
+ * structural similarity metric
+ ****************************************************************************/
static void ssim_4x4x2_core( const uint8_t *pix1, int stride1,
const uint8_t *pix2, int stride2,
int sums[2][4])
{
int x, y, z;
float ssim = 0.0;
- int sums[2][width/4+3][4];
- int (*sum0)[4] = sums[0];
- int (*sum1)[4] = sums[1];
+ int (*sum0)[4] = x264_malloc(4 * (width/4+3) * sizeof(int));
+ int (*sum1)[4] = x264_malloc(4 * (width/4+3) * sizeof(int));
width >>= 2;
height >>= 2;
z = 0;
for( x = 0; x < width-1; x += 4 )
ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
}
+ x264_free(sum0);
+ x264_free(sum1);
return ssim / ((width-1) * (height-1));
}
+/****************************************************************************
+ * successive elimination
+ ****************************************************************************/
+static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
+ uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
+{
+ int nmv=0, i;
+ for( i=0; i<width; i++, sums++ )
+ {
+ int ads = abs( enc_dc[0] - sums[0] )
+ + abs( enc_dc[1] - sums[8] )
+ + abs( enc_dc[2] - sums[delta] )
+ + abs( enc_dc[3] - sums[delta+8] )
+ + cost_mvx[i];
+ if( ads < thresh )
+ mvs[nmv++] = i;
+ }
+ return nmv;
+}
+
+static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
+ uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
+{
+ int nmv=0, i;
+ for( i=0; i<width; i++, sums++ )
+ {
+ int ads = abs( enc_dc[0] - sums[0] )
+ + abs( enc_dc[1] - sums[delta] )
+ + cost_mvx[i];
+ if( ads < thresh )
+ mvs[nmv++] = i;
+ }
+ return nmv;
+}
+
+static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
+ uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
+{
+ int nmv=0, i;
+ for( i=0; i<width; i++, sums++ )
+ {
+ int ads = abs( enc_dc[0] - sums[0] )
+ + cost_mvx[i];
+ if( ads < thresh )
+ mvs[nmv++] = i;
+ }
+ return nmv;
+}
+
+
/****************************************************************************
* x264_pixel_init:
****************************************************************************/
{
memset( pixf, 0, sizeof(*pixf) );
-#define INIT( name, cpu ) \
+#define INIT2( name, cpu ) \
pixf->name[PIXEL_16x16] = x264_pixel_##name##_16x16##cpu;\
- pixf->name[PIXEL_16x8] = x264_pixel_##name##_16x8##cpu;\
+ pixf->name[PIXEL_16x8] = x264_pixel_##name##_16x8##cpu;
+#define INIT4( name, cpu ) \
+ INIT2( name, cpu ) \
pixf->name[PIXEL_8x16] = x264_pixel_##name##_8x16##cpu;\
- pixf->name[PIXEL_8x8] = x264_pixel_##name##_8x8##cpu;\
- pixf->name[PIXEL_8x4] = x264_pixel_##name##_8x4##cpu;\
+ pixf->name[PIXEL_8x8] = x264_pixel_##name##_8x8##cpu;
+#define INIT5( name, cpu ) \
+ INIT4( name, cpu ) \
+ pixf->name[PIXEL_8x4] = x264_pixel_##name##_8x4##cpu;
+#define INIT7( name, cpu ) \
+ INIT5( name, cpu ) \
pixf->name[PIXEL_4x8] = x264_pixel_##name##_4x8##cpu;\
pixf->name[PIXEL_4x4] = x264_pixel_##name##_4x4##cpu;
- INIT( sad, );
- INIT( sad_x3, );
- INIT( sad_x4, );
- INIT( ssd, );
- INIT( satd, );
+#define INIT_ADS( cpu ) \
+ pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
+ pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
+ pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
+
+ INIT7( sad, );
+ INIT7( sad_x3, );
+ INIT7( sad_x4, );
+ INIT7( ssd, );
+ INIT7( satd, );
+ INIT7( satd_x3, );
+ INIT7( satd_x4, );
+ INIT4( sa8d, );
+ INIT_ADS( );
- pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16;
- pixf->sa8d[PIXEL_16x8] = x264_pixel_sa8d_16x8;
- pixf->sa8d[PIXEL_8x16] = x264_pixel_sa8d_8x16;
- pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8;
pixf->ssim_4x4x2_core = ssim_4x4x2_core;
pixf->ssim_end4 = ssim_end4;
-#ifdef HAVE_MMXEXT
+#ifdef HAVE_MMX
if( cpu&X264_CPU_MMX )
{
- INIT( ssd, _mmx );
+ INIT7( ssd, _mmx );
}
if( cpu&X264_CPU_MMXEXT )
{
- INIT( sad, _mmxext );
- INIT( sad_x3, _mmxext );
- INIT( sad_x4, _mmxext );
- INIT( satd, _mmxext );
-
- pixf->sad_pde[PIXEL_16x16] = x264_pixel_sad_pde_16x16_mmxext;
- pixf->sad_pde[PIXEL_16x8 ] = x264_pixel_sad_pde_16x8_mmxext;
- pixf->sad_pde[PIXEL_8x16 ] = x264_pixel_sad_pde_8x16_mmxext;
+ INIT7( sad, _mmxext );
+ INIT7( sad_x3, _mmxext );
+ INIT7( sad_x4, _mmxext );
+ INIT7( satd, _mmxext );
+ INIT7( satd_x3, _mmxext );
+ INIT7( satd_x4, _mmxext );
+ INIT_ADS( _mmxext );
#ifdef ARCH_X86
pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmxext;
pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8_mmxext;
pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmxext;
pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_mmxext;
+
+ if( cpu&X264_CPU_CACHELINE_SPLIT )
+ {
+ if( cpu&X264_CPU_CACHELINE_32 )
+ {
+ INIT5( sad, _cache32_mmxext );
+ INIT4( sad_x3, _cache32_mmxext );
+ INIT4( sad_x4, _cache32_mmxext );
+ }
+ else
+ {
+ INIT5( sad, _cache64_mmxext );
+ INIT4( sad_x3, _cache64_mmxext );
+ INIT4( sad_x4, _cache64_mmxext );
+ }
+ }
+#else
+ if( cpu&X264_CPU_CACHELINE_SPLIT )
+ {
+ pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmxext;
+ pixf->sad[PIXEL_8x8] = x264_pixel_sad_8x8_cache64_mmxext;
+ pixf->sad[PIXEL_8x4] = x264_pixel_sad_8x4_cache64_mmxext;
+ pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmxext;
+ pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_cache64_mmxext;
+ pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmxext;
+ pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_cache64_mmxext;
+ }
#endif
pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
pixf->intra_satd_x3_8x8c = x264_intra_satd_x3_8x8c_mmxext;
pixf->intra_satd_x3_4x4 = x264_intra_satd_x3_4x4_mmxext;
}
-#endif
-#ifdef HAVE_SSE2
// disable on AMD processors since it is slower
if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_3DNOW) )
{
- pixf->sad[PIXEL_16x16] = x264_pixel_sad_16x16_sse2;
- pixf->sad[PIXEL_16x8 ] = x264_pixel_sad_16x8_sse2;
-
- pixf->satd[PIXEL_16x16]= x264_pixel_satd_16x16_sse2;
- pixf->satd[PIXEL_16x8] = x264_pixel_satd_16x8_sse2;
- pixf->satd[PIXEL_8x16] = x264_pixel_satd_8x16_sse2;
- pixf->satd[PIXEL_8x8] = x264_pixel_satd_8x8_sse2;
- pixf->satd[PIXEL_8x4] = x264_pixel_satd_8x4_sse2;
+ INIT2( sad, _sse2 );
+ INIT2( sad_x3, _sse2 );
+ INIT2( sad_x4, _sse2 );
+ INIT5( satd, _sse2 );
+ INIT5( satd_x3, _sse2 );
+ INIT5( satd_x4, _sse2 );
+ INIT_ADS( _sse2 );
#ifdef ARCH_X86
- pixf->sad_x3[PIXEL_16x16] = x264_pixel_sad_x3_16x16_sse2;
- pixf->sad_x3[PIXEL_16x8 ] = x264_pixel_sad_x3_16x8_sse2;
-
- pixf->sad_x4[PIXEL_16x16] = x264_pixel_sad_x4_16x16_sse2;
- pixf->sad_x4[PIXEL_16x8 ] = x264_pixel_sad_x4_16x8_sse2;
+ if( cpu&X264_CPU_CACHELINE_SPLIT )
+ {
+ INIT2( sad, _cache64_sse2 );
+ INIT2( sad_x3, _cache64_sse2 );
+ INIT2( sad_x4, _cache64_sse2 );
+ }
#endif
}
// these are faster on both Intel and AMD
if( cpu&X264_CPU_SSE2 )
{
- pixf->ssd[PIXEL_16x16] = x264_pixel_ssd_16x16_sse2;
- pixf->ssd[PIXEL_16x8] = x264_pixel_ssd_16x8_sse2;
+ INIT5( ssd, _sse2 );
pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_sse2;
pixf->ssim_end4 = x264_pixel_ssim_end4_sse2;
-
-#ifdef ARCH_X86_64
pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8_sse2;
+#ifdef ARCH_X86_64
pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
#endif
}
+
+#ifdef HAVE_SSE3
+ if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_SPLIT) )
+ {
+ INIT2( sad, _sse3 );
+ INIT2( sad_x3, _sse3 );
+ INIT2( sad_x4, _sse3 );
+ }
+
+ if( cpu&X264_CPU_SSSE3 )
+ {
+ INIT7( satd, _ssse3 );
+ INIT7( satd_x3, _ssse3 );
+ INIT7( satd_x4, _ssse3 );
+ INIT_ADS( _ssse3 );
+ pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
+ pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8_ssse3;
+ pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
+ pixf->intra_satd_x3_8x8c = x264_intra_satd_x3_8x8c_ssse3;
+ pixf->intra_satd_x3_4x4 = x264_intra_satd_x3_4x4_ssse3;
+#ifdef ARCH_X86_64
+ pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_ssse3;
#endif
+ if( cpu&X264_CPU_CACHELINE_SPLIT )
+ {
+ INIT2( sad, _cache64_ssse3 );
+ INIT2( sad_x3, _cache64_ssse3 );
+ INIT2( sad_x4, _cache64_ssse3 );
+ }
+ }
+#endif //HAVE_SSE3
+#endif //HAVE_MMX
#ifdef ARCH_PPC
if( cpu&X264_CPU_ALTIVEC )
}
#endif
#ifdef ARCH_UltraSparc
- pixf->sad[PIXEL_8x8] = x264_pixel_sad_8x8_vis;
- pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_vis;
- pixf->sad[PIXEL_16x8] = x264_pixel_sad_16x8_vis;
- pixf->sad[PIXEL_16x16] = x264_pixel_sad_16x16_vis;
-
- pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_vis;
- pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_vis;
- pixf->sad_x3[PIXEL_16x8] = x264_pixel_sad_x3_16x8_vis;
- pixf->sad_x3[PIXEL_16x16] = x264_pixel_sad_x3_16x16_vis;
-
- pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_vis;
- pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_vis;
- pixf->sad_x4[PIXEL_16x8] = x264_pixel_sad_x4_16x8_vis;
- pixf->sad_x4[PIXEL_16x16] = x264_pixel_sad_x4_16x16_vis;
+ INIT4( sad, _vis );
+ INIT4( sad_x3, _vis );
+ INIT4( sad_x4, _vis );
#endif
+
+ pixf->ads[PIXEL_8x16] =
+ pixf->ads[PIXEL_8x4] =
+ pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
+ pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
}