/*****************************************************************************
* pixel.c: pixel metrics
*****************************************************************************
- * Copyright (C) 2003-2013 x264 project
+ * Copyright (C) 2003-2016 x264 project
*
* Authors: Loren Merritt <lorenm@u.washington.edu>
* Laurent Aimar <fenrir@via.ecp.fr>
#endif
#if ARCH_ARM
# include "arm/pixel.h"
+# include "arm/predict.h"
#endif
-#if ARCH_UltraSPARC
-# include "sparc/pixel.h"
+#if ARCH_AARCH64
+# include "aarch64/pixel.h"
+# include "aarch64/predict.h"
+#endif
+#if ARCH_MIPS
+# include "mips/pixel.h"
#endif
#define PIXEL_VAR2_C( name, w, h, shift ) \
static int name( pixel *pix1, intptr_t i_stride1, pixel *pix2, intptr_t i_stride2, int *ssd ) \
{ \
- uint32_t var = 0, sum = 0, sqr = 0; \
+ int var = 0, sum = 0, sqr = 0; \
for( int y = 0; y < h; y++ ) \
{ \
for( int x = 0; x < w; x++ ) \
pix1 += i_stride1; \
pix2 += i_stride2; \
} \
- sum = abs(sum); \
- var = sqr - ((uint64_t)sum * sum >> shift); \
+ var = sqr - ((int64_t)sum * sum >> shift); \
*ssd = sqr; \
return var; \
}
SAD_X( 4x8 )
SAD_X( 4x4 )
-#if !HIGH_BIT_DEPTH
-#if ARCH_UltraSPARC
-SAD_X( 16x16_vis )
-SAD_X( 16x8_vis )
-SAD_X( 8x16_vis )
-SAD_X( 8x8_vis )
-#endif
-#endif // !HIGH_BIT_DEPTH
-
/****************************************************************************
* pixel_satd_x4
* no faster than single satd, but needed for satd to be a drop-in replacement for sad
#endif
#if !HIGH_BIT_DEPTH
-#if HAVE_ARMV6
+#if HAVE_ARMV6 || ARCH_AARCH64
SATD_X_DECL7( _neon )
#endif
#endif // !HIGH_BIT_DEPTH
INTRA_MBCMP_8x8( sad, _mmx2, _c )
INTRA_MBCMP_8x8(sa8d, _sse2, _sse2 )
#endif
+#if !HIGH_BIT_DEPTH && (HAVE_ARMV6 || ARCH_AARCH64)
+INTRA_MBCMP_8x8( sad, _neon, _neon )
+INTRA_MBCMP_8x8(sa8d, _neon, _neon )
+#endif
#define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma, cpu, cpu2 )\
void x264_intra_##mbcmp##_x3_##size##chroma##cpu( pixel *fenc, pixel *fdec, int res[3] )\
#if HAVE_MMX
#if HIGH_BIT_DEPTH
+#define x264_predict_8x8c_v_mmx2 x264_predict_8x8c_v_mmx
+#define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_c
#define x264_predict_8x8c_v_sse2 x264_predict_8x8c_v_sse
#define x264_predict_8x16c_v_sse2 x264_predict_8x16c_v_sse
#define x264_predict_16x16_v_sse2 x264_predict_16x16_v_sse
INTRA_MBCMP( sad, 4x4, v, h, dc, , _mmx2, _c )
-INTRA_MBCMP( sad, 8x8, dc, h, v, c, _mmx2, _c )
+INTRA_MBCMP( sad, 8x8, dc, h, v, c, _mmx2, _mmx2 )
+INTRA_MBCMP( sad, 8x16, dc, h, v, c, _mmx2, _mmx2 )
+INTRA_MBCMP(satd, 8x16, dc, h, v, c, _mmx2, _mmx2 )
INTRA_MBCMP( sad, 16x16, v, h, dc, , _mmx2, _mmx2 )
INTRA_MBCMP( sad, 8x8, dc, h, v, c, _sse2, _sse2 )
+INTRA_MBCMP( sad, 8x16, dc, h, v, c, _sse2, _sse2 )
+INTRA_MBCMP(satd, 8x16, dc, h, v, c, _sse2, _sse2 )
INTRA_MBCMP( sad, 16x16, v, h, dc, , _sse2, _sse2 )
INTRA_MBCMP( sad, 8x8, dc, h, v, c, _ssse3, _sse2 )
+INTRA_MBCMP( sad, 8x16, dc, h, v, c, _ssse3, _sse2 )
+INTRA_MBCMP(satd, 8x16, dc, h, v, c, _ssse3, _sse2 )
INTRA_MBCMP( sad, 16x16, v, h, dc, , _ssse3, _sse2 )
+INTRA_MBCMP(satd, 8x16, dc, h, v, c, _sse4, _sse2 )
+INTRA_MBCMP(satd, 8x16, dc, h, v, c, _avx, _sse2 )
#else
#define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_mmx
INTRA_MBCMP( sad, 8x16, dc, h, v, c, _mmx2, _mmx2 )
INTRA_MBCMP(satd, 8x16, dc, h, v, c, _xop, _mmx2 )
#endif
#endif
+#if !HIGH_BIT_DEPTH && HAVE_ARMV6
+INTRA_MBCMP( sad, 4x4, v, h, dc, , _neon, _armv6 )
+INTRA_MBCMP(satd, 4x4, v, h, dc, , _neon, _armv6 )
+INTRA_MBCMP( sad, 8x8, dc, h, v, c, _neon, _neon )
+INTRA_MBCMP(satd, 8x8, dc, h, v, c, _neon, _neon )
+INTRA_MBCMP( sad, 8x16, dc, h, v, c, _neon, _c )
+INTRA_MBCMP(satd, 8x16, dc, h, v, c, _neon, _c )
+INTRA_MBCMP( sad, 16x16, v, h, dc, , _neon, _neon )
+INTRA_MBCMP(satd, 16x16, v, h, dc, , _neon, _neon )
+#endif
+#if !HIGH_BIT_DEPTH && ARCH_AARCH64
+INTRA_MBCMP( sad, 4x4, v, h, dc, , _neon, _neon )
+INTRA_MBCMP(satd, 4x4, v, h, dc, , _neon, _neon )
+INTRA_MBCMP( sad, 8x8, dc, h, v, c, _neon, _neon )
+INTRA_MBCMP(satd, 8x8, dc, h, v, c, _neon, _neon )
+INTRA_MBCMP( sad, 8x16, dc, h, v, c, _neon, _neon )
+INTRA_MBCMP(satd, 8x16, dc, h, v, c, _neon, _neon )
+INTRA_MBCMP( sad, 16x16, v, h, dc, , _neon, _neon )
+INTRA_MBCMP(satd, 16x16, v, h, dc, , _neon, _neon )
+#endif
// No C implementation of intra_satd_x9. See checkasm for its behavior,
// or see x264_mb_analyse_intra for the entirely different algorithm we
pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_mmx2;
pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_mmx2;
pixf->intra_satd_x3_8x8c = x264_intra_satd_x3_8x8c_mmx2;
+ pixf->intra_sad_x3_8x16c = x264_intra_sad_x3_8x16c_mmx2;
+ pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_mmx2;
pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
}
pixf->asd8 = x264_pixel_asd8_sse2;
pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_sse2;
pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_sse2;
+ pixf->intra_sad_x3_8x16c = x264_intra_sad_x3_8x16c_sse2;
+ pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
}
if( cpu&X264_CPU_SSE2_IS_FAST )
pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_ssse3;
pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_ssse3;
pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_ssse3;
+ pixf->intra_sad_x3_8x16c = x264_intra_sad_x3_8x16c_ssse3;
+ pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_ssse3;
}
if( cpu&X264_CPU_SSE4 )
#if ARCH_X86_64
pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
#endif
+ pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
}
if( cpu&X264_CPU_AVX )
{
+ INIT5_NAME( sad_aligned, sad, _ssse3 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
INIT_ADS( _avx );
INIT6( satd, _avx );
pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_avx;
#if ARCH_X86_64
pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
#endif
+ pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
}
if( cpu&X264_CPU_XOP )
{
+ INIT5( sad_x3, _xop );
+ INIT5( sad_x4, _xop );
+ pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_xop;
+ pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
+ pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_xop;
pixf->vsad = x264_pixel_vsad_xop;
pixf->asd8 = x264_pixel_asd8_xop;
+#if ARCH_X86_64
+ pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_xop;
+#endif
}
if( cpu&X264_CPU_AVX2 )
{
INIT2_NAME( sad_aligned, sad, _avx2 );
INIT2( sad_x3, _avx2 );
INIT2( sad_x4, _avx2 );
+ pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx2;
pixf->vsad = x264_pixel_vsad_avx2;
pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
+ pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_avx2;
}
#endif // HAVE_MMX
#else // !HIGH_BIT_DEPTH
pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
}
}
-
- if( cpu&X264_CPU_SSE_MISALIGN )
- {
- INIT2( sad_x3, _sse2_misalign );
- INIT2( sad_x4, _sse2_misalign );
- }
}
if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
}
else
{
- pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_ssse3;
- pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_ssse3;
- pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_ssse3;
+ INIT2( sad_x3, _ssse3 );
+ INIT5( sad_x4, _ssse3 );
}
if( (cpu&X264_CPU_SLOW_ATOM) || (cpu&X264_CPU_SLOW_SHUFFLE) )
{
if( cpu&X264_CPU_AVX )
{
+ INIT2_NAME( sad_aligned, sad, _sse2 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
+ INIT2( sad_x3, _avx );
+ INIT2( sad_x4, _avx );
INIT8( satd, _avx );
INIT7( satd_x3, _avx );
INIT7( satd_x4, _avx );
pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_xop;
pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8_xop;
pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_xop;
+ pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_xop;
pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
pixf->var[PIXEL_8x16] = x264_pixel_var_8x16_xop;
pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_xop;
pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_avx2;
pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_avx2;
pixf->intra_sad_x9_8x8 = x264_intra_sad_x9_8x8_avx2;
+ pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_avx2;
pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
#if ARCH_X86_64
pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx2;
INIT4( hadamard_ac, _neon );
pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8_neon;
pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
+ pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_neon;
pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_neon;
+ pixf->var[PIXEL_8x16] = x264_pixel_var_8x16_neon;
pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_neon;
pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_neon;
-
+ pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_neon;
+ pixf->vsad = x264_pixel_vsad_neon;
+ pixf->asd8 = x264_pixel_asd8_neon;
+
+ pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_neon;
+ pixf->intra_satd_x3_4x4 = x264_intra_satd_x3_4x4_neon;
+ pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_neon;
+ pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_neon;
+ pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_neon;
+ pixf->intra_satd_x3_8x8c = x264_intra_satd_x3_8x8c_neon;
+ pixf->intra_sad_x3_8x16c = x264_intra_sad_x3_8x16c_neon;
+ pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_neon;
+ pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_neon;
+ pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_neon;
+
+ pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_neon;
pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_neon;
pixf->ssim_end4 = x264_pixel_ssim_end4_neon;
}
}
#endif
+
+#if ARCH_AARCH64
+ if( cpu&X264_CPU_NEON )
+ {
+ INIT8( sad, _neon );
+ // AArch64 has no distinct instructions for aligned load/store
+ INIT8_NAME( sad_aligned, sad, _neon );
+ INIT7( sad_x3, _neon );
+ INIT7( sad_x4, _neon );
+ INIT8( ssd, _neon );
+ INIT8( satd, _neon );
+ INIT7( satd_x3, _neon );
+ INIT7( satd_x4, _neon );
+ INIT4( hadamard_ac, _neon );
+
+ pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8_neon;
+ pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
+ pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_neon;
+
+ pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_neon;
+ pixf->var[PIXEL_8x16] = x264_pixel_var_8x16_neon;
+ pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_neon;
+ pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_neon;
+ pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_neon;
+ pixf->vsad = x264_pixel_vsad_neon;
+ pixf->asd8 = x264_pixel_asd8_neon;
+
+ pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_neon;
+ pixf->intra_satd_x3_4x4 = x264_intra_satd_x3_4x4_neon;
+ pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_neon;
+ pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_neon;
+ pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_neon;
+ pixf->intra_satd_x3_8x8c = x264_intra_satd_x3_8x8c_neon;
+ pixf->intra_sad_x3_8x16c = x264_intra_sad_x3_8x16c_neon;
+ pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_neon;
+ pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_neon;
+ pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_neon;
+
+ pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_neon;
+ pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_neon;
+ pixf->ssim_end4 = x264_pixel_ssim_end4_neon;
+ }
+#endif // ARCH_AARCH64
+
+#if HAVE_MSA
+ if( cpu&X264_CPU_MSA )
+ {
+ INIT8( sad, _msa );
+ INIT8_NAME( sad_aligned, sad, _msa );
+ INIT8( ssd, _msa );
+ INIT7( sad_x3, _msa );
+ INIT7( sad_x4, _msa );
+ INIT8( satd, _msa );
+ INIT4( hadamard_ac, _msa );
+
+ pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_msa;
+ pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_msa;
+ pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_msa;
+ pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_msa;
+ pixf->intra_satd_x3_4x4 = x264_intra_satd_x3_4x4_msa;
+ pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_msa;
+ pixf->intra_satd_x3_8x8c = x264_intra_satd_x3_8x8c_msa;
+ pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_msa;
+
+ pixf->ssim_4x4x2_core = x264_ssim_4x4x2_core_msa;
+
+ pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_msa;
+ pixf->var[PIXEL_8x16] = x264_pixel_var_8x16_msa;
+ pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_msa;
+ pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_msa;
+ pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_msa;
+ pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
+ pixf->sa8d[PIXEL_8x8] = x264_pixel_sa8d_8x8;
+ }
+#endif // HAVE_MSA
+
#endif // HIGH_BIT_DEPTH
#if HAVE_ALTIVEC
if( cpu&X264_CPU_ALTIVEC )
x264_pixel_altivec_init( pixf );
}
#endif
-#if !HIGH_BIT_DEPTH
-#if ARCH_UltraSPARC
- INIT4( sad, _vis );
- INIT4( sad_x3, _vis );
- INIT4( sad_x4, _vis );
-#endif
-#endif // !HIGH_BIT_DEPTH
pixf->ads[PIXEL_8x16] =
pixf->ads[PIXEL_8x4] =