1 /*****************************************************************************
2 * pixel.h: x86 pixel metrics
3 *****************************************************************************
4 * Copyright (C) 2003-2011 x264 project
6 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
7 * Loren Merritt <lorenm@u.washington.edu>
8 * Fiona Glaser <fiona@x264.com>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
24 * This program is also available under a commercial proprietary license.
25 * For more information, contact us at licensing@x264.com.
26 *****************************************************************************/
28 #ifndef X264_I386_PIXEL_H
29 #define X264_I386_PIXEL_H
31 #define DECL_PIXELS( ret, name, suffix, args ) \
32 ret x264_pixel_##name##_16x16_##suffix args;\
33 ret x264_pixel_##name##_16x8_##suffix args;\
34 ret x264_pixel_##name##_8x16_##suffix args;\
35 ret x264_pixel_##name##_8x8_##suffix args;\
36 ret x264_pixel_##name##_8x4_##suffix args;\
37 ret x264_pixel_##name##_4x16_##suffix args;\
38 ret x264_pixel_##name##_4x8_##suffix args;\
39 ret x264_pixel_##name##_4x4_##suffix args;\
41 #define DECL_X1( name, suffix ) \
42 DECL_PIXELS( int, name, suffix, ( pixel *, int, pixel *, int ) )
44 #define DECL_X4( name, suffix ) \
45 DECL_PIXELS( void, name##_x3, suffix, ( pixel *, pixel *, pixel *, pixel *, int, int * ) )\
46 DECL_PIXELS( void, name##_x4, suffix, ( pixel *, pixel *, pixel *, pixel *, pixel *, int, int * ) )
50 DECL_X4( sad, sse2_misalign )
52 DECL_X1( sad, sse2_aligned )
54 DECL_X1( sad, ssse3_aligned )
61 DECL_X1( ssd, sse2slow )
68 DECL_X1( satd, ssse3 )
74 DECL_X1( sa8d, ssse3 )
78 DECL_X1( sad, cache32_mmx2 );
79 DECL_X1( sad, cache64_mmx2 );
80 DECL_X1( sad, cache64_sse2 );
81 DECL_X1( sad, cache64_ssse3 );
82 DECL_X4( sad, cache32_mmx2 );
83 DECL_X4( sad, cache64_mmx2 );
84 DECL_X4( sad, cache64_sse2 );
85 DECL_X4( sad, cache64_ssse3 );
87 DECL_PIXELS( uint64_t, var, mmx2, ( pixel *pix, int i_stride ))
88 DECL_PIXELS( uint64_t, var, sse2, ( pixel *pix, int i_stride ))
89 DECL_PIXELS( uint64_t, var, avx, ( pixel *pix, int i_stride ))
90 DECL_PIXELS( uint64_t, var, xop, ( pixel *pix, int i_stride ))
91 DECL_PIXELS( uint64_t, hadamard_ac, mmx2, ( pixel *pix, int i_stride ))
92 DECL_PIXELS( uint64_t, hadamard_ac, sse2, ( pixel *pix, int i_stride ))
93 DECL_PIXELS( uint64_t, hadamard_ac, ssse3, ( pixel *pix, int i_stride ))
94 DECL_PIXELS( uint64_t, hadamard_ac, sse4, ( pixel *pix, int i_stride ))
95 DECL_PIXELS( uint64_t, hadamard_ac, avx, ( pixel *pix, int i_stride ))
96 DECL_PIXELS( uint64_t, hadamard_ac, xop, ( pixel *pix, int i_stride ))
99 void x264_intra_satd_x3_4x4_mmx2 ( pixel *, pixel *, int * );
100 void x264_intra_satd_x3_4x4_ssse3 ( uint8_t *, uint8_t *, int * );
101 void x264_intra_sad_x3_4x4_mmx2 ( pixel *, pixel *, int * );
102 void x264_intra_sad_x3_4x4_sse4 ( uint8_t *, uint8_t *, int * );
103 void x264_intra_sad_x3_4x4_avx ( uint8_t *, uint8_t *, int * );
104 void x264_intra_satd_x3_8x8c_mmx2 ( pixel *, pixel *, int * );
105 void x264_intra_satd_x3_8x8c_ssse3 ( uint8_t *, uint8_t *, int * );
106 void x264_intra_sad_x3_8x8c_mmx2 ( pixel *, pixel *, int * );
107 void x264_intra_sad_x3_8x8c_sse2 ( pixel *, pixel *, int * );
108 void x264_intra_sad_x3_8x8c_ssse3 ( pixel *, pixel *, int * );
109 void x264_intra_satd_x3_16x16_mmx2 ( pixel *, pixel *, int * );
110 void x264_intra_satd_x3_16x16_ssse3( uint8_t *, uint8_t *, int * );
111 void x264_intra_sad_x3_16x16_mmx2 ( pixel *, pixel *, int * );
112 void x264_intra_sad_x3_16x16_sse2 ( pixel *, pixel *, int * );
113 void x264_intra_sad_x3_16x16_ssse3 ( pixel *, pixel *, int * );
114 void x264_intra_sa8d_x3_8x8_mmx2 ( uint8_t *, uint8_t *, int * );
115 void x264_intra_sa8d_x3_8x8_sse2 ( pixel *, pixel *, int * );
116 void x264_intra_sa8d_x3_8x8_ssse3 ( uint8_t *, uint8_t *, int * );
117 void x264_intra_sa8d_x3_8x8_avx ( uint8_t *, uint8_t *, int * );
118 void x264_intra_sa8d_x3_8x8_xop ( uint8_t *, uint8_t *, int * );
119 void x264_intra_sad_x3_8x8_mmx2 ( pixel *, pixel *, int * );
120 void x264_intra_sad_x3_8x8_sse2 ( pixel *, pixel *, int * );
121 void x264_intra_sad_x3_8x8_ssse3 ( pixel *, pixel *, int * );
122 void x264_intra_sad_x3_8x8_avx ( pixel *, pixel *, int * );
123 int x264_intra_satd_x9_4x4_ssse3( uint8_t *, uint8_t *, uint16_t * );
124 int x264_intra_satd_x9_4x4_sse4 ( uint8_t *, uint8_t *, uint16_t * );
125 int x264_intra_satd_x9_4x4_avx ( uint8_t *, uint8_t *, uint16_t * );
126 int x264_intra_satd_x9_4x4_xop ( uint8_t *, uint8_t *, uint16_t * );
127 int x264_intra_sad_x9_4x4_ssse3 ( uint8_t *, uint8_t *, uint16_t * );
128 int x264_intra_sad_x9_4x4_sse4 ( uint8_t *, uint8_t *, uint16_t * );
129 int x264_intra_sad_x9_4x4_avx ( uint8_t *, uint8_t *, uint16_t * );
131 void x264_pixel_ssd_nv12_core_mmx2( pixel *pixuv1, int stride1,
132 pixel *pixuv2, int stride2, int width,
133 int height, uint64_t *ssd_u, uint64_t *ssd_v );
134 void x264_pixel_ssd_nv12_core_sse2( pixel *pixuv1, int stride1,
135 pixel *pixuv2, int stride2, int width,
136 int height, uint64_t *ssd_u, uint64_t *ssd_v );
137 void x264_pixel_ssd_nv12_core_avx ( pixel *pixuv1, int stride1,
138 pixel *pixuv2, int stride2, int width,
139 int height, uint64_t *ssd_u, uint64_t *ssd_v );
140 void x264_pixel_ssim_4x4x2_core_mmx2( const uint8_t *pix1, int stride1,
141 const uint8_t *pix2, int stride2, int sums[2][4] );
142 void x264_pixel_ssim_4x4x2_core_sse2( const pixel *pix1, int stride1,
143 const pixel *pix2, int stride2, int sums[2][4] );
144 void x264_pixel_ssim_4x4x2_core_avx ( const pixel *pix1, int stride1,
145 const pixel *pix2, int stride2, int sums[2][4] );
146 float x264_pixel_ssim_end4_sse2( int sum0[5][4], int sum1[5][4], int width );
147 float x264_pixel_ssim_end4_avx( int sum0[5][4], int sum1[5][4], int width );
148 int x264_pixel_var2_8x8_mmx2( pixel *, int, pixel *, int, int * );
149 int x264_pixel_var2_8x8_sse2( pixel *, int, pixel *, int, int * );
150 int x264_pixel_var2_8x8_ssse3( uint8_t *, int, uint8_t *, int, int * );
151 int x264_pixel_var2_8x8_xop( uint8_t *, int, uint8_t *, int, int * );
152 int x264_pixel_vsad_mmx2( pixel *src, int stride, int height );
153 int x264_pixel_vsad_sse2( pixel *src, int stride, int height );
155 #define DECL_ADS( size, suffix ) \
156 int x264_pixel_ads##size##_##suffix( int enc_dc[size], uint16_t *sums, int delta,\
157 uint16_t *cost_mvx, int16_t *mvs, int width, int thresh );