Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
Download
52868 views
1
/*****************************************************************************
2
* pixel.h: x86 pixel metrics
3
*****************************************************************************
4
* Copyright (C) 2003-2016 x264 project
5
*
6
* Authors: Laurent Aimar <[email protected]>
7
* Loren Merritt <[email protected]>
8
* Fiona Glaser <[email protected]>
9
*
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.
14
*
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.
19
*
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.
23
*
24
* This program is also available under a commercial proprietary license.
25
* For more information, contact us at [email protected].
26
*****************************************************************************/
27
28
#ifndef X264_I386_PIXEL_H
29
#define X264_I386_PIXEL_H
30
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;\
40
41
#define DECL_X1( name, suffix ) \
42
DECL_PIXELS( int, name, suffix, ( pixel *, intptr_t, pixel *, intptr_t ) )
43
44
#define DECL_X4( name, suffix ) \
45
DECL_PIXELS( void, name##_x3, suffix, ( pixel *, pixel *, pixel *, pixel *, intptr_t, int * ) )\
46
DECL_PIXELS( void, name##_x4, suffix, ( pixel *, pixel *, pixel *, pixel *, pixel *, intptr_t, int * ) )
47
48
DECL_X1( sad, mmx2 )
49
DECL_X1( sad, sse2 )
50
DECL_X1( sad, sse3 )
51
DECL_X1( sad, sse2_aligned )
52
DECL_X1( sad, ssse3 )
53
DECL_X1( sad, ssse3_aligned )
54
DECL_X1( sad, avx2 )
55
DECL_X4( sad, mmx2 )
56
DECL_X4( sad, sse2 )
57
DECL_X4( sad, sse3 )
58
DECL_X4( sad, ssse3 )
59
DECL_X4( sad, xop )
60
DECL_X4( sad, avx )
61
DECL_X4( sad, avx2 )
62
DECL_X1( ssd, mmx )
63
DECL_X1( ssd, mmx2 )
64
DECL_X1( ssd, sse2slow )
65
DECL_X1( ssd, sse2 )
66
DECL_X1( ssd, ssse3 )
67
DECL_X1( ssd, avx )
68
DECL_X1( ssd, xop )
69
DECL_X1( ssd, avx2 )
70
DECL_X1( satd, mmx2 )
71
DECL_X1( satd, sse2 )
72
DECL_X1( satd, ssse3 )
73
DECL_X1( satd, ssse3_atom )
74
DECL_X1( satd, sse4 )
75
DECL_X1( satd, avx )
76
DECL_X1( satd, xop )
77
DECL_X1( satd, avx2 )
78
DECL_X1( sa8d, mmx2 )
79
DECL_X1( sa8d, sse2 )
80
DECL_X1( sa8d, ssse3 )
81
DECL_X1( sa8d, ssse3_atom )
82
DECL_X1( sa8d, sse4 )
83
DECL_X1( sa8d, avx )
84
DECL_X1( sa8d, xop )
85
DECL_X1( sa8d, avx2 )
86
DECL_X1( sad, cache32_mmx2 );
87
DECL_X1( sad, cache64_mmx2 );
88
DECL_X1( sad, cache64_sse2 );
89
DECL_X1( sad, cache64_ssse3 );
90
DECL_X4( sad, cache32_mmx2 );
91
DECL_X4( sad, cache64_mmx2 );
92
DECL_X4( sad, cache64_sse2 );
93
DECL_X4( sad, cache64_ssse3 );
94
95
DECL_PIXELS( uint64_t, var, mmx2, ( pixel *pix, intptr_t i_stride ))
96
DECL_PIXELS( uint64_t, var, sse2, ( pixel *pix, intptr_t i_stride ))
97
DECL_PIXELS( uint64_t, var, avx, ( pixel *pix, intptr_t i_stride ))
98
DECL_PIXELS( uint64_t, var, xop, ( pixel *pix, intptr_t i_stride ))
99
DECL_PIXELS( uint64_t, var, avx2, ( pixel *pix, intptr_t i_stride ))
100
DECL_PIXELS( uint64_t, hadamard_ac, mmx2, ( pixel *pix, intptr_t i_stride ))
101
DECL_PIXELS( uint64_t, hadamard_ac, sse2, ( pixel *pix, intptr_t i_stride ))
102
DECL_PIXELS( uint64_t, hadamard_ac, ssse3, ( pixel *pix, intptr_t i_stride ))
103
DECL_PIXELS( uint64_t, hadamard_ac, ssse3_atom, ( pixel *pix, intptr_t i_stride ))
104
DECL_PIXELS( uint64_t, hadamard_ac, sse4, ( pixel *pix, intptr_t i_stride ))
105
DECL_PIXELS( uint64_t, hadamard_ac, avx, ( pixel *pix, intptr_t i_stride ))
106
DECL_PIXELS( uint64_t, hadamard_ac, xop, ( pixel *pix, intptr_t i_stride ))
107
DECL_PIXELS( uint64_t, hadamard_ac, avx2, ( pixel *pix, intptr_t i_stride ))
108
109
110
void x264_intra_satd_x3_4x4_mmx2 ( pixel *, pixel *, int * );
111
void x264_intra_sad_x3_4x4_mmx2 ( pixel *, pixel *, int * );
112
void x264_intra_sad_x3_4x4_sse2 ( pixel *, pixel *, int * );
113
void x264_intra_sad_x3_4x4_ssse3 ( pixel *, pixel *, int * );
114
void x264_intra_sad_x3_4x4_avx ( pixel *, pixel *, int * );
115
void x264_intra_satd_x3_8x8c_mmx2 ( pixel *, pixel *, int * );
116
void x264_intra_satd_x3_8x8c_ssse3 ( uint8_t *, uint8_t *, int * );
117
void x264_intra_sad_x3_8x8c_mmx2 ( pixel *, pixel *, int * );
118
void x264_intra_sad_x3_8x8c_sse2 ( pixel *, pixel *, int * );
119
void x264_intra_sad_x3_8x8c_ssse3 ( pixel *, pixel *, int * );
120
void x264_intra_sad_x3_8x8c_avx2 ( pixel *, pixel *, int * );
121
void x264_intra_satd_x3_16x16_mmx2 ( pixel *, pixel *, int * );
122
void x264_intra_satd_x3_16x16_ssse3( uint8_t *, uint8_t *, int * );
123
void x264_intra_sad_x3_16x16_mmx2 ( pixel *, pixel *, int * );
124
void x264_intra_sad_x3_16x16_sse2 ( pixel *, pixel *, int * );
125
void x264_intra_sad_x3_16x16_ssse3 ( pixel *, pixel *, int * );
126
void x264_intra_sad_x3_16x16_avx2 ( pixel *, pixel *, int * );
127
void x264_intra_sa8d_x3_8x8_mmx2 ( uint8_t *, uint8_t *, int * );
128
void x264_intra_sa8d_x3_8x8_sse2 ( pixel *, pixel *, int * );
129
void x264_intra_sad_x3_8x8_mmx2 ( pixel *, pixel *, int * );
130
void x264_intra_sad_x3_8x8_sse2 ( pixel *, pixel *, int * );
131
void x264_intra_sad_x3_8x8_ssse3 ( pixel *, pixel *, int * );
132
void x264_intra_sad_x3_8x8_avx2 ( uint16_t*, uint16_t*, int * );
133
int x264_intra_satd_x9_4x4_ssse3( uint8_t *, uint8_t *, uint16_t * );
134
int x264_intra_satd_x9_4x4_sse4 ( uint8_t *, uint8_t *, uint16_t * );
135
int x264_intra_satd_x9_4x4_avx ( uint8_t *, uint8_t *, uint16_t * );
136
int x264_intra_satd_x9_4x4_xop ( uint8_t *, uint8_t *, uint16_t * );
137
int x264_intra_sad_x9_4x4_ssse3 ( uint8_t *, uint8_t *, uint16_t * );
138
int x264_intra_sad_x9_4x4_sse4 ( uint8_t *, uint8_t *, uint16_t * );
139
int x264_intra_sad_x9_4x4_avx ( uint8_t *, uint8_t *, uint16_t * );
140
int x264_intra_sa8d_x9_8x8_ssse3( uint8_t *, uint8_t *, uint8_t *, uint16_t *, uint16_t * );
141
int x264_intra_sa8d_x9_8x8_sse4 ( uint8_t *, uint8_t *, uint8_t *, uint16_t *, uint16_t * );
142
int x264_intra_sa8d_x9_8x8_avx ( uint8_t *, uint8_t *, uint8_t *, uint16_t *, uint16_t * );
143
int x264_intra_sad_x9_8x8_ssse3 ( uint8_t *, uint8_t *, uint8_t *, uint16_t *, uint16_t * );
144
int x264_intra_sad_x9_8x8_sse4 ( uint8_t *, uint8_t *, uint8_t *, uint16_t *, uint16_t * );
145
int x264_intra_sad_x9_8x8_avx ( uint8_t *, uint8_t *, uint8_t *, uint16_t *, uint16_t * );
146
int x264_intra_sad_x9_8x8_avx2 ( uint8_t *, uint8_t *, uint8_t *, uint16_t *, uint16_t * );
147
148
void x264_pixel_ssd_nv12_core_mmx2( pixel *pixuv1, intptr_t stride1,
149
pixel *pixuv2, intptr_t stride2, int width,
150
int height, uint64_t *ssd_u, uint64_t *ssd_v );
151
void x264_pixel_ssd_nv12_core_sse2( pixel *pixuv1, intptr_t stride1,
152
pixel *pixuv2, intptr_t stride2, int width,
153
int height, uint64_t *ssd_u, uint64_t *ssd_v );
154
void x264_pixel_ssd_nv12_core_avx ( pixel *pixuv1, intptr_t stride1,
155
pixel *pixuv2, intptr_t stride2, int width,
156
int height, uint64_t *ssd_u, uint64_t *ssd_v );
157
void x264_pixel_ssd_nv12_core_xop ( pixel *pixuv1, intptr_t stride1,
158
pixel *pixuv2, intptr_t stride2, int width,
159
int height, uint64_t *ssd_u, uint64_t *ssd_v );
160
void x264_pixel_ssd_nv12_core_avx2( pixel *pixuv1, intptr_t stride1,
161
pixel *pixuv2, intptr_t stride2, int width,
162
int height, uint64_t *ssd_u, uint64_t *ssd_v );
163
void x264_pixel_ssim_4x4x2_core_mmx2( const uint8_t *pix1, intptr_t stride1,
164
const uint8_t *pix2, intptr_t stride2, int sums[2][4] );
165
void x264_pixel_ssim_4x4x2_core_sse2( const pixel *pix1, intptr_t stride1,
166
const pixel *pix2, intptr_t stride2, int sums[2][4] );
167
void x264_pixel_ssim_4x4x2_core_avx ( const pixel *pix1, intptr_t stride1,
168
const pixel *pix2, intptr_t stride2, int sums[2][4] );
169
float x264_pixel_ssim_end4_sse2( int sum0[5][4], int sum1[5][4], int width );
170
float x264_pixel_ssim_end4_avx ( int sum0[5][4], int sum1[5][4], int width );
171
int x264_pixel_var2_8x8_mmx2 ( pixel *, intptr_t, pixel *, intptr_t, int * );
172
int x264_pixel_var2_8x8_sse2 ( pixel *, intptr_t, pixel *, intptr_t, int * );
173
int x264_pixel_var2_8x8_ssse3 ( uint8_t *, intptr_t, uint8_t *, intptr_t, int * );
174
int x264_pixel_var2_8x8_xop ( uint8_t *, intptr_t, uint8_t *, intptr_t, int * );
175
int x264_pixel_var2_8x8_avx2 ( uint8_t *, intptr_t, uint8_t *, intptr_t, int * );
176
int x264_pixel_var2_8x16_mmx2 ( pixel *, intptr_t, pixel *, intptr_t, int * );
177
int x264_pixel_var2_8x16_sse2 ( pixel *, intptr_t, pixel *, intptr_t, int * );
178
int x264_pixel_var2_8x16_ssse3( uint8_t *, intptr_t, uint8_t *, intptr_t, int * );
179
int x264_pixel_var2_8x16_xop ( uint8_t *, intptr_t, uint8_t *, intptr_t, int * );
180
int x264_pixel_var2_8x16_avx2 ( uint8_t *, intptr_t, uint8_t *, intptr_t, int * );
181
int x264_pixel_vsad_mmx2 ( pixel *src, intptr_t stride, int height );
182
int x264_pixel_vsad_sse2 ( pixel *src, intptr_t stride, int height );
183
int x264_pixel_vsad_ssse3( pixel *src, intptr_t stride, int height );
184
int x264_pixel_vsad_xop ( pixel *src, intptr_t stride, int height );
185
int x264_pixel_vsad_avx2 ( uint16_t *src, intptr_t stride, int height );
186
int x264_pixel_asd8_sse2 ( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2, int height );
187
int x264_pixel_asd8_ssse3( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2, int height );
188
int x264_pixel_asd8_xop ( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2, int height );
189
uint64_t x264_pixel_sa8d_satd_16x16_sse2 ( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2 );
190
uint64_t x264_pixel_sa8d_satd_16x16_ssse3 ( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2 );
191
uint64_t x264_pixel_sa8d_satd_16x16_ssse3_atom( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2 );
192
uint64_t x264_pixel_sa8d_satd_16x16_sse4 ( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2 );
193
uint64_t x264_pixel_sa8d_satd_16x16_avx ( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2 );
194
uint64_t x264_pixel_sa8d_satd_16x16_xop ( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2 );
195
uint64_t x264_pixel_sa8d_satd_16x16_avx2 ( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2 );
196
197
198
#define DECL_ADS( size, suffix ) \
199
int x264_pixel_ads##size##_##suffix( int enc_dc[size], uint16_t *sums, int delta,\
200
uint16_t *cost_mvx, int16_t *mvs, int width, int thresh );
201
DECL_ADS( 4, mmx2 )
202
DECL_ADS( 2, mmx2 )
203
DECL_ADS( 1, mmx2 )
204
DECL_ADS( 4, sse2 )
205
DECL_ADS( 2, sse2 )
206
DECL_ADS( 1, sse2 )
207
DECL_ADS( 4, ssse3 )
208
DECL_ADS( 2, ssse3 )
209
DECL_ADS( 1, ssse3 )
210
DECL_ADS( 4, avx )
211
DECL_ADS( 2, avx )
212
DECL_ADS( 1, avx )
213
DECL_ADS( 4, avx2 )
214
DECL_ADS( 2, avx2 )
215
DECL_ADS( 1, avx2 )
216
217
#undef DECL_PIXELS
218
#undef DECL_X1
219
#undef DECL_X4
220
#undef DECL_ADS
221
222
#endif
223
224