1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
|
/*
* Copyright 2012 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/basic_types.h"
#include "libyuv/compare_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// This module is for 32 bit Visual C x86 and clangcl
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
__declspec(naked)
uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
__asm {
mov eax, [esp + 4] // src_a
mov edx, [esp + 8] // src_b
mov ecx, [esp + 12] // count
pxor xmm0, xmm0
pxor xmm5, xmm5
wloop:
movdqu xmm1, [eax]
lea eax, [eax + 16]
movdqu xmm2, [edx]
lea edx, [edx + 16]
movdqa xmm3, xmm1 // abs trick
psubusb xmm1, xmm2
psubusb xmm2, xmm3
por xmm1, xmm2
movdqa xmm2, xmm1
punpcklbw xmm1, xmm5
punpckhbw xmm2, xmm5
pmaddwd xmm1, xmm1
pmaddwd xmm2, xmm2
paddd xmm0, xmm1
paddd xmm0, xmm2
sub ecx, 16
jg wloop
pshufd xmm1, xmm0, 0xee
paddd xmm0, xmm1
pshufd xmm1, xmm0, 0x01
paddd xmm0, xmm1
movd eax, xmm0
ret
}
}
// Visual C 2012 required for AVX2.
#if _MSC_VER >= 1700
// C4752: found Intel(R) Advanced Vector Extensions; consider using /arch:AVX.
#pragma warning(disable: 4752)
__declspec(naked)
uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count) {
__asm {
mov eax, [esp + 4] // src_a
mov edx, [esp + 8] // src_b
mov ecx, [esp + 12] // count
vpxor ymm0, ymm0, ymm0 // sum
vpxor ymm5, ymm5, ymm5 // constant 0 for unpck
sub edx, eax
wloop:
vmovdqu ymm1, [eax]
vmovdqu ymm2, [eax + edx]
lea eax, [eax + 32]
vpsubusb ymm3, ymm1, ymm2 // abs difference trick
vpsubusb ymm2, ymm2, ymm1
vpor ymm1, ymm2, ymm3
vpunpcklbw ymm2, ymm1, ymm5 // u16. mutates order.
vpunpckhbw ymm1, ymm1, ymm5
vpmaddwd ymm2, ymm2, ymm2 // square + hadd to u32.
vpmaddwd ymm1, ymm1, ymm1
vpaddd ymm0, ymm0, ymm1
vpaddd ymm0, ymm0, ymm2
sub ecx, 32
jg wloop
vpshufd ymm1, ymm0, 0xee // 3, 2 + 1, 0 both lanes.
vpaddd ymm0, ymm0, ymm1
vpshufd ymm1, ymm0, 0x01 // 1 + 0 both lanes.
vpaddd ymm0, ymm0, ymm1
vpermq ymm1, ymm0, 0x02 // high + low lane.
vpaddd ymm0, ymm0, ymm1
vmovd eax, xmm0
vzeroupper
ret
}
}
#endif // _MSC_VER >= 1700
uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 }; // 33 ^ 16
uvec32 kHashMul0 = {
0x0c3525e1, // 33 ^ 15
0xa3476dc1, // 33 ^ 14
0x3b4039a1, // 33 ^ 13
0x4f5f0981, // 33 ^ 12
};
uvec32 kHashMul1 = {
0x30f35d61, // 33 ^ 11
0x855cb541, // 33 ^ 10
0x040a9121, // 33 ^ 9
0x747c7101, // 33 ^ 8
};
uvec32 kHashMul2 = {
0xec41d4e1, // 33 ^ 7
0x4cfa3cc1, // 33 ^ 6
0x025528a1, // 33 ^ 5
0x00121881, // 33 ^ 4
};
uvec32 kHashMul3 = {
0x00008c61, // 33 ^ 3
0x00000441, // 33 ^ 2
0x00000021, // 33 ^ 1
0x00000001, // 33 ^ 0
};
__declspec(naked)
uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
__asm {
mov eax, [esp + 4] // src
mov ecx, [esp + 8] // count
movd xmm0, [esp + 12] // seed
pxor xmm7, xmm7 // constant 0 for unpck
movdqa xmm6, xmmword ptr kHash16x33
wloop:
movdqu xmm1, [eax] // src[0-15]
lea eax, [eax + 16]
pmulld xmm0, xmm6 // hash *= 33 ^ 16
movdqa xmm5, xmmword ptr kHashMul0
movdqa xmm2, xmm1
punpcklbw xmm2, xmm7 // src[0-7]
movdqa xmm3, xmm2
punpcklwd xmm3, xmm7 // src[0-3]
pmulld xmm3, xmm5
movdqa xmm5, xmmword ptr kHashMul1
movdqa xmm4, xmm2
punpckhwd xmm4, xmm7 // src[4-7]
pmulld xmm4, xmm5
movdqa xmm5, xmmword ptr kHashMul2
punpckhbw xmm1, xmm7 // src[8-15]
movdqa xmm2, xmm1
punpcklwd xmm2, xmm7 // src[8-11]
pmulld xmm2, xmm5
movdqa xmm5, xmmword ptr kHashMul3
punpckhwd xmm1, xmm7 // src[12-15]
pmulld xmm1, xmm5
paddd xmm3, xmm4 // add 16 results
paddd xmm1, xmm2
paddd xmm1, xmm3
pshufd xmm2, xmm1, 0x0e // upper 2 dwords
paddd xmm1, xmm2
pshufd xmm2, xmm1, 0x01
paddd xmm1, xmm2
paddd xmm0, xmm1
sub ecx, 16
jg wloop
movd eax, xmm0 // return hash
ret
}
}
// Visual C 2012 required for AVX2.
#if _MSC_VER >= 1700
__declspec(naked)
uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed) {
__asm {
mov eax, [esp + 4] // src
mov ecx, [esp + 8] // count
vmovd xmm0, [esp + 12] // seed
wloop:
vpmovzxbd xmm3, [eax] // src[0-3]
vpmulld xmm0, xmm0, xmmword ptr kHash16x33 // hash *= 33 ^ 16
vpmovzxbd xmm4, [eax + 4] // src[4-7]
vpmulld xmm3, xmm3, xmmword ptr kHashMul0
vpmovzxbd xmm2, [eax + 8] // src[8-11]
vpmulld xmm4, xmm4, xmmword ptr kHashMul1
vpmovzxbd xmm1, [eax + 12] // src[12-15]
vpmulld xmm2, xmm2, xmmword ptr kHashMul2
lea eax, [eax + 16]
vpmulld xmm1, xmm1, xmmword ptr kHashMul3
vpaddd xmm3, xmm3, xmm4 // add 16 results
vpaddd xmm1, xmm1, xmm2
vpaddd xmm1, xmm1, xmm3
vpshufd xmm2, xmm1, 0x0e // upper 2 dwords
vpaddd xmm1, xmm1,xmm2
vpshufd xmm2, xmm1, 0x01
vpaddd xmm1, xmm1, xmm2
vpaddd xmm0, xmm0, xmm1
sub ecx, 16
jg wloop
vmovd eax, xmm0 // return hash
vzeroupper
ret
}
}
#endif // _MSC_VER >= 1700
#endif // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
|
