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
|
/*
* Copyright 2011, Ben Langmead <langmea@cs.jhu.edu>
*
* This file is part of Bowtie 2.
*
* Bowtie 2 is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* Bowtie 2 is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with Bowtie 2. If not, see <http://www.gnu.org/licenses/>.
*/
/*
* sse_wrap.h
*
* Routines to wrap Streaming SIMD Extensions (SSE) emmintrin.h
* for an Intel x86 CPU and SIMD Everywhere (simde) for other CPUs.
*/
#ifndef SSE_WRAP_H_
#define SSE_WRAP_H_
#define SIMDE_ENABLE_NATIVE_ALIASES
#ifdef SSE_AVX2
#include <simde/x86/avx2.h>
#define NBYTES_PER_REG 32
#define BYTES_LOG2_PER_REG 5
#define SSE_MASK_ALL ((int) 0xffffffff)
typedef __m256i SSERegI;
#define sse_adds_epi16(x, y) _mm256_adds_epi16(x, y)
#define sse_adds_epu8(x, y) _mm256_adds_epu8(x, y)
#define sse_cmpeq_epi16(x, y) _mm256_cmpeq_epi16(x, y)
#define sse_cmpeq_epi8(x, y) _mm256_cmpeq_epi8(x, y)
#define sse_cmpgt_epi16(x, y) _mm256_cmpgt_epi16(x, y)
#define sse_cmpgt_epi8(x, y) _mm256_cmpgt_epi8(x, y)
#define sse_cmplt_epi16(x, y) _mm256_cmpgt_epi16(y,x)
#define sse_extract_epi16(x, y) _mm256_extract_epi16(x, y)
#define sse_insert_epi16(x, y, z) _mm256_insert_epi16(x, y, z)
#define sse_load_siall(x) _mm256_load_si256(x)
#define sse_max_epi16(x, y) _mm256_max_epi16(x, y)
#define sse_max_epu8(x, y) _mm256_max_epu8(x, y)
#define sse_movemask_epi8(x) _mm256_movemask_epi8(x)
#define sse_or_siall(x, y) _mm256_or_si256(x, y)
#define sse_setzero_siall() _mm256_setzero_si256()
#define sse_slli_epi16(x, y) _mm256_slli_epi16(x, y)
#define sse_srli_epi16(x, y) _mm256_srli_epi16(x, y)
#define sse_srli_epu8(x, y) _mm256_srli_epu8(x, y)
#define sse_store_siall(x, y) _mm256_store_si256(x, y)
#define sse_subs_epi16(x, y) _mm256_subs_epi16(x, y)
#define sse_subs_epu8(x, y) _mm256_subs_epu8(x, y)
#define sse_xor_siall(x, y) _mm256_xor_si256(x, y)
#define sse_set1_epi16(x) _mm256_set1_epi16(x)
/* AVX2 does not have a native 256-bit shift instruction */
/* Note only works for y<=16, which is OK for this code */
#define sse_slli_siall(x, y) \
_mm256_alignr_epi8(x, _mm256_permute2x128_si256(x, x, _MM_SHUFFLE(0, 0, 2, 0)), 16-y)
#define sse_srli_siall(x, y) \
_mm256_alignr_epi8(_mm256_permute2x128_si256(x, x, _MM_SHUFFLE(2, 0, 0, 1)), x, y)
/* we can avoid one instruction, when y==16 */
#define sse_slli_siall_16(x) \
_mm256_permute2x128_si256(x, x, _MM_SHUFFLE(0, 0, 2, 0))
#define sse_srli_siall_16(x) \
_mm256_permute2x128_si256(x, x, _MM_SHUFFLE(2, 0, 0, 1))
/* this operates on the 2x 128-bit lanes independenty */
#define sse_slli_si128(x, y) _mm256_slli_si256(x, y)
#define sse_srli_si128(x, y) _mm256_srli_si256(x, y)
/* compute the max val of a vector */
#define sse_max_score_i16(inval, outval) { \
SSERegI vlmax = inval; \
SSERegI vltmp = sse_srli_siall(vlmax, 16); \
vlmax = sse_max_epi16(vlmax, vltmp); \
/* we use only 128-bit hers, use the fast version */ \
vltmp = sse_srli_si128(vlmax, 8); \
vlmax = sse_max_epi16(vlmax, vltmp); \
vltmp = sse_srli_si128(vlmax, 4); \
vlmax = sse_max_epi16(vlmax, vltmp); \
vltmp = sse_srli_si128(vlmax, 2); \
vlmax = sse_max_epi16(vlmax, vltmp); \
outval = sse_extract_epi16(vlmax, 0); \
}
#define sse_max_score_u8(inval, outval) { \
SSERegI vlmax = inval; \
SSERegI vltmp = sse_srli_siall(vlmax, 16); \
vlmax = sse_max_epu8(vlmax, vltmp); \
/* we use only 128-bit hers, use the fast version */ \
vltmp = sse_srli_si128(vlmax, 8); \
vlmax = sse_max_epu8(vlmax, vltmp); \
vltmp = sse_srli_si128(vlmax, 4); \
vlmax = sse_max_epu8(vlmax, vltmp); \
vltmp = sse_srli_si128(vlmax, 2); \
vlmax = sse_max_epu8(vlmax, vltmp); \
vltmp = sse_srli_si128(vlmax, 1); \
vlmax = sse_max_epu8(vlmax, vltmp); \
outval = sse_extract_epi16(vlmax, 0); \
outval = outval & 0x00ff; \
}
#else /* no SSE_AVX2 */
#include <simde/x86/sse2.h>
#define NBYTES_PER_REG 16
#define BYTES_LOG2_PER_REG 4
#define SSE_MASK_ALL 0xffff
typedef __m128i SSERegI;
#define sse_adds_epi16(x, y) _mm_adds_epi16(x, y)
#define sse_adds_epu8(x, y) _mm_adds_epu8(x, y)
#define sse_cmpeq_epi16(x, y) _mm_cmpeq_epi16(x, y)
#define sse_cmpeq_epi8(x, y) _mm_cmpeq_epi8(x, y)
#define sse_cmpgt_epi16(x, y) _mm_cmpgt_epi16(x, y)
#define sse_cmpgt_epi8(x, y) _mm_cmpgt_epi8(x, y)
#define sse_cmplt_epi16(x, y) _mm_cmplt_epi16(x, y)
#define sse_cmplt_epu8(x, y) _mm_cmplt_epu8(x, y)
#define sse_extract_epi16(x, y) _mm_extract_epi16(x, y)
#define sse_insert_epi16(x, y, z) _mm_insert_epi16(x, y, z)
#define sse_load_siall(x) _mm_load_si128(x)
#define sse_max_epi16(x, y) _mm_max_epi16(x, y)
#define sse_max_epu8(x, y) _mm_max_epu8(x, y)
#define sse_movemask_epi8(x) _mm_movemask_epi8(x)
#define sse_or_siall(x, y) _mm_or_si128(x, y)
#define sse_setzero_siall() _mm_setzero_si128()
#define sse_slli_epi16(x, y) _mm_slli_epi16(x, y)
#define sse_slli_siall(x, y) _mm_slli_si128(x, y)
#define sse_srli_epi16(x, y) _mm_srli_epi16(x, y)
#define sse_srli_epu8(x, y) _mm_srli_epu8(x, y)
#define sse_srli_siall(x, y) _mm_srli_si128(x, y)
#define sse_store_siall(x, y) _mm_store_si128(x, y)
#define sse_subs_epi16(x, y) _mm_subs_epi16(x, y)
#define sse_subs_epu8(x, y) _mm_subs_epu8(x, y)
#define sse_xor_siall(x, y) _mm_xor_si128(x, y)
#define sse_set1_epi16(x) _mm_set1_epi16(x)
/* compute the max val of a vector */
#define sse_max_score_i16(inval, outval) { \
SSERegI vlmax = inval; \
SSERegI vltmp = sse_srli_siall(vlmax, 8); \
vlmax = sse_max_epi16(vlmax, vltmp); \
vltmp = sse_srli_siall(vlmax, 4); \
vlmax = sse_max_epi16(vlmax, vltmp); \
vltmp = sse_srli_siall(vlmax, 2); \
vlmax = sse_max_epi16(vlmax, vltmp); \
outval = sse_extract_epi16(vlmax, 0); \
}
#define sse_max_score_u8(inval, outval) { \
SSERegI vlmax = inval; \
SSERegI vltmp = sse_srli_siall(vlmax, 8); \
vlmax = sse_max_epu8(vlmax, vltmp); \
vltmp = sse_srli_siall(vlmax, 4); \
vlmax = sse_max_epu8(vlmax, vltmp); \
vltmp = sse_srli_siall(vlmax, 2); \
vlmax = sse_max_epu8(vlmax, vltmp); \
vltmp = sse_srli_siall(vlmax, 1); \
vlmax = sse_max_epu8(vlmax, vltmp); \
outval = sse_extract_epi16(vlmax, 0); \
outval = outval & 0x00ff; \
}
#endif /* SSE_AVX2 */
/* Fill all elements in outval with inval */
/* opt version will check for special ivals that can use shortcuts */
#define sse_fill_i16(inval, outval) outval=sse_set1_epi16(inval)
#define sse_fill_i16_opt(inval, outval) { \
if (inval==0xffff) outval = sse_cmpeq_epi16(outval, outval); \
else if (inval==0) outval = sse_xor_siall(outval, outval); \
else sse_fill_i16(inval, outval); \
}
#define sse_fill_u8(inval, outval) {\
int invalloc = inval; \
int dup = (invalloc << 8) | (invalloc & 0x00ff); \
sse_fill_i16(dup, outval); \
}
#define sse_fill_u8_opt(inval, outval) {\
if (inval==0xff) outval = sse_cmpeq_epi16(outval, outval); \
else if (inval==0) outval = sse_xor_siall(outval, outval); \
else sse_fill_u8(inval, outval); \
}
/* Set the low element with invl, all others to 0 */
#define sse_set_low_i16(inval, outval) { \
outval = sse_setzero_siall(); \
outval = sse_insert_epi16(outval, inval, 0); \
}
#define sse_set_low_u8(inval, outval) { \
outval = sse_setzero_siall(); \
outval = sse_insert_epi16(outval, inval, 0); \
}
#endif /* SSE_WRAP_H_ */
|
