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
223
224
225
226
227
228
229
230
|
/* strcspn with SSE4.2 intrinsics
Copyright (C) 2009-2014 Free Software Foundation, Inc.
Contributed by Intel Corporation.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library 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
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include "config.h"
#ifdef HAVE_SSE4_2
#include <glib.h>
#ifdef WIN32
#include <tmmintrin.h>
#endif
#include <nmmintrin.h>
#include <string.h>
#include "ws_mempbrk.h"
/* __has_feature(address_sanitizer) is used later for Clang, this is for
* compatibility with other compilers (such as GCC and MSVC) */
#ifndef __has_feature
# define __has_feature(x) 0
#endif
#define cast_128aligned__m128i(p) ((const __m128i *) (const void *) (p))
/* Helper for variable shifts of SSE registers.
Copyright (C) 2010 Free Software Foundation, Inc.
*/
static const gint8 ___m128i_shift_right[31] =
{
0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15,
-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
};
static inline __m128i
__m128i_shift_right (__m128i value, unsigned long int offset)
{
/* _mm_loadu_si128() works with unaligned data, cast safe */
return _mm_shuffle_epi8 (value,
_mm_loadu_si128 (cast_128aligned__m128i(___m128i_shift_right + offset)));
}
/* We use 0x2:
_SIDD_SBYTE_OPS
| _SIDD_CMP_EQUAL_ANY
| _SIDD_POSITIVE_POLARITY
| _SIDD_LEAST_SIGNIFICANT
on pcmpistri to compare xmm/mem128
0 1 2 3 4 5 6 7 8 9 A B C D E F
X X X X X X X X X X X X X X X X
against xmm
0 1 2 3 4 5 6 7 8 9 A B C D E F
A A A A A A A A A A A A A A A A
to find out if the first 16byte data element has any byte A and
the offset of the first byte. There are 3 cases:
1. The first 16byte data element has the byte A at the offset X.
2. The first 16byte data element has EOS and doesn't have the byte A.
3. The first 16byte data element is valid and doesn't have the byte A.
Here is the table of ECX, CFlag, ZFlag and SFlag for 2 cases:
1 X 1 0/1 0
2 16 0 1 0
3 16 0 0 0
We exit from the loop for cases 1 and 2 with jbe which branches
when either CFlag or ZFlag is 1. If CFlag == 1, ECX has the offset
X for case 1. */
const char *
_ws_mempbrk_sse42(const char *s, size_t slen, const char *a)
{
const char *aligned;
__m128i mask;
int offset;
#if __has_feature(address_sanitizer) || defined(__SANITIZE_ADDRESS__)
{
/* As 'a' is not guarantueed to have a size of at least 16 bytes, and is not
* aligned, _mm_load_si128() cannot be used when ASAN is enabled. That
* triggers a buffer overflow which is harmless as 'a' is guaranteed to be
* '\0' terminated, and the PCMISTRI instruction always ignored everything
* starting from EOS ('\0'). A false positive indeed. */
size_t length;
length = strlen(a);
/* Don't use SSE4.2 if the length of A > 16. */
if (length > 16)
return _ws_mempbrk(s, slen, a);
mask = _mm_setzero_si128();
memcpy(&mask, a, length);
}
#else /* else if ASAN is disabled */
offset = (int) ((size_t) a & 15);
aligned = (const char *) ((size_t) a & -16L);
if (offset != 0)
{
int length;
/* Load masks. */
/* cast safe - _mm_load_si128() it's 16B aligned */
mask = __m128i_shift_right(_mm_load_si128 (cast_128aligned__m128i(aligned)), offset);
/* Find where the NULL terminator is. */
length = _mm_cmpistri (mask, mask, 0x3a);
if (length == 16 - offset)
{
/* There is no NULL terminator. */
__m128i mask1 = _mm_load_si128 (cast_128aligned__m128i(aligned + 16));
int idx = _mm_cmpistri (mask1, mask1, 0x3a);
length += idx;
/* Don't use SSE4.2 if the length of A > 16. */
if (length > 16)
return _ws_mempbrk(s, slen, a);
if (idx != 0)
{
/* Combine mask0 and mask1. We could play games with
palignr, but frankly this data should be in L1 now
so do the merge via an unaligned load. */
mask = _mm_loadu_si128 (cast_128aligned__m128i(a));
}
}
}
else
{
int length;
/* A is aligned. (cast safe) */
mask = _mm_load_si128 (cast_128aligned__m128i(a));
/* Find where the NULL terminator is. */
length = _mm_cmpistri (mask, mask, 0x3a);
if (length == 16)
{
/* There is no NULL terminator. Don't use SSE4.2 if the length
of A > 16. */
if (a[16] != 0)
return _ws_mempbrk(s, slen, a);
}
}
#endif /* ASAN disabled */
offset = (int) ((size_t) s & 15);
aligned = (const char *) ((size_t) s & -16L);
if (offset != 0)
{
/* Check partial string. cast safe it's 16B aligned */
__m128i value = __m128i_shift_right (_mm_load_si128 (cast_128aligned__m128i(aligned)), offset);
int length = _mm_cmpistri (mask, value, 0x2);
/* No need to check ZFlag since ZFlag is always 1. */
int cflag = _mm_cmpistrc (mask, value, 0x2);
int idx = _mm_cmpistri (value, value, 0x3a);
if (cflag)
return s + length;
/* Find where the NULL terminator is. */
if (idx < 16 - offset)
{
/* fond NUL @ 'idx', need to switch to slower mempbrk */
return _ws_mempbrk(s + idx + 1, slen - idx - 1, a); /* slen is bigger than 16 & idx < 16 so no undeflow here */
}
aligned += 16;
slen -= (16 - offset);
}
else
aligned = s;
while (slen >= 16)
{
__m128i value = _mm_load_si128 (cast_128aligned__m128i(aligned));
int idx = _mm_cmpistri (mask, value, 0x2);
int cflag = _mm_cmpistrc (mask, value, 0x2);
int zflag = _mm_cmpistrz (mask, value, 0x2);
if (cflag)
return aligned + idx;
if (zflag)
{
/* found NUL, need to switch to slower mempbrk */
return _ws_mempbrk(aligned, slen, a);
}
aligned += 16;
slen -= 16;
}
/* XXX, use mempbrk_slow here? */
return _ws_mempbrk(aligned, slen, a);
}
#endif /* HAVE_SSE4_2 */
/*
* Editor modelines
*
* Local Variables:
* c-basic-offset: 2
* tab-width: 8
* indent-tabs-mode: nil
* End:
*
* ex: set shiftwidth=2 tabstop=8 expandtab:
* :indentSize=2:tabSize=8:noTabs=true:
*/
|
