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
|
/* SPDX-License-Identifier: BSD-3-Clause */
/* Copyright 2017-2020, Intel Corporation */
#ifndef PMEM2_MEMCPY_SSE2_H
#define PMEM2_MEMCPY_SSE2_H
#include <xmmintrin.h>
#include <stddef.h>
#include <stdint.h>
#include "out.h"
static force_inline void
memmove_small_sse2_noflush(char *dest, const char *src, size_t len)
{
ASSERT(len <= 64);
if (len <= 8)
goto le8;
if (len <= 32)
goto le32;
if (len > 48) {
/* 49..64 */
__m128i xmm0 = _mm_loadu_si128((__m128i *)src);
__m128i xmm1 = _mm_loadu_si128((__m128i *)(src + 16));
__m128i xmm2 = _mm_loadu_si128((__m128i *)(src + 32));
__m128i xmm3 = _mm_loadu_si128((__m128i *)(src + len - 16));
_mm_storeu_si128((__m128i *)dest, xmm0);
_mm_storeu_si128((__m128i *)(dest + 16), xmm1);
_mm_storeu_si128((__m128i *)(dest + 32), xmm2);
_mm_storeu_si128((__m128i *)(dest + len - 16), xmm3);
return;
}
/* 33..48 */
__m128i xmm0 = _mm_loadu_si128((__m128i *)src);
__m128i xmm1 = _mm_loadu_si128((__m128i *)(src + 16));
__m128i xmm2 = _mm_loadu_si128((__m128i *)(src + len - 16));
_mm_storeu_si128((__m128i *)dest, xmm0);
_mm_storeu_si128((__m128i *)(dest + 16), xmm1);
_mm_storeu_si128((__m128i *)(dest + len - 16), xmm2);
return;
le32:
if (len > 16) {
/* 17..32 */
__m128i xmm0 = _mm_loadu_si128((__m128i *)src);
__m128i xmm1 = _mm_loadu_si128((__m128i *)(src + len - 16));
_mm_storeu_si128((__m128i *)dest, xmm0);
_mm_storeu_si128((__m128i *)(dest + len - 16), xmm1);
return;
}
/* 9..16 */
uint64_t d80 = *(ua_uint64_t *)src;
uint64_t d81 = *(ua_uint64_t *)(src + len - 8);
*(ua_uint64_t *)dest = d80;
*(ua_uint64_t *)(dest + len - 8) = d81;
return;
le8:
if (len <= 2)
goto le2;
if (len > 4) {
/* 5..8 */
uint32_t d40 = *(ua_uint32_t *)src;
uint32_t d41 = *(ua_uint32_t *)(src + len - 4);
*(ua_uint32_t *)dest = d40;
*(ua_uint32_t *)(dest + len - 4) = d41;
return;
}
/* 3..4 */
uint16_t d20 = *(ua_uint16_t *)src;
uint16_t d21 = *(ua_uint16_t *)(src + len - 2);
*(ua_uint16_t *)dest = d20;
*(ua_uint16_t *)(dest + len - 2) = d21;
return;
le2:
if (len == 2) {
*(ua_uint16_t *)dest = *(ua_uint16_t *)src;
return;
}
*(uint8_t *)dest = *(uint8_t *)src;
}
static force_inline void
memmove_small_sse2(char *dest, const char *src, size_t len, flush_fn flush)
{
/*
* pmemcheck complains about "overwritten stores before they were made
* persistent" for overlapping stores (last instruction in each code
* path) in the optimized version.
* libc's memcpy also does that, so we can't use it here.
*/
if (On_pmemcheck) {
memmove_nodrain_generic(dest, src, len, PMEM2_F_MEM_NOFLUSH,
NULL);
} else {
memmove_small_sse2_noflush(dest, src, len);
}
flush(dest, len);
}
#endif
|
