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
|
/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#include "nscore.h"
#include "nsAlgorithm.h"
#include <emmintrin.h>
#include <nsUTF8Utils.h>
void
LossyConvertEncoding16to8::write_sse2(const char16_t* aSource,
uint32_t aSourceLength)
{
char* dest = mDestination;
// Align source to a 16-byte boundary.
uint32_t i = 0;
uint32_t alignLen =
XPCOM_MIN<uint32_t>(aSourceLength,
uint32_t(-NS_PTR_TO_INT32(aSource) & 0xf) / sizeof(char16_t));
for (; i < alignLen; ++i) {
dest[i] = static_cast<unsigned char>(aSource[i]);
}
// Walk 64 bytes (four XMM registers) at a time.
__m128i vectmask = _mm_set1_epi16(0x00ff);
for (; aSourceLength - i > 31; i += 32) {
__m128i source1 = _mm_load_si128(reinterpret_cast<const __m128i*>(aSource + i));
source1 = _mm_and_si128(source1, vectmask);
__m128i source2 = _mm_load_si128(reinterpret_cast<const __m128i*>(aSource + i + 8));
source2 = _mm_and_si128(source2, vectmask);
__m128i source3 = _mm_load_si128(reinterpret_cast<const __m128i*>(aSource + i + 16));
source3 = _mm_and_si128(source3, vectmask);
__m128i source4 = _mm_load_si128(reinterpret_cast<const __m128i*>(aSource + i + 24));
source4 = _mm_and_si128(source4, vectmask);
// Pack the source data. SSE2 views this as a saturating uint16_t to
// uint8_t conversion, but since we masked off the high-order byte of every
// uint16_t, we're really just grabbing the low-order bytes of source1 and
// source2.
__m128i packed1 = _mm_packus_epi16(source1, source2);
__m128i packed2 = _mm_packus_epi16(source3, source4);
// This store needs to be unaligned since there's no guarantee that the
// alignment we did above for the source will align the destination.
_mm_storeu_si128(reinterpret_cast<__m128i*>(dest + i), packed1);
_mm_storeu_si128(reinterpret_cast<__m128i*>(dest + i + 16), packed2);
}
// Finish up the rest.
for (; i < aSourceLength; ++i) {
dest[i] = static_cast<unsigned char>(aSource[i]);
}
mDestination += i;
}
void
LossyConvertEncoding8to16::write_sse2(const char* aSource,
uint32_t aSourceLength)
{
char16_t* dest = mDestination;
// Align source to a 16-byte boundary. We choose to align source rather than
// dest because we'd rather have our loads than our stores be fast. You have
// to wait for a load to complete, but you can keep on moving after issuing a
// store.
uint32_t i = 0;
uint32_t alignLen = XPCOM_MIN(aSourceLength,
uint32_t(-NS_PTR_TO_INT32(aSource) & 0xf));
for (; i < alignLen; ++i) {
dest[i] = static_cast<unsigned char>(aSource[i]);
}
// Walk 32 bytes (two XMM registers) at a time.
for (; aSourceLength - i > 31; i += 32) {
__m128i source1 = _mm_load_si128(reinterpret_cast<const __m128i*>(aSource + i));
__m128i source2 = _mm_load_si128(reinterpret_cast<const __m128i*>(aSource + i + 16));
// Interleave 0s in with the bytes of source to create lo and hi.
__m128i lo1 = _mm_unpacklo_epi8(source1, _mm_setzero_si128());
__m128i hi1 = _mm_unpackhi_epi8(source1, _mm_setzero_si128());
__m128i lo2 = _mm_unpacklo_epi8(source2, _mm_setzero_si128());
__m128i hi2 = _mm_unpackhi_epi8(source2, _mm_setzero_si128());
// store lo and hi into dest.
_mm_storeu_si128(reinterpret_cast<__m128i*>(dest + i), lo1);
_mm_storeu_si128(reinterpret_cast<__m128i*>(dest + i + 8), hi1);
_mm_storeu_si128(reinterpret_cast<__m128i*>(dest + i + 16), lo2);
_mm_storeu_si128(reinterpret_cast<__m128i*>(dest + i + 24), hi2);
}
// Finish up whatever's left.
for (; i < aSourceLength; ++i) {
dest[i] = static_cast<unsigned char>(aSource[i]);
}
mDestination += i;
}
|
