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
|
static inline void
enc_loop_avx2_inner_first (const uint8_t **s, uint8_t **o)
{
// First load is done at s - 0 to not get a segfault:
__m256i src = _mm256_loadu_si256((__m256i *) *s);
// Shift by 4 bytes, as required by enc_reshuffle:
src = _mm256_permutevar8x32_epi32(src, _mm256_setr_epi32(0, 0, 1, 2, 3, 4, 5, 6));
// Reshuffle, translate, store:
src = enc_reshuffle(src);
src = enc_translate(src);
_mm256_storeu_si256((__m256i *) *o, src);
// Subsequent loads will be done at s - 4, set pointer for next round:
*s += 20;
*o += 32;
}
static inline void
enc_loop_avx2_inner (const uint8_t **s, uint8_t **o)
{
// Load input:
__m256i src = _mm256_loadu_si256((__m256i *) *s);
// Reshuffle, translate, store:
src = enc_reshuffle(src);
src = enc_translate(src);
_mm256_storeu_si256((__m256i *) *o, src);
*s += 24;
*o += 32;
}
static inline void
enc_loop_avx2 (const uint8_t **s, size_t *slen, uint8_t **o, size_t *olen)
{
if (*slen < 32) {
return;
}
// Process blocks of 24 bytes at a time. Because blocks are loaded 32
// bytes at a time an offset of -4, ensure that there will be at least
// 4 remaining bytes after the last round, so that the final read will
// not pass beyond the bounds of the input buffer:
size_t rounds = (*slen - 4) / 24;
*slen -= rounds * 24; // 24 bytes consumed per round
*olen += rounds * 32; // 32 bytes produced per round
// The first loop iteration requires special handling to ensure that
// the read, which is done at an offset, does not underflow the buffer:
enc_loop_avx2_inner_first(s, o);
rounds--;
while (rounds > 0) {
if (rounds >= 8) {
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
rounds -= 8;
continue;
}
if (rounds >= 4) {
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
rounds -= 4;
continue;
}
if (rounds >= 2) {
enc_loop_avx2_inner(s, o);
enc_loop_avx2_inner(s, o);
rounds -= 2;
continue;
}
enc_loop_avx2_inner(s, o);
break;
}
// Add the offset back:
*s += 4;
}
|
