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
|
/* compare256_sse2.c -- SSE2 version of compare256
* Copyright Adam Stylinski <kungfujesus06@gmail.com>
* For conditions of distribution and use, see copyright notice in zlib.h
*/
#include "zbuild.h"
#include "zendian.h"
#include "zmemory.h"
#include "deflate.h"
#include "fallback_builtins.h"
#ifdef X86_SSE2
#include <emmintrin.h>
static inline uint32_t compare256_sse2_static(const uint8_t *src0, const uint8_t *src1) {
__m128i xmm_src0, xmm_src1, xmm_cmp;
/* Do the first load unaligned, than all subsequent ones we have at least
* one aligned load. Sadly aligning both loads is probably unrealistic */
xmm_src0 = _mm_loadu_si128((__m128i*)src0);
xmm_src1 = _mm_loadu_si128((__m128i*)src1);
xmm_cmp = _mm_cmpeq_epi8(xmm_src0, xmm_src1);
unsigned mask = (unsigned)_mm_movemask_epi8(xmm_cmp);
/* Compiler _may_ turn this branch into a ptest + movemask,
* since a lot of those uops are shared and fused */
if (mask != 0xFFFF)
return zng_ctz32(~mask);
const uint8_t *last0 = src0 + 240;
const uint8_t *last1 = src1 + 240;
int align_offset = ((uintptr_t)src0) & 15;
int align_adv = 16 - align_offset;
uint32_t len = align_adv;
src0 += align_adv;
src1 += align_adv;
for (int i = 0; i < 15; ++i) {
xmm_src0 = _mm_load_si128((__m128i*)src0);
xmm_src1 = _mm_loadu_si128((__m128i*)src1);
xmm_cmp = _mm_cmpeq_epi8(xmm_src0, xmm_src1);
mask = (unsigned)_mm_movemask_epi8(xmm_cmp);
/* Compiler _may_ turn this branch into a ptest + movemask,
* since a lot of those uops are shared and fused */
if (mask != 0xFFFF)
return len + zng_ctz32(~mask);
len += 16, src0 += 16, src1 += 16;
}
if (align_offset) {
xmm_src0 = _mm_loadu_si128((__m128i*)last0);
xmm_src1 = _mm_loadu_si128((__m128i*)last1);
xmm_cmp = _mm_cmpeq_epi8(xmm_src0, xmm_src1);
mask = (unsigned)_mm_movemask_epi8(xmm_cmp);
if (mask != 0xFFFF)
return 240 + zng_ctz32(~mask);
}
return 256;
}
Z_INTERNAL uint32_t compare256_sse2(const uint8_t *src0, const uint8_t *src1) {
return compare256_sse2_static(src0, src1);
}
#define LONGEST_MATCH longest_match_sse2
#define COMPARE256 compare256_sse2_static
#include "match_tpl.h"
#define LONGEST_MATCH_SLOW
#define LONGEST_MATCH longest_match_slow_sse2
#define COMPARE256 compare256_sse2_static
#include "match_tpl.h"
#endif
|
