Merge branch 'jk/strbuf-tolower'
[git/git.git] / compat / bswap.h
1 /*
2 * Let's make sure we always have a sane definition for ntohl()/htonl().
3 * Some libraries define those as a function call, just to perform byte
4 * shifting, bringing significant overhead to what should be a simple
5 * operation.
6 */
7
8 /*
9 * Default version that the compiler ought to optimize properly with
10 * constant values.
11 */
12 static inline uint32_t default_swab32(uint32_t val)
13 {
14 return (((val & 0xff000000) >> 24) |
15 ((val & 0x00ff0000) >> 8) |
16 ((val & 0x0000ff00) << 8) |
17 ((val & 0x000000ff) << 24));
18 }
19
20 static inline uint64_t default_bswap64(uint64_t val)
21 {
22 return (((val & (uint64_t)0x00000000000000ffULL) << 56) |
23 ((val & (uint64_t)0x000000000000ff00ULL) << 40) |
24 ((val & (uint64_t)0x0000000000ff0000ULL) << 24) |
25 ((val & (uint64_t)0x00000000ff000000ULL) << 8) |
26 ((val & (uint64_t)0x000000ff00000000ULL) >> 8) |
27 ((val & (uint64_t)0x0000ff0000000000ULL) >> 24) |
28 ((val & (uint64_t)0x00ff000000000000ULL) >> 40) |
29 ((val & (uint64_t)0xff00000000000000ULL) >> 56));
30 }
31
32 #undef bswap32
33 #undef bswap64
34
35 #if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
36
37 #define bswap32 git_bswap32
38 static inline uint32_t git_bswap32(uint32_t x)
39 {
40 uint32_t result;
41 if (__builtin_constant_p(x))
42 result = default_swab32(x);
43 else
44 __asm__("bswap %0" : "=r" (result) : "0" (x));
45 return result;
46 }
47
48 #define bswap64 git_bswap64
49 #if defined(__x86_64__)
50 static inline uint64_t git_bswap64(uint64_t x)
51 {
52 uint64_t result;
53 if (__builtin_constant_p(x))
54 result = default_bswap64(x);
55 else
56 __asm__("bswap %q0" : "=r" (result) : "0" (x));
57 return result;
58 }
59 #else
60 static inline uint64_t git_bswap64(uint64_t x)
61 {
62 union { uint64_t i64; uint32_t i32[2]; } tmp, result;
63 if (__builtin_constant_p(x))
64 result.i64 = default_bswap64(x);
65 else {
66 tmp.i64 = x;
67 result.i32[0] = git_bswap32(tmp.i32[1]);
68 result.i32[1] = git_bswap32(tmp.i32[0]);
69 }
70 return result.i64;
71 }
72 #endif
73
74 #elif defined(_MSC_VER) && (defined(_M_IX86) || defined(_M_X64))
75
76 #include <stdlib.h>
77
78 #define bswap32(x) _byteswap_ulong(x)
79 #define bswap64(x) _byteswap_uint64(x)
80
81 #endif
82
83 #if defined(bswap32)
84
85 #undef ntohl
86 #undef htonl
87 #define ntohl(x) bswap32(x)
88 #define htonl(x) bswap32(x)
89
90 #endif
91
92 #if defined(bswap64)
93
94 #undef ntohll
95 #undef htonll
96 #define ntohll(x) bswap64(x)
97 #define htonll(x) bswap64(x)
98
99 #else
100
101 #undef ntohll
102 #undef htonll
103
104 #if !defined(__BYTE_ORDER)
105 # if defined(BYTE_ORDER) && defined(LITTLE_ENDIAN) && defined(BIG_ENDIAN)
106 # define __BYTE_ORDER BYTE_ORDER
107 # define __LITTLE_ENDIAN LITTLE_ENDIAN
108 # define __BIG_ENDIAN BIG_ENDIAN
109 # endif
110 #endif
111
112 #if !defined(__BYTE_ORDER)
113 # error "Cannot determine endianness"
114 #endif
115
116 #if __BYTE_ORDER == __BIG_ENDIAN
117 # define ntohll(n) (n)
118 # define htonll(n) (n)
119 #else
120 # define ntohll(n) default_bswap64(n)
121 # define htonll(n) default_bswap64(n)
122 #endif
123
124 #endif
125
126 /*
127 * Performance might be improved if the CPU architecture is OK with
128 * unaligned 32-bit loads and a fast ntohl() is available.
129 * Otherwise fall back to byte loads and shifts which is portable,
130 * and is faster on architectures with memory alignment issues.
131 */
132
133 #if defined(__i386__) || defined(__x86_64__) || \
134 defined(_M_IX86) || defined(_M_X64) || \
135 defined(__ppc__) || defined(__ppc64__) || \
136 defined(__powerpc__) || defined(__powerpc64__) || \
137 defined(__s390__) || defined(__s390x__)
138
139 #define get_be16(p) ntohs(*(unsigned short *)(p))
140 #define get_be32(p) ntohl(*(unsigned int *)(p))
141 #define put_be32(p, v) do { *(unsigned int *)(p) = htonl(v); } while (0)
142
143 #else
144
145 #define get_be16(p) ( \
146 (*((unsigned char *)(p) + 0) << 8) | \
147 (*((unsigned char *)(p) + 1) << 0) )
148 #define get_be32(p) ( \
149 (*((unsigned char *)(p) + 0) << 24) | \
150 (*((unsigned char *)(p) + 1) << 16) | \
151 (*((unsigned char *)(p) + 2) << 8) | \
152 (*((unsigned char *)(p) + 3) << 0) )
153 #define put_be32(p, v) do { \
154 unsigned int __v = (v); \
155 *((unsigned char *)(p) + 0) = __v >> 24; \
156 *((unsigned char *)(p) + 1) = __v >> 16; \
157 *((unsigned char *)(p) + 2) = __v >> 8; \
158 *((unsigned char *)(p) + 3) = __v >> 0; } while (0)
159
160 #endif