1 | /* |
1 | /* |
2 | * libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
3 | * |
3 | * |
4 | * Copyright (©) 2009-2011 Marc Alexander Lehmann |
4 | * Copyright (©) 2009-2011 Marc Alexander Lehmann <libecb@schmorp.de> |
|
|
5 | * Copyright (©) 2011 Emanuele Giaquinta |
5 | * All rights reserved. |
6 | * All rights reserved. |
6 | * |
7 | * |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * tion, are permitted provided that the following conditions are met: |
9 | * tion, are permitted provided that the following conditions are met: |
9 | * |
10 | * |
… | |
… | |
27 | */ |
28 | */ |
28 | |
29 | |
29 | #ifndef ECB_H |
30 | #ifndef ECB_H |
30 | #define ECB_H |
31 | #define ECB_H |
31 | |
32 | |
|
|
33 | #ifdef _WIN32 |
|
|
34 | typedef signed char int8_t; |
|
|
35 | typedef unsigned char uint8_t; |
|
|
36 | typedef signed short int16_t; |
|
|
37 | typedef unsigned short uint16_t; |
|
|
38 | typedef signed int int32_t; |
|
|
39 | typedef unsigned int uint32_t; |
|
|
40 | #if __GNUC__ |
|
|
41 | typedef signed long long int64_t; |
|
|
42 | typedef unsigned long long uint64_t; |
|
|
43 | #else /* _MSC_VER || __BORLANDC__ */ |
|
|
44 | typedef signed __int64 int64_t; |
|
|
45 | typedef unsigned __int64 uint64_t; |
|
|
46 | #endif |
|
|
47 | #else |
|
|
48 | #include <inttypes.h> |
|
|
49 | #endif |
|
|
50 | |
|
|
51 | /* many compilers define _GNUC_ to some versions but then only implement |
|
|
52 | * what their idiot authors think are the "more important" extensions, |
|
|
53 | * causing enourmous grief in return for some better fake benchmark numbers. |
|
|
54 | * or so. |
|
|
55 | * we try to detect these and simply assume they are not gcc - if they have |
|
|
56 | * an issue with that they should have done it right in the first place. |
|
|
57 | */ |
|
|
58 | #ifndef ECB_GCC_VERSION |
|
|
59 | #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) |
|
|
60 | #define ECB_GCC_VERSION(major,minor) 0 |
|
|
61 | #else |
32 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
|
|
63 | #endif |
|
|
64 | #endif |
|
|
65 | |
|
|
66 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
67 | |
|
|
68 | #if __cplusplus |
|
|
69 | #define ecb_inline static inline |
|
|
70 | #elif ECB_GCC_VERSION(2,5) |
|
|
71 | #define ecb_inline static __inline__ |
|
|
72 | #elif ECB_C99 |
|
|
73 | #define ecb_inline static inline |
|
|
74 | #else |
|
|
75 | #define ecb_inline static |
|
|
76 | #endif |
|
|
77 | |
|
|
78 | #if ECB_GCC_VERSION(3,3) |
|
|
79 | #define ecb_restrict __restrict__ |
|
|
80 | #elif ECB_C99 |
|
|
81 | #define ecb_restrict restrict |
|
|
82 | #else |
|
|
83 | #define ecb_restrict |
|
|
84 | #endif |
|
|
85 | |
|
|
86 | typedef int ecb_bool; |
|
|
87 | |
|
|
88 | #define ECB_CONCAT_(a, b) a ## b |
|
|
89 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
|
|
90 | #define ECB_STRINGIFY_(a) # a |
|
|
91 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
|
|
92 | |
|
|
93 | #define ecb_function_ ecb_inline |
33 | |
94 | |
34 | #if ECB_GCC_VERSION(3,1) |
95 | #if ECB_GCC_VERSION(3,1) |
35 | # define ecb_attribute(attrlist) __attribute__(attrlist) |
96 | #define ecb_attribute(attrlist) __attribute__(attrlist) |
36 | # define ecb_is_constant(c) __builtin_constant_p (c) |
97 | #define ecb_is_constant(expr) __builtin_constant_p (expr) |
37 | # define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
98 | #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) |
38 | # define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
99 | #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) |
39 | #else |
100 | #else |
40 | # define ecb_attribute(attrlist) |
101 | #define ecb_attribute(attrlist) |
41 | # define ecb_is_constant(c) 0 |
102 | #define ecb_is_constant(expr) 0 |
42 | # define ecb_expect(expr,value) (expr) |
103 | #define ecb_expect(expr,value) (expr) |
43 | # define ecb_prefetch(addr,rw,locality) |
104 | #define ecb_prefetch(addr,rw,locality) |
44 | #endif |
105 | #endif |
45 | |
106 | |
46 | /* no emulation for ecb_decltype */ |
107 | /* no emulation for ecb_decltype */ |
47 | #if ECB_GCC_VERSION(4,5) |
108 | #if ECB_GCC_VERSION(4,5) |
48 | # define ecb_decltype(x) __decltype(x) |
109 | #define ecb_decltype(x) __decltype(x) |
49 | #elif ECB_GCC_VERSION(3,0) |
110 | #elif ECB_GCC_VERSION(3,0) |
50 | # define ecb_decltype(x) typeof(x) |
111 | #define ecb_decltype(x) __typeof(x) |
51 | #endif |
112 | #endif |
52 | |
113 | |
53 | #define ecb_noinline ecb_attribute ((noinline)) |
114 | #define ecb_noinline ecb_attribute ((__noinline__)) |
54 | #define ecb_noreturn ecb_attribute ((noreturn)) |
115 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
55 | #define ecb_unused ecb_attribute ((unused)) |
116 | #define ecb_unused ecb_attribute ((__unused__)) |
56 | #define ecb_const ecb_attribute ((const)) |
117 | #define ecb_const ecb_attribute ((__const__)) |
57 | #define ecb_pure ecb_attribute ((pure)) |
118 | #define ecb_pure ecb_attribute ((__pure__)) |
|
|
119 | |
|
|
120 | #if ECB_GCC_VERSION(4,3) |
|
|
121 | #define ecb_artificial ecb_attribute ((__artificial__)) |
58 | #define ecb_hot ecb_attribute ((hot)) // 4.3 |
122 | #define ecb_hot ecb_attribute ((__hot__)) |
59 | #define ecb_cold ecb_attribute ((cold)) // 4.3 |
123 | #define ecb_cold ecb_attribute ((__cold__)) |
|
|
124 | #else |
|
|
125 | #define ecb_artificial |
|
|
126 | #define ecb_hot |
|
|
127 | #define ecb_cold |
|
|
128 | #endif |
|
|
129 | |
|
|
130 | /* put around conditional expressions if you are very sure that the */ |
|
|
131 | /* expression is mostly true or mostly false. note that these return */ |
|
|
132 | /* booleans, not the expression. */ |
|
|
133 | #define ecb_expect_false(expr) ecb_expect (!!(expr), 0) |
|
|
134 | #define ecb_expect_true(expr) ecb_expect (!!(expr), 1) |
|
|
135 | /* for compatibility to the rest of the world */ |
|
|
136 | #define ecb_likely(expr) ecb_expect_true (expr) |
|
|
137 | #define ecb_unlikely(expr) ecb_expect_false (expr) |
|
|
138 | |
|
|
139 | /* count trailing zero bits and count # of one bits */ |
|
|
140 | #if ECB_GCC_VERSION(3,4) |
|
|
141 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
|
|
142 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
|
|
143 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
|
|
144 | #define ecb_ctz32(x) __builtin_ctz (x) |
|
|
145 | #define ecb_ctz64(x) __builtin_ctzll (x) |
|
|
146 | #define ecb_popcount32(x) __builtin_popcount (x) |
|
|
147 | /* no popcountll */ |
|
|
148 | #else |
|
|
149 | ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; |
|
|
150 | ecb_function_ int |
|
|
151 | ecb_ctz32 (uint32_t x) |
|
|
152 | { |
|
|
153 | int r = 0; |
|
|
154 | |
|
|
155 | x &= ~x + 1; /* this isolates the lowest bit */ |
|
|
156 | |
|
|
157 | #if ECB_branchless_on_i386 |
|
|
158 | r += !!(x & 0xaaaaaaaa) << 0; |
|
|
159 | r += !!(x & 0xcccccccc) << 1; |
|
|
160 | r += !!(x & 0xf0f0f0f0) << 2; |
|
|
161 | r += !!(x & 0xff00ff00) << 3; |
|
|
162 | r += !!(x & 0xffff0000) << 4; |
|
|
163 | #else |
|
|
164 | if (x & 0xaaaaaaaa) r += 1; |
|
|
165 | if (x & 0xcccccccc) r += 2; |
|
|
166 | if (x & 0xf0f0f0f0) r += 4; |
|
|
167 | if (x & 0xff00ff00) r += 8; |
|
|
168 | if (x & 0xffff0000) r += 16; |
|
|
169 | #endif |
|
|
170 | |
|
|
171 | return r; |
|
|
172 | } |
|
|
173 | |
|
|
174 | ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
|
|
175 | ecb_function_ int |
|
|
176 | ecb_ctz64 (uint64_t x) |
|
|
177 | { |
|
|
178 | int shift = x & 0xffffffffU ? 0 : 32; |
|
|
179 | return ecb_ctz32 (x >> shift) + shift; |
|
|
180 | } |
|
|
181 | |
|
|
182 | ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
|
|
183 | ecb_function_ int |
|
|
184 | ecb_popcount32 (uint32_t x) |
|
|
185 | { |
|
|
186 | x -= (x >> 1) & 0x55555555; |
|
|
187 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
|
|
188 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
|
|
189 | x *= 0x01010101; |
|
|
190 | |
|
|
191 | return x >> 24; |
|
|
192 | } |
|
|
193 | |
|
|
194 | /* you have the choice beetween something with a table lookup, */ |
|
|
195 | /* something using lots of bit arithmetic and a simple loop */ |
|
|
196 | /* we went for the loop */ |
|
|
197 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
|
|
198 | ecb_function_ int ecb_ld32 (uint32_t x) |
|
|
199 | { |
|
|
200 | int r = 0; |
|
|
201 | |
|
|
202 | if (x >> 16) { x >>= 16; r += 16; } |
|
|
203 | if (x >> 8) { x >>= 8; r += 8; } |
|
|
204 | if (x >> 4) { x >>= 4; r += 4; } |
|
|
205 | if (x >> 2) { x >>= 2; r += 2; } |
|
|
206 | if (x >> 1) { r += 1; } |
|
|
207 | |
|
|
208 | return r; |
|
|
209 | } |
|
|
210 | |
|
|
211 | ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
|
|
212 | ecb_function_ int ecb_ld64 (uint64_t x) |
|
|
213 | { |
|
|
214 | int r = 0; |
|
|
215 | |
|
|
216 | if (x >> 32) { x >>= 32; r += 32; } |
|
|
217 | |
|
|
218 | return r + ecb_ld32 (x); |
|
|
219 | } |
|
|
220 | #endif |
|
|
221 | |
|
|
222 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
|
|
223 | /* so for this version we are lazy */ |
|
|
224 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
|
|
225 | ecb_function_ int |
|
|
226 | ecb_popcount64 (uint64_t x) |
|
|
227 | { |
|
|
228 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
|
|
229 | } |
|
|
230 | |
|
|
231 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; |
|
|
232 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; |
|
|
233 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; |
|
|
234 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; |
|
|
235 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
|
|
236 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
|
|
237 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
|
|
238 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
|
|
239 | |
|
|
240 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
|
|
241 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
|
|
242 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
|
|
243 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
|
|
244 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
|
|
245 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
|
|
246 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
|
|
247 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
|
|
248 | |
|
|
249 | #if ECB_GCC_VERSION(4,3) |
|
|
250 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
|
|
251 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
|
|
252 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
|
|
253 | #else |
|
|
254 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
|
|
255 | ecb_function_ uint16_t |
|
|
256 | ecb_bswap16 (uint16_t x) |
|
|
257 | { |
|
|
258 | return ecb_rotl16 (x, 8); |
|
|
259 | } |
|
|
260 | |
|
|
261 | ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
|
|
262 | ecb_function_ uint32_t |
|
|
263 | ecb_bswap32 (uint32_t x) |
|
|
264 | { |
|
|
265 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
|
|
266 | } |
|
|
267 | |
|
|
268 | ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
|
|
269 | ecb_function_ uint64_t |
|
|
270 | ecb_bswap64 (uint64_t x) |
|
|
271 | { |
|
|
272 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
|
|
273 | } |
|
|
274 | #endif |
60 | |
275 | |
61 | #if ECB_GCC_VERSION(4,5) |
276 | #if ECB_GCC_VERSION(4,5) |
62 | # define ecb_unreachable() __builtin_unreachable () |
277 | #define ecb_unreachable() __builtin_unreachable () |
63 | #else |
278 | #else |
64 | // this seems to work fine, but gcc always emits a warning for it :/ |
279 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
65 | static inline void ecb_unreachable () ecb_attribute ((noreturn)); |
280 | ecb_function_ void ecb_unreachable (void) ecb_noreturn; |
66 | static inline void ecb_unreachable () { } |
281 | ecb_function_ void ecb_unreachable (void) { } |
67 | #endif |
282 | #endif |
68 | |
283 | |
69 | // put into ifs if you are very sure that the expression |
|
|
70 | // is mostly true or mosty false. note that these return |
|
|
71 | // booleans, not the expression. |
|
|
72 | #define ecb_expect_false(expr) ecb_expect ((expr) ? 1 : 0, 0) |
|
|
73 | #define ecb_expect_true(expr) ecb_expect ((expr) ? 1 : 0, 1) |
|
|
74 | |
|
|
75 | // try to tell the compiler that some condition is definitely true |
284 | /* try to tell the compiler that some condition is definitely true */ |
76 | #define ecb_assume(cond) do { if (!(cond)) unreachable (); } while (0) |
285 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
77 | |
286 | |
78 | // count trailing zero bits and count # of one bits |
287 | ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; |
79 | #if ECB_GCC_VERSION(3,4) |
288 | ecb_function_ unsigned char |
80 | static inline int ecb_ctz (unsigned int x) { return __builtin_ctz (x); } |
289 | ecb_byteorder_helper (void) |
81 | static inline int ecb_popcount (unsigned int x) { return __builtin_popcount (x); } |
|
|
82 | #else |
|
|
83 | static int ecb_ctz (unsigned int x) ecb_const |
|
|
84 | { |
290 | { |
85 | int r = 0; |
291 | const uint32_t u = 0x11223344; |
86 | |
292 | return *(unsigned char *)&u; |
87 | x &= -x; // this isolates the lowest bit |
|
|
88 | |
|
|
89 | if (x & 0xaaaaaaaa) r += 1; |
|
|
90 | if (x & 0xcccccccc) r += 2; |
|
|
91 | if (x & 0xf0f0f0f0) r += 4; |
|
|
92 | if (x & 0xff00ff00) r += 8; |
|
|
93 | if (x & 0xffff0000) r += 16; |
|
|
94 | |
|
|
95 | return r; |
|
|
96 | } |
293 | } |
97 | |
294 | |
98 | static int ecb_popcount (unsigned int x) ecb_const |
295 | ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; |
99 | { |
296 | ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
100 | x -= (x >> 1) & 0x55555555; |
297 | ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; |
101 | x = ((x >> 2) & 0x33333333) + (x & 0x33333333); |
298 | ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
102 | x = ((x >> 4) + x) & 0x0f0f0f0f; |
|
|
103 | x *= 0x01010101; |
|
|
104 | |
299 | |
105 | return x >> 24; |
300 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
106 | } |
301 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
|
|
302 | #else |
|
|
303 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
107 | #endif |
304 | #endif |
108 | |
305 | |
|
|
306 | #if ecb_cplusplus_does_not_suck |
|
|
307 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
|
|
308 | template<typename T, int N> |
|
|
309 | static inline int ecb_array_length (const T (&arr)[N]) |
|
|
310 | { |
|
|
311 | return N; |
|
|
312 | } |
|
|
313 | #else |
|
|
314 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
109 | #endif |
315 | #endif |
110 | |
316 | |
|
|
317 | #endif |
|
|
318 | |