| 1 | /* |
1 | /* |
| 2 | * libecb - http://software.schmorp.de/pkg/libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
| 3 | * |
3 | * |
| 4 | * Copyright (©) 2009-2011 Marc Alexander Lehmann <libecb@schmorp.de> |
4 | * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> |
| 5 | * Copyright (©) 2011 Emanuele Giaquinta |
5 | * Copyright (©) 2011 Emanuele Giaquinta |
| 6 | * All rights reserved. |
6 | * All rights reserved. |
| 7 | * |
7 | * |
| 8 | * 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- |
| 9 | * tion, are permitted provided that the following conditions are met: |
9 | * tion, are permitted provided that the following conditions are met: |
| … | |
… | |
| 28 | */ |
28 | */ |
| 29 | |
29 | |
| 30 | #ifndef ECB_H |
30 | #ifndef ECB_H |
| 31 | #define ECB_H |
31 | #define ECB_H |
| 32 | |
32 | |
|
|
33 | /* 16 bits major, 16 bits minor */ |
|
|
34 | #define ECB_VERSION 0x00010001 |
|
|
35 | |
| 33 | #ifdef _WIN32 |
36 | #ifdef _WIN32 |
| 34 | typedef signed char int8_t; |
37 | typedef signed char int8_t; |
| 35 | typedef unsigned char uint8_t; |
38 | typedef unsigned char uint8_t; |
| 36 | typedef signed short int16_t; |
39 | typedef signed short int16_t; |
| 37 | typedef unsigned short uint16_t; |
40 | typedef unsigned short uint16_t; |
| 38 | typedef signed int int32_t; |
41 | typedef signed int int32_t; |
| 39 | typedef unsigned int uint32_t; |
42 | typedef unsigned int uint32_t; |
| 40 | #if __GNUC__ |
43 | #if __GNUC__ |
| 41 | typedef signed long long int64_t; |
44 | typedef signed long long int64_t; |
| 42 | typedef unsigned long long uint64_t; |
45 | typedef unsigned long long uint64_t; |
| 43 | #else |
46 | #else /* _MSC_VER || __BORLANDC__ */ |
| 44 | typedef signed __int64 int64_t; |
47 | typedef signed __int64 int64_t; |
| 45 | typedef unsigned __int64 uint64_t; |
48 | typedef unsigned __int64 uint64_t; |
| 46 | #endif |
49 | #endif |
|
|
50 | #ifdef _WIN64 |
|
|
51 | #define ECB_PTRSIZE 8 |
|
|
52 | typedef uint64_t uintptr_t; |
|
|
53 | typedef int64_t intptr_t; |
|
|
54 | #else |
|
|
55 | #define ECB_PTRSIZE 4 |
|
|
56 | typedef uint32_t uintptr_t; |
|
|
57 | typedef int32_t intptr_t; |
|
|
58 | #endif |
|
|
59 | typedef intptr_t ptrdiff_t; |
| 47 | #else |
60 | #else |
| 48 | #include <inttypes.h> |
61 | #include <inttypes.h> |
|
|
62 | #if UINTMAX_MAX > 0xffffffffU |
|
|
63 | #define ECB_PTRSIZE 8 |
|
|
64 | #else |
|
|
65 | #define ECB_PTRSIZE 4 |
|
|
66 | #endif |
| 49 | #endif |
67 | #endif |
| 50 | |
68 | |
| 51 | /* many compilers define _GNUC_ to some versions but then only implement |
69 | /* many compilers define _GNUC_ to some versions but then only implement |
| 52 | * what their idiot authors think are the "more important" extensions, |
70 | * what their idiot authors think are the "more important" extensions, |
| 53 | * causing enourmous grief in return for some better fake benchmark numbers. |
71 | * causing enormous grief in return for some better fake benchmark numbers. |
| 54 | * or so. |
72 | * or so. |
| 55 | * we try to detect these and simply assume they are not gcc - if they have |
73 | * 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. |
74 | * an issue with that they should have done it right in the first place. |
| 57 | */ |
75 | */ |
| 58 | #ifndef ECB_GCC_VERSION |
76 | #ifndef ECB_GCC_VERSION |
| 59 | #if !defined(__GNUC_MINOR__) || defined(__INTEL_COMPILER) || defined(__SUNPRO_C) || defined(__SUNPRO_CC) || defined(__llvm__) || defined(__clang__) |
77 | #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 |
78 | #define ECB_GCC_VERSION(major,minor) 0 |
| 61 | #else |
79 | #else |
| 62 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
80 | #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) |
| 63 | #endif |
81 | #endif |
| 64 | #endif |
82 | #endif |
| 65 | |
83 | |
|
|
84 | #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ |
| 66 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
85 | #define ECB_C99 (__STDC_VERSION__ >= 199901L) |
|
|
86 | #define ECB_C11 (__STDC_VERSION__ >= 201112L) |
|
|
87 | #define ECB_CPP (__cplusplus+0) |
|
|
88 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
89 | |
|
|
90 | /*****************************************************************************/ |
|
|
91 | |
|
|
92 | /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */ |
|
|
93 | /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */ |
|
|
94 | |
|
|
95 | #if ECB_NO_THREADS |
|
|
96 | #define ECB_NO_SMP 1 |
|
|
97 | #endif |
|
|
98 | |
|
|
99 | #if ECB_NO_SMP |
|
|
100 | #define ECB_MEMORY_FENCE do { } while (0) |
|
|
101 | #endif |
|
|
102 | |
|
|
103 | #ifndef ECB_MEMORY_FENCE |
|
|
104 | #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
105 | #if __i386 || __i386__ |
|
|
106 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") |
|
|
107 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
|
|
108 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
109 | #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ |
|
|
110 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
|
|
111 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
|
|
112 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
113 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
|
|
114 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
115 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
|
|
116 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
|
|
117 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") |
|
|
118 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
|
|
119 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
|
|
120 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
|
|
121 | #elif __sparc || __sparc__ |
|
|
122 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") |
|
|
123 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") |
|
|
124 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") |
|
|
125 | #elif defined __s390__ || defined __s390x__ |
|
|
126 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") |
|
|
127 | #elif defined __mips__ |
|
|
128 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") |
|
|
129 | #elif defined __alpha__ |
|
|
130 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory") |
|
|
131 | #elif defined __hppa__ |
|
|
132 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory") |
|
|
133 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
|
|
134 | #elif defined __ia64__ |
|
|
135 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory") |
|
|
136 | #endif |
|
|
137 | #endif |
|
|
138 | #endif |
|
|
139 | |
|
|
140 | #ifndef ECB_MEMORY_FENCE |
|
|
141 | #if ECB_GCC_VERSION(4,7) |
|
|
142 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
143 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
144 | #elif defined __clang && __has_feature (cxx_atomic) |
|
|
145 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
|
|
146 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
|
|
147 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
|
|
148 | #define ECB_MEMORY_FENCE __sync_synchronize () |
|
|
149 | #elif _MSC_VER >= 1400 /* VC++ 2005 */ |
|
|
150 | #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) |
|
|
151 | #define ECB_MEMORY_FENCE _ReadWriteBarrier () |
|
|
152 | #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ |
|
|
153 | #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () |
|
|
154 | #elif defined _WIN32 |
|
|
155 | #include <WinNT.h> |
|
|
156 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
|
|
157 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
|
|
158 | #include <mbarrier.h> |
|
|
159 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
|
|
160 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
|
|
161 | #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () |
|
|
162 | #elif __xlC__ |
|
|
163 | #define ECB_MEMORY_FENCE __sync () |
|
|
164 | #endif |
|
|
165 | #endif |
|
|
166 | |
|
|
167 | #ifndef ECB_MEMORY_FENCE |
|
|
168 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
|
|
169 | /* we assume that these memory fences work on all variables/all memory accesses, */ |
|
|
170 | /* not just C11 atomics and atomic accesses */ |
|
|
171 | #include <stdatomic.h> |
|
|
172 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
173 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
174 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
175 | /* for most usages, or gcc and clang have a bug */ |
|
|
176 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
177 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
178 | #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) |
|
|
179 | #endif |
|
|
180 | #endif |
|
|
181 | |
|
|
182 | #ifndef ECB_MEMORY_FENCE |
|
|
183 | #if !ECB_AVOID_PTHREADS |
|
|
184 | /* |
|
|
185 | * if you get undefined symbol references to pthread_mutex_lock, |
|
|
186 | * or failure to find pthread.h, then you should implement |
|
|
187 | * the ECB_MEMORY_FENCE operations for your cpu/compiler |
|
|
188 | * OR provide pthread.h and link against the posix thread library |
|
|
189 | * of your system. |
|
|
190 | */ |
|
|
191 | #include <pthread.h> |
|
|
192 | #define ECB_NEEDS_PTHREADS 1 |
|
|
193 | #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1 |
|
|
194 | |
|
|
195 | static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER; |
|
|
196 | #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0) |
|
|
197 | #endif |
|
|
198 | #endif |
|
|
199 | |
|
|
200 | #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE |
|
|
201 | #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE |
|
|
202 | #endif |
|
|
203 | |
|
|
204 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
|
|
205 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
|
|
206 | #endif |
|
|
207 | |
|
|
208 | /*****************************************************************************/ |
| 67 | |
209 | |
| 68 | #if __cplusplus |
210 | #if __cplusplus |
| 69 | #define ecb_inline static inline |
211 | #define ecb_inline static inline |
| 70 | #elif ECB_GCC_VERSION(2,5) |
212 | #elif ECB_GCC_VERSION(2,5) |
| 71 | #define ecb_inline static __inline__ |
213 | #define ecb_inline static __inline__ |
| … | |
… | |
| 110 | #elif ECB_GCC_VERSION(3,0) |
252 | #elif ECB_GCC_VERSION(3,0) |
| 111 | #define ecb_decltype(x) __typeof(x) |
253 | #define ecb_decltype(x) __typeof(x) |
| 112 | #endif |
254 | #endif |
| 113 | |
255 | |
| 114 | #define ecb_noinline ecb_attribute ((__noinline__)) |
256 | #define ecb_noinline ecb_attribute ((__noinline__)) |
| 115 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
| 116 | #define ecb_unused ecb_attribute ((__unused__)) |
257 | #define ecb_unused ecb_attribute ((__unused__)) |
| 117 | #define ecb_const ecb_attribute ((__const__)) |
258 | #define ecb_const ecb_attribute ((__const__)) |
| 118 | #define ecb_pure ecb_attribute ((__pure__)) |
259 | #define ecb_pure ecb_attribute ((__pure__)) |
|
|
260 | |
|
|
261 | #if ECB_C11 |
|
|
262 | #define ecb_noreturn _Noreturn |
|
|
263 | #else |
|
|
264 | #define ecb_noreturn ecb_attribute ((__noreturn__)) |
|
|
265 | #endif |
| 119 | |
266 | |
| 120 | #if ECB_GCC_VERSION(4,3) |
267 | #if ECB_GCC_VERSION(4,3) |
| 121 | #define ecb_artificial ecb_attribute ((__artificial__)) |
268 | #define ecb_artificial ecb_attribute ((__artificial__)) |
| 122 | #define ecb_hot ecb_attribute ((__hot__)) |
269 | #define ecb_hot ecb_attribute ((__hot__)) |
| 123 | #define ecb_cold ecb_attribute ((__cold__)) |
270 | #define ecb_cold ecb_attribute ((__cold__)) |
| … | |
… | |
| 152 | { |
299 | { |
| 153 | int r = 0; |
300 | int r = 0; |
| 154 | |
301 | |
| 155 | x &= ~x + 1; /* this isolates the lowest bit */ |
302 | x &= ~x + 1; /* this isolates the lowest bit */ |
| 156 | |
303 | |
|
|
304 | #if ECB_branchless_on_i386 |
|
|
305 | r += !!(x & 0xaaaaaaaa) << 0; |
|
|
306 | r += !!(x & 0xcccccccc) << 1; |
|
|
307 | r += !!(x & 0xf0f0f0f0) << 2; |
|
|
308 | r += !!(x & 0xff00ff00) << 3; |
|
|
309 | r += !!(x & 0xffff0000) << 4; |
|
|
310 | #else |
| 157 | if (x & 0xaaaaaaaa) r += 1; |
311 | if (x & 0xaaaaaaaa) r += 1; |
| 158 | if (x & 0xcccccccc) r += 2; |
312 | if (x & 0xcccccccc) r += 2; |
| 159 | if (x & 0xf0f0f0f0) r += 4; |
313 | if (x & 0xf0f0f0f0) r += 4; |
| 160 | if (x & 0xff00ff00) r += 8; |
314 | if (x & 0xff00ff00) r += 8; |
| 161 | if (x & 0xffff0000) r += 16; |
315 | if (x & 0xffff0000) r += 16; |
|
|
316 | #endif |
| 162 | |
317 | |
| 163 | return r; |
318 | return r; |
| 164 | } |
319 | } |
| 165 | |
320 | |
| 166 | ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
321 | ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; |
| 167 | ecb_function_ int |
322 | ecb_function_ int |
| 168 | ecb_ctz64 (uint64_t x) |
323 | ecb_ctz64 (uint64_t x) |
| 169 | { |
324 | { |
| 170 | int shift = x & 0xffffffffU ? 0 : 32; |
325 | int shift = x & 0xffffffffU ? 0 : 32; |
| 171 | return ecb_ctz (x >> shift) + shift; |
326 | return ecb_ctz32 (x >> shift) + shift; |
| 172 | } |
327 | } |
| 173 | |
328 | |
| 174 | ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
329 | ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; |
| 175 | ecb_function_ int |
330 | ecb_function_ int |
| 176 | ecb_popcount32 (uint32_t x) |
331 | ecb_popcount32 (uint32_t x) |
| … | |
… | |
| 181 | x *= 0x01010101; |
336 | x *= 0x01010101; |
| 182 | |
337 | |
| 183 | return x >> 24; |
338 | return x >> 24; |
| 184 | } |
339 | } |
| 185 | |
340 | |
| 186 | /* you have the choice beetween something with a table lookup, */ |
|
|
| 187 | /* something using lots of bit arithmetic and a simple loop */ |
|
|
| 188 | /* we went for the loop */ |
|
|
| 189 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
341 | ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; |
| 190 | ecb_function_ int ecb_ld32 (uint32_t x) |
342 | ecb_function_ int ecb_ld32 (uint32_t x) |
| 191 | { |
343 | { |
| 192 | int r = -1; |
344 | int r = 0; |
| 193 | |
345 | |
| 194 | do |
346 | if (x >> 16) { x >>= 16; r += 16; } |
| 195 | { |
347 | if (x >> 8) { x >>= 8; r += 8; } |
| 196 | x >>= 1; |
348 | if (x >> 4) { x >>= 4; r += 4; } |
| 197 | ++r; |
349 | if (x >> 2) { x >>= 2; r += 2; } |
| 198 | } |
350 | if (x >> 1) { r += 1; } |
| 199 | while (x); |
|
|
| 200 | |
351 | |
| 201 | return r; |
352 | return r; |
| 202 | } |
353 | } |
| 203 | |
354 | |
| 204 | ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
355 | ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; |
| 205 | ecb_function_ int ecb_ld64 (uint64_t x) |
356 | ecb_function_ int ecb_ld64 (uint64_t x) |
| 206 | { |
357 | { |
| 207 | int r = -1; |
358 | int r = 0; |
| 208 | |
359 | |
| 209 | do |
360 | if (x >> 32) { x >>= 32; r += 32; } |
| 210 | { |
|
|
| 211 | x >>= 1; |
|
|
| 212 | ++r; |
|
|
| 213 | } |
|
|
| 214 | while (x); |
|
|
| 215 | |
361 | |
|
|
362 | return r + ecb_ld32 (x); |
|
|
363 | } |
|
|
364 | #endif |
|
|
365 | |
|
|
366 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; |
|
|
367 | ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } |
|
|
368 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; |
|
|
369 | ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } |
|
|
370 | |
|
|
371 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; |
|
|
372 | ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) |
|
|
373 | { |
|
|
374 | return ( (x * 0x0802U & 0x22110U) |
|
|
375 | | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; |
|
|
376 | } |
|
|
377 | |
|
|
378 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; |
|
|
379 | ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) |
|
|
380 | { |
|
|
381 | x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); |
|
|
382 | x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); |
|
|
383 | x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); |
|
|
384 | x = ( x >> 8 ) | ( x << 8); |
|
|
385 | |
| 216 | return r; |
386 | return x; |
| 217 | } |
387 | } |
| 218 | #endif |
388 | |
|
|
389 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; |
|
|
390 | ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) |
|
|
391 | { |
|
|
392 | x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); |
|
|
393 | x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); |
|
|
394 | x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); |
|
|
395 | x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); |
|
|
396 | x = ( x >> 16 ) | ( x << 16); |
|
|
397 | |
|
|
398 | return x; |
|
|
399 | } |
| 219 | |
400 | |
| 220 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
401 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
| 221 | /* so for this version we are lazy */ |
402 | /* so for this version we are lazy */ |
| 222 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
403 | ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; |
| 223 | ecb_function_ int |
404 | ecb_function_ int |
| 224 | ecb_popcount64 (uint64_t x) |
405 | ecb_popcount64 (uint64_t x) |
| 225 | { |
406 | { |
| 226 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
407 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
| 227 | } |
408 | } |
| 228 | |
409 | |
|
|
410 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; |
|
|
411 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; |
|
|
412 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; |
|
|
413 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; |
|
|
414 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
|
|
415 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
|
|
416 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
|
|
417 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
|
|
418 | |
|
|
419 | ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
|
|
420 | ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
|
|
421 | ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
|
|
422 | ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
|
|
423 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
|
|
424 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
|
|
425 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
|
|
426 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
|
|
427 | |
| 229 | #if ECB_GCC_VERSION(4,3) |
428 | #if ECB_GCC_VERSION(4,3) |
| 230 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
429 | #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) |
| 231 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
430 | #define ecb_bswap32(x) __builtin_bswap32 (x) |
| 232 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
431 | #define ecb_bswap64(x) __builtin_bswap64 (x) |
| 233 | #else |
432 | #else |
| 234 | ecb_function_ uint32_t ecb_bswap16 (uint32_t x) ecb_const; |
433 | ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; |
| 235 | ecb_function_ uint32_t |
434 | ecb_function_ uint16_t |
| 236 | ecb_bswap16 (uint32_t x) |
435 | ecb_bswap16 (uint16_t x) |
| 237 | { |
436 | { |
| 238 | return ((x >> 8) & 0xff) |
437 | return ecb_rotl16 (x, 8); |
| 239 | | ((x << 8) & 0x00ff0000) |
|
|
| 240 | | (x << 24); |
|
|
| 241 | } |
438 | } |
| 242 | |
439 | |
| 243 | ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
440 | ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; |
| 244 | ecb_function_ uint32_t |
441 | ecb_function_ uint32_t |
| 245 | ecb_bswap32 (uint32_t x) |
442 | ecb_bswap32 (uint32_t x) |
| 246 | { |
443 | { |
| 247 | return (x >> 24) |
444 | return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); |
| 248 | | ((x >> 8) & 0x0000ff00) |
|
|
| 249 | | ((x << 8) & 0x00ff0000) |
|
|
| 250 | | (x << 24); |
|
|
| 251 | } |
445 | } |
| 252 | |
446 | |
| 253 | ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
447 | ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; |
| 254 | ecb_function_ uint64_t |
448 | ecb_function_ uint64_t |
| 255 | ecb_bswap64 (uint64_t x) |
449 | ecb_bswap64 (uint64_t x) |
| 256 | { |
450 | { |
| 257 | return (((uint64_t)ecb_bswap32 (x)) << 32) |
451 | return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); |
| 258 | | ecb_bswap32 (x >> 32); |
|
|
| 259 | } |
452 | } |
| 260 | #endif |
453 | #endif |
| 261 | |
454 | |
| 262 | #if ECB_GCC_VERSION(4,5) |
455 | #if ECB_GCC_VERSION(4,5) |
| 263 | #define ecb_unreachable() __builtin_unreachable () |
456 | #define ecb_unreachable() __builtin_unreachable () |
| 264 | #else |
457 | #else |
| 265 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
458 | /* this seems to work fine, but gcc always emits a warning for it :/ */ |
| 266 | ecb_function_ void ecb_unreachable (void) ecb_noreturn; |
459 | ecb_inline void ecb_unreachable (void) ecb_noreturn; |
| 267 | ecb_function_ void ecb_unreachable (void) { } |
460 | ecb_inline void ecb_unreachable (void) { } |
| 268 | #endif |
461 | #endif |
| 269 | |
462 | |
| 270 | /* try to tell the compiler that some condition is definitely true */ |
463 | /* try to tell the compiler that some condition is definitely true */ |
| 271 | #define ecb_assume(cond) do { if (!(cond)) ecb_unreachable (); } while (0) |
464 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
| 272 | |
465 | |
| 273 | ecb_function_ unsigned char ecb_byteorder_helper (void) ecb_const; |
466 | ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; |
| 274 | ecb_function_ unsigned char |
467 | ecb_inline unsigned char |
| 275 | ecb_byteorder_helper (void) |
468 | ecb_byteorder_helper (void) |
| 276 | { |
469 | { |
| 277 | const uint32_t u = 0x11223344; |
470 | /* the union code still generates code under pressure in gcc, */ |
| 278 | return *(unsigned char *)&u; |
471 | /* but less than using pointers, and always seem to */ |
|
|
472 | /* successfully return a constant. */ |
|
|
473 | /* the reason why we have this horrible preprocessor mess */ |
|
|
474 | /* is to avoid it in all cases, at least on common architectures */ |
|
|
475 | /* and yes, gcc defines __BYTE_ORDER__, g++ does not */ |
|
|
476 | #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64 |
|
|
477 | return 0x44; |
|
|
478 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
|
|
479 | return 0x44; |
|
|
480 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
|
|
481 | retrurn 0x11; |
|
|
482 | #else |
|
|
483 | union |
|
|
484 | { |
|
|
485 | uint32_t i; |
|
|
486 | uint8_t c; |
|
|
487 | } u = { 0x11223344 }; |
|
|
488 | return u.c; |
|
|
489 | #endif |
| 279 | } |
490 | } |
| 280 | |
491 | |
| 281 | ecb_function_ ecb_bool ecb_big_endian (void) ecb_const; |
492 | ecb_inline ecb_bool ecb_big_endian (void) ecb_const; |
| 282 | ecb_function_ ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
493 | ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
| 283 | ecb_function_ ecb_bool ecb_little_endian (void) ecb_const; |
494 | ecb_inline ecb_bool ecb_little_endian (void) ecb_const; |
| 284 | ecb_function_ ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
495 | ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
| 285 | |
496 | |
| 286 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
497 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
| 287 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
498 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
| 288 | #else |
499 | #else |
| 289 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
500 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
|
|
501 | #endif |
|
|
502 | |
|
|
503 | #if __cplusplus |
|
|
504 | template<typename T> |
|
|
505 | static inline T ecb_div_rd (T val, T div) |
|
|
506 | { |
|
|
507 | return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; |
|
|
508 | } |
|
|
509 | template<typename T> |
|
|
510 | static inline T ecb_div_ru (T val, T div) |
|
|
511 | { |
|
|
512 | return val < 0 ? - ((-val ) / div) : (val + div - 1) / div; |
|
|
513 | } |
|
|
514 | #else |
|
|
515 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
|
|
516 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
| 290 | #endif |
517 | #endif |
| 291 | |
518 | |
| 292 | #if ecb_cplusplus_does_not_suck |
519 | #if ecb_cplusplus_does_not_suck |
| 293 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
520 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
| 294 | template<typename T, int N> |
521 | template<typename T, int N> |
| … | |
… | |
| 298 | } |
525 | } |
| 299 | #else |
526 | #else |
| 300 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
527 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
| 301 | #endif |
528 | #endif |
| 302 | |
529 | |
| 303 | ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; |
|
|
| 304 | ecb_inline uint32_t |
|
|
| 305 | ecb_rotr32 (uint32_t x, unsigned int count) |
|
|
| 306 | { |
|
|
| 307 | return (x << (32 - count)) | (x >> count); |
|
|
| 308 | } |
|
|
| 309 | |
|
|
| 310 | ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; |
|
|
| 311 | ecb_inline uint32_t |
|
|
| 312 | ecb_rotl32 (uint32_t x, unsigned int count) |
|
|
| 313 | { |
|
|
| 314 | return (x >> (32 - count)) | (x << count); |
|
|
| 315 | } |
|
|
| 316 | |
|
|
| 317 | ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; |
|
|
| 318 | ecb_inline uint64_t |
|
|
| 319 | ecb_rotr64 (uint64_t x, unsigned int count) |
|
|
| 320 | { |
|
|
| 321 | return (x << (64 - count)) | (x >> count); |
|
|
| 322 | } |
|
|
| 323 | |
|
|
| 324 | ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; |
|
|
| 325 | ecb_inline uint64_t |
|
|
| 326 | ecb_rotl64 (uint64_t x, unsigned int count) |
|
|
| 327 | { |
|
|
| 328 | return (x >> (64 - count)) | (x << count); |
|
|
| 329 | } |
|
|
| 330 | |
|
|
| 331 | #endif |
530 | #endif |
| 332 | |
531 | |
