ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.115
Committed: Wed Dec 19 23:51:24 2012 UTC (11 years, 6 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.114: +2 -2 lines
Log Message:
§

File Contents

# User Rev Content
1 root 1.1 /*
2 root 1.17 * libecb - http://software.schmorp.de/pkg/libecb
3 root 1.1 *
4 root 1.72 * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de>
5 root 1.7 * Copyright (©) 2011 Emanuele Giaquinta
6 root 1.1 * All rights reserved.
7     *
8     * Redistribution and use in source and binary forms, with or without modifica-
9     * tion, are permitted provided that the following conditions are met:
10     *
11     * 1. Redistributions of source code must retain the above copyright notice,
12     * this list of conditions and the following disclaimer.
13     *
14     * 2. Redistributions in binary form must reproduce the above copyright
15     * notice, this list of conditions and the following disclaimer in the
16     * documentation and/or other materials provided with the distribution.
17     *
18     * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
19     * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
20     * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
21     * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
22     * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
23     * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
24     * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25     * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
26     * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
27     * OF THE POSSIBILITY OF SUCH DAMAGE.
28     */
29    
30     #ifndef ECB_H
31     #define ECB_H
32    
33 root 1.87 /* 16 bits major, 16 bits minor */
34 root 1.114 #define ECB_VERSION 0x00010003
35 root 1.87
36 root 1.44 #ifdef _WIN32
37     typedef signed char int8_t;
38     typedef unsigned char uint8_t;
39     typedef signed short int16_t;
40     typedef unsigned short uint16_t;
41     typedef signed int int32_t;
42     typedef unsigned int uint32_t;
43     #if __GNUC__
44     typedef signed long long int64_t;
45     typedef unsigned long long uint64_t;
46 root 1.51 #else /* _MSC_VER || __BORLANDC__ */
47 root 1.44 typedef signed __int64 int64_t;
48     typedef unsigned __int64 uint64_t;
49     #endif
50 root 1.87 #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 root 1.44 #else
60     #include <inttypes.h>
61 root 1.87 #if UINTMAX_MAX > 0xffffffffU
62     #define ECB_PTRSIZE 8
63     #else
64     #define ECB_PTRSIZE 4
65     #endif
66 root 1.44 #endif
67 root 1.6
68 root 1.114 /* work around x32 idiocy by defining proper macros */
69     #if __x86_64 || _M_AMD64
70     #if __ILP32
71 root 1.115 #define ECB_AMD64_X32 1
72 root 1.114 #else
73 root 1.115 #define ECB_AMD64 1
74 root 1.114 #endif
75     #endif
76    
77 root 1.12 /* many compilers define _GNUC_ to some versions but then only implement
78     * what their idiot authors think are the "more important" extensions,
79 sf-exg 1.59 * causing enormous grief in return for some better fake benchmark numbers.
80 root 1.18 * or so.
81 root 1.12 * we try to detect these and simply assume they are not gcc - if they have
82     * an issue with that they should have done it right in the first place.
83     */
84 root 1.14 #ifndef ECB_GCC_VERSION
85 root 1.85 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
86 root 1.37 #define ECB_GCC_VERSION(major,minor) 0
87     #else
88     #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
89     #endif
90 root 1.12 #endif
91 root 1.1
92 root 1.91 #define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */
93     #define ECB_C99 (__STDC_VERSION__ >= 199901L)
94     #define ECB_C11 (__STDC_VERSION__ >= 201112L)
95     #define ECB_CPP (__cplusplus+0)
96     #define ECB_CPP11 (__cplusplus >= 201103L)
97 root 1.90
98 root 1.102 #if ECB_CPP
99     #define ECB_EXTERN_C extern "C"
100     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
101     #define ECB_EXTERN_C_END }
102     #else
103     #define ECB_EXTERN_C extern
104     #define ECB_EXTERN_C_BEG
105     #define ECB_EXTERN_C_END
106     #endif
107    
108 root 1.52 /*****************************************************************************/
109    
110 root 1.58 /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
111     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
112    
113 root 1.79 #if ECB_NO_THREADS
114 root 1.95 #define ECB_NO_SMP 1
115 root 1.79 #endif
116    
117 root 1.93 #if ECB_NO_SMP
118 root 1.64 #define ECB_MEMORY_FENCE do { } while (0)
119 root 1.58 #endif
120    
121 root 1.52 #ifndef ECB_MEMORY_FENCE
122 root 1.85 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
123 root 1.73 #if __i386 || __i386__
124 root 1.54 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
125 root 1.94 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
126     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
127 root 1.73 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
128 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
129     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
130     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
131 root 1.63 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
132 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
133 root 1.85 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
134     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
135 root 1.84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
136 root 1.85 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
137 root 1.94 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
138     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
139 root 1.73 #elif __sparc || __sparc__
140 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
141     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
142     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
143 root 1.85 #elif defined __s390__ || defined __s390x__
144 root 1.77 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
145 root 1.85 #elif defined __mips__
146 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
147 root 1.86 #elif defined __alpha__
148 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
149     #elif defined __hppa__
150     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
151     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
152     #elif defined __ia64__
153     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
154 root 1.52 #endif
155     #endif
156     #endif
157    
158     #ifndef ECB_MEMORY_FENCE
159 root 1.93 #if ECB_GCC_VERSION(4,7)
160 root 1.97 /* see comment below (stdatomic.h) about the C11 memory model. */
161 root 1.93 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
162 root 1.110
163     /* The __has_feature syntax from clang is so misdesigned that we cannot use it
164     * without risking compile time errors with other compilers. We *could*
165     * define our own ecb_clang_has_feature, but I just can't be bothered to work
166 root 1.112 * around this shit time and again.
167 root 1.110 * #elif defined __clang && __has_feature (cxx_atomic)
168     * // see comment below (stdatomic.h) about the C11 memory model.
169     * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
170     */
171    
172 root 1.93 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
173 root 1.52 #define ECB_MEMORY_FENCE __sync_synchronize ()
174 root 1.57 #elif _MSC_VER >= 1400 /* VC++ 2005 */
175     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
176     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
177     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
178     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
179 root 1.85 #elif defined _WIN32
180 root 1.55 #include <WinNT.h>
181 root 1.57 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
182 root 1.72 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
183     #include <mbarrier.h>
184     #define ECB_MEMORY_FENCE __machine_rw_barrier ()
185     #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
186     #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
187 root 1.82 #elif __xlC__
188 root 1.83 #define ECB_MEMORY_FENCE __sync ()
189 root 1.52 #endif
190     #endif
191    
192 root 1.53 #ifndef ECB_MEMORY_FENCE
193 root 1.94 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
194     /* we assume that these memory fences work on all variables/all memory accesses, */
195     /* not just C11 atomics and atomic accesses */
196     #include <stdatomic.h>
197 root 1.97 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
198     /* any fence other than seq_cst, which isn't very efficient for us. */
199     /* Why that is, we don't know - either the C11 memory model is quite useless */
200     /* for most usages, or gcc and clang have a bug */
201     /* I *currently* lean towards the latter, and inefficiently implement */
202     /* all three of ecb's fences as a seq_cst fence */
203 root 1.94 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
204     #endif
205     #endif
206    
207     #ifndef ECB_MEMORY_FENCE
208 root 1.62 #if !ECB_AVOID_PTHREADS
209     /*
210     * if you get undefined symbol references to pthread_mutex_lock,
211     * or failure to find pthread.h, then you should implement
212     * the ECB_MEMORY_FENCE operations for your cpu/compiler
213     * OR provide pthread.h and link against the posix thread library
214     * of your system.
215     */
216     #include <pthread.h>
217     #define ECB_NEEDS_PTHREADS 1
218     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
219 root 1.52
220 root 1.62 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
221     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
222     #endif
223     #endif
224    
225 root 1.85 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
226 root 1.52 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
227 root 1.62 #endif
228    
229 root 1.85 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
230 root 1.52 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
231     #endif
232    
233     /*****************************************************************************/
234    
235 root 1.39 #if __cplusplus
236 root 1.46 #define ecb_inline static inline
237 root 1.38 #elif ECB_GCC_VERSION(2,5)
238 root 1.46 #define ecb_inline static __inline__
239 root 1.39 #elif ECB_C99
240 root 1.46 #define ecb_inline static inline
241 root 1.29 #else
242 root 1.46 #define ecb_inline static
243 root 1.38 #endif
244    
245     #if ECB_GCC_VERSION(3,3)
246     #define ecb_restrict __restrict__
247 root 1.39 #elif ECB_C99
248 root 1.38 #define ecb_restrict restrict
249     #else
250     #define ecb_restrict
251 root 1.4 #endif
252    
253 root 1.38 typedef int ecb_bool;
254    
255 root 1.8 #define ECB_CONCAT_(a, b) a ## b
256     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
257     #define ECB_STRINGIFY_(a) # a
258     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
259    
260 root 1.46 #define ecb_function_ ecb_inline
261 root 1.3
262 root 1.1 #if ECB_GCC_VERSION(3,1)
263 root 1.37 #define ecb_attribute(attrlist) __attribute__(attrlist)
264     #define ecb_is_constant(expr) __builtin_constant_p (expr)
265     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
266     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
267     #else
268     #define ecb_attribute(attrlist)
269     #define ecb_is_constant(expr) 0
270     #define ecb_expect(expr,value) (expr)
271     #define ecb_prefetch(addr,rw,locality)
272 root 1.1 #endif
273    
274 root 1.2 /* no emulation for ecb_decltype */
275 root 1.1 #if ECB_GCC_VERSION(4,5)
276 root 1.37 #define ecb_decltype(x) __decltype(x)
277 root 1.1 #elif ECB_GCC_VERSION(3,0)
278 root 1.39 #define ecb_decltype(x) __typeof(x)
279 root 1.1 #endif
280    
281 root 1.24 #define ecb_noinline ecb_attribute ((__noinline__))
282     #define ecb_unused ecb_attribute ((__unused__))
283     #define ecb_const ecb_attribute ((__const__))
284     #define ecb_pure ecb_attribute ((__pure__))
285 root 1.35
286 root 1.90 #if ECB_C11
287     #define ecb_noreturn _Noreturn
288     #else
289     #define ecb_noreturn ecb_attribute ((__noreturn__))
290     #endif
291    
292 root 1.35 #if ECB_GCC_VERSION(4,3)
293 root 1.39 #define ecb_artificial ecb_attribute ((__artificial__))
294     #define ecb_hot ecb_attribute ((__hot__))
295     #define ecb_cold ecb_attribute ((__cold__))
296 root 1.35 #else
297     #define ecb_artificial
298     #define ecb_hot
299     #define ecb_cold
300     #endif
301 root 1.1
302 root 1.39 /* put around conditional expressions if you are very sure that the */
303     /* expression is mostly true or mostly false. note that these return */
304     /* booleans, not the expression. */
305 root 1.33 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
306     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
307 root 1.36 /* for compatibility to the rest of the world */
308 root 1.33 #define ecb_likely(expr) ecb_expect_true (expr)
309     #define ecb_unlikely(expr) ecb_expect_false (expr)
310 root 1.1
311 root 1.3 /* count trailing zero bits and count # of one bits */
312 root 1.1 #if ECB_GCC_VERSION(3,4)
313 root 1.49 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
314     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
315     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
316 root 1.35 #define ecb_ctz32(x) __builtin_ctz (x)
317 root 1.49 #define ecb_ctz64(x) __builtin_ctzll (x)
318 root 1.35 #define ecb_popcount32(x) __builtin_popcount (x)
319 root 1.49 /* no popcountll */
320 root 1.1 #else
321 root 1.39 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
322     ecb_function_ int
323 root 1.35 ecb_ctz32 (uint32_t x)
324     {
325     int r = 0;
326    
327 root 1.48 x &= ~x + 1; /* this isolates the lowest bit */
328 root 1.35
329 root 1.50 #if ECB_branchless_on_i386
330     r += !!(x & 0xaaaaaaaa) << 0;
331     r += !!(x & 0xcccccccc) << 1;
332     r += !!(x & 0xf0f0f0f0) << 2;
333     r += !!(x & 0xff00ff00) << 3;
334     r += !!(x & 0xffff0000) << 4;
335     #else
336 root 1.35 if (x & 0xaaaaaaaa) r += 1;
337     if (x & 0xcccccccc) r += 2;
338     if (x & 0xf0f0f0f0) r += 4;
339     if (x & 0xff00ff00) r += 8;
340     if (x & 0xffff0000) r += 16;
341 root 1.50 #endif
342 root 1.35
343     return r;
344     }
345    
346 root 1.49 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
347     ecb_function_ int
348     ecb_ctz64 (uint64_t x)
349     {
350     int shift = x & 0xffffffffU ? 0 : 32;
351 root 1.50 return ecb_ctz32 (x >> shift) + shift;
352 root 1.49 }
353    
354 root 1.39 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
355     ecb_function_ int
356 root 1.35 ecb_popcount32 (uint32_t x)
357     {
358     x -= (x >> 1) & 0x55555555;
359     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
360     x = ((x >> 4) + x) & 0x0f0f0f0f;
361     x *= 0x01010101;
362 root 1.1
363 root 1.35 return x >> 24;
364     }
365 root 1.49
366     ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
367     ecb_function_ int ecb_ld32 (uint32_t x)
368     {
369 root 1.50 int r = 0;
370 root 1.49
371 root 1.50 if (x >> 16) { x >>= 16; r += 16; }
372     if (x >> 8) { x >>= 8; r += 8; }
373     if (x >> 4) { x >>= 4; r += 4; }
374     if (x >> 2) { x >>= 2; r += 2; }
375     if (x >> 1) { r += 1; }
376 root 1.49
377     return r;
378     }
379    
380     ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
381     ecb_function_ int ecb_ld64 (uint64_t x)
382     {
383 root 1.50 int r = 0;
384 root 1.49
385 root 1.50 if (x >> 32) { x >>= 32; r += 32; }
386 root 1.49
387 root 1.50 return r + ecb_ld32 (x);
388 root 1.49 }
389 root 1.1 #endif
390    
391 root 1.88 ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
392     ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
393     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
394     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
395    
396 root 1.71 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
397 root 1.70 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
398     {
399     return ( (x * 0x0802U & 0x22110U)
400     | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
401     }
402    
403 root 1.71 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
404 root 1.70 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
405     {
406     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
407     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
408     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
409     x = ( x >> 8 ) | ( x << 8);
410    
411     return x;
412     }
413    
414 root 1.71 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
415 root 1.70 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
416     {
417     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
418     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
419     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
420     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
421     x = ( x >> 16 ) | ( x << 16);
422    
423     return x;
424     }
425    
426 root 1.49 /* popcount64 is only available on 64 bit cpus as gcc builtin */
427     /* so for this version we are lazy */
428     ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
429     ecb_function_ int
430     ecb_popcount64 (uint64_t x)
431     {
432     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
433     }
434    
435 root 1.50 ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
436     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
437     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
438     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
439     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
440     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
441     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
442     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
443    
444     ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
445     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
446     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
447     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
448     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
449     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
450     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
451     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
452    
453 root 1.13 #if ECB_GCC_VERSION(4,3)
454 root 1.49 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
455     #define ecb_bswap32(x) __builtin_bswap32 (x)
456     #define ecb_bswap64(x) __builtin_bswap64 (x)
457 root 1.13 #else
458 root 1.50 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
459     ecb_function_ uint16_t
460     ecb_bswap16 (uint16_t x)
461 root 1.49 {
462 root 1.50 return ecb_rotl16 (x, 8);
463 root 1.49 }
464    
465 root 1.39 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
466     ecb_function_ uint32_t
467 root 1.35 ecb_bswap32 (uint32_t x)
468     {
469 root 1.50 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
470 root 1.35 }
471    
472 root 1.49 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
473     ecb_function_ uint64_t
474     ecb_bswap64 (uint64_t x)
475 root 1.35 {
476 root 1.50 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
477 root 1.35 }
478 root 1.13 #endif
479    
480     #if ECB_GCC_VERSION(4,5)
481 root 1.35 #define ecb_unreachable() __builtin_unreachable ()
482 root 1.13 #else
483 root 1.35 /* this seems to work fine, but gcc always emits a warning for it :/ */
484 root 1.75 ecb_inline void ecb_unreachable (void) ecb_noreturn;
485     ecb_inline void ecb_unreachable (void) { }
486 root 1.13 #endif
487    
488 root 1.41 /* try to tell the compiler that some condition is definitely true */
489 root 1.100 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
490 root 1.41
491 root 1.76 ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
492     ecb_inline unsigned char
493 root 1.23 ecb_byteorder_helper (void)
494 root 1.3 {
495 root 1.98 /* the union code still generates code under pressure in gcc, */
496 sf-exg 1.111 /* but less than using pointers, and always seems to */
497 root 1.98 /* successfully return a constant. */
498     /* the reason why we have this horrible preprocessor mess */
499 root 1.99 /* is to avoid it in all cases, at least on common architectures */
500 sf-exg 1.111 /* or when using a recent enough gcc version (>= 4.6) */
501 root 1.98 #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
502     return 0x44;
503     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
504     return 0x44;
505     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
506 root 1.113 return 0x11;
507 root 1.98 #else
508     union
509     {
510     uint32_t i;
511     uint8_t c;
512     } u = { 0x11223344 };
513     return u.c;
514     #endif
515 root 1.3 }
516    
517 root 1.76 ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
518     ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
519     ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
520     ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
521 root 1.3
522 root 1.39 #if ECB_GCC_VERSION(3,0) || ECB_C99
523 root 1.35 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
524 root 1.31 #else
525 root 1.35 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
526 root 1.31 #endif
527 root 1.21
528 sf-exg 1.68 #if __cplusplus
529     template<typename T>
530     static inline T ecb_div_rd (T val, T div)
531     {
532     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
533     }
534     template<typename T>
535     static inline T ecb_div_ru (T val, T div)
536     {
537     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
538     }
539     #else
540     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
541     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
542     #endif
543 sf-exg 1.67
544 root 1.5 #if ecb_cplusplus_does_not_suck
545 root 1.40 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
546 root 1.35 template<typename T, int N>
547     static inline int ecb_array_length (const T (&arr)[N])
548     {
549     return N;
550     }
551 root 1.5 #else
552 root 1.35 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
553 root 1.5 #endif
554    
555 root 1.104 /*******************************************************************************/
556     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
557    
558     /* basically, everything uses "ieee pure-endian" floating point numbers */
559     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
560     #if 0 \
561     || __i386 || __i386__ \
562     || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
563     || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
564     || defined __arm__ && defined __ARM_EABI__ \
565     || defined __s390__ || defined __s390x__ \
566     || defined __mips__ \
567     || defined __alpha__ \
568     || defined __hppa__ \
569     || defined __ia64__ \
570     || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64
571     #define ECB_STDFP 1
572     #include <string.h> /* for memcpy */
573 root 1.102 #else
574 root 1.104 #define ECB_STDFP 0
575 root 1.109 #include <math.h> /* for frexp*, ldexp* */
576 root 1.102 #endif
577    
578 root 1.104 #ifndef ECB_NO_LIBM
579 root 1.103
580 root 1.104 /* convert a float to ieee single/binary32 */
581 root 1.103 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
582     ecb_function_ uint32_t
583     ecb_float_to_binary32 (float x)
584     {
585     uint32_t r;
586    
587     #if ECB_STDFP
588 root 1.104 memcpy (&r, &x, 4);
589 root 1.103 #else
590 root 1.105 /* slow emulation, works for anything but -0 */
591 root 1.103 uint32_t m;
592     int e;
593    
594 root 1.108 if (x == 0e0f ) return 0x00000000U;
595 root 1.103 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
596     if (x < -3.40282346638528860e+38f) return 0xff800000U;
597 root 1.105 if (x != x ) return 0x7fbfffffU;
598 root 1.103
599     m = frexpf (x, &e) * 0x1000000U;
600    
601     r = m & 0x80000000U;
602    
603     if (r)
604     m = -m;
605    
606 root 1.108 if (e <= -126)
607 root 1.103 {
608     m &= 0xffffffU;
609     m >>= (-125 - e);
610     e = -126;
611     }
612    
613     r |= (e + 126) << 23;
614     r |= m & 0x7fffffU;
615     #endif
616    
617     return r;
618     }
619    
620 root 1.104 /* converts an ieee single/binary32 to a float */
621 root 1.103 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
622     ecb_function_ float
623     ecb_binary32_to_float (uint32_t x)
624     {
625     float r;
626    
627     #if ECB_STDFP
628 root 1.104 memcpy (&r, &x, 4);
629 root 1.103 #else
630     /* emulation, only works for normals and subnormals and +0 */
631     int neg = x >> 31;
632     int e = (x >> 23) & 0xffU;
633    
634     x &= 0x7fffffU;
635    
636     if (e)
637     x |= 0x800000U;
638 root 1.104 else
639     e = 1;
640 root 1.103
641     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
642 root 1.108 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
643 root 1.103
644     r = neg ? -r : r;
645     #endif
646    
647     return r;
648     }
649    
650 root 1.104 /* convert a double to ieee double/binary64 */
651 root 1.103 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
652     ecb_function_ uint64_t
653     ecb_double_to_binary64 (double x)
654     {
655 root 1.104 uint64_t r;
656    
657     #if ECB_STDFP
658     memcpy (&r, &x, 8);
659     #else
660 root 1.105 /* slow emulation, works for anything but -0 */
661 root 1.104 uint64_t m;
662     int e;
663    
664 root 1.108 if (x == 0e0 ) return 0x0000000000000000U;
665 root 1.104 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
666     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
667 root 1.105 if (x != x ) return 0X7ff7ffffffffffffU;
668 root 1.104
669     m = frexp (x, &e) * 0x20000000000000U;
670    
671     r = m & 0x8000000000000000;;
672    
673     if (r)
674     m = -m;
675    
676 root 1.108 if (e <= -1022)
677 root 1.104 {
678     m &= 0x1fffffffffffffU;
679     m >>= (-1021 - e);
680     e = -1022;
681     }
682    
683     r |= ((uint64_t)(e + 1022)) << 52;
684     r |= m & 0xfffffffffffffU;
685     #endif
686    
687     return r;
688     }
689    
690     /* converts an ieee double/binary64 to a double */
691     ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
692     ecb_function_ double
693     ecb_binary64_to_double (uint64_t x)
694     {
695     double r;
696    
697     #if ECB_STDFP
698     memcpy (&r, &x, 8);
699     #else
700     /* emulation, only works for normals and subnormals and +0 */
701     int neg = x >> 63;
702     int e = (x >> 52) & 0x7ffU;
703    
704     x &= 0xfffffffffffffU;
705    
706     if (e)
707     x |= 0x10000000000000U;
708     else
709     e = 1;
710    
711 root 1.107 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
712 root 1.108 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
713 root 1.104
714     r = neg ? -r : r;
715     #endif
716    
717     return r;
718 root 1.103 }
719    
720     #endif
721 root 1.102
722 root 1.1 #endif
723