ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.127
Committed: Sun Jan 5 15:54:09 2014 UTC (10 years, 5 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.126: +16 -3 lines
Log Message:
*** empty 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 1.123 * Copyright (©) 2009-2013 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 root 1.125 #if __amd64 || __x86_64 || _M_AMD64 || _M_X64
70 root 1.119 #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_CPP (__cplusplus+0)
93     #define ECB_CPP11 (__cplusplus >= 201103L)
94 root 1.90
95 root 1.102 #if ECB_CPP
96 root 1.127 #define ECB_C 0
97     #define ECB_STDC_VERSION 0
98     #else
99     #define ECB_C 1
100     #define ECB_STDC_VERSION __STDC_VERSION__
101     #endif
102    
103     #define ECB_C99 (ECB_STDC_VERSION >= 199901L)
104     #define ECB_C11 (ECB_STDC_VERSION >= 201112L)
105    
106     #if ECB_CPP
107 root 1.102 #define ECB_EXTERN_C extern "C"
108     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
109     #define ECB_EXTERN_C_END }
110     #else
111     #define ECB_EXTERN_C extern
112     #define ECB_EXTERN_C_BEG
113     #define ECB_EXTERN_C_END
114     #endif
115    
116 root 1.52 /*****************************************************************************/
117    
118 root 1.58 /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
119     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
120    
121 root 1.79 #if ECB_NO_THREADS
122 root 1.95 #define ECB_NO_SMP 1
123 root 1.79 #endif
124    
125 root 1.93 #if ECB_NO_SMP
126 root 1.64 #define ECB_MEMORY_FENCE do { } while (0)
127 root 1.58 #endif
128    
129 root 1.52 #ifndef ECB_MEMORY_FENCE
130 root 1.85 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
131 root 1.73 #if __i386 || __i386__
132 root 1.54 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
133 root 1.94 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
134     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
135 root 1.73 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
136 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
137     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
138     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
139 root 1.63 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
140 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
141 root 1.85 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
142     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
143 root 1.84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
144 root 1.85 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
145 root 1.94 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
146     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
147 root 1.124 #elif (__sparc || __sparc__) && !__sparcv8
148 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
149     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
150     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
151 root 1.85 #elif defined __s390__ || defined __s390x__
152 root 1.77 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
153 root 1.85 #elif defined __mips__
154 root 1.118 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
155 root 1.116 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
156     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
157 root 1.86 #elif defined __alpha__
158 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
159     #elif defined __hppa__
160     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
161     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
162     #elif defined __ia64__
163     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
164 root 1.117 #elif defined __m68k__
165     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
166     #elif defined __m88k__
167     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
168     #elif defined __sh__
169     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
170 root 1.52 #endif
171     #endif
172     #endif
173    
174     #ifndef ECB_MEMORY_FENCE
175 root 1.93 #if ECB_GCC_VERSION(4,7)
176 root 1.97 /* see comment below (stdatomic.h) about the C11 memory model. */
177 root 1.93 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
178 root 1.110
179     /* The __has_feature syntax from clang is so misdesigned that we cannot use it
180     * without risking compile time errors with other compilers. We *could*
181     * define our own ecb_clang_has_feature, but I just can't be bothered to work
182 root 1.112 * around this shit time and again.
183 root 1.110 * #elif defined __clang && __has_feature (cxx_atomic)
184     * // see comment below (stdatomic.h) about the C11 memory model.
185     * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
186     */
187    
188 root 1.93 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
189 root 1.52 #define ECB_MEMORY_FENCE __sync_synchronize ()
190 root 1.126 #elif _MSC_VER >= 1500 /* VC++ 2008 */
191     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
192     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
193     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
194     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
195     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
196 root 1.57 #elif _MSC_VER >= 1400 /* VC++ 2005 */
197     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
198     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
199     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
200     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
201 root 1.85 #elif defined _WIN32
202 root 1.55 #include <WinNT.h>
203 root 1.57 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
204 root 1.72 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
205     #include <mbarrier.h>
206     #define ECB_MEMORY_FENCE __machine_rw_barrier ()
207     #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
208     #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
209 root 1.82 #elif __xlC__
210 root 1.83 #define ECB_MEMORY_FENCE __sync ()
211 root 1.52 #endif
212     #endif
213    
214 root 1.53 #ifndef ECB_MEMORY_FENCE
215 root 1.94 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
216     /* we assume that these memory fences work on all variables/all memory accesses, */
217     /* not just C11 atomics and atomic accesses */
218     #include <stdatomic.h>
219 root 1.97 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
220     /* any fence other than seq_cst, which isn't very efficient for us. */
221     /* Why that is, we don't know - either the C11 memory model is quite useless */
222     /* for most usages, or gcc and clang have a bug */
223     /* I *currently* lean towards the latter, and inefficiently implement */
224     /* all three of ecb's fences as a seq_cst fence */
225 root 1.94 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
226     #endif
227     #endif
228    
229     #ifndef ECB_MEMORY_FENCE
230 root 1.62 #if !ECB_AVOID_PTHREADS
231     /*
232     * if you get undefined symbol references to pthread_mutex_lock,
233     * or failure to find pthread.h, then you should implement
234     * the ECB_MEMORY_FENCE operations for your cpu/compiler
235     * OR provide pthread.h and link against the posix thread library
236     * of your system.
237     */
238     #include <pthread.h>
239     #define ECB_NEEDS_PTHREADS 1
240     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
241 root 1.52
242 root 1.62 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
243     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
244     #endif
245     #endif
246    
247 root 1.85 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
248 root 1.52 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
249 root 1.62 #endif
250    
251 root 1.85 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
252 root 1.52 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
253     #endif
254    
255     /*****************************************************************************/
256    
257 root 1.39 #if __cplusplus
258 root 1.46 #define ecb_inline static inline
259 root 1.38 #elif ECB_GCC_VERSION(2,5)
260 root 1.46 #define ecb_inline static __inline__
261 root 1.39 #elif ECB_C99
262 root 1.46 #define ecb_inline static inline
263 root 1.29 #else
264 root 1.46 #define ecb_inline static
265 root 1.38 #endif
266    
267     #if ECB_GCC_VERSION(3,3)
268     #define ecb_restrict __restrict__
269 root 1.39 #elif ECB_C99
270 root 1.38 #define ecb_restrict restrict
271     #else
272     #define ecb_restrict
273 root 1.4 #endif
274    
275 root 1.38 typedef int ecb_bool;
276    
277 root 1.8 #define ECB_CONCAT_(a, b) a ## b
278     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
279     #define ECB_STRINGIFY_(a) # a
280     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
281    
282 root 1.46 #define ecb_function_ ecb_inline
283 root 1.3
284 root 1.1 #if ECB_GCC_VERSION(3,1)
285 root 1.37 #define ecb_attribute(attrlist) __attribute__(attrlist)
286     #define ecb_is_constant(expr) __builtin_constant_p (expr)
287     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
288     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
289     #else
290     #define ecb_attribute(attrlist)
291 root 1.127
292     /* possible C11 impl for integral types
293     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
294     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
295    
296 root 1.37 #define ecb_is_constant(expr) 0
297     #define ecb_expect(expr,value) (expr)
298     #define ecb_prefetch(addr,rw,locality)
299 root 1.1 #endif
300    
301 root 1.2 /* no emulation for ecb_decltype */
302 root 1.1 #if ECB_GCC_VERSION(4,5)
303 root 1.37 #define ecb_decltype(x) __decltype(x)
304 root 1.1 #elif ECB_GCC_VERSION(3,0)
305 root 1.39 #define ecb_decltype(x) __typeof(x)
306 root 1.1 #endif
307    
308 root 1.24 #define ecb_noinline ecb_attribute ((__noinline__))
309     #define ecb_unused ecb_attribute ((__unused__))
310     #define ecb_const ecb_attribute ((__const__))
311     #define ecb_pure ecb_attribute ((__pure__))
312 root 1.35
313 root 1.90 #if ECB_C11
314     #define ecb_noreturn _Noreturn
315     #else
316     #define ecb_noreturn ecb_attribute ((__noreturn__))
317     #endif
318    
319 root 1.35 #if ECB_GCC_VERSION(4,3)
320 root 1.39 #define ecb_artificial ecb_attribute ((__artificial__))
321     #define ecb_hot ecb_attribute ((__hot__))
322     #define ecb_cold ecb_attribute ((__cold__))
323 root 1.35 #else
324     #define ecb_artificial
325     #define ecb_hot
326     #define ecb_cold
327     #endif
328 root 1.1
329 root 1.39 /* put around conditional expressions if you are very sure that the */
330     /* expression is mostly true or mostly false. note that these return */
331     /* booleans, not the expression. */
332 root 1.33 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
333     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
334 root 1.36 /* for compatibility to the rest of the world */
335 root 1.33 #define ecb_likely(expr) ecb_expect_true (expr)
336     #define ecb_unlikely(expr) ecb_expect_false (expr)
337 root 1.1
338 root 1.3 /* count trailing zero bits and count # of one bits */
339 root 1.1 #if ECB_GCC_VERSION(3,4)
340 root 1.49 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
341     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
342     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
343 root 1.35 #define ecb_ctz32(x) __builtin_ctz (x)
344 root 1.49 #define ecb_ctz64(x) __builtin_ctzll (x)
345 root 1.35 #define ecb_popcount32(x) __builtin_popcount (x)
346 root 1.49 /* no popcountll */
347 root 1.1 #else
348 root 1.39 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
349     ecb_function_ int
350 root 1.35 ecb_ctz32 (uint32_t x)
351     {
352     int r = 0;
353    
354 root 1.48 x &= ~x + 1; /* this isolates the lowest bit */
355 root 1.35
356 root 1.50 #if ECB_branchless_on_i386
357     r += !!(x & 0xaaaaaaaa) << 0;
358     r += !!(x & 0xcccccccc) << 1;
359     r += !!(x & 0xf0f0f0f0) << 2;
360     r += !!(x & 0xff00ff00) << 3;
361     r += !!(x & 0xffff0000) << 4;
362     #else
363 root 1.35 if (x & 0xaaaaaaaa) r += 1;
364     if (x & 0xcccccccc) r += 2;
365     if (x & 0xf0f0f0f0) r += 4;
366     if (x & 0xff00ff00) r += 8;
367     if (x & 0xffff0000) r += 16;
368 root 1.50 #endif
369 root 1.35
370     return r;
371     }
372    
373 root 1.49 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
374     ecb_function_ int
375     ecb_ctz64 (uint64_t x)
376     {
377     int shift = x & 0xffffffffU ? 0 : 32;
378 root 1.50 return ecb_ctz32 (x >> shift) + shift;
379 root 1.49 }
380    
381 root 1.39 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
382     ecb_function_ int
383 root 1.35 ecb_popcount32 (uint32_t x)
384     {
385     x -= (x >> 1) & 0x55555555;
386     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
387     x = ((x >> 4) + x) & 0x0f0f0f0f;
388     x *= 0x01010101;
389 root 1.1
390 root 1.35 return x >> 24;
391     }
392 root 1.49
393     ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
394     ecb_function_ int ecb_ld32 (uint32_t x)
395     {
396 root 1.50 int r = 0;
397 root 1.49
398 root 1.50 if (x >> 16) { x >>= 16; r += 16; }
399     if (x >> 8) { x >>= 8; r += 8; }
400     if (x >> 4) { x >>= 4; r += 4; }
401     if (x >> 2) { x >>= 2; r += 2; }
402     if (x >> 1) { r += 1; }
403 root 1.49
404     return r;
405     }
406    
407     ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
408     ecb_function_ int ecb_ld64 (uint64_t x)
409     {
410 root 1.50 int r = 0;
411 root 1.49
412 root 1.50 if (x >> 32) { x >>= 32; r += 32; }
413 root 1.49
414 root 1.50 return r + ecb_ld32 (x);
415 root 1.49 }
416 root 1.1 #endif
417    
418 root 1.88 ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
419     ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
420     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
421     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
422    
423 root 1.71 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
424 root 1.70 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
425     {
426     return ( (x * 0x0802U & 0x22110U)
427     | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
428     }
429    
430 root 1.71 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
431 root 1.70 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
432     {
433     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
434     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
435     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
436     x = ( x >> 8 ) | ( x << 8);
437    
438     return x;
439     }
440    
441 root 1.71 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
442 root 1.70 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
443     {
444     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
445     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
446     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
447     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
448     x = ( x >> 16 ) | ( x << 16);
449    
450     return x;
451     }
452    
453 root 1.49 /* popcount64 is only available on 64 bit cpus as gcc builtin */
454     /* so for this version we are lazy */
455     ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
456     ecb_function_ int
457     ecb_popcount64 (uint64_t x)
458     {
459     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
460     }
461    
462 root 1.50 ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
463     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
464     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
465     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
466     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
467     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
468     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
469     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
470    
471     ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
472     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
473     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
474     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
475     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
476     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
477     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
478     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
479    
480 root 1.13 #if ECB_GCC_VERSION(4,3)
481 root 1.49 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
482     #define ecb_bswap32(x) __builtin_bswap32 (x)
483     #define ecb_bswap64(x) __builtin_bswap64 (x)
484 root 1.13 #else
485 root 1.50 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
486     ecb_function_ uint16_t
487     ecb_bswap16 (uint16_t x)
488 root 1.49 {
489 root 1.50 return ecb_rotl16 (x, 8);
490 root 1.49 }
491    
492 root 1.39 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
493     ecb_function_ uint32_t
494 root 1.35 ecb_bswap32 (uint32_t x)
495     {
496 root 1.50 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
497 root 1.35 }
498    
499 root 1.49 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
500     ecb_function_ uint64_t
501     ecb_bswap64 (uint64_t x)
502 root 1.35 {
503 root 1.50 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
504 root 1.35 }
505 root 1.13 #endif
506    
507     #if ECB_GCC_VERSION(4,5)
508 root 1.35 #define ecb_unreachable() __builtin_unreachable ()
509 root 1.13 #else
510 root 1.35 /* this seems to work fine, but gcc always emits a warning for it :/ */
511 root 1.75 ecb_inline void ecb_unreachable (void) ecb_noreturn;
512     ecb_inline void ecb_unreachable (void) { }
513 root 1.13 #endif
514    
515 root 1.41 /* try to tell the compiler that some condition is definitely true */
516 root 1.100 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
517 root 1.41
518 root 1.76 ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
519     ecb_inline unsigned char
520 root 1.23 ecb_byteorder_helper (void)
521 root 1.3 {
522 root 1.98 /* the union code still generates code under pressure in gcc, */
523 sf-exg 1.111 /* but less than using pointers, and always seems to */
524 root 1.98 /* successfully return a constant. */
525     /* the reason why we have this horrible preprocessor mess */
526 root 1.99 /* is to avoid it in all cases, at least on common architectures */
527 sf-exg 1.111 /* or when using a recent enough gcc version (>= 4.6) */
528 root 1.98 #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
529     return 0x44;
530     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
531     return 0x44;
532     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
533 root 1.113 return 0x11;
534 root 1.98 #else
535     union
536     {
537     uint32_t i;
538     uint8_t c;
539     } u = { 0x11223344 };
540     return u.c;
541     #endif
542 root 1.3 }
543    
544 root 1.76 ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
545     ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
546     ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
547     ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
548 root 1.3
549 root 1.39 #if ECB_GCC_VERSION(3,0) || ECB_C99
550 root 1.35 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
551 root 1.31 #else
552 root 1.35 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
553 root 1.31 #endif
554 root 1.21
555 sf-exg 1.68 #if __cplusplus
556     template<typename T>
557     static inline T ecb_div_rd (T val, T div)
558     {
559     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
560     }
561     template<typename T>
562     static inline T ecb_div_ru (T val, T div)
563     {
564     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
565     }
566     #else
567     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
568     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
569     #endif
570 sf-exg 1.67
571 root 1.5 #if ecb_cplusplus_does_not_suck
572 root 1.40 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
573 root 1.35 template<typename T, int N>
574     static inline int ecb_array_length (const T (&arr)[N])
575     {
576     return N;
577     }
578 root 1.5 #else
579 root 1.35 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
580 root 1.5 #endif
581    
582 root 1.104 /*******************************************************************************/
583     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
584    
585     /* basically, everything uses "ieee pure-endian" floating point numbers */
586     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
587     #if 0 \
588     || __i386 || __i386__ \
589     || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
590     || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
591     || defined __arm__ && defined __ARM_EABI__ \
592     || defined __s390__ || defined __s390x__ \
593     || defined __mips__ \
594     || defined __alpha__ \
595     || defined __hppa__ \
596     || defined __ia64__ \
597 root 1.117 || defined __m68k__ \
598     || defined __m88k__ \
599     || defined __sh__ \
600 root 1.104 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64
601     #define ECB_STDFP 1
602     #include <string.h> /* for memcpy */
603 root 1.102 #else
604 root 1.104 #define ECB_STDFP 0
605 root 1.102 #endif
606    
607 root 1.104 #ifndef ECB_NO_LIBM
608 root 1.103
609 root 1.121 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
610    
611 root 1.122 /* only the oldest of old doesn't have this one. solaris. */
612     #ifdef INFINITY
613     #define ECB_INFINITY INFINITY
614     #else
615     #define ECB_INFINITY HUGE_VAL
616     #endif
617    
618     #ifdef NAN
619 root 1.121 #define ECB_NAN NAN
620     #else
621 root 1.122 #define ECB_NAN ECB_INFINITY
622 root 1.121 #endif
623 root 1.120
624     /* converts an ieee half/binary16 to a float */
625     ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
626     ecb_function_ float
627     ecb_binary16_to_float (uint16_t x)
628     {
629     int e = (x >> 10) & 0x1f;
630     int m = x & 0x3ff;
631     float r;
632    
633 root 1.121 if (!e ) r = ldexpf (m , -24);
634     else if (e != 31) r = ldexpf (m + 0x400, e - 25);
635     else if (m ) r = ECB_NAN;
636 root 1.123 else r = ECB_INFINITY;
637 root 1.120
638     return x & 0x8000 ? -r : r;
639     }
640    
641 root 1.104 /* convert a float to ieee single/binary32 */
642 root 1.103 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
643     ecb_function_ uint32_t
644     ecb_float_to_binary32 (float x)
645     {
646     uint32_t r;
647    
648     #if ECB_STDFP
649 root 1.104 memcpy (&r, &x, 4);
650 root 1.103 #else
651 root 1.105 /* slow emulation, works for anything but -0 */
652 root 1.103 uint32_t m;
653     int e;
654    
655 root 1.108 if (x == 0e0f ) return 0x00000000U;
656 root 1.103 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
657     if (x < -3.40282346638528860e+38f) return 0xff800000U;
658 root 1.105 if (x != x ) return 0x7fbfffffU;
659 root 1.103
660     m = frexpf (x, &e) * 0x1000000U;
661    
662     r = m & 0x80000000U;
663    
664     if (r)
665     m = -m;
666    
667 root 1.108 if (e <= -126)
668 root 1.103 {
669     m &= 0xffffffU;
670     m >>= (-125 - e);
671     e = -126;
672     }
673    
674     r |= (e + 126) << 23;
675     r |= m & 0x7fffffU;
676     #endif
677    
678     return r;
679     }
680    
681 root 1.104 /* converts an ieee single/binary32 to a float */
682 root 1.103 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
683     ecb_function_ float
684     ecb_binary32_to_float (uint32_t x)
685     {
686     float r;
687    
688     #if ECB_STDFP
689 root 1.104 memcpy (&r, &x, 4);
690 root 1.103 #else
691     /* emulation, only works for normals and subnormals and +0 */
692     int neg = x >> 31;
693     int e = (x >> 23) & 0xffU;
694    
695     x &= 0x7fffffU;
696    
697     if (e)
698     x |= 0x800000U;
699 root 1.104 else
700     e = 1;
701 root 1.103
702     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
703 root 1.108 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
704 root 1.103
705     r = neg ? -r : r;
706     #endif
707    
708     return r;
709     }
710    
711 root 1.104 /* convert a double to ieee double/binary64 */
712 root 1.103 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
713     ecb_function_ uint64_t
714     ecb_double_to_binary64 (double x)
715     {
716 root 1.104 uint64_t r;
717    
718     #if ECB_STDFP
719     memcpy (&r, &x, 8);
720     #else
721 root 1.105 /* slow emulation, works for anything but -0 */
722 root 1.104 uint64_t m;
723     int e;
724    
725 root 1.108 if (x == 0e0 ) return 0x0000000000000000U;
726 root 1.104 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
727     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
728 root 1.105 if (x != x ) return 0X7ff7ffffffffffffU;
729 root 1.104
730     m = frexp (x, &e) * 0x20000000000000U;
731    
732     r = m & 0x8000000000000000;;
733    
734     if (r)
735     m = -m;
736    
737 root 1.108 if (e <= -1022)
738 root 1.104 {
739     m &= 0x1fffffffffffffU;
740     m >>= (-1021 - e);
741     e = -1022;
742     }
743    
744     r |= ((uint64_t)(e + 1022)) << 52;
745     r |= m & 0xfffffffffffffU;
746     #endif
747    
748     return r;
749     }
750    
751     /* converts an ieee double/binary64 to a double */
752     ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
753     ecb_function_ double
754     ecb_binary64_to_double (uint64_t x)
755     {
756     double r;
757    
758     #if ECB_STDFP
759     memcpy (&r, &x, 8);
760     #else
761     /* emulation, only works for normals and subnormals and +0 */
762     int neg = x >> 63;
763     int e = (x >> 52) & 0x7ffU;
764    
765     x &= 0xfffffffffffffU;
766    
767     if (e)
768     x |= 0x10000000000000U;
769     else
770     e = 1;
771    
772 root 1.107 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
773 root 1.108 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
774 root 1.104
775     r = neg ? -r : r;
776     #endif
777    
778     return r;
779 root 1.103 }
780    
781     #endif
782 root 1.102
783 root 1.1 #endif
784