ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libeio/ecb.h
Revision: 1.19
Committed: Fri Jan 24 23:46:17 2014 UTC (10 years, 3 months ago) by root
Content type: text/plain
Branch: MAIN
CVS Tags: rel-4_2, rel-4_3
Changes since 1.18: +74 -9 lines
Log Message: 
*** empty log message ***

File Contents

# User Rev Content
1 root 1.1 /*
2     * libecb - http://software.schmorp.de/pkg/libecb
3     *
4 root 1.19 * Copyright (©) 2009-2013 Marc Alexander Lehmann <libecb@schmorp.de>
5 root 1.1 * Copyright (©) 2011 Emanuele Giaquinta
6     * 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 1.12 /* 16 bits major, 16 bits minor */
34 root 1.17 #define ECB_VERSION 0x00010003
35 root 1.12
36 root 1.2 #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 1.3 #else /* _MSC_VER || __BORLANDC__ */
47 root 1.2 typedef signed __int64 int64_t;
48     typedef unsigned __int64 uint64_t;
49     #endif
50 root 1.12 #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 1.2 #else
60     #include <inttypes.h>
61 root 1.12 #if UINTMAX_MAX > 0xffffffffU
62     #define ECB_PTRSIZE 8
63     #else
64     #define ECB_PTRSIZE 4
65     #endif
66 root 1.2 #endif
67 root 1.1
68 root 1.17 /* work around x32 idiocy by defining proper macros */
69 root 1.19 #if __amd64 || __x86_64 || _M_AMD64 || _M_X64
70     #if _ILP32
71 root 1.17 #define ECB_AMD64_X32 1
72     #else
73     #define ECB_AMD64 1
74     #endif
75     #endif
76    
77 root 1.1 /* many compilers define _GNUC_ to some versions but then only implement
78     * what their idiot authors think are the "more important" extensions,
79 root 1.8 * causing enormous grief in return for some better fake benchmark numbers.
80 root 1.1 * or so.
81     * 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     #ifndef ECB_GCC_VERSION
85 root 1.11 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
86 root 1.2 #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 1.1 #endif
91    
92 root 1.13 #define ECB_CPP (__cplusplus+0)
93     #define ECB_CPP11 (__cplusplus >= 201103L)
94    
95 root 1.15 #if ECB_CPP
96 root 1.19 #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 1.15 #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 1.4 /*****************************************************************************/
117    
118 root 1.8 /* 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 1.11 #if ECB_NO_THREADS
122 root 1.13 #define ECB_NO_SMP 1
123 root 1.11 #endif
124    
125 root 1.13 #if ECB_NO_SMP
126 root 1.8 #define ECB_MEMORY_FENCE do { } while (0)
127     #endif
128    
129 root 1.4 #ifndef ECB_MEMORY_FENCE
130 root 1.11 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
131     #if __i386 || __i386__
132 root 1.5 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
133 root 1.13 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
134     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
135 root 1.11 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
136 root 1.13 #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 1.8 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
140 root 1.13 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
141 root 1.11 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
142     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
143     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
144     #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
145 root 1.13 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
146     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
147 root 1.19 #elif (__sparc || __sparc__) && !__sparcv8
148 root 1.13 #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 1.11 #elif defined __s390__ || defined __s390x__
152     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
153     #elif defined __mips__
154 root 1.19 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
155 root 1.18 /* 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 1.11 #elif defined __alpha__
158 root 1.13 #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 1.19 #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 1.4 #endif
171     #endif
172     #endif
173    
174     #ifndef ECB_MEMORY_FENCE
175 root 1.13 #if ECB_GCC_VERSION(4,7)
176     /* see comment below (stdatomic.h) about the C11 memory model. */
177     #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
178 root 1.19 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
179     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
180 root 1.15
181     /* The __has_feature syntax from clang is so misdesigned that we cannot use it
182     * without risking compile time errors with other compilers. We *could*
183     * define our own ecb_clang_has_feature, but I just can't be bothered to work
184 root 1.16 * around this shit time and again.
185 root 1.15 * #elif defined __clang && __has_feature (cxx_atomic)
186     * // see comment below (stdatomic.h) about the C11 memory model.
187     * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
188 root 1.19 * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
189     * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
190 root 1.15 */
191    
192 root 1.13 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
193 root 1.4 #define ECB_MEMORY_FENCE __sync_synchronize ()
194 root 1.19 #elif _MSC_VER >= 1500 /* VC++ 2008 */
195     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
196     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
197     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
198     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
199     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
200 root 1.7 #elif _MSC_VER >= 1400 /* VC++ 2005 */
201     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
202     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
203     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
204     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
205 root 1.11 #elif defined _WIN32
206 root 1.6 #include <WinNT.h>
207 root 1.7 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
208 root 1.11 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
209     #include <mbarrier.h>
210     #define ECB_MEMORY_FENCE __machine_rw_barrier ()
211     #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
212     #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
213     #elif __xlC__
214     #define ECB_MEMORY_FENCE __sync ()
215 root 1.4 #endif
216     #endif
217    
218     #ifndef ECB_MEMORY_FENCE
219 root 1.13 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
220     /* we assume that these memory fences work on all variables/all memory accesses, */
221     /* not just C11 atomics and atomic accesses */
222     #include <stdatomic.h>
223     /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
224     /* any fence other than seq_cst, which isn't very efficient for us. */
225     /* Why that is, we don't know - either the C11 memory model is quite useless */
226     /* for most usages, or gcc and clang have a bug */
227     /* I *currently* lean towards the latter, and inefficiently implement */
228     /* all three of ecb's fences as a seq_cst fence */
229 root 1.19 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
230     /* for all __atomic_thread_fence's except seq_cst */
231 root 1.13 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
232     #endif
233     #endif
234    
235     #ifndef ECB_MEMORY_FENCE
236 root 1.8 #if !ECB_AVOID_PTHREADS
237     /*
238     * if you get undefined symbol references to pthread_mutex_lock,
239     * or failure to find pthread.h, then you should implement
240     * the ECB_MEMORY_FENCE operations for your cpu/compiler
241     * OR provide pthread.h and link against the posix thread library
242     * of your system.
243     */
244     #include <pthread.h>
245     #define ECB_NEEDS_PTHREADS 1
246     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
247    
248     static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
249     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
250     #endif
251     #endif
252 root 1.4
253 root 1.11 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
254 root 1.4 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
255 root 1.8 #endif
256    
257 root 1.11 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
258 root 1.4 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
259     #endif
260    
261     /*****************************************************************************/
262    
263 root 1.2 #if __cplusplus
264     #define ecb_inline static inline
265     #elif ECB_GCC_VERSION(2,5)
266     #define ecb_inline static __inline__
267     #elif ECB_C99
268     #define ecb_inline static inline
269     #else
270     #define ecb_inline static
271     #endif
272    
273     #if ECB_GCC_VERSION(3,3)
274     #define ecb_restrict __restrict__
275     #elif ECB_C99
276     #define ecb_restrict restrict
277 root 1.1 #else
278 root 1.2 #define ecb_restrict
279 root 1.1 #endif
280    
281 root 1.2 typedef int ecb_bool;
282    
283 root 1.1 #define ECB_CONCAT_(a, b) a ## b
284     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
285     #define ECB_STRINGIFY_(a) # a
286     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
287    
288 root 1.2 #define ecb_function_ ecb_inline
289 root 1.1
290     #if ECB_GCC_VERSION(3,1)
291 root 1.2 #define ecb_attribute(attrlist) __attribute__(attrlist)
292     #define ecb_is_constant(expr) __builtin_constant_p (expr)
293     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
294     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
295     #else
296     #define ecb_attribute(attrlist)
297 root 1.19
298     /* possible C11 impl for integral types
299     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
300     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
301    
302 root 1.2 #define ecb_is_constant(expr) 0
303     #define ecb_expect(expr,value) (expr)
304     #define ecb_prefetch(addr,rw,locality)
305 root 1.1 #endif
306    
307     /* no emulation for ecb_decltype */
308     #if ECB_GCC_VERSION(4,5)
309 root 1.2 #define ecb_decltype(x) __decltype(x)
310 root 1.1 #elif ECB_GCC_VERSION(3,0)
311 root 1.2 #define ecb_decltype(x) __typeof(x)
312 root 1.1 #endif
313    
314     #define ecb_noinline ecb_attribute ((__noinline__))
315     #define ecb_unused ecb_attribute ((__unused__))
316     #define ecb_const ecb_attribute ((__const__))
317     #define ecb_pure ecb_attribute ((__pure__))
318    
319 root 1.13 #if ECB_C11
320     #define ecb_noreturn _Noreturn
321     #else
322     #define ecb_noreturn ecb_attribute ((__noreturn__))
323     #endif
324    
325 root 1.2 #if ECB_GCC_VERSION(4,3)
326     #define ecb_artificial ecb_attribute ((__artificial__))
327     #define ecb_hot ecb_attribute ((__hot__))
328     #define ecb_cold ecb_attribute ((__cold__))
329     #else
330     #define ecb_artificial
331     #define ecb_hot
332     #define ecb_cold
333     #endif
334    
335     /* put around conditional expressions if you are very sure that the */
336     /* expression is mostly true or mostly false. note that these return */
337     /* booleans, not the expression. */
338 root 1.1 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
339     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
340 root 1.2 /* for compatibility to the rest of the world */
341 root 1.1 #define ecb_likely(expr) ecb_expect_true (expr)
342     #define ecb_unlikely(expr) ecb_expect_false (expr)
343    
344     /* count trailing zero bits and count # of one bits */
345     #if ECB_GCC_VERSION(3,4)
346 root 1.2 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
347     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
348     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
349     #define ecb_ctz32(x) __builtin_ctz (x)
350     #define ecb_ctz64(x) __builtin_ctzll (x)
351     #define ecb_popcount32(x) __builtin_popcount (x)
352     /* no popcountll */
353 root 1.1 #else
354 root 1.2 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
355     ecb_function_ int
356     ecb_ctz32 (uint32_t x)
357     {
358     int r = 0;
359    
360     x &= ~x + 1; /* this isolates the lowest bit */
361    
362 root 1.3 #if ECB_branchless_on_i386
363     r += !!(x & 0xaaaaaaaa) << 0;
364     r += !!(x & 0xcccccccc) << 1;
365     r += !!(x & 0xf0f0f0f0) << 2;
366     r += !!(x & 0xff00ff00) << 3;
367     r += !!(x & 0xffff0000) << 4;
368     #else
369 root 1.2 if (x & 0xaaaaaaaa) r += 1;
370     if (x & 0xcccccccc) r += 2;
371     if (x & 0xf0f0f0f0) r += 4;
372     if (x & 0xff00ff00) r += 8;
373     if (x & 0xffff0000) r += 16;
374 root 1.3 #endif
375 root 1.2
376     return r;
377     }
378    
379     ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
380     ecb_function_ int
381     ecb_ctz64 (uint64_t x)
382     {
383     int shift = x & 0xffffffffU ? 0 : 32;
384 root 1.3 return ecb_ctz32 (x >> shift) + shift;
385 root 1.2 }
386    
387     ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
388     ecb_function_ int
389     ecb_popcount32 (uint32_t x)
390     {
391     x -= (x >> 1) & 0x55555555;
392     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
393     x = ((x >> 4) + x) & 0x0f0f0f0f;
394     x *= 0x01010101;
395    
396     return x >> 24;
397     }
398    
399     ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
400     ecb_function_ int ecb_ld32 (uint32_t x)
401     {
402 root 1.3 int r = 0;
403 root 1.2
404 root 1.3 if (x >> 16) { x >>= 16; r += 16; }
405     if (x >> 8) { x >>= 8; r += 8; }
406     if (x >> 4) { x >>= 4; r += 4; }
407     if (x >> 2) { x >>= 2; r += 2; }
408     if (x >> 1) { r += 1; }
409 root 1.2
410     return r;
411     }
412    
413     ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
414     ecb_function_ int ecb_ld64 (uint64_t x)
415     {
416 root 1.3 int r = 0;
417 root 1.2
418 root 1.3 if (x >> 32) { x >>= 32; r += 32; }
419 root 1.1
420 root 1.3 return r + ecb_ld32 (x);
421 root 1.2 }
422     #endif
423 root 1.1
424 root 1.12 ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
425     ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
426     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
427     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
428    
429 root 1.11 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
430     ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
431     {
432     return ( (x * 0x0802U & 0x22110U)
433     | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
434     }
435    
436     ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
437     ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
438     {
439     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
440     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
441     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
442     x = ( x >> 8 ) | ( x << 8);
443    
444     return x;
445     }
446    
447     ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
448     ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
449     {
450     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
451     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
452     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
453     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
454     x = ( x >> 16 ) | ( x << 16);
455    
456     return x;
457     }
458    
459 root 1.2 /* popcount64 is only available on 64 bit cpus as gcc builtin */
460     /* so for this version we are lazy */
461     ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
462     ecb_function_ int
463     ecb_popcount64 (uint64_t x)
464 root 1.1 {
465 root 1.2 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
466 root 1.1 }
467    
468 root 1.3 ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
469     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
470     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
471     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
472     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
473     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
474     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
475     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
476    
477     ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
478     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
479     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
480     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
481     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
482     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
483     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
484     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
485    
486 root 1.1 #if ECB_GCC_VERSION(4,3)
487 root 1.2 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
488     #define ecb_bswap32(x) __builtin_bswap32 (x)
489     #define ecb_bswap64(x) __builtin_bswap64 (x)
490 root 1.1 #else
491 root 1.3 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
492     ecb_function_ uint16_t
493     ecb_bswap16 (uint16_t x)
494 root 1.2 {
495 root 1.3 return ecb_rotl16 (x, 8);
496 root 1.2 }
497    
498     ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
499     ecb_function_ uint32_t
500     ecb_bswap32 (uint32_t x)
501     {
502 root 1.3 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
503 root 1.2 }
504    
505     ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
506     ecb_function_ uint64_t
507     ecb_bswap64 (uint64_t x)
508     {
509 root 1.3 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
510 root 1.2 }
511 root 1.1 #endif
512    
513     #if ECB_GCC_VERSION(4,5)
514 root 1.2 #define ecb_unreachable() __builtin_unreachable ()
515 root 1.1 #else
516 root 1.2 /* this seems to work fine, but gcc always emits a warning for it :/ */
517 root 1.11 ecb_inline void ecb_unreachable (void) ecb_noreturn;
518     ecb_inline void ecb_unreachable (void) { }
519 root 1.1 #endif
520    
521 root 1.2 /* try to tell the compiler that some condition is definitely true */
522 root 1.14 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
523 root 1.2
524 root 1.11 ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
525     ecb_inline unsigned char
526 root 1.1 ecb_byteorder_helper (void)
527     {
528 root 1.14 /* the union code still generates code under pressure in gcc, */
529 root 1.16 /* but less than using pointers, and always seems to */
530 root 1.14 /* successfully return a constant. */
531     /* the reason why we have this horrible preprocessor mess */
532     /* is to avoid it in all cases, at least on common architectures */
533 root 1.16 /* or when using a recent enough gcc version (>= 4.6) */
534 root 1.14 #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
535     return 0x44;
536     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
537     return 0x44;
538     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
539 root 1.16 return 0x11;
540 root 1.14 #else
541     union
542     {
543     uint32_t i;
544     uint8_t c;
545     } u = { 0x11223344 };
546     return u.c;
547     #endif
548 root 1.1 }
549    
550 root 1.11 ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
551     ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
552     ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
553     ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
554 root 1.1
555 root 1.2 #if ECB_GCC_VERSION(3,0) || ECB_C99
556     #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
557 root 1.1 #else
558 root 1.2 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
559 root 1.1 #endif
560    
561 root 1.10 #if __cplusplus
562     template<typename T>
563     static inline T ecb_div_rd (T val, T div)
564     {
565     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
566     }
567     template<typename T>
568     static inline T ecb_div_ru (T val, T div)
569     {
570     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
571     }
572     #else
573     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
574     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
575     #endif
576    
577 root 1.1 #if ecb_cplusplus_does_not_suck
578 root 1.2 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
579     template<typename T, int N>
580     static inline int ecb_array_length (const T (&arr)[N])
581     {
582     return N;
583     }
584 root 1.1 #else
585 root 1.2 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
586 root 1.1 #endif
587    
588 root 1.15 /*******************************************************************************/
589     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
590    
591     /* basically, everything uses "ieee pure-endian" floating point numbers */
592     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
593     #if 0 \
594     || __i386 || __i386__ \
595     || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
596     || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
597     || defined __arm__ && defined __ARM_EABI__ \
598     || defined __s390__ || defined __s390x__ \
599     || defined __mips__ \
600     || defined __alpha__ \
601     || defined __hppa__ \
602     || defined __ia64__ \
603 root 1.19 || defined __m68k__ \
604     || defined __m88k__ \
605     || defined __sh__ \
606 root 1.15 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64
607     #define ECB_STDFP 1
608     #include <string.h> /* for memcpy */
609     #else
610     #define ECB_STDFP 0
611     #endif
612    
613     #ifndef ECB_NO_LIBM
614    
615 root 1.19 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
616    
617     /* only the oldest of old doesn't have this one. solaris. */
618     #ifdef INFINITY
619     #define ECB_INFINITY INFINITY
620     #else
621     #define ECB_INFINITY HUGE_VAL
622     #endif
623    
624     #ifdef NAN
625     #define ECB_NAN NAN
626     #else
627     #define ECB_NAN ECB_INFINITY
628     #endif
629    
630     /* converts an ieee half/binary16 to a float */
631     ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
632     ecb_function_ float
633     ecb_binary16_to_float (uint16_t x)
634     {
635     int e = (x >> 10) & 0x1f;
636     int m = x & 0x3ff;
637     float r;
638    
639     if (!e ) r = ldexpf (m , -24);
640     else if (e != 31) r = ldexpf (m + 0x400, e - 25);
641     else if (m ) r = ECB_NAN;
642     else r = ECB_INFINITY;
643    
644     return x & 0x8000 ? -r : r;
645     }
646    
647 root 1.15 /* convert a float to ieee single/binary32 */
648     ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
649     ecb_function_ uint32_t
650     ecb_float_to_binary32 (float x)
651     {
652     uint32_t r;
653    
654     #if ECB_STDFP
655     memcpy (&r, &x, 4);
656     #else
657     /* slow emulation, works for anything but -0 */
658     uint32_t m;
659     int e;
660    
661     if (x == 0e0f ) return 0x00000000U;
662     if (x > +3.40282346638528860e+38f) return 0x7f800000U;
663     if (x < -3.40282346638528860e+38f) return 0xff800000U;
664     if (x != x ) return 0x7fbfffffU;
665    
666     m = frexpf (x, &e) * 0x1000000U;
667    
668     r = m & 0x80000000U;
669    
670     if (r)
671     m = -m;
672    
673     if (e <= -126)
674     {
675     m &= 0xffffffU;
676     m >>= (-125 - e);
677     e = -126;
678     }
679    
680     r |= (e + 126) << 23;
681     r |= m & 0x7fffffU;
682     #endif
683    
684     return r;
685     }
686    
687     /* converts an ieee single/binary32 to a float */
688     ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
689     ecb_function_ float
690     ecb_binary32_to_float (uint32_t x)
691     {
692     float r;
693    
694     #if ECB_STDFP
695     memcpy (&r, &x, 4);
696     #else
697     /* emulation, only works for normals and subnormals and +0 */
698     int neg = x >> 31;
699     int e = (x >> 23) & 0xffU;
700    
701     x &= 0x7fffffU;
702    
703     if (e)
704     x |= 0x800000U;
705     else
706     e = 1;
707    
708     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
709     r = ldexpf (x * (0.5f / 0x800000U), e - 126);
710    
711     r = neg ? -r : r;
712     #endif
713    
714     return r;
715     }
716    
717     /* convert a double to ieee double/binary64 */
718     ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
719     ecb_function_ uint64_t
720     ecb_double_to_binary64 (double x)
721     {
722     uint64_t r;
723    
724     #if ECB_STDFP
725     memcpy (&r, &x, 8);
726     #else
727     /* slow emulation, works for anything but -0 */
728     uint64_t m;
729     int e;
730    
731     if (x == 0e0 ) return 0x0000000000000000U;
732     if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
733     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
734     if (x != x ) return 0X7ff7ffffffffffffU;
735    
736     m = frexp (x, &e) * 0x20000000000000U;
737    
738     r = m & 0x8000000000000000;;
739    
740     if (r)
741     m = -m;
742    
743     if (e <= -1022)
744     {
745     m &= 0x1fffffffffffffU;
746     m >>= (-1021 - e);
747     e = -1022;
748     }
749    
750     r |= ((uint64_t)(e + 1022)) << 52;
751     r |= m & 0xfffffffffffffU;
752     #endif
753    
754     return r;
755     }
756    
757     /* converts an ieee double/binary64 to a double */
758     ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
759     ecb_function_ double
760     ecb_binary64_to_double (uint64_t x)
761     {
762     double r;
763    
764     #if ECB_STDFP
765     memcpy (&r, &x, 8);
766     #else
767     /* emulation, only works for normals and subnormals and +0 */
768     int neg = x >> 63;
769     int e = (x >> 52) & 0x7ffU;
770    
771     x &= 0xfffffffffffffU;
772    
773     if (e)
774     x |= 0x10000000000000U;
775     else
776     e = 1;
777    
778     /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
779     r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
780    
781     r = neg ? -r : r;
782     #endif
783    
784     return r;
785     }
786    
787     #endif
788    
789 root 1.1 #endif
790