ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.137
Committed: Tue Oct 14 14:01:07 2014 UTC (9 years, 8 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.136: +4 -6 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 1.129 * Copyright (©) 2009-2014 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 root 1.133 *
29     * Alternatively, the contents of this file may be used under the terms of
30     * the GNU General Public License ("GPL") version 2 or any later version,
31     * in which case the provisions of the GPL are applicable instead of
32     * the above. If you wish to allow the use of your version of this file
33     * only under the terms of the GPL and not to allow others to use your
34     * version of this file under the BSD license, indicate your decision
35     * by deleting the provisions above and replace them with the notice
36     * and other provisions required by the GPL. If you do not delete the
37     * provisions above, a recipient may use your version of this file under
38     * either the BSD or the GPL.
39 root 1.1 */
40    
41     #ifndef ECB_H
42     #define ECB_H
43    
44 root 1.87 /* 16 bits major, 16 bits minor */
45 root 1.114 #define ECB_VERSION 0x00010003
46 root 1.87
47 root 1.44 #ifdef _WIN32
48     typedef signed char int8_t;
49     typedef unsigned char uint8_t;
50     typedef signed short int16_t;
51     typedef unsigned short uint16_t;
52     typedef signed int int32_t;
53     typedef unsigned int uint32_t;
54     #if __GNUC__
55     typedef signed long long int64_t;
56     typedef unsigned long long uint64_t;
57 root 1.51 #else /* _MSC_VER || __BORLANDC__ */
58 root 1.44 typedef signed __int64 int64_t;
59     typedef unsigned __int64 uint64_t;
60     #endif
61 root 1.87 #ifdef _WIN64
62     #define ECB_PTRSIZE 8
63     typedef uint64_t uintptr_t;
64     typedef int64_t intptr_t;
65     #else
66     #define ECB_PTRSIZE 4
67     typedef uint32_t uintptr_t;
68     typedef int32_t intptr_t;
69     #endif
70 root 1.44 #else
71     #include <inttypes.h>
72 root 1.87 #if UINTMAX_MAX > 0xffffffffU
73     #define ECB_PTRSIZE 8
74     #else
75     #define ECB_PTRSIZE 4
76     #endif
77 root 1.44 #endif
78 root 1.6
79 root 1.114 /* work around x32 idiocy by defining proper macros */
80 root 1.125 #if __amd64 || __x86_64 || _M_AMD64 || _M_X64
81 root 1.119 #if _ILP32
82 root 1.115 #define ECB_AMD64_X32 1
83 root 1.114 #else
84 root 1.115 #define ECB_AMD64 1
85 root 1.114 #endif
86     #endif
87    
88 root 1.12 /* many compilers define _GNUC_ to some versions but then only implement
89     * what their idiot authors think are the "more important" extensions,
90 sf-exg 1.59 * causing enormous grief in return for some better fake benchmark numbers.
91 root 1.18 * or so.
92 root 1.12 * we try to detect these and simply assume they are not gcc - if they have
93     * an issue with that they should have done it right in the first place.
94     */
95 root 1.137 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
96     #define ECB_GCC_VERSION(major,minor) 0
97     #else
98     #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
99 root 1.12 #endif
100 root 1.1
101 root 1.91 #define ECB_CPP (__cplusplus+0)
102     #define ECB_CPP11 (__cplusplus >= 201103L)
103 root 1.90
104 root 1.102 #if ECB_CPP
105 root 1.127 #define ECB_C 0
106     #define ECB_STDC_VERSION 0
107     #else
108     #define ECB_C 1
109     #define ECB_STDC_VERSION __STDC_VERSION__
110     #endif
111    
112     #define ECB_C99 (ECB_STDC_VERSION >= 199901L)
113     #define ECB_C11 (ECB_STDC_VERSION >= 201112L)
114    
115     #if ECB_CPP
116 root 1.102 #define ECB_EXTERN_C extern "C"
117     #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
118     #define ECB_EXTERN_C_END }
119     #else
120     #define ECB_EXTERN_C extern
121     #define ECB_EXTERN_C_BEG
122     #define ECB_EXTERN_C_END
123     #endif
124    
125 root 1.52 /*****************************************************************************/
126    
127 root 1.58 /* ECB_NO_THREADS - ecb is not used by multiple threads, ever */
128     /* ECB_NO_SMP - ecb might be used in multiple threads, but only on a single cpu */
129    
130 root 1.79 #if ECB_NO_THREADS
131 root 1.95 #define ECB_NO_SMP 1
132 root 1.79 #endif
133    
134 root 1.93 #if ECB_NO_SMP
135 root 1.64 #define ECB_MEMORY_FENCE do { } while (0)
136 root 1.58 #endif
137    
138 root 1.52 #ifndef ECB_MEMORY_FENCE
139 root 1.85 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
140 root 1.73 #if __i386 || __i386__
141 root 1.54 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
142 root 1.94 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
143     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
144 root 1.73 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__
145 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
146     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
147     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
148 root 1.63 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
149 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
150 root 1.85 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
151     || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__
152 root 1.84 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
153 root 1.85 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
154 root 1.94 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__
155     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
156 root 1.129 #elif __aarch64__
157     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
158 root 1.124 #elif (__sparc || __sparc__) && !__sparcv8
159 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
160     #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
161     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
162 root 1.85 #elif defined __s390__ || defined __s390x__
163 root 1.77 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
164 root 1.85 #elif defined __mips__
165 root 1.118 /* GNU/Linux emulates sync on mips1 architectures, so we force its use */
166 root 1.116 /* anybody else who still uses mips1 is supposed to send in their version, with detection code. */
167     #define ECB_MEMORY_FENCE __asm__ __volatile__ (".set mips2; sync; .set mips0" : : : "memory")
168 root 1.86 #elif defined __alpha__
169 root 1.94 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mb" : : : "memory")
170     #elif defined __hppa__
171     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
172     #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("")
173     #elif defined __ia64__
174     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mf" : : : "memory")
175 root 1.117 #elif defined __m68k__
176     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
177     #elif defined __m88k__
178     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("tb1 0,%%r0,128" : : : "memory")
179     #elif defined __sh__
180     #define ECB_MEMORY_FENCE __asm__ __volatile__ ("" : : : "memory")
181 root 1.52 #endif
182     #endif
183     #endif
184    
185     #ifndef ECB_MEMORY_FENCE
186 root 1.93 #if ECB_GCC_VERSION(4,7)
187 root 1.97 /* see comment below (stdatomic.h) about the C11 memory model. */
188 root 1.93 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
189 root 1.128 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
190     #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
191 root 1.110
192     /* The __has_feature syntax from clang is so misdesigned that we cannot use it
193     * without risking compile time errors with other compilers. We *could*
194     * define our own ecb_clang_has_feature, but I just can't be bothered to work
195 root 1.112 * around this shit time and again.
196 root 1.110 * #elif defined __clang && __has_feature (cxx_atomic)
197     * // see comment below (stdatomic.h) about the C11 memory model.
198     * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
199 root 1.128 * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
200     * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
201 root 1.110 */
202    
203 root 1.93 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
204 root 1.52 #define ECB_MEMORY_FENCE __sync_synchronize ()
205 root 1.126 #elif _MSC_VER >= 1500 /* VC++ 2008 */
206     /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
207     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
208     #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
209     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
210     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
211 root 1.57 #elif _MSC_VER >= 1400 /* VC++ 2005 */
212     #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
213     #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
214     #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
215     #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
216 root 1.85 #elif defined _WIN32
217 root 1.55 #include <WinNT.h>
218 root 1.57 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
219 root 1.72 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
220     #include <mbarrier.h>
221     #define ECB_MEMORY_FENCE __machine_rw_barrier ()
222     #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier ()
223     #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier ()
224 root 1.82 #elif __xlC__
225 root 1.83 #define ECB_MEMORY_FENCE __sync ()
226 root 1.52 #endif
227     #endif
228    
229 root 1.53 #ifndef ECB_MEMORY_FENCE
230 root 1.94 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
231     /* we assume that these memory fences work on all variables/all memory accesses, */
232     /* not just C11 atomics and atomic accesses */
233     #include <stdatomic.h>
234 root 1.97 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
235     /* any fence other than seq_cst, which isn't very efficient for us. */
236     /* Why that is, we don't know - either the C11 memory model is quite useless */
237     /* for most usages, or gcc and clang have a bug */
238     /* I *currently* lean towards the latter, and inefficiently implement */
239     /* all three of ecb's fences as a seq_cst fence */
240 root 1.128 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
241     /* for all __atomic_thread_fence's except seq_cst */
242 root 1.94 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
243     #endif
244     #endif
245    
246     #ifndef ECB_MEMORY_FENCE
247 root 1.62 #if !ECB_AVOID_PTHREADS
248     /*
249     * if you get undefined symbol references to pthread_mutex_lock,
250     * or failure to find pthread.h, then you should implement
251     * the ECB_MEMORY_FENCE operations for your cpu/compiler
252     * OR provide pthread.h and link against the posix thread library
253     * of your system.
254     */
255     #include <pthread.h>
256     #define ECB_NEEDS_PTHREADS 1
257     #define ECB_MEMORY_FENCE_NEEDS_PTHREADS 1
258 root 1.52
259 root 1.62 static pthread_mutex_t ecb_mf_lock = PTHREAD_MUTEX_INITIALIZER;
260     #define ECB_MEMORY_FENCE do { pthread_mutex_lock (&ecb_mf_lock); pthread_mutex_unlock (&ecb_mf_lock); } while (0)
261     #endif
262     #endif
263    
264 root 1.85 #if !defined ECB_MEMORY_FENCE_ACQUIRE && defined ECB_MEMORY_FENCE
265 root 1.52 #define ECB_MEMORY_FENCE_ACQUIRE ECB_MEMORY_FENCE
266 root 1.62 #endif
267    
268 root 1.85 #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
269 root 1.52 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
270     #endif
271    
272     /*****************************************************************************/
273    
274 root 1.39 #if __cplusplus
275 root 1.46 #define ecb_inline static inline
276 root 1.38 #elif ECB_GCC_VERSION(2,5)
277 root 1.46 #define ecb_inline static __inline__
278 root 1.39 #elif ECB_C99
279 root 1.46 #define ecb_inline static inline
280 root 1.29 #else
281 root 1.46 #define ecb_inline static
282 root 1.38 #endif
283    
284     #if ECB_GCC_VERSION(3,3)
285     #define ecb_restrict __restrict__
286 root 1.39 #elif ECB_C99
287 root 1.38 #define ecb_restrict restrict
288     #else
289     #define ecb_restrict
290 root 1.4 #endif
291    
292 root 1.38 typedef int ecb_bool;
293    
294 root 1.8 #define ECB_CONCAT_(a, b) a ## b
295     #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
296     #define ECB_STRINGIFY_(a) # a
297     #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
298    
299 root 1.46 #define ecb_function_ ecb_inline
300 root 1.3
301 root 1.1 #if ECB_GCC_VERSION(3,1)
302 root 1.37 #define ecb_attribute(attrlist) __attribute__(attrlist)
303     #define ecb_is_constant(expr) __builtin_constant_p (expr)
304     #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
305     #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
306     #else
307     #define ecb_attribute(attrlist)
308 root 1.127
309     /* possible C11 impl for integral types
310     typedef struct ecb_is_constant_struct ecb_is_constant_struct;
311     #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
312    
313 root 1.37 #define ecb_is_constant(expr) 0
314     #define ecb_expect(expr,value) (expr)
315     #define ecb_prefetch(addr,rw,locality)
316 root 1.1 #endif
317    
318 root 1.2 /* no emulation for ecb_decltype */
319 root 1.1 #if ECB_GCC_VERSION(4,5)
320 root 1.37 #define ecb_decltype(x) __decltype(x)
321 root 1.1 #elif ECB_GCC_VERSION(3,0)
322 root 1.39 #define ecb_decltype(x) __typeof(x)
323 root 1.1 #endif
324    
325 root 1.135 #if _MSC_VER >= 1300
326     #define ecb_deprecated __declspec(deprecated)
327     #else
328     #define ecb_deprecated ecb_attribute ((__deprecated__))
329     #endif
330    
331 root 1.24 #define ecb_noinline ecb_attribute ((__noinline__))
332     #define ecb_unused ecb_attribute ((__unused__))
333     #define ecb_const ecb_attribute ((__const__))
334     #define ecb_pure ecb_attribute ((__pure__))
335 root 1.35
336 root 1.136 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx __declspec(noreturn) */
337 root 1.90 #if ECB_C11
338     #define ecb_noreturn _Noreturn
339     #else
340     #define ecb_noreturn ecb_attribute ((__noreturn__))
341     #endif
342    
343 root 1.35 #if ECB_GCC_VERSION(4,3)
344 root 1.39 #define ecb_artificial ecb_attribute ((__artificial__))
345     #define ecb_hot ecb_attribute ((__hot__))
346     #define ecb_cold ecb_attribute ((__cold__))
347 root 1.35 #else
348     #define ecb_artificial
349     #define ecb_hot
350     #define ecb_cold
351     #endif
352 root 1.1
353 root 1.39 /* put around conditional expressions if you are very sure that the */
354     /* expression is mostly true or mostly false. note that these return */
355     /* booleans, not the expression. */
356 root 1.33 #define ecb_expect_false(expr) ecb_expect (!!(expr), 0)
357     #define ecb_expect_true(expr) ecb_expect (!!(expr), 1)
358 root 1.36 /* for compatibility to the rest of the world */
359 root 1.33 #define ecb_likely(expr) ecb_expect_true (expr)
360     #define ecb_unlikely(expr) ecb_expect_false (expr)
361 root 1.1
362 root 1.3 /* count trailing zero bits and count # of one bits */
363 root 1.1 #if ECB_GCC_VERSION(3,4)
364 root 1.49 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */
365     #define ecb_ld32(x) (__builtin_clz (x) ^ 31)
366     #define ecb_ld64(x) (__builtin_clzll (x) ^ 63)
367 root 1.35 #define ecb_ctz32(x) __builtin_ctz (x)
368 root 1.49 #define ecb_ctz64(x) __builtin_ctzll (x)
369 root 1.35 #define ecb_popcount32(x) __builtin_popcount (x)
370 root 1.49 /* no popcountll */
371 root 1.1 #else
372 root 1.39 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const;
373     ecb_function_ int
374 root 1.35 ecb_ctz32 (uint32_t x)
375     {
376     int r = 0;
377    
378 root 1.48 x &= ~x + 1; /* this isolates the lowest bit */
379 root 1.35
380 root 1.50 #if ECB_branchless_on_i386
381     r += !!(x & 0xaaaaaaaa) << 0;
382     r += !!(x & 0xcccccccc) << 1;
383     r += !!(x & 0xf0f0f0f0) << 2;
384     r += !!(x & 0xff00ff00) << 3;
385     r += !!(x & 0xffff0000) << 4;
386     #else
387 root 1.35 if (x & 0xaaaaaaaa) r += 1;
388     if (x & 0xcccccccc) r += 2;
389     if (x & 0xf0f0f0f0) r += 4;
390     if (x & 0xff00ff00) r += 8;
391     if (x & 0xffff0000) r += 16;
392 root 1.50 #endif
393 root 1.35
394     return r;
395     }
396    
397 root 1.49 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const;
398     ecb_function_ int
399     ecb_ctz64 (uint64_t x)
400     {
401     int shift = x & 0xffffffffU ? 0 : 32;
402 root 1.50 return ecb_ctz32 (x >> shift) + shift;
403 root 1.49 }
404    
405 root 1.39 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const;
406     ecb_function_ int
407 root 1.35 ecb_popcount32 (uint32_t x)
408     {
409     x -= (x >> 1) & 0x55555555;
410     x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
411     x = ((x >> 4) + x) & 0x0f0f0f0f;
412     x *= 0x01010101;
413 root 1.1
414 root 1.35 return x >> 24;
415     }
416 root 1.49
417     ecb_function_ int ecb_ld32 (uint32_t x) ecb_const;
418     ecb_function_ int ecb_ld32 (uint32_t x)
419     {
420 root 1.50 int r = 0;
421 root 1.49
422 root 1.50 if (x >> 16) { x >>= 16; r += 16; }
423     if (x >> 8) { x >>= 8; r += 8; }
424     if (x >> 4) { x >>= 4; r += 4; }
425     if (x >> 2) { x >>= 2; r += 2; }
426     if (x >> 1) { r += 1; }
427 root 1.49
428     return r;
429     }
430    
431     ecb_function_ int ecb_ld64 (uint64_t x) ecb_const;
432     ecb_function_ int ecb_ld64 (uint64_t x)
433     {
434 root 1.50 int r = 0;
435 root 1.49
436 root 1.50 if (x >> 32) { x >>= 32; r += 32; }
437 root 1.49
438 root 1.50 return r + ecb_ld32 (x);
439 root 1.49 }
440 root 1.1 #endif
441    
442 root 1.88 ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const;
443     ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
444     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const;
445     ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
446    
447 root 1.71 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const;
448 root 1.70 ecb_function_ uint8_t ecb_bitrev8 (uint8_t x)
449     {
450     return ( (x * 0x0802U & 0x22110U)
451     | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
452     }
453    
454 root 1.71 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const;
455 root 1.70 ecb_function_ uint16_t ecb_bitrev16 (uint16_t x)
456     {
457     x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
458     x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
459     x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
460     x = ( x >> 8 ) | ( x << 8);
461    
462     return x;
463     }
464    
465 root 1.71 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const;
466 root 1.70 ecb_function_ uint32_t ecb_bitrev32 (uint32_t x)
467     {
468     x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
469     x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
470     x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
471     x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
472     x = ( x >> 16 ) | ( x << 16);
473    
474     return x;
475     }
476    
477 root 1.49 /* popcount64 is only available on 64 bit cpus as gcc builtin */
478     /* so for this version we are lazy */
479     ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const;
480     ecb_function_ int
481     ecb_popcount64 (uint64_t x)
482     {
483     return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
484     }
485    
486 root 1.50 ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const;
487     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const;
488     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const;
489     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const;
490     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const;
491     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const;
492     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const;
493     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const;
494    
495     ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); }
496     ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); }
497     ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); }
498     ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
499     ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
500     ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
501     ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
502     ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
503    
504 root 1.13 #if ECB_GCC_VERSION(4,3)
505 root 1.49 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
506     #define ecb_bswap32(x) __builtin_bswap32 (x)
507     #define ecb_bswap64(x) __builtin_bswap64 (x)
508 root 1.13 #else
509 root 1.50 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const;
510     ecb_function_ uint16_t
511     ecb_bswap16 (uint16_t x)
512 root 1.49 {
513 root 1.50 return ecb_rotl16 (x, 8);
514 root 1.49 }
515    
516 root 1.39 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const;
517     ecb_function_ uint32_t
518 root 1.35 ecb_bswap32 (uint32_t x)
519     {
520 root 1.50 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
521 root 1.35 }
522    
523 root 1.49 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const;
524     ecb_function_ uint64_t
525     ecb_bswap64 (uint64_t x)
526 root 1.35 {
527 root 1.50 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
528 root 1.35 }
529 root 1.13 #endif
530    
531     #if ECB_GCC_VERSION(4,5)
532 root 1.35 #define ecb_unreachable() __builtin_unreachable ()
533 root 1.13 #else
534 root 1.35 /* this seems to work fine, but gcc always emits a warning for it :/ */
535 root 1.75 ecb_inline void ecb_unreachable (void) ecb_noreturn;
536     ecb_inline void ecb_unreachable (void) { }
537 root 1.13 #endif
538    
539 root 1.41 /* try to tell the compiler that some condition is definitely true */
540 root 1.100 #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
541 root 1.41
542 root 1.76 ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const;
543     ecb_inline unsigned char
544 root 1.23 ecb_byteorder_helper (void)
545 root 1.3 {
546 root 1.98 /* the union code still generates code under pressure in gcc, */
547 sf-exg 1.111 /* but less than using pointers, and always seems to */
548 root 1.98 /* successfully return a constant. */
549     /* the reason why we have this horrible preprocessor mess */
550 root 1.99 /* is to avoid it in all cases, at least on common architectures */
551 sf-exg 1.111 /* or when using a recent enough gcc version (>= 4.6) */
552 root 1.98 #if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
553     return 0x44;
554     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
555     return 0x44;
556     #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
557 root 1.113 return 0x11;
558 root 1.98 #else
559     union
560     {
561     uint32_t i;
562     uint8_t c;
563     } u = { 0x11223344 };
564     return u.c;
565     #endif
566 root 1.3 }
567    
568 root 1.76 ecb_inline ecb_bool ecb_big_endian (void) ecb_const;
569     ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; }
570     ecb_inline ecb_bool ecb_little_endian (void) ecb_const;
571     ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; }
572 root 1.3
573 root 1.39 #if ECB_GCC_VERSION(3,0) || ECB_C99
574 root 1.35 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
575 root 1.31 #else
576 root 1.35 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
577 root 1.31 #endif
578 root 1.21
579 sf-exg 1.68 #if __cplusplus
580     template<typename T>
581     static inline T ecb_div_rd (T val, T div)
582     {
583     return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
584     }
585     template<typename T>
586     static inline T ecb_div_ru (T val, T div)
587     {
588     return val < 0 ? - ((-val ) / div) : (val + div - 1) / div;
589     }
590     #else
591     #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
592     #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
593     #endif
594 sf-exg 1.67
595 root 1.5 #if ecb_cplusplus_does_not_suck
596 root 1.40 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
597 root 1.35 template<typename T, int N>
598     static inline int ecb_array_length (const T (&arr)[N])
599     {
600     return N;
601     }
602 root 1.5 #else
603 root 1.35 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
604 root 1.5 #endif
605    
606 root 1.104 /*******************************************************************************/
607     /* floating point stuff, can be disabled by defining ECB_NO_LIBM */
608    
609     /* basically, everything uses "ieee pure-endian" floating point numbers */
610     /* the only noteworthy exception is ancient armle, which uses order 43218765 */
611     #if 0 \
612     || __i386 || __i386__ \
613     || __amd64 || __amd64__ || __x86_64 || __x86_64__ \
614     || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
615     || defined __s390__ || defined __s390x__ \
616     || defined __mips__ \
617     || defined __alpha__ \
618     || defined __hppa__ \
619     || defined __ia64__ \
620 root 1.117 || defined __m68k__ \
621     || defined __m88k__ \
622     || defined __sh__ \
623 root 1.130 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 \
624 root 1.131 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
625 root 1.132 || defined __aarch64__
626 root 1.104 #define ECB_STDFP 1
627     #include <string.h> /* for memcpy */
628 root 1.102 #else
629 root 1.104 #define ECB_STDFP 0
630 root 1.102 #endif
631    
632 root 1.104 #ifndef ECB_NO_LIBM
633 root 1.103
634 root 1.121 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
635    
636 root 1.122 /* only the oldest of old doesn't have this one. solaris. */
637     #ifdef INFINITY
638     #define ECB_INFINITY INFINITY
639     #else
640     #define ECB_INFINITY HUGE_VAL
641     #endif
642    
643     #ifdef NAN
644 root 1.121 #define ECB_NAN NAN
645     #else
646 root 1.122 #define ECB_NAN ECB_INFINITY
647 root 1.121 #endif
648 root 1.120
649     /* converts an ieee half/binary16 to a float */
650     ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const;
651     ecb_function_ float
652     ecb_binary16_to_float (uint16_t x)
653     {
654     int e = (x >> 10) & 0x1f;
655     int m = x & 0x3ff;
656     float r;
657    
658 root 1.121 if (!e ) r = ldexpf (m , -24);
659     else if (e != 31) r = ldexpf (m + 0x400, e - 25);
660     else if (m ) r = ECB_NAN;
661 root 1.123 else r = ECB_INFINITY;
662 root 1.120
663     return x & 0x8000 ? -r : r;
664     }
665    
666 root 1.104 /* convert a float to ieee single/binary32 */
667 root 1.103 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const;
668     ecb_function_ uint32_t
669     ecb_float_to_binary32 (float x)
670     {
671     uint32_t r;
672    
673     #if ECB_STDFP
674 root 1.104 memcpy (&r, &x, 4);
675 root 1.103 #else
676 root 1.105 /* slow emulation, works for anything but -0 */
677 root 1.103 uint32_t m;
678     int e;
679    
680 root 1.108 if (x == 0e0f ) return 0x00000000U;
681 root 1.103 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
682     if (x < -3.40282346638528860e+38f) return 0xff800000U;
683 root 1.105 if (x != x ) return 0x7fbfffffU;
684 root 1.103
685     m = frexpf (x, &e) * 0x1000000U;
686    
687     r = m & 0x80000000U;
688    
689     if (r)
690     m = -m;
691    
692 root 1.108 if (e <= -126)
693 root 1.103 {
694     m &= 0xffffffU;
695     m >>= (-125 - e);
696     e = -126;
697     }
698    
699     r |= (e + 126) << 23;
700     r |= m & 0x7fffffU;
701     #endif
702    
703     return r;
704     }
705    
706 root 1.104 /* converts an ieee single/binary32 to a float */
707 root 1.103 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const;
708     ecb_function_ float
709     ecb_binary32_to_float (uint32_t x)
710     {
711     float r;
712    
713     #if ECB_STDFP
714 root 1.104 memcpy (&r, &x, 4);
715 root 1.103 #else
716     /* emulation, only works for normals and subnormals and +0 */
717     int neg = x >> 31;
718     int e = (x >> 23) & 0xffU;
719    
720     x &= 0x7fffffU;
721    
722     if (e)
723     x |= 0x800000U;
724 root 1.104 else
725     e = 1;
726 root 1.103
727     /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
728 root 1.108 r = ldexpf (x * (0.5f / 0x800000U), e - 126);
729 root 1.103
730     r = neg ? -r : r;
731     #endif
732    
733     return r;
734     }
735    
736 root 1.104 /* convert a double to ieee double/binary64 */
737 root 1.103 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const;
738     ecb_function_ uint64_t
739     ecb_double_to_binary64 (double x)
740     {
741 root 1.104 uint64_t r;
742    
743     #if ECB_STDFP
744     memcpy (&r, &x, 8);
745     #else
746 root 1.105 /* slow emulation, works for anything but -0 */
747 root 1.104 uint64_t m;
748     int e;
749    
750 root 1.108 if (x == 0e0 ) return 0x0000000000000000U;
751 root 1.104 if (x > +1.79769313486231470e+308) return 0x7ff0000000000000U;
752     if (x < -1.79769313486231470e+308) return 0xfff0000000000000U;
753 root 1.105 if (x != x ) return 0X7ff7ffffffffffffU;
754 root 1.104
755     m = frexp (x, &e) * 0x20000000000000U;
756    
757     r = m & 0x8000000000000000;;
758    
759     if (r)
760     m = -m;
761    
762 root 1.108 if (e <= -1022)
763 root 1.104 {
764     m &= 0x1fffffffffffffU;
765     m >>= (-1021 - e);
766     e = -1022;
767     }
768    
769     r |= ((uint64_t)(e + 1022)) << 52;
770     r |= m & 0xfffffffffffffU;
771     #endif
772    
773     return r;
774     }
775    
776     /* converts an ieee double/binary64 to a double */
777     ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const;
778     ecb_function_ double
779     ecb_binary64_to_double (uint64_t x)
780     {
781     double r;
782    
783     #if ECB_STDFP
784     memcpy (&r, &x, 8);
785     #else
786     /* emulation, only works for normals and subnormals and +0 */
787     int neg = x >> 63;
788     int e = (x >> 52) & 0x7ffU;
789    
790     x &= 0xfffffffffffffU;
791    
792     if (e)
793     x |= 0x10000000000000U;
794     else
795     e = 1;
796    
797 root 1.107 /* we distrust ldexp a bit and do the 2**-53 scaling by an extra multiply */
798 root 1.108 r = ldexp (x * (0.5 / 0x10000000000000U), e - 1022);
799 root 1.104
800     r = neg ? -r : r;
801     #endif
802    
803     return r;
804 root 1.103 }
805    
806     #endif
807 root 1.102
808 root 1.1 #endif
809