ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
Revision: 1.129
Committed: Thu Mar 20 22:04:29 2014 UTC (10 years, 3 months ago) by root
Content type: text/plain
Branch: MAIN
Changes since 1.128: +3 -1 lines
Log Message:
patch by Riku Voipio <riku.voipio@linaro.org>

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