ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libecb/ecb.h
(Generate patch)

Comparing libecb/ecb.h (file contents):
Revision 1.120 by root, Sat Oct 26 09:13:54 2013 UTC vs.
Revision 1.206 by root, Fri Mar 25 14:33:02 2022 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2012 Marc Alexander Lehmann <libecb@schmorp.de> 4 * Copyright (©) 2009-2015,2018-2021 Marc Alexander Lehmann <libecb@schmorp.de>
5 * Copyright (©) 2011 Emanuele Giaquinta 5 * Copyright (©) 2011 Emanuele Giaquinta
6 * All rights reserved. 6 * All rights reserved.
7 * 7 *
8 * Redistribution and use in source and binary forms, with or without modifica- 8 * Redistribution and use in source and binary forms, with or without modifica-
9 * tion, are permitted provided that the following conditions are met: 9 * tion, are permitted provided that the following conditions are met:
… …
23 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 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, 24 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
25 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- 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 26 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
27 * OF THE POSSIBILITY OF SUCH DAMAGE. 27 * OF THE POSSIBILITY OF SUCH DAMAGE.
28 *
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.
28 */ 39 */
29 40
30#ifndef ECB_H 41#ifndef ECB_H
31#define ECB_H 42#define ECB_H
32 43
33/* 16 bits major, 16 bits minor */ 44/* 16 bits major, 16 bits minor */
34#define ECB_VERSION 0x00010003 45#define ECB_VERSION 0x0001000c
35 46
36#ifdef _WIN32 47#include <string.h> /* for memcpy */
48
49#if defined (_WIN32) && !defined (__MINGW32__)
37 typedef signed char int8_t; 50 typedef signed char int8_t;
38 typedef unsigned char uint8_t; 51 typedef unsigned char uint8_t;
52 typedef signed char int_fast8_t;
53 typedef unsigned char uint_fast8_t;
39 typedef signed short int16_t; 54 typedef signed short int16_t;
40 typedef unsigned short uint16_t; 55 typedef unsigned short uint16_t;
56 typedef signed int int_fast16_t;
57 typedef unsigned int uint_fast16_t;
41 typedef signed int int32_t; 58 typedef signed int int32_t;
42 typedef unsigned int uint32_t; 59 typedef unsigned int uint32_t;
60 typedef signed int int_fast32_t;
61 typedef unsigned int uint_fast32_t;
43 #if __GNUC__ 62 #if __GNUC__
44 typedef signed long long int64_t; 63 typedef signed long long int64_t;
45 typedef unsigned long long uint64_t; 64 typedef unsigned long long uint64_t;
46 #else /* _MSC_VER || __BORLANDC__ */ 65 #else /* _MSC_VER || __BORLANDC__ */
47 typedef signed __int64 int64_t; 66 typedef signed __int64 int64_t;
48 typedef unsigned __int64 uint64_t; 67 typedef unsigned __int64 uint64_t;
49 #endif 68 #endif
69 typedef int64_t int_fast64_t;
70 typedef uint64_t uint_fast64_t;
50 #ifdef _WIN64 71 #ifdef _WIN64
51 #define ECB_PTRSIZE 8 72 #define ECB_PTRSIZE 8
52 typedef uint64_t uintptr_t; 73 typedef uint64_t uintptr_t;
53 typedef int64_t intptr_t; 74 typedef int64_t intptr_t;
54 #else 75 #else
… …
56 typedef uint32_t uintptr_t; 77 typedef uint32_t uintptr_t;
57 typedef int32_t intptr_t; 78 typedef int32_t intptr_t;
58 #endif 79 #endif
59#else 80#else
60 #include <inttypes.h> 81 #include <inttypes.h>
61 #if UINTMAX_MAX > 0xffffffffU 82 #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU
62 #define ECB_PTRSIZE 8 83 #define ECB_PTRSIZE 8
63 #else 84 #else
64 #define ECB_PTRSIZE 4 85 #define ECB_PTRSIZE 4
65 #endif 86 #endif
66#endif 87#endif
67 88
89#define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__)
90#define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64)
91
92#ifndef ECB_OPTIMIZE_SIZE
93 #if __OPTIMIZE_SIZE__
94 #define ECB_OPTIMIZE_SIZE 1
95 #else
96 #define ECB_OPTIMIZE_SIZE 0
97 #endif
98#endif
99
68/* work around x32 idiocy by defining proper macros */ 100/* work around x32 idiocy by defining proper macros */
69#if __x86_64 || _M_AMD64 101#if ECB_GCC_AMD64 || ECB_MSVC_AMD64
70 #if _ILP32 102 #if _ILP32
71 #define ECB_AMD64_X32 1 103 #define ECB_AMD64_X32 1
72 #else 104 #else
73 #define ECB_AMD64 1 105 #define ECB_AMD64 1
74 #endif 106 #endif
107#endif
108
109#if ECB_PTRSIZE >= 8 || ECB_AMD64_X32
110 #define ECB_64BIT_NATIVE 1
111#else
112 #define ECB_64BIT_NATIVE 0
75#endif 113#endif
76 114
77/* many compilers define _GNUC_ to some versions but then only implement 115/* many compilers define _GNUC_ to some versions but then only implement
78 * what their idiot authors think are the "more important" extensions, 116 * what their idiot authors think are the "more important" extensions,
79 * causing enormous grief in return for some better fake benchmark numbers. 117 * causing enormous grief in return for some better fake benchmark numbers.
80 * or so. 118 * or so.
81 * we try to detect these and simply assume they are not gcc - if they have 119 * 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. 120 * an issue with that they should have done it right in the first place.
83 */ 121 */
84#ifndef ECB_GCC_VERSION
85 #if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__ 122#if !defined __GNUC_MINOR__ || defined __INTEL_COMPILER || defined __SUNPRO_C || defined __SUNPRO_CC || defined __llvm__ || defined __clang__
86 #define ECB_GCC_VERSION(major,minor) 0 123 #define ECB_GCC_VERSION(major,minor) 0
87 #else 124#else
88 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor))) 125 #define ECB_GCC_VERSION(major,minor) (__GNUC__ > (major) || (__GNUC__ == (major) && __GNUC_MINOR__ >= (minor)))
89 #endif 126#endif
90#endif
91 127
92#define ECB_C (__STDC__+0) /* this assumes that __STDC__ is either empty or a number */ 128#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
93#define ECB_C99 (__STDC_VERSION__ >= 199901L) 129
94#define ECB_C11 (__STDC_VERSION__ >= 201112L) 130#if __clang__ && defined __has_builtin
131 #define ECB_CLANG_BUILTIN(x) __has_builtin (x)
132#else
133 #define ECB_CLANG_BUILTIN(x) 0
134#endif
135
136#if __clang__ && defined __has_extension
137 #define ECB_CLANG_EXTENSION(x) __has_extension (x)
138#else
139 #define ECB_CLANG_EXTENSION(x) 0
140#endif
141
95#define ECB_CPP (__cplusplus+0) 142#define ECB_CPP (__cplusplus+0)
96#define ECB_CPP11 (__cplusplus >= 201103L) 143#define ECB_CPP11 (__cplusplus >= 201103L)
144#define ECB_CPP14 (__cplusplus >= 201402L)
145#define ECB_CPP17 (__cplusplus >= 201703L)
146
147#if ECB_CPP
148 #define ECB_C 0
149 #define ECB_STDC_VERSION 0
150#else
151 #define ECB_C 1
152 #define ECB_STDC_VERSION __STDC_VERSION__
153#endif
154
155#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
156#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
157#define ECB_C17 (ECB_STDC_VERSION >= 201710L)
97 158
98#if ECB_CPP 159#if ECB_CPP
99 #define ECB_EXTERN_C extern "C" 160 #define ECB_EXTERN_C extern "C"
100 #define ECB_EXTERN_C_BEG ECB_EXTERN_C { 161 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
101 #define ECB_EXTERN_C_END } 162 #define ECB_EXTERN_C_END }
… …
116 177
117#if ECB_NO_SMP 178#if ECB_NO_SMP
118 #define ECB_MEMORY_FENCE do { } while (0) 179 #define ECB_MEMORY_FENCE do { } while (0)
119#endif 180#endif
120 181
182/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
183#if __xlC__ && ECB_CPP
184 #include <builtins.h>
185#endif
186
187#if 1400 <= _MSC_VER
188 #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
189#endif
190
121#ifndef ECB_MEMORY_FENCE 191#ifndef ECB_MEMORY_FENCE
122 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 192 #if ECB_GCC_VERSION(2,5) || defined __INTEL_COMPILER || (__llvm__ && __GNUC__) || __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
193 #define ECB_MEMORY_FENCE_RELAXED __asm__ __volatile__ ("" : : : "memory")
123 #if __i386 || __i386__ 194 #if __i386 || __i386__
124 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory") 195 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("lock; orb $0, -1(%%esp)" : : : "memory")
125 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 196 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
126 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 197 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
127 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 198 #elif ECB_GCC_AMD64
128 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 199 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
129 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 200 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
130 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 201 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
131 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 202 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
132 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory") 203 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("sync" : : : "memory")
204 #elif defined __ARM_ARCH_2__ \
205 || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \
206 || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \
207 || defined __ARM_ARCH_5__ || defined __ARM_ARCH_5E__ \
208 || defined __ARM_ARCH_5T__ || defined __ARM_ARCH_5TE__ \
209 || defined __ARM_ARCH_5TEJ__
210 /* should not need any, unless running old code on newer cpu - arm doesn't support that */
133 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ 211 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
134 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 212 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
213 || defined __ARM_ARCH_6T2__
135 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory") 214 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mcr p15,0,%0,c7,c10,5" : : "r" (0) : "memory")
136 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 215 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
137 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 216 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
138 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 217 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
139 #elif __sparc || __sparc__ 218 #elif __aarch64__
219 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
220 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
140 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory") 221 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad | #StoreStore | #StoreLoad" : : : "memory")
141 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 222 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
142 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 223 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
143 #elif defined __s390__ || defined __s390x__ 224 #elif defined __s390__ || defined __s390x__
144 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 225 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
… …
165 246
166#ifndef ECB_MEMORY_FENCE 247#ifndef ECB_MEMORY_FENCE
167 #if ECB_GCC_VERSION(4,7) 248 #if ECB_GCC_VERSION(4,7)
168 /* see comment below (stdatomic.h) about the C11 memory model. */ 249 /* see comment below (stdatomic.h) about the C11 memory model. */
169 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 250 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
251 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
252 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
253 #undef ECB_MEMORY_FENCE_RELAXED
254 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
170 255
171 /* The __has_feature syntax from clang is so misdesigned that we cannot use it 256 #elif ECB_CLANG_EXTENSION(c_atomic)
172 * without risking compile time errors with other compilers. We *could*
173 * define our own ecb_clang_has_feature, but I just can't be bothered to work
174 * around this shit time and again.
175 * #elif defined __clang && __has_feature (cxx_atomic)
176 * // see comment below (stdatomic.h) about the C11 memory model. 257 /* see comment below (stdatomic.h) about the C11 memory model. */
177 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 258 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
178 */ 259 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
260 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
261 #undef ECB_MEMORY_FENCE_RELAXED
262 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
179 263
180 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 264 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
181 #define ECB_MEMORY_FENCE __sync_synchronize () 265 #define ECB_MEMORY_FENCE __sync_synchronize ()
266 #elif _MSC_VER >= 1500 /* VC++ 2008 */
267 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
268 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
269 #define ECB_MEMORY_FENCE _ReadWriteBarrier (); MemoryBarrier()
270 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier (); MemoryBarrier() /* according to msdn, _ReadBarrier is not a load fence */
271 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier (); MemoryBarrier()
182 #elif _MSC_VER >= 1400 /* VC++ 2005 */ 272 #elif _MSC_VER >= 1400 /* VC++ 2005 */
183 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier) 273 #pragma intrinsic(_ReadBarrier,_WriteBarrier,_ReadWriteBarrier)
184 #define ECB_MEMORY_FENCE _ReadWriteBarrier () 274 #define ECB_MEMORY_FENCE _ReadWriteBarrier ()
185 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */ 275 #define ECB_MEMORY_FENCE_ACQUIRE _ReadWriteBarrier () /* according to msdn, _ReadBarrier is not a load fence */
186 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier () 276 #define ECB_MEMORY_FENCE_RELEASE _WriteBarrier ()
187 #elif defined _WIN32 277 #elif defined _WIN32
188 #include <WinNT.h> 278 #include <WinNT.h>
189 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 279 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
190 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 280 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
191 #include <mbarrier.h> 281 #include <mbarrier.h>
192 #define ECB_MEMORY_FENCE __machine_rw_barrier () 282 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
193 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () 283 #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
194 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () 284 #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
285 #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
195 #elif __xlC__ 286 #elif __xlC__
196 #define ECB_MEMORY_FENCE __sync () 287 #define ECB_MEMORY_FENCE __sync ()
197 #endif 288 #endif
198#endif 289#endif
199 290
200#ifndef ECB_MEMORY_FENCE 291#ifndef ECB_MEMORY_FENCE
201 #if ECB_C11 && !defined __STDC_NO_ATOMICS__ 292 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
202 /* we assume that these memory fences work on all variables/all memory accesses, */ 293 /* we assume that these memory fences work on all variables/all memory accesses, */
203 /* not just C11 atomics and atomic accesses */ 294 /* not just C11 atomics and atomic accesses */
204 #include <stdatomic.h> 295 #include <stdatomic.h>
205 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
206 /* any fence other than seq_cst, which isn't very efficient for us. */
207 /* Why that is, we don't know - either the C11 memory model is quite useless */
208 /* for most usages, or gcc and clang have a bug */
209 /* I *currently* lean towards the latter, and inefficiently implement */
210 /* all three of ecb's fences as a seq_cst fence */
211 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) 296 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
297 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
298 #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
212 #endif 299 #endif
213#endif 300#endif
214 301
215#ifndef ECB_MEMORY_FENCE 302#ifndef ECB_MEMORY_FENCE
216 #if !ECB_AVOID_PTHREADS 303 #if !ECB_AVOID_PTHREADS
… …
236 323
237#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE 324#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
238 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 325 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
239#endif 326#endif
240 327
328#if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE
329 #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */
330#endif
331
241/*****************************************************************************/ 332/*****************************************************************************/
242 333
243#if __cplusplus 334#if ECB_CPP
244 #define ecb_inline static inline 335 #define ecb_inline static inline
245#elif ECB_GCC_VERSION(2,5) 336#elif ECB_GCC_VERSION(2,5)
246 #define ecb_inline static __inline__ 337 #define ecb_inline static __inline__
247#elif ECB_C99 338#elif ECB_C99
248 #define ecb_inline static inline 339 #define ecb_inline static inline
… …
262 353
263#define ECB_CONCAT_(a, b) a ## b 354#define ECB_CONCAT_(a, b) a ## b
264#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 355#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
265#define ECB_STRINGIFY_(a) # a 356#define ECB_STRINGIFY_(a) # a
266#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) 357#define ECB_STRINGIFY(a) ECB_STRINGIFY_(a)
358#define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr))
267 359
268#define ecb_function_ ecb_inline 360#define ecb_function_ ecb_inline
269 361
270#if ECB_GCC_VERSION(3,1) 362#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
271 #define ecb_attribute(attrlist) __attribute__(attrlist) 363 #define ecb_attribute(attrlist) __attribute__ (attrlist)
364#else
365 #define ecb_attribute(attrlist)
366#endif
367
368#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_constant_p)
272 #define ecb_is_constant(expr) __builtin_constant_p (expr) 369 #define ecb_is_constant(expr) __builtin_constant_p (expr)
370#else
371 /* possible C11 impl for integral types
372 typedef struct ecb_is_constant_struct ecb_is_constant_struct;
373 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
374
375 #define ecb_is_constant(expr) 0
376#endif
377
378#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
273 #define ecb_expect(expr,value) __builtin_expect ((expr),(value)) 379 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
380#else
381 #define ecb_expect(expr,value) (expr)
382#endif
383
384#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
274 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality) 385 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
275#else 386#else
276 #define ecb_attribute(attrlist)
277 #define ecb_is_constant(expr) 0
278 #define ecb_expect(expr,value) (expr)
279 #define ecb_prefetch(addr,rw,locality) 387 #define ecb_prefetch(addr,rw,locality)
280#endif 388#endif
281 389
282/* no emulation for ecb_decltype */ 390/* no emulation for ecb_decltype */
283#if ECB_GCC_VERSION(4,5) 391#if ECB_CPP11
392 // older implementations might have problems with decltype(x)::type, work around it
393 template<class T> struct ecb_decltype_t { typedef T type; };
284 #define ecb_decltype(x) __decltype(x) 394 #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
285#elif ECB_GCC_VERSION(3,0) 395#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
286 #define ecb_decltype(x) __typeof(x) 396 #define ecb_decltype(x) __typeof__ (x)
287#endif 397#endif
288 398
399#if _MSC_VER >= 1300
400 #define ecb_deprecated __declspec (deprecated)
401#else
402 #define ecb_deprecated ecb_attribute ((__deprecated__))
403#endif
404
405#if _MSC_VER >= 1500
406 #define ecb_deprecated_message(msg) __declspec (deprecated (msg))
407#elif ECB_GCC_VERSION(4,5)
408 #define ecb_deprecated_message(msg) ecb_attribute ((__deprecated__ (msg))
409#else
410 #define ecb_deprecated_message(msg) ecb_deprecated
411#endif
412
413#if _MSC_VER >= 1400
414 #define ecb_noinline __declspec (noinline)
415#else
289#define ecb_noinline ecb_attribute ((__noinline__)) 416 #define ecb_noinline ecb_attribute ((__noinline__))
417#endif
418
290#define ecb_unused ecb_attribute ((__unused__)) 419#define ecb_unused ecb_attribute ((__unused__))
291#define ecb_const ecb_attribute ((__const__)) 420#define ecb_const ecb_attribute ((__const__))
292#define ecb_pure ecb_attribute ((__pure__)) 421#define ecb_pure ecb_attribute ((__pure__))
293 422
294#if ECB_C11 423#if ECB_C11 || __IBMC_NORETURN
424 /* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/language_ref/noreturn.html */
295 #define ecb_noreturn _Noreturn 425 #define ecb_noreturn _Noreturn
426#elif ECB_CPP11
427 #define ecb_noreturn [[noreturn]]
428#elif _MSC_VER >= 1200
429 /* http://msdn.microsoft.com/en-us/library/k6ktzx3s.aspx */
430 #define ecb_noreturn __declspec (noreturn)
296#else 431#else
297 #define ecb_noreturn ecb_attribute ((__noreturn__)) 432 #define ecb_noreturn ecb_attribute ((__noreturn__))
298#endif 433#endif
299 434
300#if ECB_GCC_VERSION(4,3) 435#if ECB_GCC_VERSION(4,3)
… …
315/* for compatibility to the rest of the world */ 450/* for compatibility to the rest of the world */
316#define ecb_likely(expr) ecb_expect_true (expr) 451#define ecb_likely(expr) ecb_expect_true (expr)
317#define ecb_unlikely(expr) ecb_expect_false (expr) 452#define ecb_unlikely(expr) ecb_expect_false (expr)
318 453
319/* count trailing zero bits and count # of one bits */ 454/* count trailing zero bits and count # of one bits */
320#if ECB_GCC_VERSION(3,4) 455#if ECB_GCC_VERSION(3,4) \
321 /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ 456 || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \
322 #define ecb_ld32(x) (__builtin_clz (x) ^ 31) 457 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
323 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) 458 && ECB_CLANG_BUILTIN(__builtin_popcount))
324 #define ecb_ctz32(x) __builtin_ctz (x) 459 #define ecb_ctz32(x) __builtin_ctz (x)
460 #define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x))
325 #define ecb_ctz64(x) __builtin_ctzll (x) 461 #define ecb_clz32(x) __builtin_clz (x)
462 #define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x))
463 #define ecb_ld32(x) (ecb_clz32 (x) ^ 31)
464 #define ecb_ld64(x) (ecb_clz64 (x) ^ 63)
326 #define ecb_popcount32(x) __builtin_popcount (x) 465 #define ecb_popcount32(x) __builtin_popcount (x)
327 /* no popcountll */ 466 /* ecb_popcount64 is more difficult, see below */
328#else 467#else
329 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 468 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
330 ecb_function_ int 469 ecb_function_ ecb_const int
331 ecb_ctz32 (uint32_t x) 470 ecb_ctz32 (uint32_t x)
332 { 471 {
472#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
473 unsigned long r;
474 _BitScanForward (&r, x);
475 return (int)r;
476#else
333 int r = 0; 477 int r = 0;
478
479 /* todo: use david seal's algorithm */
334 480
335 x &= ~x + 1; /* this isolates the lowest bit */ 481 x &= ~x + 1; /* this isolates the lowest bit */
336 482
337#if ECB_branchless_on_i386 483#if ECB_branchless_on_i386
338 r += !!(x & 0xaaaaaaaa) << 0; 484 r += !!(x & 0xaaaaaaaa) << 0;
… …
347 if (x & 0xff00ff00) r += 8; 493 if (x & 0xff00ff00) r += 8;
348 if (x & 0xffff0000) r += 16; 494 if (x & 0xffff0000) r += 16;
349#endif 495#endif
350 496
351 return r; 497 return r;
498#endif
352 } 499 }
353 500
354 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; 501 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
355 ecb_function_ int 502 ecb_function_ ecb_const int
356 ecb_ctz64 (uint64_t x) 503 ecb_ctz64 (uint64_t x)
357 { 504 {
505#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
506 unsigned long r;
507 _BitScanForward64 (&r, x);
508 return (int)r;
509#else
358 int shift = x & 0xffffffffU ? 0 : 32; 510 int shift = x & 0xffffffff ? 0 : 32;
359 return ecb_ctz32 (x >> shift) + shift; 511 return ecb_ctz32 (x >> shift) + shift;
512#endif
360 } 513 }
361 514
362 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 515 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
363 ecb_function_ int 516 ecb_function_ ecb_const int
364 ecb_popcount32 (uint32_t x) 517 ecb_popcount32 (uint32_t x)
365 { 518 {
366 x -= (x >> 1) & 0x55555555; 519 x -= (x >> 1) & 0x55555555;
367 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 520 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
368 x = ((x >> 4) + x) & 0x0f0f0f0f; 521 x = ((x >> 4) + x) & 0x0f0f0f0f;
369 x *= 0x01010101; 522 x *= 0x01010101;
370 523
371 return x >> 24; 524 return x >> 24;
372 } 525 }
373 526
374 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 527 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
375 ecb_function_ int ecb_ld32 (uint32_t x) 528 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
376 { 529 {
530#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
531 unsigned long r;
532 _BitScanReverse (&r, x);
533 return (int)r;
534#else
377 int r = 0; 535 int r = 0;
378 536
379 if (x >> 16) { x >>= 16; r += 16; } 537 if (x >> 16) { x >>= 16; r += 16; }
380 if (x >> 8) { x >>= 8; r += 8; } 538 if (x >> 8) { x >>= 8; r += 8; }
381 if (x >> 4) { x >>= 4; r += 4; } 539 if (x >> 4) { x >>= 4; r += 4; }
382 if (x >> 2) { x >>= 2; r += 2; } 540 if (x >> 2) { x >>= 2; r += 2; }
383 if (x >> 1) { r += 1; } 541 if (x >> 1) { r += 1; }
384 542
385 return r; 543 return r;
544#endif
386 } 545 }
387 546
388 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; 547 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
389 ecb_function_ int ecb_ld64 (uint64_t x) 548 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
390 { 549 {
550#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
551 unsigned long r;
552 _BitScanReverse64 (&r, x);
553 return (int)r;
554#else
391 int r = 0; 555 int r = 0;
392 556
393 if (x >> 32) { x >>= 32; r += 32; } 557 if (x >> 32) { x >>= 32; r += 32; }
394 558
395 return r + ecb_ld32 (x); 559 return r + ecb_ld32 (x);
560#endif
396 } 561 }
397#endif 562#endif
398 563
399ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; 564ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
400ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } 565ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
401ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; 566ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
402ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } 567ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
403 568
404ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; 569ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
405ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) 570ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
406{ 571{
407 return ( (x * 0x0802U & 0x22110U) 572 return ( (x * 0x0802U & 0x22110U)
408 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 573 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
409} 574}
410 575
411ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; 576ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
412ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) 577ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
413{ 578{
414 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); 579 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
415 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); 580 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
416 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); 581 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
417 x = ( x >> 8 ) | ( x << 8); 582 x = ( x >> 8 ) | ( x << 8);
418 583
419 return x; 584 return x;
420} 585}
421 586
422ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; 587ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
423ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) 588ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
424{ 589{
425 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); 590 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
426 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); 591 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
427 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); 592 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
428 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); 593 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
429 x = ( x >> 16 ) | ( x << 16); 594 x = ( x >> 16 ) | ( x << 16);
430 595
431 return x; 596 return x;
432} 597}
433 598
434/* popcount64 is only available on 64 bit cpus as gcc builtin */
435/* so for this version we are lazy */
436ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 599ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
437ecb_function_ int 600ecb_function_ ecb_const int
438ecb_popcount64 (uint64_t x) 601ecb_popcount64 (uint64_t x)
439{ 602{
603 /* popcount64 is only available on 64 bit cpus as gcc builtin. */
604 /* also, gcc/clang make this surprisingly difficult to use */
605#if __LP64__ && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
606 return __builtin_popcountl (x);
607#else
440 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 608 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
609#endif
441} 610}
442 611
443ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; 612ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
444ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; 613ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
445ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; 614ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
446ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; 615ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
447ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; 616ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
448ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; 617ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
449ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; 618ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
450ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; 619ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
451 620
452ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } 621ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
453ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } 622ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
454ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } 623ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
455ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 624ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
456ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 625ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
457ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 626ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
458ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 627ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
459ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 628ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
460 629
461#if ECB_GCC_VERSION(4,3) 630#if ECB_CPP
631
632inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
633inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
634inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
635inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
636
637inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
638inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
639inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
640inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
641
642inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
643inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
644inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
645inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
646
647inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
648inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
649inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
650inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
651
652inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
653inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
654inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
655
656inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
657inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
658inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
659inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
660
661inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
662inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
663inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
664inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
665
666#endif
667
668#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
669 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
670 #define ecb_bswap16(x) __builtin_bswap16 (x)
671 #else
462 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 672 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
673 #endif
463 #define ecb_bswap32(x) __builtin_bswap32 (x) 674 #define ecb_bswap32(x) __builtin_bswap32 (x)
464 #define ecb_bswap64(x) __builtin_bswap64 (x) 675 #define ecb_bswap64(x) __builtin_bswap64 (x)
676#elif _MSC_VER
677 #include <stdlib.h>
678 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
679 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
680 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
465#else 681#else
466 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; 682 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
467 ecb_function_ uint16_t 683 ecb_function_ ecb_const uint16_t
468 ecb_bswap16 (uint16_t x) 684 ecb_bswap16 (uint16_t x)
469 { 685 {
470 return ecb_rotl16 (x, 8); 686 return ecb_rotl16 (x, 8);
471 } 687 }
472 688
473 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 689 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
474 ecb_function_ uint32_t 690 ecb_function_ ecb_const uint32_t
475 ecb_bswap32 (uint32_t x) 691 ecb_bswap32 (uint32_t x)
476 { 692 {
477 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); 693 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
478 } 694 }
479 695
480 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; 696 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
481 ecb_function_ uint64_t 697 ecb_function_ ecb_const uint64_t
482 ecb_bswap64 (uint64_t x) 698 ecb_bswap64 (uint64_t x)
483 { 699 {
484 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); 700 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
485 } 701 }
486#endif 702#endif
487 703
488#if ECB_GCC_VERSION(4,5) 704#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
489 #define ecb_unreachable() __builtin_unreachable () 705 #define ecb_unreachable() __builtin_unreachable ()
490#else 706#else
491 /* this seems to work fine, but gcc always emits a warning for it :/ */ 707 /* this seems to work fine, but gcc always emits a warning for it :/ */
492 ecb_inline void ecb_unreachable (void) ecb_noreturn; 708 ecb_inline ecb_noreturn void ecb_unreachable (void);
493 ecb_inline void ecb_unreachable (void) { } 709 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
494#endif 710#endif
495 711
496/* try to tell the compiler that some condition is definitely true */ 712/* try to tell the compiler that some condition is definitely true */
497#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 713#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
498 714
499ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 715ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
500ecb_inline unsigned char 716ecb_inline ecb_const uint32_t
501ecb_byteorder_helper (void) 717ecb_byteorder_helper (void)
502{ 718{
503 /* the union code still generates code under pressure in gcc, */ 719 /* the union code still generates code under pressure in gcc, */
504 /* but less than using pointers, and always seems to */ 720 /* but less than using pointers, and always seems to */
505 /* successfully return a constant. */ 721 /* successfully return a constant. */
506 /* the reason why we have this horrible preprocessor mess */ 722 /* the reason why we have this horrible preprocessor mess */
507 /* is to avoid it in all cases, at least on common architectures */ 723 /* is to avoid it in all cases, at least on common architectures */
508 /* or when using a recent enough gcc version (>= 4.6) */ 724 /* or when using a recent enough gcc version (>= 4.6) */
509#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
510 return 0x44;
511#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 725#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
726 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
727 #define ECB_LITTLE_ENDIAN 1
512 return 0x44; 728 return 0x44332211;
513#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 729#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
730 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
731 #define ECB_BIG_ENDIAN 1
514 return 0x11; 732 return 0x11223344;
515#else 733#else
516 union 734 union
517 { 735 {
736 uint8_t c[4];
518 uint32_t i; 737 uint32_t u;
519 uint8_t c;
520 } u = { 0x11223344 }; 738 } u = { 0x11, 0x22, 0x33, 0x44 };
521 return u.c; 739 return u.u;
522#endif 740#endif
523} 741}
524 742
525ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 743ecb_inline ecb_const ecb_bool ecb_big_endian (void);
526ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 744ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
527ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 745ecb_inline ecb_const ecb_bool ecb_little_endian (void);
528ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 746ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
747
748/*****************************************************************************/
749/* unaligned load/store */
750
751ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
752ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
753ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
754
755ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
756ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
757ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
758
759ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
760ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
761ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
762
763ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
764ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
765ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
766
767ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
768ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
769ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
770
771ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
772ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
773ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
774
775ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
776ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
777ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
778
779ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
780ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
781ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
782
783ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
784ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
785ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
786
787ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
788ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
789ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
790
791#if ECB_CPP
792
793inline uint8_t ecb_bswap (uint8_t v) { return v; }
794inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
795inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
796inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
797
798template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
799template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
800template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
801template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
802template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
803template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
804template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
805template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
806
807template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
808template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
809template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
810template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
811template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
812template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
813template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
814template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
815
816#endif
817
818/*****************************************************************************/
819/* pointer/integer hashing */
820
821/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
822ecb_function_ uint32_t ecb_mix32 (uint32_t v);
823ecb_function_ uint32_t ecb_mix32 (uint32_t v)
824{
825 v ^= v >> 16; v *= 0x7feb352dU;
826 v ^= v >> 15; v *= 0x846ca68bU;
827 v ^= v >> 16;
828 return v;
829}
830
831ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
832ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
833{
834 v ^= v >> 16 ; v *= 0x43021123U;
835 v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
836 v ^= v >> 16 ;
837 return v;
838}
839
840/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
841ecb_function_ uint64_t ecb_mix64 (uint64_t v);
842ecb_function_ uint64_t ecb_mix64 (uint64_t v)
843{
844 v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
845 v ^= v >> 27; v *= 0x94d049bb133111ebU;
846 v ^= v >> 31;
847 return v;
848}
849
850ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
851ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
852{
853 v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
854 v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
855 v ^= v >> 30 ^ v >> 60;
856 return v;
857}
858
859ecb_function_ uintptr_t ecb_ptrmix (void *p);
860ecb_function_ uintptr_t ecb_ptrmix (void *p)
861{
862 #if ECB_PTRSIZE <= 4
863 return ecb_mix32 ((uint32_t)p);
864 #else
865 return ecb_mix64 ((uint64_t)p);
866 #endif
867}
868
869ecb_function_ void *ecb_ptrunmix (uintptr_t v);
870ecb_function_ void *ecb_ptrunmix (uintptr_t v)
871{
872 #if ECB_PTRSIZE <= 4
873 return (void *)ecb_unmix32 (v);
874 #else
875 return (void *)ecb_unmix64 (v);
876 #endif
877}
878
879#if ECB_CPP
880
881template<typename T>
882inline uintptr_t ecb_ptrmix (T *p)
883{
884 return ecb_ptrmix (static_cast<void *>(p));
885}
886
887template<typename T>
888inline T *ecb_ptrunmix (uintptr_t v)
889{
890 return static_cast<T *>(ecb_ptrunmix (v));
891}
892
893#endif
894
895/*****************************************************************************/
896/* gray code */
897
898ecb_function_ uint_fast8_t ecb_gray8_encode (uint_fast8_t b) { return b ^ (b >> 1); }
899ecb_function_ uint_fast16_t ecb_gray16_encode (uint_fast16_t b) { return b ^ (b >> 1); }
900ecb_function_ uint_fast32_t ecb_gray32_encode (uint_fast32_t b) { return b ^ (b >> 1); }
901ecb_function_ uint_fast64_t ecb_gray64_encode (uint_fast64_t b) { return b ^ (b >> 1); }
902
903ecb_function_ uint8_t ecb_gray8_decode (uint8_t g)
904{
905 g ^= g >> 1;
906 g ^= g >> 2;
907 g ^= g >> 4;
908
909 return g;
910}
911
912ecb_function_ uint16_t ecb_gray16_decode (uint16_t g)
913{
914 g ^= g >> 1;
915 g ^= g >> 2;
916 g ^= g >> 4;
917 g ^= g >> 8;
918
919 return g;
920}
921
922ecb_function_ uint32_t ecb_gray32_decode (uint32_t g)
923{
924 g ^= g >> 1;
925 g ^= g >> 2;
926 g ^= g >> 4;
927 g ^= g >> 8;
928 g ^= g >> 16;
929
930 return g;
931}
932
933ecb_function_ uint64_t ecb_gray64_decode (uint64_t g)
934{
935 g ^= g >> 1;
936 g ^= g >> 2;
937 g ^= g >> 4;
938 g ^= g >> 8;
939 g ^= g >> 16;
940 g ^= g >> 32;
941
942 return g;
943}
944
945#if ECB_CPP
946
947ecb_function_ uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray8_encode (b); }
948ecb_function_ uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray16_encode (b); }
949ecb_function_ uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray32_encode (b); }
950ecb_function_ uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray64_encode (b); }
951
952ecb_function_ uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray8_decode (g); }
953ecb_function_ uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray16_decode (g); }
954ecb_function_ uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray32_decode (g); }
955ecb_function_ uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray64_decode (g); }
956
957#endif
958
959/*****************************************************************************/
960/* 2d hilbert curves */
961
962/* algorithm from the book Hacker's Delight, modified to not */
963/* run into undefined behaviour for n==16 */
964static uint32_t
965ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
966{
967 uint32_t comp, swap, cs, t, sr;
968
969 /* pad s on the left (unused) bits with 01 (no change groups) */
970 s |= 0x55555555U << n << n;
971 /* "s shift right" */
972 sr = (s >> 1) & 0x55555555U;
973 /* compute complement and swap info in two-bit groups */
974 cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
975
976 /* parallel prefix xor op to propagate both complement
977 * and swap info together from left to right (there is
978 * no step "cs ^= cs >> 1", so in effect it computes
979 * two independent parallel prefix operations on two
980 * interleaved sets of sixteen bits).
981 */
982 cs ^= cs >> 2;
983 cs ^= cs >> 4;
984 cs ^= cs >> 8;
985 cs ^= cs >> 16;
986
987 /* separate swap and complement bits */
988 swap = cs & 0x55555555U;
989 comp = (cs >> 1) & 0x55555555U;
990
991 /* calculate coordinates in odd and even bit positions */
992 t = (s & swap) ^ comp;
993 s = s ^ sr ^ t ^ (t << 1);
994
995 /* unpad/clear out any junk on the left */
996 s = s & ((1 << n << n) - 1);
997
998 /* Now "unshuffle" to separate the x and y bits. */
999 t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
1000 t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
1001 t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
1002 t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
1003
1004 /* now s contains two 16-bit coordinates */
1005 return s;
1006}
1007
1008/* 64 bit, a straightforward extension to the 32 bit case */
1009static uint64_t
1010ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
1011{
1012 uint64_t comp, swap, cs, t, sr;
1013
1014 /* pad s on the left (unused) bits with 01 (no change groups) */
1015 s |= 0x5555555555555555U << n << n;
1016 /* "s shift right" */
1017 sr = (s >> 1) & 0x5555555555555555U;
1018 /* compute complement and swap info in two-bit groups */
1019 cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
1020
1021 /* parallel prefix xor op to propagate both complement
1022 * and swap info together from left to right (there is
1023 * no step "cs ^= cs >> 1", so in effect it computes
1024 * two independent parallel prefix operations on two
1025 * interleaved sets of thirty-two bits).
1026 */
1027 cs ^= cs >> 2;
1028 cs ^= cs >> 4;
1029 cs ^= cs >> 8;
1030 cs ^= cs >> 16;
1031 cs ^= cs >> 32;
1032
1033 /* separate swap and complement bits */
1034 swap = cs & 0x5555555555555555U;
1035 comp = (cs >> 1) & 0x5555555555555555U;
1036
1037 /* calculate coordinates in odd and even bit positions */
1038 t = (s & swap) ^ comp;
1039 s = s ^ sr ^ t ^ (t << 1);
1040
1041 /* unpad/clear out any junk on the left */
1042 s = s & ((1 << n << n) - 1);
1043
1044 /* Now "unshuffle" to separate the x and y bits. */
1045 t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
1046 t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
1047 t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
1048 t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
1049 t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
1050
1051 /* now s contains two 32-bit coordinates */
1052 return s;
1053}
1054
1055/* algorithm from the book Hacker's Delight, but a similar algorithm*/
1056/* is given in https://doi.org/10.1002/spe.4380160103 */
1057/* this has been slightly improved over the original version */
1058ecb_function_ uint32_t
1059ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
1060{
1061 uint32_t row;
1062 uint32_t state = 0;
1063 uint32_t s = 0;
1064
1065 do
1066 {
1067 --n;
1068
1069 row = 4 * state
1070 | (2 & (xy >> n >> 15))
1071 | (1 & (xy >> n ));
1072
1073 /* these funky constants are lookup tables for two-bit values */
1074 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1075 state = (0x8fe65831U >> 2 * row) & 3;
1076 }
1077 while (n > 0);
1078
1079 return s;
1080}
1081
1082/* 64 bit, essentially the same as 32 bit */
1083ecb_function_ uint64_t
1084ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
1085{
1086 uint32_t row;
1087 uint32_t state = 0;
1088 uint64_t s = 0;
1089
1090 do
1091 {
1092 --n;
1093
1094 row = 4 * state
1095 | (2 & (xy >> n >> 31))
1096 | (1 & (xy >> n ));
1097
1098 /* these funky constants are lookup tables for two-bit values */
1099 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1100 state = (0x8fe65831U >> 2 * row) & 3;
1101 }
1102 while (n > 0);
1103
1104 return s;
1105}
1106
1107/*****************************************************************************/
1108/* division */
529 1109
530#if ECB_GCC_VERSION(3,0) || ECB_C99 1110#if ECB_GCC_VERSION(3,0) || ECB_C99
1111 /* C99 tightened the definition of %, so we can use a more efficient version */
531 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 1112 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
532#else 1113#else
533 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 1114 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
534#endif 1115#endif
535 1116
536#if __cplusplus 1117#if ECB_CPP
537 template<typename T> 1118 template<typename T>
538 static inline T ecb_div_rd (T val, T div) 1119 static inline T ecb_div_rd (T val, T div)
539 { 1120 {
540 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; 1121 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
541 } 1122 }
… …
547#else 1128#else
548 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) 1129 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
549 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) 1130 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
550#endif 1131#endif
551 1132
1133/*****************************************************************************/
1134/* array length */
1135
552#if ecb_cplusplus_does_not_suck 1136#if ecb_cplusplus_does_not_suck
553 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 1137 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
554 template<typename T, int N> 1138 template<typename T, int N>
555 static inline int ecb_array_length (const T (&arr)[N]) 1139 static inline int ecb_array_length (const T (&arr)[N])
556 { 1140 {
… …
558 } 1142 }
559#else 1143#else
560 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 1144 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
561#endif 1145#endif
562 1146
1147/*****************************************************************************/
1148/* IEEE 754-2008 half float conversions */
1149
1150ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
1151ecb_function_ ecb_const uint32_t
1152ecb_binary16_to_binary32 (uint32_t x)
1153{
1154 unsigned int s = (x & 0x8000) << (31 - 15);
1155 int e = (x >> 10) & 0x001f;
1156 unsigned int m = x & 0x03ff;
1157
1158 if (ecb_expect_false (e == 31))
1159 /* infinity or NaN */
1160 e = 255 - (127 - 15);
1161 else if (ecb_expect_false (!e))
1162 {
1163 if (ecb_expect_true (!m))
1164 /* zero, handled by code below by forcing e to 0 */
1165 e = 0 - (127 - 15);
1166 else
1167 {
1168 /* subnormal, renormalise */
1169 unsigned int s = 10 - ecb_ld32 (m);
1170
1171 m = (m << s) & 0x3ff; /* mask implicit bit */
1172 e -= s - 1;
1173 }
1174 }
1175
1176 /* e and m now are normalised, or zero, (or inf or nan) */
1177 e += 127 - 15;
1178
1179 return s | (e << 23) | (m << (23 - 10));
1180}
1181
1182ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
1183ecb_function_ ecb_const uint16_t
1184ecb_binary32_to_binary16 (uint32_t x)
1185{
1186 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
1187 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
1188 unsigned int m = x & 0x007fffff;
1189
1190 x &= 0x7fffffff;
1191
1192 /* if it's within range of binary16 normals, use fast path */
1193 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
1194 {
1195 /* mantissa round-to-even */
1196 m += 0x00000fff + ((m >> (23 - 10)) & 1);
1197
1198 /* handle overflow */
1199 if (ecb_expect_false (m >= 0x00800000))
1200 {
1201 m >>= 1;
1202 e += 1;
1203 }
1204
1205 return s | (e << 10) | (m >> (23 - 10));
1206 }
1207
1208 /* handle large numbers and infinity */
1209 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
1210 return s | 0x7c00;
1211
1212 /* handle zero, subnormals and small numbers */
1213 if (ecb_expect_true (x < 0x38800000))
1214 {
1215 /* zero */
1216 if (ecb_expect_true (!x))
1217 return s;
1218
1219 /* handle subnormals */
1220
1221 /* too small, will be zero */
1222 if (e < (14 - 24)) /* might not be sharp, but is good enough */
1223 return s;
1224
1225 m |= 0x00800000; /* make implicit bit explicit */
1226
1227 /* very tricky - we need to round to the nearest e (+10) bit value */
1228 {
1229 unsigned int bits = 14 - e;
1230 unsigned int half = (1 << (bits - 1)) - 1;
1231 unsigned int even = (m >> bits) & 1;
1232
1233 /* if this overflows, we will end up with a normalised number */
1234 m = (m + half + even) >> bits;
1235 }
1236
1237 return s | m;
1238 }
1239
1240 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
1241 m >>= 13;
1242
1243 return s | 0x7c00 | m | !m;
1244}
1245
1246/*******************************************************************************/
1247/* fast integer to ascii */
1248
1249/*
1250 * This code is pretty complicated because it is general. The idea behind it,
1251 * however, is pretty simple: first, the number is multiplied with a scaling
1252 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
1253 * number with the first digit in the upper bits.
1254 * Then this digit is converted to text and masked out. The resulting number
1255 * is then multiplied by 10, by multiplying the fixed point representation
1256 * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
1257 * format becomes 5.27, 6.26 and so on.
1258 * The rest involves only advancing the pointer if we already generated a
1259 * non-zero digit, so leading zeroes are overwritten.
1260 */
1261
1262/* simply return a mask with "bits" bits set */
1263#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
1264
1265/* oputput a single digit. maskvalue is 10**digitidx */
1266#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
1267 if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
1268 { \
1269 char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
1270 *ptr = digit + '0'; /* output it */ \
1271 nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
1272 ptr += nz; /* output digit only if non-zero digit seen */ \
1273 x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
1274 }
1275
1276/* convert integer to fixed point format and multiply out digits, highest first */
1277/* requires magic constants: max. digits and number of bits after the decimal point */
1278#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
1279ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
1280{ \
1281 char nz = lz; /* non-zero digit seen? */ \
1282 /* convert to x.bits fixed-point */ \
1283 type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
1284 /* output up to 10 digits */ \
1285 ecb_i2a_digit (type,bits,digitmask, 1, 0); \
1286 ecb_i2a_digit (type,bits,digitmask, 10, 1); \
1287 ecb_i2a_digit (type,bits,digitmask, 100, 2); \
1288 ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
1289 ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
1290 ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
1291 ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
1292 ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
1293 ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
1294 ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
1295 return ptr; \
1296}
1297
1298/* predefined versions of the above, for various digits */
1299/* ecb_i2a_xN = almost N digits, limit defined by macro */
1300/* ecb_i2a_N = up to N digits, leading zeroes suppressed */
1301/* ecb_i2a_0N = exactly N digits, including leading zeroes */
1302
1303/* non-leading-zero versions, limited range */
1304#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
1305#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
1306ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
1307ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
1308
1309/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
1310ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1311ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1312ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1313ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1314ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1315ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1316ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1317ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1318
1319/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
1320ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1321ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1322ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1323ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1324ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1325ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1326ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1327ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1328
1329#define ECB_I2A_I32_DIGITS 11
1330#define ECB_I2A_U32_DIGITS 10
1331#define ECB_I2A_I64_DIGITS 20
1332#define ECB_I2A_U64_DIGITS 21
1333#define ECB_I2A_MAX_DIGITS 21
1334
1335ecb_inline char *
1336ecb_i2a_u32 (char *ptr, uint32_t u)
1337{
1338 #if ECB_64BIT_NATIVE
1339 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1340 ptr = ecb_i2a_x10 (ptr, u);
1341 else /* x10 almost, but not fully, covers 32 bit */
1342 {
1343 uint32_t u1 = u % 1000000000;
1344 uint32_t u2 = u / 1000000000;
1345
1346 *ptr++ = u2 + '0';
1347 ptr = ecb_i2a_09 (ptr, u1);
1348 }
1349 #else
1350 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1351 ecb_i2a_x5 (ptr, u);
1352 else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
1353 {
1354 uint32_t u1 = u % 10000;
1355 uint32_t u2 = u / 10000;
1356
1357 ptr = ecb_i2a_x5 (ptr, u2);
1358 ptr = ecb_i2a_04 (ptr, u1);
1359 }
1360 else
1361 {
1362 uint32_t u1 = u % 10000;
1363 uint32_t ua = u / 10000;
1364 uint32_t u2 = ua % 10000;
1365 uint32_t u3 = ua / 10000;
1366
1367 ptr = ecb_i2a_2 (ptr, u3);
1368 ptr = ecb_i2a_04 (ptr, u2);
1369 ptr = ecb_i2a_04 (ptr, u1);
1370 }
1371 #endif
1372
1373 return ptr;
1374}
1375
1376ecb_inline char *
1377ecb_i2a_i32 (char *ptr, int32_t v)
1378{
1379 *ptr = '-'; ptr += v < 0;
1380 uint32_t u = v < 0 ? -(uint32_t)v : v;
1381
1382 #if ECB_64BIT_NATIVE
1383 ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
1384 #else
1385 ptr = ecb_i2a_u32 (ptr, u);
1386 #endif
1387
1388 return ptr;
1389}
1390
1391ecb_inline char *
1392ecb_i2a_u64 (char *ptr, uint64_t u)
1393{
1394 #if ECB_64BIT_NATIVE
1395 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1396 ptr = ecb_i2a_x10 (ptr, u);
1397 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1398 {
1399 uint64_t u1 = u % 1000000000;
1400 uint64_t u2 = u / 1000000000;
1401
1402 ptr = ecb_i2a_x10 (ptr, u2);
1403 ptr = ecb_i2a_09 (ptr, u1);
1404 }
1405 else
1406 {
1407 uint64_t u1 = u % 1000000000;
1408 uint64_t ua = u / 1000000000;
1409 uint64_t u2 = ua % 1000000000;
1410 uint64_t u3 = ua / 1000000000;
1411
1412 ptr = ecb_i2a_2 (ptr, u3);
1413 ptr = ecb_i2a_09 (ptr, u2);
1414 ptr = ecb_i2a_09 (ptr, u1);
1415 }
1416 #else
1417 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1418 ptr = ecb_i2a_x5 (ptr, u);
1419 else
1420 {
1421 uint64_t u1 = u % 10000;
1422 uint64_t u2 = u / 10000;
1423
1424 ptr = ecb_i2a_u64 (ptr, u2);
1425 ptr = ecb_i2a_04 (ptr, u1);
1426 }
1427 #endif
1428
1429 return ptr;
1430}
1431
1432ecb_inline char *
1433ecb_i2a_i64 (char *ptr, int64_t v)
1434{
1435 *ptr = '-'; ptr += v < 0;
1436 uint64_t u = v < 0 ? -(uint64_t)v : v;
1437
1438 #if ECB_64BIT_NATIVE
1439 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1440 ptr = ecb_i2a_x10 (ptr, u);
1441 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1442 {
1443 uint64_t u1 = u % 1000000000;
1444 uint64_t u2 = u / 1000000000;
1445
1446 ptr = ecb_i2a_x10 (ptr, u2);
1447 ptr = ecb_i2a_09 (ptr, u1);
1448 }
1449 else
1450 {
1451 uint64_t u1 = u % 1000000000;
1452 uint64_t ua = u / 1000000000;
1453 uint64_t u2 = ua % 1000000000;
1454 uint64_t u3 = ua / 1000000000;
1455
1456 /* 2**31 is 19 digits, so the top is exactly one digit */
1457 *ptr++ = u3 + '0';
1458 ptr = ecb_i2a_09 (ptr, u2);
1459 ptr = ecb_i2a_09 (ptr, u1);
1460 }
1461 #else
1462 ptr = ecb_i2a_u64 (ptr, u);
1463 #endif
1464
1465 return ptr;
1466}
1467
563/*******************************************************************************/ 1468/*******************************************************************************/
564/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ 1469/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
565 1470
566/* basically, everything uses "ieee pure-endian" floating point numbers */ 1471/* basically, everything uses "ieee pure-endian" floating point numbers */
567/* the only noteworthy exception is ancient armle, which uses order 43218765 */ 1472/* the only noteworthy exception is ancient armle, which uses order 43218765 */
568#if 0 \ 1473#if 0 \
569 || __i386 || __i386__ \ 1474 || __i386 || __i386__ \
570 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ 1475 || ECB_GCC_AMD64 \
571 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ 1476 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
572 || defined __arm__ && defined __ARM_EABI__ \
573 || defined __s390__ || defined __s390x__ \ 1477 || defined __s390__ || defined __s390x__ \
574 || defined __mips__ \ 1478 || defined __mips__ \
575 || defined __alpha__ \ 1479 || defined __alpha__ \
576 || defined __hppa__ \ 1480 || defined __hppa__ \
577 || defined __ia64__ \ 1481 || defined __ia64__ \
578 || defined __m68k__ \ 1482 || defined __m68k__ \
579 || defined __m88k__ \ 1483 || defined __m88k__ \
580 || defined __sh__ \ 1484 || defined __sh__ \
581 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 1485 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
1486 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
1487 || defined __aarch64__
582 #define ECB_STDFP 1 1488 #define ECB_STDFP 1
583 #include <string.h> /* for memcpy */
584#else 1489#else
585 #define ECB_STDFP 0 1490 #define ECB_STDFP 0
586#endif 1491#endif
587 1492
588#ifndef ECB_NO_LIBM 1493#ifndef ECB_NO_LIBM
589 1494
590 #include <math.h> /* for frexp*, ldexp* */ 1495 #include <math.h> /* for frexp*, ldexp*, INFINITY, NAN */
591 1496
592 /* converts an ieee half/binary16 to a float */ 1497 /* only the oldest of old doesn't have this one. solaris. */
593 ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; 1498 #ifdef INFINITY
594 ecb_function_ float 1499 #define ECB_INFINITY INFINITY
595 ecb_binary16_to_float (uint16_t x) 1500 #else
596 { 1501 #define ECB_INFINITY HUGE_VAL
597 int e = (x >> 10) & 0x1f; 1502 #endif
598 int m = x & 0x3ff;
599 float r;
600 1503
601 if (!e)
602 r = ldexpf (m , -24);
603 else if (e != 31)
604 r = ldexpf (m + 0x400, e - 25);
605 else if (m)
606 #ifdef NAN 1504 #ifdef NAN
607 r = NAN; 1505 #define ECB_NAN NAN
608 #endif
609 else 1506 #else
610 r = INFINITY; 1507 #define ECB_NAN ECB_INFINITY
1508 #endif
611 1509
612 return x & 0x8000 ? -r : r; 1510 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
613 } 1511 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
1512 #define ecb_frexpf(x,e) frexpf ((x), (e))
1513 #else
1514 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
1515 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
1516 #endif
614 1517
615 /* convert a float to ieee single/binary32 */ 1518 /* convert a float to ieee single/binary32 */
616 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; 1519 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
617 ecb_function_ uint32_t 1520 ecb_function_ ecb_const uint32_t
618 ecb_float_to_binary32 (float x) 1521 ecb_float_to_binary32 (float x)
619 { 1522 {
620 uint32_t r; 1523 uint32_t r;
621 1524
622 #if ECB_STDFP 1525 #if ECB_STDFP
… …
629 if (x == 0e0f ) return 0x00000000U; 1532 if (x == 0e0f ) return 0x00000000U;
630 if (x > +3.40282346638528860e+38f) return 0x7f800000U; 1533 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
631 if (x < -3.40282346638528860e+38f) return 0xff800000U; 1534 if (x < -3.40282346638528860e+38f) return 0xff800000U;
632 if (x != x ) return 0x7fbfffffU; 1535 if (x != x ) return 0x7fbfffffU;
633 1536
634 m = frexpf (x, &e) * 0x1000000U; 1537 m = ecb_frexpf (x, &e) * 0x1000000U;
635 1538
636 r = m & 0x80000000U; 1539 r = m & 0x80000000U;
637 1540
638 if (r) 1541 if (r)
639 m = -m; 1542 m = -m;
… …
651 1554
652 return r; 1555 return r;
653 } 1556 }
654 1557
655 /* converts an ieee single/binary32 to a float */ 1558 /* converts an ieee single/binary32 to a float */
656 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; 1559 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
657 ecb_function_ float 1560 ecb_function_ ecb_const float
658 ecb_binary32_to_float (uint32_t x) 1561 ecb_binary32_to_float (uint32_t x)
659 { 1562 {
660 float r; 1563 float r;
661 1564
662 #if ECB_STDFP 1565 #if ECB_STDFP
… …
672 x |= 0x800000U; 1575 x |= 0x800000U;
673 else 1576 else
674 e = 1; 1577 e = 1;
675 1578
676 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ 1579 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
677 r = ldexpf (x * (0.5f / 0x800000U), e - 126); 1580 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
678 1581
679 r = neg ? -r : r; 1582 r = neg ? -r : r;
680 #endif 1583 #endif
681 1584
682 return r; 1585 return r;
683 } 1586 }
684 1587
685 /* convert a double to ieee double/binary64 */ 1588 /* convert a double to ieee double/binary64 */
686 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; 1589 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
687 ecb_function_ uint64_t 1590 ecb_function_ ecb_const uint64_t
688 ecb_double_to_binary64 (double x) 1591 ecb_double_to_binary64 (double x)
689 { 1592 {
690 uint64_t r; 1593 uint64_t r;
691 1594
692 #if ECB_STDFP 1595 #if ECB_STDFP
… …
721 1624
722 return r; 1625 return r;
723 } 1626 }
724 1627
725 /* converts an ieee double/binary64 to a double */ 1628 /* converts an ieee double/binary64 to a double */
726 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; 1629 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
727 ecb_function_ double 1630 ecb_function_ ecb_const double
728 ecb_binary64_to_double (uint64_t x) 1631 ecb_binary64_to_double (uint64_t x)
729 { 1632 {
730 double r; 1633 double r;
731 1634
732 #if ECB_STDFP 1635 #if ECB_STDFP
… …
750 #endif 1653 #endif
751 1654
752 return r; 1655 return r;
753 } 1656 }
754 1657
755#endif 1658 /* convert a float to ieee half/binary16 */
1659 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1660 ecb_function_ ecb_const uint16_t
1661 ecb_float_to_binary16 (float x)
1662 {
1663 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1664 }
756 1665
757#endif 1666 /* convert an ieee half/binary16 to float */
1667 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1668 ecb_function_ ecb_const float
1669 ecb_binary16_to_float (uint16_t x)
1670 {
1671 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1672 }
758 1673
1674#endif
1675
1676#endif
1677

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines