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

Comparing libecb/ecb.h (file contents):
Revision 1.129 by root, Thu Mar 20 22:04:29 2014 UTC vs.
Revision 1.209 by root, Fri Mar 25 15:23:14 2022 UTC

1/* 1/*
2 * libecb - http://software.schmorp.de/pkg/libecb 2 * libecb - http://software.schmorp.de/pkg/libecb
3 * 3 *
4 * Copyright (©) 2009-2014 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 __amd64 || __x86_64 || _M_AMD64 || _M_X64 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
127
128#define ECB_CLANG_VERSION(major,minor) (__clang_major__ > (major) || (__clang_major__ == (major) && __clang_minor__ >= (minor)))
129
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
90#endif 140#endif
91 141
92#define ECB_CPP (__cplusplus+0) 142#define ECB_CPP (__cplusplus+0)
93#define ECB_CPP11 (__cplusplus >= 201103L) 143#define ECB_CPP11 (__cplusplus >= 201103L)
144#define ECB_CPP14 (__cplusplus >= 201402L)
145#define ECB_CPP17 (__cplusplus >= 201703L)
94 146
95#if ECB_CPP 147#if ECB_CPP
96 #define ECB_C 0 148 #define ECB_C 0
97 #define ECB_STDC_VERSION 0 149 #define ECB_STDC_VERSION 0
98#else 150#else
… …
100 #define ECB_STDC_VERSION __STDC_VERSION__ 152 #define ECB_STDC_VERSION __STDC_VERSION__
101#endif 153#endif
102 154
103#define ECB_C99 (ECB_STDC_VERSION >= 199901L) 155#define ECB_C99 (ECB_STDC_VERSION >= 199901L)
104#define ECB_C11 (ECB_STDC_VERSION >= 201112L) 156#define ECB_C11 (ECB_STDC_VERSION >= 201112L)
157#define ECB_C17 (ECB_STDC_VERSION >= 201710L)
105 158
106#if ECB_CPP 159#if ECB_CPP
107 #define ECB_EXTERN_C extern "C" 160 #define ECB_EXTERN_C extern "C"
108 #define ECB_EXTERN_C_BEG ECB_EXTERN_C { 161 #define ECB_EXTERN_C_BEG ECB_EXTERN_C {
109 #define ECB_EXTERN_C_END } 162 #define ECB_EXTERN_C_END }
… …
124 177
125#if ECB_NO_SMP 178#if ECB_NO_SMP
126 #define ECB_MEMORY_FENCE do { } while (0) 179 #define ECB_MEMORY_FENCE do { } while (0)
127#endif 180#endif
128 181
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
129#ifndef ECB_MEMORY_FENCE 191#ifndef ECB_MEMORY_FENCE
130 #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")
131 #if __i386 || __i386__ 194 #if __i386 || __i386__
132 #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")
133 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 196 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
134 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 197 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
135 #elif __amd64 || __amd64__ || __x86_64 || __x86_64__ 198 #elif ECB_GCC_AMD64
136 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 199 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
137 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 200 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
138 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 201 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
139 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 202 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
140 #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 */
141 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ 211 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
142 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 212 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
213 || defined __ARM_ARCH_6T2__
143 #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")
144 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 215 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
145 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 216 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
146 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 217 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
147 #elif __aarch64__ 218 #elif __aarch64__
148 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") 219 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
149 #elif (__sparc || __sparc__) && !__sparcv8 220 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
150 #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")
151 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory") 222 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("membar #LoadStore | #LoadLoad" : : : "memory")
152 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore") 223 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("membar #LoadStore | #StoreStore")
153 #elif defined __s390__ || defined __s390x__ 224 #elif defined __s390__ || defined __s390x__
154 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory") 225 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("bcr 15,0" : : : "memory")
… …
177 #if ECB_GCC_VERSION(4,7) 248 #if ECB_GCC_VERSION(4,7)
178 /* see comment below (stdatomic.h) about the C11 memory model. */ 249 /* see comment below (stdatomic.h) about the C11 memory model. */
179 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 250 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
180 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) 251 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
181 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) 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)
182 255
183 /* The __has_feature syntax from clang is so misdesigned that we cannot use it 256 #elif ECB_CLANG_EXTENSION(c_atomic)
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 * around this shit time and again.
187 * #elif defined __clang && __has_feature (cxx_atomic)
188 * // see comment below (stdatomic.h) about the C11 memory model. 257 /* see comment below (stdatomic.h) about the C11 memory model. */
189 * #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 258 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
190 * #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) 259 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
191 * #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) 260 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
192 */ 261 #undef ECB_MEMORY_FENCE_RELAXED
262 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
193 263
194 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 264 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
195 #define ECB_MEMORY_FENCE __sync_synchronize () 265 #define ECB_MEMORY_FENCE __sync_synchronize ()
196 #elif _MSC_VER >= 1500 /* VC++ 2008 */ 266 #elif _MSC_VER >= 1500 /* VC++ 2008 */
197 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ 267 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
… …
207 #elif defined _WIN32 277 #elif defined _WIN32
208 #include <WinNT.h> 278 #include <WinNT.h>
209 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 279 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
210 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 280 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
211 #include <mbarrier.h> 281 #include <mbarrier.h>
212 #define ECB_MEMORY_FENCE __machine_rw_barrier () 282 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
213 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () 283 #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
214 #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 ()
215 #elif __xlC__ 286 #elif __xlC__
216 #define ECB_MEMORY_FENCE __sync () 287 #define ECB_MEMORY_FENCE __sync ()
217 #endif 288 #endif
218#endif 289#endif
219 290
220#ifndef ECB_MEMORY_FENCE 291#ifndef ECB_MEMORY_FENCE
221 #if ECB_C11 && !defined __STDC_NO_ATOMICS__ 292 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
222 /* 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, */
223 /* not just C11 atomics and atomic accesses */ 294 /* not just C11 atomics and atomic accesses */
224 #include <stdatomic.h> 295 #include <stdatomic.h>
225 /* 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 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
232 /* for all __atomic_thread_fence's except seq_cst */
233 #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)
234 #endif 299 #endif
235#endif 300#endif
236 301
237#ifndef ECB_MEMORY_FENCE 302#ifndef ECB_MEMORY_FENCE
238 #if !ECB_AVOID_PTHREADS 303 #if !ECB_AVOID_PTHREADS
… …
258 323
259#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE 324#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
260 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 325 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
261#endif 326#endif
262 327
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
263/*****************************************************************************/ 332/*****************************************************************************/
264 333
265#if __cplusplus 334#if ECB_CPP
266 #define ecb_inline static inline 335 #define ecb_inline static inline
267#elif ECB_GCC_VERSION(2,5) 336#elif ECB_GCC_VERSION(2,5)
268 #define ecb_inline static __inline__ 337 #define ecb_inline static __inline__
269#elif ECB_C99 338#elif ECB_C99
270 #define ecb_inline static inline 339 #define ecb_inline static inline
… …
284 353
285#define ECB_CONCAT_(a, b) a ## b 354#define ECB_CONCAT_(a, b) a ## b
286#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) 355#define ECB_CONCAT(a, b) ECB_CONCAT_(a, b)
287#define ECB_STRINGIFY_(a) # a 356#define ECB_STRINGIFY_(a) # a
288#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))
289 359
290#define ecb_function_ ecb_inline 360#define ecb_function_ ecb_inline
291 361
292#if ECB_GCC_VERSION(3,1) 362#if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8)
293 #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)
294 #define ecb_is_constant(expr) __builtin_constant_p (expr) 369 #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 370#else
298 #define ecb_attribute(attrlist)
299
300 /* possible C11 impl for integral types 371 /* possible C11 impl for integral types
301 typedef struct ecb_is_constant_struct ecb_is_constant_struct; 372 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)) */ 373 #define ecb_is_constant(expr) _Generic ((1 ? (struct ecb_is_constant_struct *)0 : (void *)((expr) - (expr)), ecb_is_constant_struct *: 0, default: 1)) */
303 374
304 #define ecb_is_constant(expr) 0 375 #define ecb_is_constant(expr) 0
376#endif
377
378#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_expect)
379 #define ecb_expect(expr,value) __builtin_expect ((expr),(value))
380#else
305 #define ecb_expect(expr,value) (expr) 381 #define ecb_expect(expr,value) (expr)
382#endif
383
384#if ECB_GCC_VERSION(3,1) || ECB_CLANG_BUILTIN(__builtin_prefetch)
385 #define ecb_prefetch(addr,rw,locality) __builtin_prefetch (addr, rw, locality)
386#else
306 #define ecb_prefetch(addr,rw,locality) 387 #define ecb_prefetch(addr,rw,locality)
307#endif 388#endif
308 389
309/* no emulation for ecb_decltype */ 390/* no emulation for ecb_decltype */
310#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; };
311 #define ecb_decltype(x) __decltype(x) 394 #define ecb_decltype(x) ecb_decltype_t<decltype (x)>::type
312#elif ECB_GCC_VERSION(3,0) 395#elif ECB_GCC_VERSION(3,0) || ECB_CLANG_VERSION(2,8)
313 #define ecb_decltype(x) __typeof(x) 396 #define ecb_decltype(x) __typeof__ (x)
314#endif 397#endif
315 398
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
316#define ecb_noinline ecb_attribute ((__noinline__)) 416 #define ecb_noinline ecb_attribute ((__noinline__))
417#endif
418
317#define ecb_unused ecb_attribute ((__unused__)) 419#define ecb_unused ecb_attribute ((__unused__))
318#define ecb_const ecb_attribute ((__const__)) 420#define ecb_const ecb_attribute ((__const__))
319#define ecb_pure ecb_attribute ((__pure__)) 421#define ecb_pure ecb_attribute ((__pure__))
320 422
321#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 */
322 #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)
323#else 431#else
324 #define ecb_noreturn ecb_attribute ((__noreturn__)) 432 #define ecb_noreturn ecb_attribute ((__noreturn__))
325#endif 433#endif
326 434
327#if ECB_GCC_VERSION(4,3) 435#if ECB_GCC_VERSION(4,3)
… …
342/* for compatibility to the rest of the world */ 450/* for compatibility to the rest of the world */
343#define ecb_likely(expr) ecb_expect_true (expr) 451#define ecb_likely(expr) ecb_expect_true (expr)
344#define ecb_unlikely(expr) ecb_expect_false (expr) 452#define ecb_unlikely(expr) ecb_expect_false (expr)
345 453
346/* count trailing zero bits and count # of one bits */ 454/* count trailing zero bits and count # of one bits */
347#if ECB_GCC_VERSION(3,4) 455#if ECB_GCC_VERSION(3,4) \
348 /* 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) \
349 #define ecb_ld32(x) (__builtin_clz (x) ^ 31) 457 && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \
350 #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) 458 && ECB_CLANG_BUILTIN(__builtin_popcount))
351 #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))
352 #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)
353 #define ecb_popcount32(x) __builtin_popcount (x) 465 #define ecb_popcount32(x) __builtin_popcount (x)
354 /* no popcountll */ 466 /* ecb_popcount64 is more difficult, see below */
355#else 467#else
356 ecb_function_ int ecb_ctz32 (uint32_t x) ecb_const; 468 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
357 ecb_function_ int 469 ecb_function_ ecb_const int
358 ecb_ctz32 (uint32_t x) 470 ecb_ctz32 (uint32_t x)
359 { 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
360 int r = 0; 477 int r;
361 478
362 x &= ~x + 1; /* this isolates the lowest bit */ 479 x &= ~x + 1; /* this isolates the lowest bit */
363 480
364#if ECB_branchless_on_i386 481 #if 1
482 /* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */
483 /* This happens to return 32 for x == 0, but the API does not support this */
484
485 /* -0 marks unused entries */
486 static unsigned char table[64] =
487 {
488 32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14,
489 10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15,
490 31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26,
491 30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0
492 };
493
494 /* magic constant results in 33 unique values in the upper 6 bits */
495 x *= 0x0450fbafU; /* == 17 * 65 * 65535 */
496
497 r = table [x >> 26];
498 #elif 0 /* branchless on i386, typically */
499 r = 0;
365 r += !!(x & 0xaaaaaaaa) << 0; 500 r += !!(x & 0xaaaaaaaa) << 0;
366 r += !!(x & 0xcccccccc) << 1; 501 r += !!(x & 0xcccccccc) << 1;
367 r += !!(x & 0xf0f0f0f0) << 2; 502 r += !!(x & 0xf0f0f0f0) << 2;
368 r += !!(x & 0xff00ff00) << 3; 503 r += !!(x & 0xff00ff00) << 3;
369 r += !!(x & 0xffff0000) << 4; 504 r += !!(x & 0xffff0000) << 4;
370#else 505 #else /* branchless on modern compilers, typically */
506 r = 0;
371 if (x & 0xaaaaaaaa) r += 1; 507 if (x & 0xaaaaaaaa) r += 1;
372 if (x & 0xcccccccc) r += 2; 508 if (x & 0xcccccccc) r += 2;
373 if (x & 0xf0f0f0f0) r += 4; 509 if (x & 0xf0f0f0f0) r += 4;
374 if (x & 0xff00ff00) r += 8; 510 if (x & 0xff00ff00) r += 8;
375 if (x & 0xffff0000) r += 16; 511 if (x & 0xffff0000) r += 16;
376#endif 512#endif
377 513
378 return r; 514 return r;
515#endif
379 } 516 }
380 517
381 ecb_function_ int ecb_ctz64 (uint64_t x) ecb_const; 518 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
382 ecb_function_ int 519 ecb_function_ ecb_const int
383 ecb_ctz64 (uint64_t x) 520 ecb_ctz64 (uint64_t x)
384 { 521 {
522#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
523 unsigned long r;
524 _BitScanForward64 (&r, x);
525 return (int)r;
526#else
385 int shift = x & 0xffffffffU ? 0 : 32; 527 int shift = x & 0xffffffff ? 0 : 32;
386 return ecb_ctz32 (x >> shift) + shift; 528 return ecb_ctz32 (x >> shift) + shift;
529#endif
387 } 530 }
388 531
532 ecb_function_ ecb_const int ecb_clz32 (uint32_t x);
533 ecb_function_ ecb_const int
534 ecb_clz32 (uint32_t x)
535 {
536#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
537 unsigned long r;
538 _BitScanReverse (&r, x);
539 return (int)r;
540#else
541
542 /* Robert Harley's algorithm from comp.arch 1996-12-07 */
543 /* This happens to return 32 for x == 0, but the API does not support this */
544
545 /* -0 marks unused table elements */
546 static unsigned char table[64] =
547 {
548 32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0,
549 -0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0,
550 17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18,
551 5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0
552 };
553
554 /* propagate leftmost 1 bit to the right */
555 x |= x >> 1;
556 x |= x >> 2;
557 x |= x >> 4;
558 x |= x >> 8;
559 x |= x >> 16;
560
561 /* magic constant results in 33 unique values in the upper 6 bits */
562 x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */
563
564 return table [x >> 26];
565#endif
566 }
567
568 ecb_function_ ecb_const int ecb_clz64 (uint64_t x);
569 ecb_function_ ecb_const int
570 ecb_clz64 (uint64_t x)
571 {
572#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
573 unsigned long r;
574 _BitScanReverse64 (&r, x);
575 return (int)r;
576#else
577 uint32_t l = x >> 32;
578 int shift = l ? 0 : 32;
579 return ecb_clz32 (l ? l : x) + shift;
580#endif
581 }
582
389 ecb_function_ int ecb_popcount32 (uint32_t x) ecb_const; 583 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
390 ecb_function_ int 584 ecb_function_ ecb_const int
391 ecb_popcount32 (uint32_t x) 585 ecb_popcount32 (uint32_t x)
392 { 586 {
393 x -= (x >> 1) & 0x55555555; 587 x -= (x >> 1) & 0x55555555;
394 x = ((x >> 2) & 0x33333333) + (x & 0x33333333); 588 x = ((x >> 2) & 0x33333333) + (x & 0x33333333);
395 x = ((x >> 4) + x) & 0x0f0f0f0f; 589 x = ((x >> 4) + x) & 0x0f0f0f0f;
396 x *= 0x01010101; 590 x *= 0x01010101;
397 591
398 return x >> 24; 592 return x >> 24;
399 } 593 }
400 594
401 ecb_function_ int ecb_ld32 (uint32_t x) ecb_const; 595 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
402 ecb_function_ int ecb_ld32 (uint32_t x) 596 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
403 { 597 {
598#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
599 unsigned long r;
600 _BitScanReverse (&r, x);
601 return (int)r;
602#else
404 int r = 0; 603 int r = 0;
405 604
406 if (x >> 16) { x >>= 16; r += 16; } 605 if (x >> 16) { x >>= 16; r += 16; }
407 if (x >> 8) { x >>= 8; r += 8; } 606 if (x >> 8) { x >>= 8; r += 8; }
408 if (x >> 4) { x >>= 4; r += 4; } 607 if (x >> 4) { x >>= 4; r += 4; }
409 if (x >> 2) { x >>= 2; r += 2; } 608 if (x >> 2) { x >>= 2; r += 2; }
410 if (x >> 1) { r += 1; } 609 if (x >> 1) { r += 1; }
411 610
412 return r; 611 return r;
612#endif
413 } 613 }
414 614
415 ecb_function_ int ecb_ld64 (uint64_t x) ecb_const; 615 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
416 ecb_function_ int ecb_ld64 (uint64_t x) 616 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
417 { 617 {
618#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
619 unsigned long r;
620 _BitScanReverse64 (&r, x);
621 return (int)r;
622#else
418 int r = 0; 623 int r = 0;
419 624
420 if (x >> 32) { x >>= 32; r += 32; } 625 if (x >> 32) { x >>= 32; r += 32; }
421 626
422 return r + ecb_ld32 (x); 627 return r + ecb_ld32 (x);
423 }
424#endif 628#endif
629 }
630#endif
425 631
426ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) ecb_const; 632ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
427ecb_function_ ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } 633ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
428ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) ecb_const; 634ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x);
429ecb_function_ ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); } 635ecb_function_ ecb_const ecb_bool ecb_is_pot64 (uint64_t x) { return !(x & (x - 1)); }
430 636
431ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) ecb_const; 637ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x);
432ecb_function_ uint8_t ecb_bitrev8 (uint8_t x) 638ecb_function_ ecb_const uint8_t ecb_bitrev8 (uint8_t x)
433{ 639{
434 return ( (x * 0x0802U & 0x22110U) 640 return ( (x * 0x0802U & 0x22110U)
435 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16; 641 | (x * 0x8020U & 0x88440U)) * 0x10101U >> 16;
436} 642}
437 643
438ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) ecb_const; 644ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x);
439ecb_function_ uint16_t ecb_bitrev16 (uint16_t x) 645ecb_function_ ecb_const uint16_t ecb_bitrev16 (uint16_t x)
440{ 646{
441 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1); 647 x = ((x >> 1) & 0x5555) | ((x & 0x5555) << 1);
442 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2); 648 x = ((x >> 2) & 0x3333) | ((x & 0x3333) << 2);
443 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4); 649 x = ((x >> 4) & 0x0f0f) | ((x & 0x0f0f) << 4);
444 x = ( x >> 8 ) | ( x << 8); 650 x = ( x >> 8 ) | ( x << 8);
445 651
446 return x; 652 return x;
447} 653}
448 654
449ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) ecb_const; 655ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x);
450ecb_function_ uint32_t ecb_bitrev32 (uint32_t x) 656ecb_function_ ecb_const uint32_t ecb_bitrev32 (uint32_t x)
451{ 657{
452 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1); 658 x = ((x >> 1) & 0x55555555) | ((x & 0x55555555) << 1);
453 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2); 659 x = ((x >> 2) & 0x33333333) | ((x & 0x33333333) << 2);
454 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4); 660 x = ((x >> 4) & 0x0f0f0f0f) | ((x & 0x0f0f0f0f) << 4);
455 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8); 661 x = ((x >> 8) & 0x00ff00ff) | ((x & 0x00ff00ff) << 8);
456 x = ( x >> 16 ) | ( x << 16); 662 x = ( x >> 16 ) | ( x << 16);
457 663
458 return x; 664 return x;
459} 665}
460 666
461/* popcount64 is only available on 64 bit cpus as gcc builtin */
462/* so for this version we are lazy */
463ecb_function_ int ecb_popcount64 (uint64_t x) ecb_const; 667ecb_function_ ecb_const int ecb_popcount64 (uint64_t x);
464ecb_function_ int 668ecb_function_ ecb_const int
465ecb_popcount64 (uint64_t x) 669ecb_popcount64 (uint64_t x)
466{ 670{
671 /* popcount64 is only available on 64 bit cpus as gcc builtin. */
672 /* also, gcc/clang make this surprisingly difficult to use */
673#if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl))
674 return __builtin_popcountl (x);
675#else
467 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); 676 return ecb_popcount32 (x) + ecb_popcount32 (x >> 32);
677#endif
468} 678}
469 679
470ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) ecb_const; 680ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count);
471ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) ecb_const; 681ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count);
472ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) ecb_const; 682ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count);
473ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) ecb_const; 683ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count);
474ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) ecb_const; 684ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count);
475ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) ecb_const; 685ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count);
476ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) ecb_const; 686ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count);
477ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) ecb_const; 687ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count);
478 688
479ecb_inline uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } 689ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); }
480ecb_inline uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } 690ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); }
481ecb_inline uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } 691ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); }
482ecb_inline uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 692ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); }
483ecb_inline uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 693ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); }
484ecb_inline uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 694ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); }
485ecb_inline uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 695ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); }
486ecb_inline uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 696ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); }
487 697
488#if ECB_GCC_VERSION(4,3) 698#if ECB_CPP
699
700inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
701inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
702inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
703inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
704
705inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
706inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
707inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
708inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
709
710inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
711inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
712inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
713inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
714
715inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
716inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
717inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
718inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
719
720inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
721inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
722inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
723
724inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
725inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
726inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
727inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
728
729inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
730inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
731inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
732inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
733
734#endif
735
736#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
737 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
738 #define ecb_bswap16(x) __builtin_bswap16 (x)
739 #else
489 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 740 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
741 #endif
490 #define ecb_bswap32(x) __builtin_bswap32 (x) 742 #define ecb_bswap32(x) __builtin_bswap32 (x)
491 #define ecb_bswap64(x) __builtin_bswap64 (x) 743 #define ecb_bswap64(x) __builtin_bswap64 (x)
744#elif _MSC_VER
745 #include <stdlib.h>
746 #define ecb_bswap16(x) ((uint16_t)_byteswap_ushort ((uint16_t)(x)))
747 #define ecb_bswap32(x) ((uint32_t)_byteswap_ulong ((uint32_t)(x)))
748 #define ecb_bswap64(x) ((uint64_t)_byteswap_uint64 ((uint64_t)(x)))
492#else 749#else
493 ecb_function_ uint16_t ecb_bswap16 (uint16_t x) ecb_const; 750 ecb_function_ ecb_const uint16_t ecb_bswap16 (uint16_t x);
494 ecb_function_ uint16_t 751 ecb_function_ ecb_const uint16_t
495 ecb_bswap16 (uint16_t x) 752 ecb_bswap16 (uint16_t x)
496 { 753 {
497 return ecb_rotl16 (x, 8); 754 return ecb_rotl16 (x, 8);
498 } 755 }
499 756
500 ecb_function_ uint32_t ecb_bswap32 (uint32_t x) ecb_const; 757 ecb_function_ ecb_const uint32_t ecb_bswap32 (uint32_t x);
501 ecb_function_ uint32_t 758 ecb_function_ ecb_const uint32_t
502 ecb_bswap32 (uint32_t x) 759 ecb_bswap32 (uint32_t x)
503 { 760 {
504 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16); 761 return (((uint32_t)ecb_bswap16 (x)) << 16) | ecb_bswap16 (x >> 16);
505 } 762 }
506 763
507 ecb_function_ uint64_t ecb_bswap64 (uint64_t x) ecb_const; 764 ecb_function_ ecb_const uint64_t ecb_bswap64 (uint64_t x);
508 ecb_function_ uint64_t 765 ecb_function_ ecb_const uint64_t
509 ecb_bswap64 (uint64_t x) 766 ecb_bswap64 (uint64_t x)
510 { 767 {
511 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32); 768 return (((uint64_t)ecb_bswap32 (x)) << 32) | ecb_bswap32 (x >> 32);
512 } 769 }
513#endif 770#endif
514 771
515#if ECB_GCC_VERSION(4,5) 772#if ECB_GCC_VERSION(4,5) || ECB_CLANG_BUILTIN(__builtin_unreachable)
516 #define ecb_unreachable() __builtin_unreachable () 773 #define ecb_unreachable() __builtin_unreachable ()
517#else 774#else
518 /* this seems to work fine, but gcc always emits a warning for it :/ */ 775 /* this seems to work fine, but gcc always emits a warning for it :/ */
519 ecb_inline void ecb_unreachable (void) ecb_noreturn; 776 ecb_inline ecb_noreturn void ecb_unreachable (void);
520 ecb_inline void ecb_unreachable (void) { } 777 ecb_inline ecb_noreturn void ecb_unreachable (void) { }
521#endif 778#endif
522 779
523/* try to tell the compiler that some condition is definitely true */ 780/* try to tell the compiler that some condition is definitely true */
524#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 781#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
525 782
526ecb_inline unsigned char ecb_byteorder_helper (void) ecb_const; 783ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
527ecb_inline unsigned char 784ecb_inline ecb_const uint32_t
528ecb_byteorder_helper (void) 785ecb_byteorder_helper (void)
529{ 786{
530 /* the union code still generates code under pressure in gcc, */ 787 /* the union code still generates code under pressure in gcc, */
531 /* but less than using pointers, and always seems to */ 788 /* but less than using pointers, and always seems to */
532 /* successfully return a constant. */ 789 /* successfully return a constant. */
533 /* the reason why we have this horrible preprocessor mess */ 790 /* the reason why we have this horrible preprocessor mess */
534 /* is to avoid it in all cases, at least on common architectures */ 791 /* is to avoid it in all cases, at least on common architectures */
535 /* or when using a recent enough gcc version (>= 4.6) */ 792 /* or when using a recent enough gcc version (>= 4.6) */
536#if __i386 || __i386__ || _M_X86 || __amd64 || __amd64__ || _M_X64
537 return 0x44;
538#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 793#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
794 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
795 #define ECB_LITTLE_ENDIAN 1
539 return 0x44; 796 return 0x44332211;
540#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 797#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
798 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
799 #define ECB_BIG_ENDIAN 1
541 return 0x11; 800 return 0x11223344;
542#else 801#else
543 union 802 union
544 { 803 {
804 uint8_t c[4];
545 uint32_t i; 805 uint32_t u;
546 uint8_t c;
547 } u = { 0x11223344 }; 806 } u = { 0x11, 0x22, 0x33, 0x44 };
548 return u.c; 807 return u.u;
549#endif 808#endif
550} 809}
551 810
552ecb_inline ecb_bool ecb_big_endian (void) ecb_const; 811ecb_inline ecb_const ecb_bool ecb_big_endian (void);
553ecb_inline ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 812ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
554ecb_inline ecb_bool ecb_little_endian (void) ecb_const; 813ecb_inline ecb_const ecb_bool ecb_little_endian (void);
555ecb_inline ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 814ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
815
816/*****************************************************************************/
817/* unaligned load/store */
818
819ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
820ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
821ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
822
823ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
824ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
825ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
826
827ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
828ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
829ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
830
831ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
832ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
833ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
834
835ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
836ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
837ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
838
839ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
840ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
841ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
842
843ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
844ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
845ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
846
847ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
848ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
849ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
850
851ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
852ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
853ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
854
855ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
856ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
857ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
858
859#if ECB_CPP
860
861inline uint8_t ecb_bswap (uint8_t v) { return v; }
862inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
863inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
864inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
865
866template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
867template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
868template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
869template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
870template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
871template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
872template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
873template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
874
875template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
876template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
877template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
878template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
879template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
880template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
881template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
882template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
883
884#endif
885
886/*****************************************************************************/
887/* pointer/integer hashing */
888
889/* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */
890ecb_function_ uint32_t ecb_mix32 (uint32_t v);
891ecb_function_ uint32_t ecb_mix32 (uint32_t v)
892{
893 v ^= v >> 16; v *= 0x7feb352dU;
894 v ^= v >> 15; v *= 0x846ca68bU;
895 v ^= v >> 16;
896 return v;
897}
898
899ecb_function_ uint32_t ecb_unmix32 (uint32_t v);
900ecb_function_ uint32_t ecb_unmix32 (uint32_t v)
901{
902 v ^= v >> 16 ; v *= 0x43021123U;
903 v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U;
904 v ^= v >> 16 ;
905 return v;
906}
907
908/* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */
909ecb_function_ uint64_t ecb_mix64 (uint64_t v);
910ecb_function_ uint64_t ecb_mix64 (uint64_t v)
911{
912 v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U;
913 v ^= v >> 27; v *= 0x94d049bb133111ebU;
914 v ^= v >> 31;
915 return v;
916}
917
918ecb_function_ uint64_t ecb_unmix64 (uint64_t v);
919ecb_function_ uint64_t ecb_unmix64 (uint64_t v)
920{
921 v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U;
922 v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U;
923 v ^= v >> 30 ^ v >> 60;
924 return v;
925}
926
927ecb_function_ uintptr_t ecb_ptrmix (void *p);
928ecb_function_ uintptr_t ecb_ptrmix (void *p)
929{
930 #if ECB_PTRSIZE <= 4
931 return ecb_mix32 ((uint32_t)p);
932 #else
933 return ecb_mix64 ((uint64_t)p);
934 #endif
935}
936
937ecb_function_ void *ecb_ptrunmix (uintptr_t v);
938ecb_function_ void *ecb_ptrunmix (uintptr_t v)
939{
940 #if ECB_PTRSIZE <= 4
941 return (void *)ecb_unmix32 (v);
942 #else
943 return (void *)ecb_unmix64 (v);
944 #endif
945}
946
947#if ECB_CPP
948
949template<typename T>
950inline uintptr_t ecb_ptrmix (T *p)
951{
952 return ecb_ptrmix (static_cast<void *>(p));
953}
954
955template<typename T>
956inline T *ecb_ptrunmix (uintptr_t v)
957{
958 return static_cast<T *>(ecb_ptrunmix (v));
959}
960
961#endif
962
963/*****************************************************************************/
964/* gray code */
965
966ecb_function_ uint_fast8_t ecb_gray8_encode (uint_fast8_t b) { return b ^ (b >> 1); }
967ecb_function_ uint_fast16_t ecb_gray16_encode (uint_fast16_t b) { return b ^ (b >> 1); }
968ecb_function_ uint_fast32_t ecb_gray32_encode (uint_fast32_t b) { return b ^ (b >> 1); }
969ecb_function_ uint_fast64_t ecb_gray64_encode (uint_fast64_t b) { return b ^ (b >> 1); }
970
971ecb_function_ uint8_t ecb_gray8_decode (uint8_t g)
972{
973 g ^= g >> 1;
974 g ^= g >> 2;
975 g ^= g >> 4;
976
977 return g;
978}
979
980ecb_function_ uint16_t ecb_gray16_decode (uint16_t g)
981{
982 g ^= g >> 1;
983 g ^= g >> 2;
984 g ^= g >> 4;
985 g ^= g >> 8;
986
987 return g;
988}
989
990ecb_function_ uint32_t ecb_gray32_decode (uint32_t g)
991{
992 g ^= g >> 1;
993 g ^= g >> 2;
994 g ^= g >> 4;
995 g ^= g >> 8;
996 g ^= g >> 16;
997
998 return g;
999}
1000
1001ecb_function_ uint64_t ecb_gray64_decode (uint64_t g)
1002{
1003 g ^= g >> 1;
1004 g ^= g >> 2;
1005 g ^= g >> 4;
1006 g ^= g >> 8;
1007 g ^= g >> 16;
1008 g ^= g >> 32;
1009
1010 return g;
1011}
1012
1013#if ECB_CPP
1014
1015ecb_function_ uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray8_encode (b); }
1016ecb_function_ uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray16_encode (b); }
1017ecb_function_ uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray32_encode (b); }
1018ecb_function_ uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray64_encode (b); }
1019
1020ecb_function_ uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray8_decode (g); }
1021ecb_function_ uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray16_decode (g); }
1022ecb_function_ uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray32_decode (g); }
1023ecb_function_ uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray64_decode (g); }
1024
1025#endif
1026
1027/*****************************************************************************/
1028/* 2d hilbert curves */
1029
1030/* algorithm from the book Hacker's Delight, modified to not */
1031/* run into undefined behaviour for n==16 */
1032static uint32_t
1033ecb_hilbert2d_index_to_coord32 (int n, uint32_t s)
1034{
1035 uint32_t comp, swap, cs, t, sr;
1036
1037 /* pad s on the left (unused) bits with 01 (no change groups) */
1038 s |= 0x55555555U << n << n;
1039 /* "s shift right" */
1040 sr = (s >> 1) & 0x55555555U;
1041 /* compute complement and swap info in two-bit groups */
1042 cs = ((s & 0x55555555U) + sr) ^ 0x55555555U;
1043
1044 /* parallel prefix xor op to propagate both complement
1045 * and swap info together from left to right (there is
1046 * no step "cs ^= cs >> 1", so in effect it computes
1047 * two independent parallel prefix operations on two
1048 * interleaved sets of sixteen bits).
1049 */
1050 cs ^= cs >> 2;
1051 cs ^= cs >> 4;
1052 cs ^= cs >> 8;
1053 cs ^= cs >> 16;
1054
1055 /* separate swap and complement bits */
1056 swap = cs & 0x55555555U;
1057 comp = (cs >> 1) & 0x55555555U;
1058
1059 /* calculate coordinates in odd and even bit positions */
1060 t = (s & swap) ^ comp;
1061 s = s ^ sr ^ t ^ (t << 1);
1062
1063 /* unpad/clear out any junk on the left */
1064 s = s & ((1 << n << n) - 1);
1065
1066 /* Now "unshuffle" to separate the x and y bits. */
1067 t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1);
1068 t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2);
1069 t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4);
1070 t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8);
1071
1072 /* now s contains two 16-bit coordinates */
1073 return s;
1074}
1075
1076/* 64 bit, a straightforward extension to the 32 bit case */
1077static uint64_t
1078ecb_hilbert2d_index_to_coord64 (int n, uint64_t s)
1079{
1080 uint64_t comp, swap, cs, t, sr;
1081
1082 /* pad s on the left (unused) bits with 01 (no change groups) */
1083 s |= 0x5555555555555555U << n << n;
1084 /* "s shift right" */
1085 sr = (s >> 1) & 0x5555555555555555U;
1086 /* compute complement and swap info in two-bit groups */
1087 cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U;
1088
1089 /* parallel prefix xor op to propagate both complement
1090 * and swap info together from left to right (there is
1091 * no step "cs ^= cs >> 1", so in effect it computes
1092 * two independent parallel prefix operations on two
1093 * interleaved sets of thirty-two bits).
1094 */
1095 cs ^= cs >> 2;
1096 cs ^= cs >> 4;
1097 cs ^= cs >> 8;
1098 cs ^= cs >> 16;
1099 cs ^= cs >> 32;
1100
1101 /* separate swap and complement bits */
1102 swap = cs & 0x5555555555555555U;
1103 comp = (cs >> 1) & 0x5555555555555555U;
1104
1105 /* calculate coordinates in odd and even bit positions */
1106 t = (s & swap) ^ comp;
1107 s = s ^ sr ^ t ^ (t << 1);
1108
1109 /* unpad/clear out any junk on the left */
1110 s = s & ((1 << n << n) - 1);
1111
1112 /* Now "unshuffle" to separate the x and y bits. */
1113 t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1);
1114 t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2);
1115 t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4);
1116 t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8);
1117 t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16);
1118
1119 /* now s contains two 32-bit coordinates */
1120 return s;
1121}
1122
1123/* algorithm from the book Hacker's Delight, but a similar algorithm*/
1124/* is given in https://doi.org/10.1002/spe.4380160103 */
1125/* this has been slightly improved over the original version */
1126ecb_function_ uint32_t
1127ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy)
1128{
1129 uint32_t row;
1130 uint32_t state = 0;
1131 uint32_t s = 0;
1132
1133 do
1134 {
1135 --n;
1136
1137 row = 4 * state
1138 | (2 & (xy >> n >> 15))
1139 | (1 & (xy >> n ));
1140
1141 /* these funky constants are lookup tables for two-bit values */
1142 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1143 state = (0x8fe65831U >> 2 * row) & 3;
1144 }
1145 while (n > 0);
1146
1147 return s;
1148}
1149
1150/* 64 bit, essentially the same as 32 bit */
1151ecb_function_ uint64_t
1152ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy)
1153{
1154 uint32_t row;
1155 uint32_t state = 0;
1156 uint64_t s = 0;
1157
1158 do
1159 {
1160 --n;
1161
1162 row = 4 * state
1163 | (2 & (xy >> n >> 31))
1164 | (1 & (xy >> n ));
1165
1166 /* these funky constants are lookup tables for two-bit values */
1167 s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3;
1168 state = (0x8fe65831U >> 2 * row) & 3;
1169 }
1170 while (n > 0);
1171
1172 return s;
1173}
1174
1175/*****************************************************************************/
1176/* division */
556 1177
557#if ECB_GCC_VERSION(3,0) || ECB_C99 1178#if ECB_GCC_VERSION(3,0) || ECB_C99
1179 /* C99 tightened the definition of %, so we can use a more efficient version */
558 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 1180 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
559#else 1181#else
560 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 1182 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
561#endif 1183#endif
562 1184
563#if __cplusplus 1185#if ECB_CPP
564 template<typename T> 1186 template<typename T>
565 static inline T ecb_div_rd (T val, T div) 1187 static inline T ecb_div_rd (T val, T div)
566 { 1188 {
567 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div; 1189 return val < 0 ? - ((-val + div - 1) / div) : (val ) / div;
568 } 1190 }
… …
573 } 1195 }
574#else 1196#else
575 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) 1197 #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)) 1198 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
577#endif 1199#endif
1200
1201/*****************************************************************************/
1202/* array length */
578 1203
579#if ecb_cplusplus_does_not_suck 1204#if ecb_cplusplus_does_not_suck
580 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 1205 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
581 template<typename T, int N> 1206 template<typename T, int N>
582 static inline int ecb_array_length (const T (&arr)[N]) 1207 static inline int ecb_array_length (const T (&arr)[N])
… …
585 } 1210 }
586#else 1211#else
587 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 1212 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
588#endif 1213#endif
589 1214
1215/*****************************************************************************/
1216/* IEEE 754-2008 half float conversions */
1217
1218ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
1219ecb_function_ ecb_const uint32_t
1220ecb_binary16_to_binary32 (uint32_t x)
1221{
1222 unsigned int s = (x & 0x8000) << (31 - 15);
1223 int e = (x >> 10) & 0x001f;
1224 unsigned int m = x & 0x03ff;
1225
1226 if (ecb_expect_false (e == 31))
1227 /* infinity or NaN */
1228 e = 255 - (127 - 15);
1229 else if (ecb_expect_false (!e))
1230 {
1231 if (ecb_expect_true (!m))
1232 /* zero, handled by code below by forcing e to 0 */
1233 e = 0 - (127 - 15);
1234 else
1235 {
1236 /* subnormal, renormalise */
1237 unsigned int s = 10 - ecb_ld32 (m);
1238
1239 m = (m << s) & 0x3ff; /* mask implicit bit */
1240 e -= s - 1;
1241 }
1242 }
1243
1244 /* e and m now are normalised, or zero, (or inf or nan) */
1245 e += 127 - 15;
1246
1247 return s | (e << 23) | (m << (23 - 10));
1248}
1249
1250ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
1251ecb_function_ ecb_const uint16_t
1252ecb_binary32_to_binary16 (uint32_t x)
1253{
1254 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
1255 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
1256 unsigned int m = x & 0x007fffff;
1257
1258 x &= 0x7fffffff;
1259
1260 /* if it's within range of binary16 normals, use fast path */
1261 if (ecb_expect_true (0x38800000 <= x && x <= 0x477fefff))
1262 {
1263 /* mantissa round-to-even */
1264 m += 0x00000fff + ((m >> (23 - 10)) & 1);
1265
1266 /* handle overflow */
1267 if (ecb_expect_false (m >= 0x00800000))
1268 {
1269 m >>= 1;
1270 e += 1;
1271 }
1272
1273 return s | (e << 10) | (m >> (23 - 10));
1274 }
1275
1276 /* handle large numbers and infinity */
1277 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
1278 return s | 0x7c00;
1279
1280 /* handle zero, subnormals and small numbers */
1281 if (ecb_expect_true (x < 0x38800000))
1282 {
1283 /* zero */
1284 if (ecb_expect_true (!x))
1285 return s;
1286
1287 /* handle subnormals */
1288
1289 /* too small, will be zero */
1290 if (e < (14 - 24)) /* might not be sharp, but is good enough */
1291 return s;
1292
1293 m |= 0x00800000; /* make implicit bit explicit */
1294
1295 /* very tricky - we need to round to the nearest e (+10) bit value */
1296 {
1297 unsigned int bits = 14 - e;
1298 unsigned int half = (1 << (bits - 1)) - 1;
1299 unsigned int even = (m >> bits) & 1;
1300
1301 /* if this overflows, we will end up with a normalised number */
1302 m = (m + half + even) >> bits;
1303 }
1304
1305 return s | m;
1306 }
1307
1308 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
1309 m >>= 13;
1310
1311 return s | 0x7c00 | m | !m;
1312}
1313
1314/*******************************************************************************/
1315/* fast integer to ascii */
1316
1317/*
1318 * This code is pretty complicated because it is general. The idea behind it,
1319 * however, is pretty simple: first, the number is multiplied with a scaling
1320 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
1321 * number with the first digit in the upper bits.
1322 * Then this digit is converted to text and masked out. The resulting number
1323 * is then multiplied by 10, by multiplying the fixed point representation
1324 * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
1325 * format becomes 5.27, 6.26 and so on.
1326 * The rest involves only advancing the pointer if we already generated a
1327 * non-zero digit, so leading zeroes are overwritten.
1328 */
1329
1330/* simply return a mask with "bits" bits set */
1331#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
1332
1333/* oputput a single digit. maskvalue is 10**digitidx */
1334#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
1335 if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
1336 { \
1337 char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
1338 *ptr = digit + '0'; /* output it */ \
1339 nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
1340 ptr += nz; /* output digit only if non-zero digit seen */ \
1341 x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
1342 }
1343
1344/* convert integer to fixed point format and multiply out digits, highest first */
1345/* requires magic constants: max. digits and number of bits after the decimal point */
1346#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
1347ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
1348{ \
1349 char nz = lz; /* non-zero digit seen? */ \
1350 /* convert to x.bits fixed-point */ \
1351 type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
1352 /* output up to 10 digits */ \
1353 ecb_i2a_digit (type,bits,digitmask, 1, 0); \
1354 ecb_i2a_digit (type,bits,digitmask, 10, 1); \
1355 ecb_i2a_digit (type,bits,digitmask, 100, 2); \
1356 ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
1357 ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
1358 ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
1359 ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
1360 ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
1361 ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
1362 ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
1363 return ptr; \
1364}
1365
1366/* predefined versions of the above, for various digits */
1367/* ecb_i2a_xN = almost N digits, limit defined by macro */
1368/* ecb_i2a_N = up to N digits, leading zeroes suppressed */
1369/* ecb_i2a_0N = exactly N digits, including leading zeroes */
1370
1371/* non-leading-zero versions, limited range */
1372#define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */
1373#define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */
1374ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
1375ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
1376
1377/* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */
1378ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1379ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1380ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1381ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1382ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1383ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1384ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1385ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1386
1387/* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */
1388ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1389ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1390ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1391ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1392ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1393ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1394ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1395ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1396
1397#define ECB_I2A_I32_DIGITS 11
1398#define ECB_I2A_U32_DIGITS 10
1399#define ECB_I2A_I64_DIGITS 20
1400#define ECB_I2A_U64_DIGITS 21
1401#define ECB_I2A_MAX_DIGITS 21
1402
1403ecb_inline char *
1404ecb_i2a_u32 (char *ptr, uint32_t u)
1405{
1406 #if ECB_64BIT_NATIVE
1407 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1408 ptr = ecb_i2a_x10 (ptr, u);
1409 else /* x10 almost, but not fully, covers 32 bit */
1410 {
1411 uint32_t u1 = u % 1000000000;
1412 uint32_t u2 = u / 1000000000;
1413
1414 *ptr++ = u2 + '0';
1415 ptr = ecb_i2a_09 (ptr, u1);
1416 }
1417 #else
1418 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1419 ecb_i2a_x5 (ptr, u);
1420 else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
1421 {
1422 uint32_t u1 = u % 10000;
1423 uint32_t u2 = u / 10000;
1424
1425 ptr = ecb_i2a_x5 (ptr, u2);
1426 ptr = ecb_i2a_04 (ptr, u1);
1427 }
1428 else
1429 {
1430 uint32_t u1 = u % 10000;
1431 uint32_t ua = u / 10000;
1432 uint32_t u2 = ua % 10000;
1433 uint32_t u3 = ua / 10000;
1434
1435 ptr = ecb_i2a_2 (ptr, u3);
1436 ptr = ecb_i2a_04 (ptr, u2);
1437 ptr = ecb_i2a_04 (ptr, u1);
1438 }
1439 #endif
1440
1441 return ptr;
1442}
1443
1444ecb_inline char *
1445ecb_i2a_i32 (char *ptr, int32_t v)
1446{
1447 *ptr = '-'; ptr += v < 0;
1448 uint32_t u = v < 0 ? -(uint32_t)v : v;
1449
1450 #if ECB_64BIT_NATIVE
1451 ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */
1452 #else
1453 ptr = ecb_i2a_u32 (ptr, u);
1454 #endif
1455
1456 return ptr;
1457}
1458
1459ecb_inline char *
1460ecb_i2a_u64 (char *ptr, uint64_t u)
1461{
1462 #if ECB_64BIT_NATIVE
1463 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1464 ptr = ecb_i2a_x10 (ptr, u);
1465 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1466 {
1467 uint64_t u1 = u % 1000000000;
1468 uint64_t u2 = u / 1000000000;
1469
1470 ptr = ecb_i2a_x10 (ptr, u2);
1471 ptr = ecb_i2a_09 (ptr, u1);
1472 }
1473 else
1474 {
1475 uint64_t u1 = u % 1000000000;
1476 uint64_t ua = u / 1000000000;
1477 uint64_t u2 = ua % 1000000000;
1478 uint64_t u3 = ua / 1000000000;
1479
1480 ptr = ecb_i2a_2 (ptr, u3);
1481 ptr = ecb_i2a_09 (ptr, u2);
1482 ptr = ecb_i2a_09 (ptr, u1);
1483 }
1484 #else
1485 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1486 ptr = ecb_i2a_x5 (ptr, u);
1487 else
1488 {
1489 uint64_t u1 = u % 10000;
1490 uint64_t u2 = u / 10000;
1491
1492 ptr = ecb_i2a_u64 (ptr, u2);
1493 ptr = ecb_i2a_04 (ptr, u1);
1494 }
1495 #endif
1496
1497 return ptr;
1498}
1499
1500ecb_inline char *
1501ecb_i2a_i64 (char *ptr, int64_t v)
1502{
1503 *ptr = '-'; ptr += v < 0;
1504 uint64_t u = v < 0 ? -(uint64_t)v : v;
1505
1506 #if ECB_64BIT_NATIVE
1507 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1508 ptr = ecb_i2a_x10 (ptr, u);
1509 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1510 {
1511 uint64_t u1 = u % 1000000000;
1512 uint64_t u2 = u / 1000000000;
1513
1514 ptr = ecb_i2a_x10 (ptr, u2);
1515 ptr = ecb_i2a_09 (ptr, u1);
1516 }
1517 else
1518 {
1519 uint64_t u1 = u % 1000000000;
1520 uint64_t ua = u / 1000000000;
1521 uint64_t u2 = ua % 1000000000;
1522 uint64_t u3 = ua / 1000000000;
1523
1524 /* 2**31 is 19 digits, so the top is exactly one digit */
1525 *ptr++ = u3 + '0';
1526 ptr = ecb_i2a_09 (ptr, u2);
1527 ptr = ecb_i2a_09 (ptr, u1);
1528 }
1529 #else
1530 ptr = ecb_i2a_u64 (ptr, u);
1531 #endif
1532
1533 return ptr;
1534}
1535
590/*******************************************************************************/ 1536/*******************************************************************************/
591/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ 1537/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
592 1538
593/* basically, everything uses "ieee pure-endian" floating point numbers */ 1539/* basically, everything uses "ieee pure-endian" floating point numbers */
594/* the only noteworthy exception is ancient armle, which uses order 43218765 */ 1540/* the only noteworthy exception is ancient armle, which uses order 43218765 */
595#if 0 \ 1541#if 0 \
596 || __i386 || __i386__ \ 1542 || __i386 || __i386__ \
597 || __amd64 || __amd64__ || __x86_64 || __x86_64__ \ 1543 || ECB_GCC_AMD64 \
598 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \ 1544 || __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ \
599 || defined __arm__ && defined __ARM_EABI__ \
600 || defined __s390__ || defined __s390x__ \ 1545 || defined __s390__ || defined __s390x__ \
601 || defined __mips__ \ 1546 || defined __mips__ \
602 || defined __alpha__ \ 1547 || defined __alpha__ \
603 || defined __hppa__ \ 1548 || defined __hppa__ \
604 || defined __ia64__ \ 1549 || defined __ia64__ \
605 || defined __m68k__ \ 1550 || defined __m68k__ \
606 || defined __m88k__ \ 1551 || defined __m88k__ \
607 || defined __sh__ \ 1552 || defined __sh__ \
608 || defined _M_IX86 || defined _M_AMD64 || defined _M_IA64 1553 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
1554 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
1555 || defined __aarch64__
609 #define ECB_STDFP 1 1556 #define ECB_STDFP 1
610 #include <string.h> /* for memcpy */
611#else 1557#else
612 #define ECB_STDFP 0 1558 #define ECB_STDFP 0
613#endif 1559#endif
614 1560
615#ifndef ECB_NO_LIBM 1561#ifndef ECB_NO_LIBM
… …
627 #define ECB_NAN NAN 1573 #define ECB_NAN NAN
628 #else 1574 #else
629 #define ECB_NAN ECB_INFINITY 1575 #define ECB_NAN ECB_INFINITY
630 #endif 1576 #endif
631 1577
632 /* converts an ieee half/binary16 to a float */ 1578 #if ECB_C99 || _XOPEN_VERSION >= 600 || _POSIX_VERSION >= 200112L
633 ecb_function_ float ecb_binary16_to_float (uint16_t x) ecb_const; 1579 #define ecb_ldexpf(x,e) ldexpf ((x), (e))
634 ecb_function_ float 1580 #define ecb_frexpf(x,e) frexpf ((x), (e))
635 ecb_binary16_to_float (uint16_t x) 1581 #else
636 { 1582 #define ecb_ldexpf(x,e) (float) ldexp ((double) (x), (e))
637 int e = (x >> 10) & 0x1f; 1583 #define ecb_frexpf(x,e) (float) frexp ((double) (x), (e))
638 int m = x & 0x3ff; 1584 #endif
639 float r;
640
641 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 else r = ECB_INFINITY;
645
646 return x & 0x8000 ? -r : r;
647 }
648 1585
649 /* convert a float to ieee single/binary32 */ 1586 /* convert a float to ieee single/binary32 */
650 ecb_function_ uint32_t ecb_float_to_binary32 (float x) ecb_const; 1587 ecb_function_ ecb_const uint32_t ecb_float_to_binary32 (float x);
651 ecb_function_ uint32_t 1588 ecb_function_ ecb_const uint32_t
652 ecb_float_to_binary32 (float x) 1589 ecb_float_to_binary32 (float x)
653 { 1590 {
654 uint32_t r; 1591 uint32_t r;
655 1592
656 #if ECB_STDFP 1593 #if ECB_STDFP
… …
663 if (x == 0e0f ) return 0x00000000U; 1600 if (x == 0e0f ) return 0x00000000U;
664 if (x > +3.40282346638528860e+38f) return 0x7f800000U; 1601 if (x > +3.40282346638528860e+38f) return 0x7f800000U;
665 if (x < -3.40282346638528860e+38f) return 0xff800000U; 1602 if (x < -3.40282346638528860e+38f) return 0xff800000U;
666 if (x != x ) return 0x7fbfffffU; 1603 if (x != x ) return 0x7fbfffffU;
667 1604
668 m = frexpf (x, &e) * 0x1000000U; 1605 m = ecb_frexpf (x, &e) * 0x1000000U;
669 1606
670 r = m & 0x80000000U; 1607 r = m & 0x80000000U;
671 1608
672 if (r) 1609 if (r)
673 m = -m; 1610 m = -m;
… …
685 1622
686 return r; 1623 return r;
687 } 1624 }
688 1625
689 /* converts an ieee single/binary32 to a float */ 1626 /* converts an ieee single/binary32 to a float */
690 ecb_function_ float ecb_binary32_to_float (uint32_t x) ecb_const; 1627 ecb_function_ ecb_const float ecb_binary32_to_float (uint32_t x);
691 ecb_function_ float 1628 ecb_function_ ecb_const float
692 ecb_binary32_to_float (uint32_t x) 1629 ecb_binary32_to_float (uint32_t x)
693 { 1630 {
694 float r; 1631 float r;
695 1632
696 #if ECB_STDFP 1633 #if ECB_STDFP
… …
706 x |= 0x800000U; 1643 x |= 0x800000U;
707 else 1644 else
708 e = 1; 1645 e = 1;
709 1646
710 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */ 1647 /* we distrust ldexpf a bit and do the 2**-24 scaling by an extra multiply */
711 r = ldexpf (x * (0.5f / 0x800000U), e - 126); 1648 r = ecb_ldexpf (x * (0.5f / 0x800000U), e - 126);
712 1649
713 r = neg ? -r : r; 1650 r = neg ? -r : r;
714 #endif 1651 #endif
715 1652
716 return r; 1653 return r;
717 } 1654 }
718 1655
719 /* convert a double to ieee double/binary64 */ 1656 /* convert a double to ieee double/binary64 */
720 ecb_function_ uint64_t ecb_double_to_binary64 (double x) ecb_const; 1657 ecb_function_ ecb_const uint64_t ecb_double_to_binary64 (double x);
721 ecb_function_ uint64_t 1658 ecb_function_ ecb_const uint64_t
722 ecb_double_to_binary64 (double x) 1659 ecb_double_to_binary64 (double x)
723 { 1660 {
724 uint64_t r; 1661 uint64_t r;
725 1662
726 #if ECB_STDFP 1663 #if ECB_STDFP
… …
755 1692
756 return r; 1693 return r;
757 } 1694 }
758 1695
759 /* converts an ieee double/binary64 to a double */ 1696 /* converts an ieee double/binary64 to a double */
760 ecb_function_ double ecb_binary64_to_double (uint64_t x) ecb_const; 1697 ecb_function_ ecb_const double ecb_binary64_to_double (uint64_t x);
761 ecb_function_ double 1698 ecb_function_ ecb_const double
762 ecb_binary64_to_double (uint64_t x) 1699 ecb_binary64_to_double (uint64_t x)
763 { 1700 {
764 double r; 1701 double r;
765 1702
766 #if ECB_STDFP 1703 #if ECB_STDFP
… …
784 #endif 1721 #endif
785 1722
786 return r; 1723 return r;
787 } 1724 }
788 1725
789#endif 1726 /* convert a float to ieee half/binary16 */
1727 ecb_function_ ecb_const uint16_t ecb_float_to_binary16 (float x);
1728 ecb_function_ ecb_const uint16_t
1729 ecb_float_to_binary16 (float x)
1730 {
1731 return ecb_binary32_to_binary16 (ecb_float_to_binary32 (x));
1732 }
790 1733
791#endif 1734 /* convert an ieee half/binary16 to float */
1735 ecb_function_ ecb_const float ecb_binary16_to_float (uint16_t x);
1736 ecb_function_ ecb_const float
1737 ecb_binary16_to_float (uint16_t x)
1738 {
1739 return ecb_binary32_to_float (ecb_binary16_to_binary32 (x));
1740 }
792 1741
1742#endif
1743
1744#endif
1745

Diff Legend

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