1 | /* |
1 | /* |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
2 | * libecb - http://software.schmorp.de/pkg/libecb |
3 | * |
3 | * |
4 | * Copyright (©) 2009-2015 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: |
… | |
… | |
40 | |
40 | |
41 | #ifndef ECB_H |
41 | #ifndef ECB_H |
42 | #define ECB_H |
42 | #define ECB_H |
43 | |
43 | |
44 | /* 16 bits major, 16 bits minor */ |
44 | /* 16 bits major, 16 bits minor */ |
45 | #define ECB_VERSION 0x00010005 |
45 | #define ECB_VERSION 0x0001000c |
46 | |
46 | |
47 | #ifdef _WIN32 |
47 | #include <string.h> /* for memcpy */ |
|
|
48 | |
|
|
49 | #if defined (_WIN32) && !defined (__MINGW32__) |
48 | typedef signed char int8_t; |
50 | typedef signed char int8_t; |
49 | 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; |
50 | typedef signed short int16_t; |
54 | typedef signed short int16_t; |
51 | 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; |
52 | typedef signed int int32_t; |
58 | typedef signed int int32_t; |
53 | 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; |
54 | #if __GNUC__ |
62 | #if __GNUC__ |
55 | typedef signed long long int64_t; |
63 | typedef signed long long int64_t; |
56 | typedef unsigned long long uint64_t; |
64 | typedef unsigned long long uint64_t; |
57 | #else /* _MSC_VER || __BORLANDC__ */ |
65 | #else /* _MSC_VER || __BORLANDC__ */ |
58 | typedef signed __int64 int64_t; |
66 | typedef signed __int64 int64_t; |
59 | typedef unsigned __int64 uint64_t; |
67 | typedef unsigned __int64 uint64_t; |
60 | #endif |
68 | #endif |
|
|
69 | typedef int64_t int_fast64_t; |
|
|
70 | typedef uint64_t uint_fast64_t; |
61 | #ifdef _WIN64 |
71 | #ifdef _WIN64 |
62 | #define ECB_PTRSIZE 8 |
72 | #define ECB_PTRSIZE 8 |
63 | typedef uint64_t uintptr_t; |
73 | typedef uint64_t uintptr_t; |
64 | typedef int64_t intptr_t; |
74 | typedef int64_t intptr_t; |
65 | #else |
75 | #else |
… | |
… | |
67 | typedef uint32_t uintptr_t; |
77 | typedef uint32_t uintptr_t; |
68 | typedef int32_t intptr_t; |
78 | typedef int32_t intptr_t; |
69 | #endif |
79 | #endif |
70 | #else |
80 | #else |
71 | #include <inttypes.h> |
81 | #include <inttypes.h> |
72 | #if UINTMAX_MAX > 0xffffffffU |
82 | #if (defined INTPTR_MAX ? INTPTR_MAX : ULONG_MAX) > 0xffffffffU |
73 | #define ECB_PTRSIZE 8 |
83 | #define ECB_PTRSIZE 8 |
74 | #else |
84 | #else |
75 | #define ECB_PTRSIZE 4 |
85 | #define ECB_PTRSIZE 4 |
76 | #endif |
86 | #endif |
77 | #endif |
87 | #endif |
78 | |
88 | |
79 | #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |
89 | #define ECB_GCC_AMD64 (__amd64 || __amd64__ || __x86_64 || __x86_64__) |
80 | #define ECB_MSVC_AMD64 (_M_AMD64 || _M_X64) |
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 |
81 | |
99 | |
82 | /* work around x32 idiocy by defining proper macros */ |
100 | /* work around x32 idiocy by defining proper macros */ |
83 | #if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
101 | #if ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
84 | #if _ILP32 |
102 | #if _ILP32 |
85 | #define ECB_AMD64_X32 1 |
103 | #define ECB_AMD64_X32 1 |
86 | #else |
104 | #else |
87 | #define ECB_AMD64 1 |
105 | #define ECB_AMD64 1 |
88 | #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 |
89 | #endif |
113 | #endif |
90 | |
114 | |
91 | /* many compilers define _GNUC_ to some versions but then only implement |
115 | /* many compilers define _GNUC_ to some versions but then only implement |
92 | * what their idiot authors think are the "more important" extensions, |
116 | * what their idiot authors think are the "more important" extensions, |
93 | * causing enormous grief in return for some better fake benchmark numbers. |
117 | * causing enormous grief in return for some better fake benchmark numbers. |
… | |
… | |
115 | #define ECB_CLANG_EXTENSION(x) 0 |
139 | #define ECB_CLANG_EXTENSION(x) 0 |
116 | #endif |
140 | #endif |
117 | |
141 | |
118 | #define ECB_CPP (__cplusplus+0) |
142 | #define ECB_CPP (__cplusplus+0) |
119 | #define ECB_CPP11 (__cplusplus >= 201103L) |
143 | #define ECB_CPP11 (__cplusplus >= 201103L) |
|
|
144 | #define ECB_CPP14 (__cplusplus >= 201402L) |
|
|
145 | #define ECB_CPP17 (__cplusplus >= 201703L) |
120 | |
146 | |
121 | #if ECB_CPP |
147 | #if ECB_CPP |
122 | #define ECB_C 0 |
148 | #define ECB_C 0 |
123 | #define ECB_STDC_VERSION 0 |
149 | #define ECB_STDC_VERSION 0 |
124 | #else |
150 | #else |
… | |
… | |
126 | #define ECB_STDC_VERSION __STDC_VERSION__ |
152 | #define ECB_STDC_VERSION __STDC_VERSION__ |
127 | #endif |
153 | #endif |
128 | |
154 | |
129 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
155 | #define ECB_C99 (ECB_STDC_VERSION >= 199901L) |
130 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
156 | #define ECB_C11 (ECB_STDC_VERSION >= 201112L) |
|
|
157 | #define ECB_C17 (ECB_STDC_VERSION >= 201710L) |
131 | |
158 | |
132 | #if ECB_CPP |
159 | #if ECB_CPP |
133 | #define ECB_EXTERN_C extern "C" |
160 | #define ECB_EXTERN_C extern "C" |
134 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
161 | #define ECB_EXTERN_C_BEG ECB_EXTERN_C { |
135 | #define ECB_EXTERN_C_END } |
162 | #define ECB_EXTERN_C_END } |
… | |
… | |
161 | #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
188 | #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */ |
162 | #endif |
189 | #endif |
163 | |
190 | |
164 | #ifndef ECB_MEMORY_FENCE |
191 | #ifndef ECB_MEMORY_FENCE |
165 | #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") |
166 | #if __i386 || __i386__ |
194 | #if __i386 || __i386__ |
167 | #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") |
168 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
196 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
169 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
197 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
170 | #elif ECB_GCC_AMD64 |
198 | #elif ECB_GCC_AMD64 |
171 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
199 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") |
172 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
200 | #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") |
173 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") |
201 | #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory") |
174 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
202 | #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ |
175 | #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 */ |
176 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
211 | #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ |
177 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ |
212 | || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \ |
|
|
213 | || defined __ARM_ARCH_6T2__ |
178 | #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") |
179 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
215 | #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ |
180 | || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ |
216 | || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__ |
181 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
217 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") |
182 | #elif __aarch64__ |
218 | #elif __aarch64__ |
183 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
219 | #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") |
184 | #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
220 | #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) |
185 | #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") |
… | |
… | |
212 | #if ECB_GCC_VERSION(4,7) |
248 | #if ECB_GCC_VERSION(4,7) |
213 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
249 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
214 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
250 | #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) |
215 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
251 | #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) |
216 | #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) |
217 | |
255 | |
218 | #elif ECB_CLANG_EXTENSION(c_atomic) |
256 | #elif ECB_CLANG_EXTENSION(c_atomic) |
219 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
257 | /* see comment below (stdatomic.h) about the C11 memory model. */ |
220 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
258 | #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) |
221 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
259 | #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) |
222 | #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) |
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) |
223 | |
263 | |
224 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
264 | #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ |
225 | #define ECB_MEMORY_FENCE __sync_synchronize () |
265 | #define ECB_MEMORY_FENCE __sync_synchronize () |
226 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
266 | #elif _MSC_VER >= 1500 /* VC++ 2008 */ |
227 | /* 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... */ |
… | |
… | |
237 | #elif defined _WIN32 |
277 | #elif defined _WIN32 |
238 | #include <WinNT.h> |
278 | #include <WinNT.h> |
239 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
279 | #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ |
240 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
280 | #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 |
241 | #include <mbarrier.h> |
281 | #include <mbarrier.h> |
242 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
282 | #define ECB_MEMORY_FENCE __machine_rw_barrier () |
243 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () |
283 | #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier () |
244 | #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 () |
245 | #elif __xlC__ |
286 | #elif __xlC__ |
246 | #define ECB_MEMORY_FENCE __sync () |
287 | #define ECB_MEMORY_FENCE __sync () |
247 | #endif |
288 | #endif |
248 | #endif |
289 | #endif |
249 | |
290 | |
250 | #ifndef ECB_MEMORY_FENCE |
291 | #ifndef ECB_MEMORY_FENCE |
251 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
292 | #if ECB_C11 && !defined __STDC_NO_ATOMICS__ |
252 | /* 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, */ |
253 | /* not just C11 atomics and atomic accesses */ |
294 | /* not just C11 atomics and atomic accesses */ |
254 | #include <stdatomic.h> |
295 | #include <stdatomic.h> |
255 | /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */ |
|
|
256 | /* any fence other than seq_cst, which isn't very efficient for us. */ |
|
|
257 | /* Why that is, we don't know - either the C11 memory model is quite useless */ |
|
|
258 | /* for most usages, or gcc and clang have a bug */ |
|
|
259 | /* I *currently* lean towards the latter, and inefficiently implement */ |
|
|
260 | /* all three of ecb's fences as a seq_cst fence */ |
|
|
261 | /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */ |
|
|
262 | /* for all __atomic_thread_fence's except seq_cst */ |
|
|
263 | #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) |
264 | #endif |
299 | #endif |
265 | #endif |
300 | #endif |
266 | |
301 | |
267 | #ifndef ECB_MEMORY_FENCE |
302 | #ifndef ECB_MEMORY_FENCE |
268 | #if !ECB_AVOID_PTHREADS |
303 | #if !ECB_AVOID_PTHREADS |
… | |
… | |
288 | |
323 | |
289 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
324 | #if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE |
290 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
325 | #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE |
291 | #endif |
326 | #endif |
292 | |
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 | |
293 | /*****************************************************************************/ |
332 | /*****************************************************************************/ |
294 | |
333 | |
295 | #if ECB_CPP |
334 | #if ECB_CPP |
296 | #define ecb_inline static inline |
335 | #define ecb_inline static inline |
297 | #elif ECB_GCC_VERSION(2,5) |
336 | #elif ECB_GCC_VERSION(2,5) |
… | |
… | |
316 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
355 | #define ECB_CONCAT(a, b) ECB_CONCAT_(a, b) |
317 | #define ECB_STRINGIFY_(a) # a |
356 | #define ECB_STRINGIFY_(a) # a |
318 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
357 | #define ECB_STRINGIFY(a) ECB_STRINGIFY_(a) |
319 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
358 | #define ECB_STRINGIFY_EXPR(expr) ((expr), ECB_STRINGIFY_ (expr)) |
320 | |
359 | |
|
|
360 | /* This marks larger functions that do not neccessarily need to be inlined */ |
|
|
361 | /* The idea is to possibly compile the header twice, */ |
|
|
362 | /* once exposing only the declarations, another time to define external functions */ |
|
|
363 | /* TODO: possibly static would be best for these at the moment? */ |
321 | #define ecb_function_ ecb_inline |
364 | #define ecb_function_ ecb_inline |
322 | |
365 | |
323 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
366 | #if ECB_GCC_VERSION(3,1) || ECB_CLANG_VERSION(2,8) |
324 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
367 | #define ecb_attribute(attrlist) __attribute__ (attrlist) |
325 | #else |
368 | #else |
… | |
… | |
415 | /* count trailing zero bits and count # of one bits */ |
458 | /* count trailing zero bits and count # of one bits */ |
416 | #if ECB_GCC_VERSION(3,4) \ |
459 | #if ECB_GCC_VERSION(3,4) \ |
417 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
460 | || (ECB_CLANG_BUILTIN(__builtin_clz) && ECB_CLANG_BUILTIN(__builtin_clzll) \ |
418 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
461 | && ECB_CLANG_BUILTIN(__builtin_ctz) && ECB_CLANG_BUILTIN(__builtin_ctzll) \ |
419 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
462 | && ECB_CLANG_BUILTIN(__builtin_popcount)) |
420 | /* we assume int == 32 bit, long == 32 or 64 bit and long long == 64 bit */ |
|
|
421 | #define ecb_ld32(x) (__builtin_clz (x) ^ 31) |
|
|
422 | #define ecb_ld64(x) (__builtin_clzll (x) ^ 63) |
|
|
423 | #define ecb_ctz32(x) __builtin_ctz (x) |
463 | #define ecb_ctz32(x) __builtin_ctz (x) |
|
|
464 | #define ecb_ctz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_ctzl (x) : __builtin_ctzll (x)) |
424 | #define ecb_ctz64(x) __builtin_ctzll (x) |
465 | #define ecb_clz32(x) __builtin_clz (x) |
|
|
466 | #define ecb_clz64(x) (__SIZEOF_LONG__ == 64 ? __builtin_clzl (x) : __builtin_clzll (x)) |
|
|
467 | #define ecb_ld32(x) (ecb_clz32 (x) ^ 31) |
|
|
468 | #define ecb_ld64(x) (ecb_clz64 (x) ^ 63) |
425 | #define ecb_popcount32(x) __builtin_popcount (x) |
469 | #define ecb_popcount32(x) __builtin_popcount (x) |
426 | /* no popcountll */ |
470 | /* ecb_popcount64 is more difficult, see below */ |
427 | #else |
471 | #else |
428 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
472 | ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); |
429 | ecb_function_ ecb_const int |
473 | ecb_function_ ecb_const int |
430 | ecb_ctz32 (uint32_t x) |
474 | ecb_ctz32 (uint32_t x) |
431 | { |
475 | { |
432 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
476 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
433 | unsigned long r; |
477 | unsigned long r; |
434 | _BitScanForward (&r, x); |
478 | _BitScanForward (&r, x); |
435 | return (int)r; |
479 | return (int)r; |
436 | #else |
480 | #else |
437 | int r = 0; |
481 | int r; |
438 | |
482 | |
439 | x &= ~x + 1; /* this isolates the lowest bit */ |
483 | x &= ~x + 1; /* this isolates the lowest bit */ |
440 | |
484 | |
441 | #if ECB_branchless_on_i386 |
485 | #if 1 |
|
|
486 | /* David Seal's algorithm, Message-ID: <32975@armltd.uucp> from 1994 */ |
|
|
487 | /* This happens to return 32 for x == 0, but the API does not support this */ |
|
|
488 | |
|
|
489 | /* -0 marks unused entries */ |
|
|
490 | static unsigned char table[64] = |
|
|
491 | { |
|
|
492 | 32, 0, 1, 12, 2, 6, -0, 13, 3, -0, 7, -0, -0, -0, -0, 14, |
|
|
493 | 10, 4, -0, -0, 8, -0, -0, 25, -0, -0, -0, -0, -0, 21, 27, 15, |
|
|
494 | 31, 11, 5, -0, -0, -0, -0, -0, 9, -0, -0, 24, -0, -0, 20, 26, |
|
|
495 | 30, -0, -0, -0, -0, 23, -0, 19, 29, -0, 22, 18, 28, 17, 16, -0 |
|
|
496 | }; |
|
|
497 | |
|
|
498 | /* magic constant results in 33 unique values in the upper 6 bits */ |
|
|
499 | x *= 0x0450fbafU; /* == 17 * 65 * 65535 */ |
|
|
500 | |
|
|
501 | r = table [x >> 26]; |
|
|
502 | #elif 0 /* branchless on i386, typically */ |
|
|
503 | r = 0; |
442 | r += !!(x & 0xaaaaaaaa) << 0; |
504 | r += !!(x & 0xaaaaaaaa) << 0; |
443 | r += !!(x & 0xcccccccc) << 1; |
505 | r += !!(x & 0xcccccccc) << 1; |
444 | r += !!(x & 0xf0f0f0f0) << 2; |
506 | r += !!(x & 0xf0f0f0f0) << 2; |
445 | r += !!(x & 0xff00ff00) << 3; |
507 | r += !!(x & 0xff00ff00) << 3; |
446 | r += !!(x & 0xffff0000) << 4; |
508 | r += !!(x & 0xffff0000) << 4; |
447 | #else |
509 | #else /* branchless on modern compilers, typically */ |
|
|
510 | r = 0; |
448 | if (x & 0xaaaaaaaa) r += 1; |
511 | if (x & 0xaaaaaaaa) r += 1; |
449 | if (x & 0xcccccccc) r += 2; |
512 | if (x & 0xcccccccc) r += 2; |
450 | if (x & 0xf0f0f0f0) r += 4; |
513 | if (x & 0xf0f0f0f0) r += 4; |
451 | if (x & 0xff00ff00) r += 8; |
514 | if (x & 0xff00ff00) r += 8; |
452 | if (x & 0xffff0000) r += 16; |
515 | if (x & 0xffff0000) r += 16; |
… | |
… | |
468 | int shift = x & 0xffffffff ? 0 : 32; |
531 | int shift = x & 0xffffffff ? 0 : 32; |
469 | return ecb_ctz32 (x >> shift) + shift; |
532 | return ecb_ctz32 (x >> shift) + shift; |
470 | #endif |
533 | #endif |
471 | } |
534 | } |
472 | |
535 | |
|
|
536 | ecb_function_ ecb_const int ecb_clz32 (uint32_t x); |
|
|
537 | ecb_function_ ecb_const int |
|
|
538 | ecb_clz32 (uint32_t x) |
|
|
539 | { |
|
|
540 | #if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM) |
|
|
541 | unsigned long r; |
|
|
542 | _BitScanReverse (&r, x); |
|
|
543 | return (int)r; |
|
|
544 | #else |
|
|
545 | |
|
|
546 | /* Robert Harley's algorithm from comp.arch 1996-12-07 */ |
|
|
547 | /* This happens to return 32 for x == 0, but the API does not support this */ |
|
|
548 | |
|
|
549 | /* -0 marks unused table elements */ |
|
|
550 | static unsigned char table[64] = |
|
|
551 | { |
|
|
552 | 32, 31, -0, 16, -0, 30, 3, -0, 15, -0, -0, -0, 29, 10, 2, -0, |
|
|
553 | -0, -0, 12, 14, 21, -0, 19, -0, -0, 28, -0, 25, -0, 9, 1, -0, |
|
|
554 | 17, -0, 4, -0, -0, -0, 11, -0, 13, 22, 20, -0, 26, -0, -0, 18, |
|
|
555 | 5, -0, -0, 23, -0, 27, -0, 6, -0, 24, 7, -0, 8, -0, 0, -0 |
|
|
556 | }; |
|
|
557 | |
|
|
558 | /* propagate leftmost 1 bit to the right */ |
|
|
559 | x |= x >> 1; |
|
|
560 | x |= x >> 2; |
|
|
561 | x |= x >> 4; |
|
|
562 | x |= x >> 8; |
|
|
563 | x |= x >> 16; |
|
|
564 | |
|
|
565 | /* magic constant results in 33 unique values in the upper 6 bits */ |
|
|
566 | x *= 0x06EB14F9U; /* == 7 * 255 * 255 * 255 */ |
|
|
567 | |
|
|
568 | return table [x >> 26]; |
|
|
569 | #endif |
|
|
570 | } |
|
|
571 | |
|
|
572 | ecb_function_ ecb_const int ecb_clz64 (uint64_t x); |
|
|
573 | ecb_function_ ecb_const int |
|
|
574 | ecb_clz64 (uint64_t x) |
|
|
575 | { |
|
|
576 | #if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM) |
|
|
577 | unsigned long r; |
|
|
578 | _BitScanReverse64 (&r, x); |
|
|
579 | return (int)r; |
|
|
580 | #else |
|
|
581 | uint32_t l = x >> 32; |
|
|
582 | int shift = l ? 0 : 32; |
|
|
583 | return ecb_clz32 (l ? l : x) + shift; |
|
|
584 | #endif |
|
|
585 | } |
|
|
586 | |
473 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
587 | ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); |
474 | ecb_function_ ecb_const int |
588 | ecb_function_ ecb_const int |
475 | ecb_popcount32 (uint32_t x) |
589 | ecb_popcount32 (uint32_t x) |
476 | { |
590 | { |
477 | x -= (x >> 1) & 0x55555555; |
591 | x -= (x >> 1) & 0x55555555; |
… | |
… | |
552 | x = ( x >> 16 ) | ( x << 16); |
666 | x = ( x >> 16 ) | ( x << 16); |
553 | |
667 | |
554 | return x; |
668 | return x; |
555 | } |
669 | } |
556 | |
670 | |
557 | /* popcount64 is only available on 64 bit cpus as gcc builtin */ |
|
|
558 | /* so for this version we are lazy */ |
|
|
559 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
671 | ecb_function_ ecb_const int ecb_popcount64 (uint64_t x); |
560 | ecb_function_ ecb_const int |
672 | ecb_function_ ecb_const int |
561 | ecb_popcount64 (uint64_t x) |
673 | ecb_popcount64 (uint64_t x) |
562 | { |
674 | { |
|
|
675 | /* popcount64 is only available on 64 bit cpus as gcc builtin. */ |
|
|
676 | /* also, gcc/clang make this surprisingly difficult to use */ |
|
|
677 | #if (__SIZEOF_LONG__ == 8) && (ECB_GCC_VERSION(3,4) || ECB_CLANG_BUILTIN (__builtin_popcountl)) |
|
|
678 | return __builtin_popcountl (x); |
|
|
679 | #else |
563 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
680 | return ecb_popcount32 (x) + ecb_popcount32 (x >> 32); |
|
|
681 | #endif |
564 | } |
682 | } |
565 | |
683 | |
566 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
684 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count); |
567 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
685 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count); |
568 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
686 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count); |
… | |
… | |
570 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
688 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count); |
571 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
689 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count); |
572 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
690 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count); |
573 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
691 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count); |
574 | |
692 | |
575 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> ( 8 - count)) | (x << count); } |
693 | ecb_inline ecb_const uint8_t ecb_rotl8 (uint8_t x, unsigned int count) { return (x >> (-count & 7)) | (x << (count & 7)); } |
576 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << ( 8 - count)) | (x >> count); } |
694 | ecb_inline ecb_const uint8_t ecb_rotr8 (uint8_t x, unsigned int count) { return (x << (-count & 7)) | (x >> (count & 7)); } |
577 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (16 - count)) | (x << count); } |
695 | ecb_inline ecb_const uint16_t ecb_rotl16 (uint16_t x, unsigned int count) { return (x >> (-count & 15)) | (x << (count & 15)); } |
578 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } |
696 | ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (-count & 15)) | (x >> (count & 15)); } |
579 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } |
697 | ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (-count & 31)) | (x << (count & 31)); } |
580 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } |
698 | ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (-count & 31)) | (x >> (count & 31)); } |
581 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } |
699 | ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (-count & 63)) | (x << (count & 63)); } |
582 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } |
700 | ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (-count & 63)) | (x >> (count & 63)); } |
|
|
701 | |
|
|
702 | #if ECB_CPP |
|
|
703 | |
|
|
704 | inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); } |
|
|
705 | inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); } |
|
|
706 | inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); } |
|
|
707 | inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); } |
|
|
708 | |
|
|
709 | inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); } |
|
|
710 | inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); } |
|
|
711 | inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); } |
|
|
712 | inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); } |
|
|
713 | |
|
|
714 | inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); } |
|
|
715 | inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); } |
|
|
716 | inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); } |
|
|
717 | inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); } |
|
|
718 | |
|
|
719 | inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); } |
|
|
720 | inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); } |
|
|
721 | inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); } |
|
|
722 | inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); } |
|
|
723 | |
|
|
724 | inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); } |
|
|
725 | inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); } |
|
|
726 | inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); } |
|
|
727 | |
|
|
728 | inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); } |
|
|
729 | inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); } |
|
|
730 | inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); } |
|
|
731 | inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); } |
|
|
732 | |
|
|
733 | inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); } |
|
|
734 | inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); } |
|
|
735 | inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); } |
|
|
736 | inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); } |
|
|
737 | |
|
|
738 | #endif |
583 | |
739 | |
584 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
740 | #if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) |
585 | #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
741 | #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) |
586 | #define ecb_bswap16(x) __builtin_bswap16 (x) |
742 | #define ecb_bswap16(x) __builtin_bswap16 (x) |
587 | #else |
743 | #else |
… | |
… | |
626 | #endif |
782 | #endif |
627 | |
783 | |
628 | /* try to tell the compiler that some condition is definitely true */ |
784 | /* try to tell the compiler that some condition is definitely true */ |
629 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
785 | #define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 |
630 | |
786 | |
631 | ecb_inline ecb_const unsigned char ecb_byteorder_helper (void); |
787 | ecb_inline ecb_const uint32_t ecb_byteorder_helper (void); |
632 | ecb_inline ecb_const unsigned char |
788 | ecb_inline ecb_const uint32_t |
633 | ecb_byteorder_helper (void) |
789 | ecb_byteorder_helper (void) |
634 | { |
790 | { |
635 | /* the union code still generates code under pressure in gcc, */ |
791 | /* the union code still generates code under pressure in gcc, */ |
636 | /* but less than using pointers, and always seems to */ |
792 | /* but less than using pointers, and always seems to */ |
637 | /* successfully return a constant. */ |
793 | /* successfully return a constant. */ |
638 | /* the reason why we have this horrible preprocessor mess */ |
794 | /* the reason why we have this horrible preprocessor mess */ |
639 | /* is to avoid it in all cases, at least on common architectures */ |
795 | /* is to avoid it in all cases, at least on common architectures */ |
640 | /* or when using a recent enough gcc version (>= 4.6) */ |
796 | /* or when using a recent enough gcc version (>= 4.6) */ |
641 | #if ((__i386 || __i386__) && !__VOS__) || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64 |
|
|
642 | return 0x44; |
|
|
643 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ |
797 | #if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \ |
|
|
798 | || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__) |
|
|
799 | #define ECB_LITTLE_ENDIAN 1 |
644 | return 0x44; |
800 | return 0x44332211; |
645 | #elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ |
801 | #elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \ |
|
|
802 | || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__) |
|
|
803 | #define ECB_BIG_ENDIAN 1 |
646 | return 0x11; |
804 | return 0x11223344; |
647 | #else |
805 | #else |
648 | union |
806 | union |
649 | { |
807 | { |
|
|
808 | uint8_t c[4]; |
650 | uint32_t i; |
809 | uint32_t u; |
651 | uint8_t c; |
|
|
652 | } u = { 0x11223344 }; |
810 | } u = { 0x11, 0x22, 0x33, 0x44 }; |
653 | return u.c; |
811 | return u.u; |
654 | #endif |
812 | #endif |
655 | } |
813 | } |
656 | |
814 | |
657 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
815 | ecb_inline ecb_const ecb_bool ecb_big_endian (void); |
658 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } |
816 | ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } |
659 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
817 | ecb_inline ecb_const ecb_bool ecb_little_endian (void); |
660 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } |
818 | ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } |
|
|
819 | |
|
|
820 | /*****************************************************************************/ |
|
|
821 | /* unaligned load/store */ |
|
|
822 | |
|
|
823 | ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } |
|
|
824 | ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } |
|
|
825 | ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } |
|
|
826 | |
|
|
827 | ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } |
|
|
828 | ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } |
|
|
829 | ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } |
|
|
830 | |
|
|
831 | ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
832 | ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
833 | ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
834 | |
|
|
835 | ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); } |
|
|
836 | ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); } |
|
|
837 | ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); } |
|
|
838 | |
|
|
839 | ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); } |
|
|
840 | ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); } |
|
|
841 | ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); } |
|
|
842 | |
|
|
843 | ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; } |
|
|
844 | ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; } |
|
|
845 | ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; } |
|
|
846 | |
|
|
847 | ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; } |
|
|
848 | ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; } |
|
|
849 | ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; } |
|
|
850 | |
|
|
851 | ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
852 | ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
853 | ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
854 | |
|
|
855 | ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); } |
|
|
856 | ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); } |
|
|
857 | ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); } |
|
|
858 | |
|
|
859 | ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); } |
|
|
860 | ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); } |
|
|
861 | ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); } |
|
|
862 | |
|
|
863 | #if ECB_CPP |
|
|
864 | |
|
|
865 | inline uint8_t ecb_bswap (uint8_t v) { return v; } |
|
|
866 | inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); } |
|
|
867 | inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); } |
|
|
868 | inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); } |
|
|
869 | |
|
|
870 | template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } |
|
|
871 | template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } |
|
|
872 | template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; } |
|
|
873 | template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); } |
|
|
874 | template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); } |
|
|
875 | template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; } |
|
|
876 | template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); } |
|
|
877 | template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); } |
|
|
878 | |
|
|
879 | template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; } |
|
|
880 | template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; } |
|
|
881 | template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; } |
|
|
882 | template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); } |
|
|
883 | template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); } |
|
|
884 | template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); } |
|
|
885 | template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); } |
|
|
886 | template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); } |
|
|
887 | |
|
|
888 | #endif |
|
|
889 | |
|
|
890 | /*****************************************************************************/ |
|
|
891 | /* pointer/integer hashing */ |
|
|
892 | |
|
|
893 | /* based on hash by Chris Wellons, https://nullprogram.com/blog/2018/07/31/ */ |
|
|
894 | ecb_function_ uint32_t ecb_mix32 (uint32_t v); |
|
|
895 | ecb_function_ uint32_t ecb_mix32 (uint32_t v) |
|
|
896 | { |
|
|
897 | v ^= v >> 16; v *= 0x7feb352dU; |
|
|
898 | v ^= v >> 15; v *= 0x846ca68bU; |
|
|
899 | v ^= v >> 16; |
|
|
900 | return v; |
|
|
901 | } |
|
|
902 | |
|
|
903 | ecb_function_ uint32_t ecb_unmix32 (uint32_t v); |
|
|
904 | ecb_function_ uint32_t ecb_unmix32 (uint32_t v) |
|
|
905 | { |
|
|
906 | v ^= v >> 16 ; v *= 0x43021123U; |
|
|
907 | v ^= v >> 15 ^ v >> 30; v *= 0x1d69e2a5U; |
|
|
908 | v ^= v >> 16 ; |
|
|
909 | return v; |
|
|
910 | } |
|
|
911 | |
|
|
912 | /* based on splitmix64, by Sebastiona Vigna, https://prng.di.unimi.it/splitmix64.c */ |
|
|
913 | ecb_function_ uint64_t ecb_mix64 (uint64_t v); |
|
|
914 | ecb_function_ uint64_t ecb_mix64 (uint64_t v) |
|
|
915 | { |
|
|
916 | v ^= v >> 30; v *= 0xbf58476d1ce4e5b9U; |
|
|
917 | v ^= v >> 27; v *= 0x94d049bb133111ebU; |
|
|
918 | v ^= v >> 31; |
|
|
919 | return v; |
|
|
920 | } |
|
|
921 | |
|
|
922 | ecb_function_ uint64_t ecb_unmix64 (uint64_t v); |
|
|
923 | ecb_function_ uint64_t ecb_unmix64 (uint64_t v) |
|
|
924 | { |
|
|
925 | v ^= v >> 31 ^ v >> 62; v *= 0x319642b2d24d8ec3U; |
|
|
926 | v ^= v >> 27 ^ v >> 54; v *= 0x96de1b173f119089U; |
|
|
927 | v ^= v >> 30 ^ v >> 60; |
|
|
928 | return v; |
|
|
929 | } |
|
|
930 | |
|
|
931 | ecb_function_ uintptr_t ecb_ptrmix (void *p); |
|
|
932 | ecb_function_ uintptr_t ecb_ptrmix (void *p) |
|
|
933 | { |
|
|
934 | #if ECB_PTRSIZE <= 4 |
|
|
935 | return ecb_mix32 ((uint32_t)p); |
|
|
936 | #else |
|
|
937 | return ecb_mix64 ((uint64_t)p); |
|
|
938 | #endif |
|
|
939 | } |
|
|
940 | |
|
|
941 | ecb_function_ void *ecb_ptrunmix (uintptr_t v); |
|
|
942 | ecb_function_ void *ecb_ptrunmix (uintptr_t v) |
|
|
943 | { |
|
|
944 | #if ECB_PTRSIZE <= 4 |
|
|
945 | return (void *)ecb_unmix32 (v); |
|
|
946 | #else |
|
|
947 | return (void *)ecb_unmix64 (v); |
|
|
948 | #endif |
|
|
949 | } |
|
|
950 | |
|
|
951 | #if ECB_CPP |
|
|
952 | |
|
|
953 | template<typename T> |
|
|
954 | inline uintptr_t ecb_ptrmix (T *p) |
|
|
955 | { |
|
|
956 | return ecb_ptrmix (static_cast<void *>(p)); |
|
|
957 | } |
|
|
958 | |
|
|
959 | template<typename T> |
|
|
960 | inline T *ecb_ptrunmix (uintptr_t v) |
|
|
961 | { |
|
|
962 | return static_cast<T *>(ecb_ptrunmix (v)); |
|
|
963 | } |
|
|
964 | |
|
|
965 | #endif |
|
|
966 | |
|
|
967 | /*****************************************************************************/ |
|
|
968 | /* gray code */ |
|
|
969 | |
|
|
970 | ecb_inline uint_fast8_t ecb_gray_encode8 (uint_fast8_t b) { return b ^ (b >> 1); } |
|
|
971 | ecb_inline uint_fast16_t ecb_gray_encode16 (uint_fast16_t b) { return b ^ (b >> 1); } |
|
|
972 | ecb_inline uint_fast32_t ecb_gray_encode32 (uint_fast32_t b) { return b ^ (b >> 1); } |
|
|
973 | ecb_inline uint_fast64_t ecb_gray_encode64 (uint_fast64_t b) { return b ^ (b >> 1); } |
|
|
974 | |
|
|
975 | ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g); |
|
|
976 | ecb_function_ uint8_t ecb_gray_decode8 (uint8_t g) |
|
|
977 | { |
|
|
978 | g ^= g >> 1; |
|
|
979 | g ^= g >> 2; |
|
|
980 | g ^= g >> 4; |
|
|
981 | |
|
|
982 | return g; |
|
|
983 | } |
|
|
984 | |
|
|
985 | ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g); |
|
|
986 | ecb_function_ uint16_t ecb_gray_decode16 (uint16_t g) |
|
|
987 | { |
|
|
988 | g ^= g >> 1; |
|
|
989 | g ^= g >> 2; |
|
|
990 | g ^= g >> 4; |
|
|
991 | g ^= g >> 8; |
|
|
992 | |
|
|
993 | return g; |
|
|
994 | } |
|
|
995 | |
|
|
996 | ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g); |
|
|
997 | ecb_function_ uint32_t ecb_gray_decode32 (uint32_t g) |
|
|
998 | { |
|
|
999 | g ^= g >> 1; |
|
|
1000 | g ^= g >> 2; |
|
|
1001 | g ^= g >> 4; |
|
|
1002 | g ^= g >> 8; |
|
|
1003 | g ^= g >> 16; |
|
|
1004 | |
|
|
1005 | return g; |
|
|
1006 | } |
|
|
1007 | |
|
|
1008 | ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g); |
|
|
1009 | ecb_function_ uint64_t ecb_gray_decode64 (uint64_t g) |
|
|
1010 | { |
|
|
1011 | g ^= g >> 1; |
|
|
1012 | g ^= g >> 2; |
|
|
1013 | g ^= g >> 4; |
|
|
1014 | g ^= g >> 8; |
|
|
1015 | g ^= g >> 16; |
|
|
1016 | g ^= g >> 32; |
|
|
1017 | |
|
|
1018 | return g; |
|
|
1019 | } |
|
|
1020 | |
|
|
1021 | #if ECB_CPP |
|
|
1022 | |
|
|
1023 | ecb_inline uint8_t ecb_gray_encode (uint8_t b) { return ecb_gray_encode8 (b); } |
|
|
1024 | ecb_inline uint16_t ecb_gray_encode (uint16_t b) { return ecb_gray_encode16 (b); } |
|
|
1025 | ecb_inline uint32_t ecb_gray_encode (uint32_t b) { return ecb_gray_encode32 (b); } |
|
|
1026 | ecb_inline uint64_t ecb_gray_encode (uint64_t b) { return ecb_gray_encode64 (b); } |
|
|
1027 | |
|
|
1028 | ecb_inline uint8_t ecb_gray_decode (uint8_t g) { return ecb_gray_decode8 (g); } |
|
|
1029 | ecb_inline uint16_t ecb_gray_decode (uint16_t g) { return ecb_gray_decode16 (g); } |
|
|
1030 | ecb_inline uint32_t ecb_gray_decode (uint32_t g) { return ecb_gray_decode32 (g); } |
|
|
1031 | ecb_inline uint64_t ecb_gray_decode (uint64_t g) { return ecb_gray_decode64 (g); } |
|
|
1032 | |
|
|
1033 | #endif |
|
|
1034 | |
|
|
1035 | /*****************************************************************************/ |
|
|
1036 | /* 2d hilbert curves */ |
|
|
1037 | |
|
|
1038 | /* algorithm from the book Hacker's Delight, modified to not */ |
|
|
1039 | /* run into undefined behaviour for n==16 */ |
|
|
1040 | static uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s); |
|
|
1041 | static uint32_t ecb_hilbert2d_index_to_coord32 (int n, uint32_t s) |
|
|
1042 | { |
|
|
1043 | uint32_t comp, swap, cs, t, sr; |
|
|
1044 | |
|
|
1045 | /* pad s on the left (unused) bits with 01 (no change groups) */ |
|
|
1046 | s |= 0x55555555U << n << n; |
|
|
1047 | /* "s shift right" */ |
|
|
1048 | sr = (s >> 1) & 0x55555555U; |
|
|
1049 | /* compute complement and swap info in two-bit groups */ |
|
|
1050 | cs = ((s & 0x55555555U) + sr) ^ 0x55555555U; |
|
|
1051 | |
|
|
1052 | /* parallel prefix xor op to propagate both complement |
|
|
1053 | * and swap info together from left to right (there is |
|
|
1054 | * no step "cs ^= cs >> 1", so in effect it computes |
|
|
1055 | * two independent parallel prefix operations on two |
|
|
1056 | * interleaved sets of sixteen bits). |
|
|
1057 | */ |
|
|
1058 | cs ^= cs >> 2; |
|
|
1059 | cs ^= cs >> 4; |
|
|
1060 | cs ^= cs >> 8; |
|
|
1061 | cs ^= cs >> 16; |
|
|
1062 | |
|
|
1063 | /* separate swap and complement bits */ |
|
|
1064 | swap = cs & 0x55555555U; |
|
|
1065 | comp = (cs >> 1) & 0x55555555U; |
|
|
1066 | |
|
|
1067 | /* calculate coordinates in odd and even bit positions */ |
|
|
1068 | t = (s & swap) ^ comp; |
|
|
1069 | s = s ^ sr ^ t ^ (t << 1); |
|
|
1070 | |
|
|
1071 | /* unpad/clear out any junk on the left */ |
|
|
1072 | s = s & ((1 << n << n) - 1); |
|
|
1073 | |
|
|
1074 | /* Now "unshuffle" to separate the x and y bits. */ |
|
|
1075 | t = (s ^ (s >> 1)) & 0x22222222U; s ^= t ^ (t << 1); |
|
|
1076 | t = (s ^ (s >> 2)) & 0x0c0c0c0cU; s ^= t ^ (t << 2); |
|
|
1077 | t = (s ^ (s >> 4)) & 0x00f000f0U; s ^= t ^ (t << 4); |
|
|
1078 | t = (s ^ (s >> 8)) & 0x0000ff00U; s ^= t ^ (t << 8); |
|
|
1079 | |
|
|
1080 | /* now s contains two 16-bit coordinates */ |
|
|
1081 | return s; |
|
|
1082 | } |
|
|
1083 | |
|
|
1084 | /* 64 bit, a straightforward extension to the 32 bit case */ |
|
|
1085 | static uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s); |
|
|
1086 | static uint64_t ecb_hilbert2d_index_to_coord64 (int n, uint64_t s) |
|
|
1087 | { |
|
|
1088 | uint64_t comp, swap, cs, t, sr; |
|
|
1089 | |
|
|
1090 | /* pad s on the left (unused) bits with 01 (no change groups) */ |
|
|
1091 | s |= 0x5555555555555555U << n << n; |
|
|
1092 | /* "s shift right" */ |
|
|
1093 | sr = (s >> 1) & 0x5555555555555555U; |
|
|
1094 | /* compute complement and swap info in two-bit groups */ |
|
|
1095 | cs = ((s & 0x5555555555555555U) + sr) ^ 0x5555555555555555U; |
|
|
1096 | |
|
|
1097 | /* parallel prefix xor op to propagate both complement |
|
|
1098 | * and swap info together from left to right (there is |
|
|
1099 | * no step "cs ^= cs >> 1", so in effect it computes |
|
|
1100 | * two independent parallel prefix operations on two |
|
|
1101 | * interleaved sets of thirty-two bits). |
|
|
1102 | */ |
|
|
1103 | cs ^= cs >> 2; |
|
|
1104 | cs ^= cs >> 4; |
|
|
1105 | cs ^= cs >> 8; |
|
|
1106 | cs ^= cs >> 16; |
|
|
1107 | cs ^= cs >> 32; |
|
|
1108 | |
|
|
1109 | /* separate swap and complement bits */ |
|
|
1110 | swap = cs & 0x5555555555555555U; |
|
|
1111 | comp = (cs >> 1) & 0x5555555555555555U; |
|
|
1112 | |
|
|
1113 | /* calculate coordinates in odd and even bit positions */ |
|
|
1114 | t = (s & swap) ^ comp; |
|
|
1115 | s = s ^ sr ^ t ^ (t << 1); |
|
|
1116 | |
|
|
1117 | /* unpad/clear out any junk on the left */ |
|
|
1118 | s = s & ((1 << n << n) - 1); |
|
|
1119 | |
|
|
1120 | /* Now "unshuffle" to separate the x and y bits. */ |
|
|
1121 | t = (s ^ (s >> 1)) & 0x2222222222222222U; s ^= t ^ (t << 1); |
|
|
1122 | t = (s ^ (s >> 2)) & 0x0c0c0c0c0c0c0c0cU; s ^= t ^ (t << 2); |
|
|
1123 | t = (s ^ (s >> 4)) & 0x00f000f000f000f0U; s ^= t ^ (t << 4); |
|
|
1124 | t = (s ^ (s >> 8)) & 0x0000ff000000ff00U; s ^= t ^ (t << 8); |
|
|
1125 | t = (s ^ (s >> 16)) & 0x00000000ffff0000U; s ^= t ^ (t << 16); |
|
|
1126 | |
|
|
1127 | /* now s contains two 32-bit coordinates */ |
|
|
1128 | return s; |
|
|
1129 | } |
|
|
1130 | |
|
|
1131 | /* algorithm from the book Hacker's Delight, but a similar algorithm*/ |
|
|
1132 | /* is given in https://doi.org/10.1002/spe.4380160103 */ |
|
|
1133 | /* this has been slightly improved over the original version */ |
|
|
1134 | ecb_function_ uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy); |
|
|
1135 | ecb_function_ uint32_t ecb_hilbert2d_coord_to_index32 (int n, uint32_t xy) |
|
|
1136 | { |
|
|
1137 | uint32_t row; |
|
|
1138 | uint32_t state = 0; |
|
|
1139 | uint32_t s = 0; |
|
|
1140 | |
|
|
1141 | do |
|
|
1142 | { |
|
|
1143 | --n; |
|
|
1144 | |
|
|
1145 | row = 4 * state |
|
|
1146 | | (2 & (xy >> n >> 15)) |
|
|
1147 | | (1 & (xy >> n )); |
|
|
1148 | |
|
|
1149 | /* these funky constants are lookup tables for two-bit values */ |
|
|
1150 | s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3; |
|
|
1151 | state = (0x8fe65831U >> 2 * row) & 3; |
|
|
1152 | } |
|
|
1153 | while (n > 0); |
|
|
1154 | |
|
|
1155 | return s; |
|
|
1156 | } |
|
|
1157 | |
|
|
1158 | /* 64 bit, essentially the same as 32 bit */ |
|
|
1159 | ecb_function_ uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy); |
|
|
1160 | ecb_function_ uint64_t ecb_hilbert2d_coord_to_index64 (int n, uint64_t xy) |
|
|
1161 | { |
|
|
1162 | uint32_t row; |
|
|
1163 | uint32_t state = 0; |
|
|
1164 | uint64_t s = 0; |
|
|
1165 | |
|
|
1166 | do |
|
|
1167 | { |
|
|
1168 | --n; |
|
|
1169 | |
|
|
1170 | row = 4 * state |
|
|
1171 | | (2 & (xy >> n >> 31)) |
|
|
1172 | | (1 & (xy >> n )); |
|
|
1173 | |
|
|
1174 | /* these funky constants are lookup tables for two-bit values */ |
|
|
1175 | s = (s << 2) | (0x361e9cb4U >> 2 * row) & 3; |
|
|
1176 | state = (0x8fe65831U >> 2 * row) & 3; |
|
|
1177 | } |
|
|
1178 | while (n > 0); |
|
|
1179 | |
|
|
1180 | return s; |
|
|
1181 | } |
|
|
1182 | |
|
|
1183 | /*****************************************************************************/ |
|
|
1184 | /* division */ |
661 | |
1185 | |
662 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
1186 | #if ECB_GCC_VERSION(3,0) || ECB_C99 |
|
|
1187 | /* C99 tightened the definition of %, so we can use a more efficient version */ |
663 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
1188 | #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) |
664 | #else |
1189 | #else |
665 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
1190 | #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) |
666 | #endif |
1191 | #endif |
667 | |
1192 | |
… | |
… | |
678 | } |
1203 | } |
679 | #else |
1204 | #else |
680 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
1205 | #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) |
681 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
1206 | #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) |
682 | #endif |
1207 | #endif |
|
|
1208 | |
|
|
1209 | /*****************************************************************************/ |
|
|
1210 | /* array length */ |
683 | |
1211 | |
684 | #if ecb_cplusplus_does_not_suck |
1212 | #if ecb_cplusplus_does_not_suck |
685 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
1213 | /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ |
686 | template<typename T, int N> |
1214 | template<typename T, int N> |
687 | static inline int ecb_array_length (const T (&arr)[N]) |
1215 | static inline int ecb_array_length (const T (&arr)[N]) |
… | |
… | |
689 | return N; |
1217 | return N; |
690 | } |
1218 | } |
691 | #else |
1219 | #else |
692 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
1220 | #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) |
693 | #endif |
1221 | #endif |
|
|
1222 | |
|
|
1223 | /*****************************************************************************/ |
|
|
1224 | /* IEEE 754-2008 half float conversions */ |
694 | |
1225 | |
695 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
1226 | ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); |
696 | ecb_function_ ecb_const uint32_t |
1227 | ecb_function_ ecb_const uint32_t |
697 | ecb_binary16_to_binary32 (uint32_t x) |
1228 | ecb_binary16_to_binary32 (uint32_t x) |
698 | { |
1229 | { |
… | |
… | |
727 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
1258 | ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); |
728 | ecb_function_ ecb_const uint16_t |
1259 | ecb_function_ ecb_const uint16_t |
729 | ecb_binary32_to_binary16 (uint32_t x) |
1260 | ecb_binary32_to_binary16 (uint32_t x) |
730 | { |
1261 | { |
731 | unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
1262 | unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ |
732 | unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
1263 | int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ |
733 | unsigned int m = x & 0x007fffff; |
1264 | unsigned int m = x & 0x007fffff; |
734 | |
1265 | |
735 | x &= 0x7fffffff; |
1266 | x &= 0x7fffffff; |
736 | |
1267 | |
737 | /* if it's within range of binary16 normals, use fast path */ |
1268 | /* if it's within range of binary16 normals, use fast path */ |
… | |
… | |
784 | |
1315 | |
785 | /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
1316 | /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ |
786 | m >>= 13; |
1317 | m >>= 13; |
787 | |
1318 | |
788 | return s | 0x7c00 | m | !m; |
1319 | return s | 0x7c00 | m | !m; |
|
|
1320 | } |
|
|
1321 | |
|
|
1322 | /*******************************************************************************/ |
|
|
1323 | /* fast integer to ascii */ |
|
|
1324 | |
|
|
1325 | /* |
|
|
1326 | * This code is pretty complicated because it is general. The idea behind it, |
|
|
1327 | * however, is pretty simple: first, the number is multiplied with a scaling |
|
|
1328 | * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point |
|
|
1329 | * number with the first digit in the upper bits. |
|
|
1330 | * Then this digit is converted to text and masked out. The resulting number |
|
|
1331 | * is then multiplied by 10, by multiplying the fixed point representation |
|
|
1332 | * by 5 and shifting the (binary) decimal point one to the right, so a 4.28 |
|
|
1333 | * format becomes 5.27, 6.26 and so on. |
|
|
1334 | * The rest involves only advancing the pointer if we already generated a |
|
|
1335 | * non-zero digit, so leading zeroes are overwritten. |
|
|
1336 | */ |
|
|
1337 | |
|
|
1338 | /* simply return a mask with "bits" bits set */ |
|
|
1339 | #define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1) |
|
|
1340 | |
|
|
1341 | /* oputput a single digit. maskvalue is 10**digitidx */ |
|
|
1342 | #define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \ |
|
|
1343 | if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \ |
|
|
1344 | { \ |
|
|
1345 | char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \ |
|
|
1346 | *ptr = digit + '0'; /* output it */ \ |
|
|
1347 | nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \ |
|
|
1348 | ptr += nz; /* output digit only if non-zero digit seen */ \ |
|
|
1349 | x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \ |
|
|
1350 | } |
|
|
1351 | |
|
|
1352 | /* convert integer to fixed point format and multiply out digits, highest first */ |
|
|
1353 | /* requires magic constants: max. digits and number of bits after the decimal point */ |
|
|
1354 | #define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \ |
|
|
1355 | ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \ |
|
|
1356 | { \ |
|
|
1357 | char nz = lz; /* non-zero digit seen? */ \ |
|
|
1358 | /* convert to x.bits fixed-point */ \ |
|
|
1359 | type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \ |
|
|
1360 | /* output up to 10 digits */ \ |
|
|
1361 | ecb_i2a_digit (type,bits,digitmask, 1, 0); \ |
|
|
1362 | ecb_i2a_digit (type,bits,digitmask, 10, 1); \ |
|
|
1363 | ecb_i2a_digit (type,bits,digitmask, 100, 2); \ |
|
|
1364 | ecb_i2a_digit (type,bits,digitmask, 1000, 3); \ |
|
|
1365 | ecb_i2a_digit (type,bits,digitmask, 10000, 4); \ |
|
|
1366 | ecb_i2a_digit (type,bits,digitmask, 100000, 5); \ |
|
|
1367 | ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \ |
|
|
1368 | ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \ |
|
|
1369 | ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \ |
|
|
1370 | ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \ |
|
|
1371 | return ptr; \ |
|
|
1372 | } |
|
|
1373 | |
|
|
1374 | /* predefined versions of the above, for various digits */ |
|
|
1375 | /* ecb_i2a_xN = almost N digits, limit defined by macro */ |
|
|
1376 | /* ecb_i2a_N = up to N digits, leading zeroes suppressed */ |
|
|
1377 | /* ecb_i2a_0N = exactly N digits, including leading zeroes */ |
|
|
1378 | |
|
|
1379 | /* non-leading-zero versions, limited range */ |
|
|
1380 | #define ECB_I2A_MAX_X5 59074 /* limit for ecb_i2a_x5 */ |
|
|
1381 | #define ECB_I2A_MAX_X10 2932500665 /* limit for ecb_i2a_x10 */ |
|
|
1382 | ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0) |
|
|
1383 | ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0) |
|
|
1384 | |
|
|
1385 | /* non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit */ |
|
|
1386 | ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0) |
|
|
1387 | ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0) |
|
|
1388 | ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0) |
|
|
1389 | ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0) |
|
|
1390 | ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0) |
|
|
1391 | ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0) |
|
|
1392 | ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0) |
|
|
1393 | ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0) |
|
|
1394 | |
|
|
1395 | /* leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit */ |
|
|
1396 | ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1) |
|
|
1397 | ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1) |
|
|
1398 | ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1) |
|
|
1399 | ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1) |
|
|
1400 | ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1) |
|
|
1401 | ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1) |
|
|
1402 | ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1) |
|
|
1403 | ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1) |
|
|
1404 | |
|
|
1405 | #define ECB_I2A_I32_DIGITS 11 |
|
|
1406 | #define ECB_I2A_U32_DIGITS 10 |
|
|
1407 | #define ECB_I2A_I64_DIGITS 20 |
|
|
1408 | #define ECB_I2A_U64_DIGITS 21 |
|
|
1409 | #define ECB_I2A_MAX_DIGITS 21 |
|
|
1410 | |
|
|
1411 | ecb_inline char * |
|
|
1412 | ecb_i2a_u32 (char *ptr, uint32_t u) |
|
|
1413 | { |
|
|
1414 | #if ECB_64BIT_NATIVE |
|
|
1415 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
|
|
1416 | ptr = ecb_i2a_x10 (ptr, u); |
|
|
1417 | else /* x10 almost, but not fully, covers 32 bit */ |
|
|
1418 | { |
|
|
1419 | uint32_t u1 = u % 1000000000; |
|
|
1420 | uint32_t u2 = u / 1000000000; |
|
|
1421 | |
|
|
1422 | *ptr++ = u2 + '0'; |
|
|
1423 | ptr = ecb_i2a_09 (ptr, u1); |
|
|
1424 | } |
|
|
1425 | #else |
|
|
1426 | if (ecb_expect_true (u <= ECB_I2A_MAX_X5)) |
|
|
1427 | ecb_i2a_x5 (ptr, u); |
|
|
1428 | else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000)) |
|
|
1429 | { |
|
|
1430 | uint32_t u1 = u % 10000; |
|
|
1431 | uint32_t u2 = u / 10000; |
|
|
1432 | |
|
|
1433 | ptr = ecb_i2a_x5 (ptr, u2); |
|
|
1434 | ptr = ecb_i2a_04 (ptr, u1); |
|
|
1435 | } |
|
|
1436 | else |
|
|
1437 | { |
|
|
1438 | uint32_t u1 = u % 10000; |
|
|
1439 | uint32_t ua = u / 10000; |
|
|
1440 | uint32_t u2 = ua % 10000; |
|
|
1441 | uint32_t u3 = ua / 10000; |
|
|
1442 | |
|
|
1443 | ptr = ecb_i2a_2 (ptr, u3); |
|
|
1444 | ptr = ecb_i2a_04 (ptr, u2); |
|
|
1445 | ptr = ecb_i2a_04 (ptr, u1); |
|
|
1446 | } |
|
|
1447 | #endif |
|
|
1448 | |
|
|
1449 | return ptr; |
|
|
1450 | } |
|
|
1451 | |
|
|
1452 | ecb_inline char * |
|
|
1453 | ecb_i2a_i32 (char *ptr, int32_t v) |
|
|
1454 | { |
|
|
1455 | *ptr = '-'; ptr += v < 0; |
|
|
1456 | uint32_t u = v < 0 ? -(uint32_t)v : v; |
|
|
1457 | |
|
|
1458 | #if ECB_64BIT_NATIVE |
|
|
1459 | ptr = ecb_i2a_x10 (ptr, u); /* x10 fully covers 31 bit */ |
|
|
1460 | #else |
|
|
1461 | ptr = ecb_i2a_u32 (ptr, u); |
|
|
1462 | #endif |
|
|
1463 | |
|
|
1464 | return ptr; |
|
|
1465 | } |
|
|
1466 | |
|
|
1467 | ecb_inline char * |
|
|
1468 | ecb_i2a_u64 (char *ptr, uint64_t u) |
|
|
1469 | { |
|
|
1470 | #if ECB_64BIT_NATIVE |
|
|
1471 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
|
|
1472 | ptr = ecb_i2a_x10 (ptr, u); |
|
|
1473 | else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000)) |
|
|
1474 | { |
|
|
1475 | uint64_t u1 = u % 1000000000; |
|
|
1476 | uint64_t u2 = u / 1000000000; |
|
|
1477 | |
|
|
1478 | ptr = ecb_i2a_x10 (ptr, u2); |
|
|
1479 | ptr = ecb_i2a_09 (ptr, u1); |
|
|
1480 | } |
|
|
1481 | else |
|
|
1482 | { |
|
|
1483 | uint64_t u1 = u % 1000000000; |
|
|
1484 | uint64_t ua = u / 1000000000; |
|
|
1485 | uint64_t u2 = ua % 1000000000; |
|
|
1486 | uint64_t u3 = ua / 1000000000; |
|
|
1487 | |
|
|
1488 | ptr = ecb_i2a_2 (ptr, u3); |
|
|
1489 | ptr = ecb_i2a_09 (ptr, u2); |
|
|
1490 | ptr = ecb_i2a_09 (ptr, u1); |
|
|
1491 | } |
|
|
1492 | #else |
|
|
1493 | if (ecb_expect_true (u <= ECB_I2A_MAX_X5)) |
|
|
1494 | ptr = ecb_i2a_x5 (ptr, u); |
|
|
1495 | else |
|
|
1496 | { |
|
|
1497 | uint64_t u1 = u % 10000; |
|
|
1498 | uint64_t u2 = u / 10000; |
|
|
1499 | |
|
|
1500 | ptr = ecb_i2a_u64 (ptr, u2); |
|
|
1501 | ptr = ecb_i2a_04 (ptr, u1); |
|
|
1502 | } |
|
|
1503 | #endif |
|
|
1504 | |
|
|
1505 | return ptr; |
|
|
1506 | } |
|
|
1507 | |
|
|
1508 | ecb_inline char * |
|
|
1509 | ecb_i2a_i64 (char *ptr, int64_t v) |
|
|
1510 | { |
|
|
1511 | *ptr = '-'; ptr += v < 0; |
|
|
1512 | uint64_t u = v < 0 ? -(uint64_t)v : v; |
|
|
1513 | |
|
|
1514 | #if ECB_64BIT_NATIVE |
|
|
1515 | if (ecb_expect_true (u <= ECB_I2A_MAX_X10)) |
|
|
1516 | ptr = ecb_i2a_x10 (ptr, u); |
|
|
1517 | else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000)) |
|
|
1518 | { |
|
|
1519 | uint64_t u1 = u % 1000000000; |
|
|
1520 | uint64_t u2 = u / 1000000000; |
|
|
1521 | |
|
|
1522 | ptr = ecb_i2a_x10 (ptr, u2); |
|
|
1523 | ptr = ecb_i2a_09 (ptr, u1); |
|
|
1524 | } |
|
|
1525 | else |
|
|
1526 | { |
|
|
1527 | uint64_t u1 = u % 1000000000; |
|
|
1528 | uint64_t ua = u / 1000000000; |
|
|
1529 | uint64_t u2 = ua % 1000000000; |
|
|
1530 | uint64_t u3 = ua / 1000000000; |
|
|
1531 | |
|
|
1532 | /* 2**31 is 19 digits, so the top is exactly one digit */ |
|
|
1533 | *ptr++ = u3 + '0'; |
|
|
1534 | ptr = ecb_i2a_09 (ptr, u2); |
|
|
1535 | ptr = ecb_i2a_09 (ptr, u1); |
|
|
1536 | } |
|
|
1537 | #else |
|
|
1538 | ptr = ecb_i2a_u64 (ptr, u); |
|
|
1539 | #endif |
|
|
1540 | |
|
|
1541 | return ptr; |
789 | } |
1542 | } |
790 | |
1543 | |
791 | /*******************************************************************************/ |
1544 | /*******************************************************************************/ |
792 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
1545 | /* floating point stuff, can be disabled by defining ECB_NO_LIBM */ |
793 | |
1546 | |
… | |
… | |
807 | || defined __sh__ \ |
1560 | || defined __sh__ \ |
808 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
1561 | || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ |
809 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
1562 | || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ |
810 | || defined __aarch64__ |
1563 | || defined __aarch64__ |
811 | #define ECB_STDFP 1 |
1564 | #define ECB_STDFP 1 |
812 | #include <string.h> /* for memcpy */ |
|
|
813 | #else |
1565 | #else |
814 | #define ECB_STDFP 0 |
1566 | #define ECB_STDFP 0 |
815 | #endif |
1567 | #endif |
816 | |
1568 | |
817 | #ifndef ECB_NO_LIBM |
1569 | #ifndef ECB_NO_LIBM |