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

Comparing libeio/ecb.h (file contents):
Revision 1.27 by root, Mon Jan 18 11:53:09 2016 UTC vs.
Revision 1.32 by root, Tue Jul 27 07:58:39 2021 UTC

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 0x00010009
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
… …
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)
81 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
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")
176 #elif defined __ARM_ARCH_2__ \ 204 #elif defined __ARM_ARCH_2__ \
177 || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \ 205 || defined __ARM_ARCH_3__ || defined __ARM_ARCH_3M__ \
178 || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \ 206 || defined __ARM_ARCH_4__ || defined __ARM_ARCH_4T__ \
… …
220 #if ECB_GCC_VERSION(4,7) 248 #if ECB_GCC_VERSION(4,7)
221 /* see comment below (stdatomic.h) about the C11 memory model. */ 249 /* see comment below (stdatomic.h) about the C11 memory model. */
222 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 250 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
223 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) 251 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
224 #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)
225 255
226 #elif ECB_CLANG_EXTENSION(c_atomic) 256 #elif ECB_CLANG_EXTENSION(c_atomic)
227 /* see comment below (stdatomic.h) about the C11 memory model. */ 257 /* see comment below (stdatomic.h) about the C11 memory model. */
228 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 258 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
229 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) 259 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
230 #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)
231 263
232 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 264 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
233 #define ECB_MEMORY_FENCE __sync_synchronize () 265 #define ECB_MEMORY_FENCE __sync_synchronize ()
234 #elif _MSC_VER >= 1500 /* VC++ 2008 */ 266 #elif _MSC_VER >= 1500 /* VC++ 2008 */
235 /* 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... */
… …
245 #elif defined _WIN32 277 #elif defined _WIN32
246 #include <WinNT.h> 278 #include <WinNT.h>
247 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 279 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
248 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 280 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
249 #include <mbarrier.h> 281 #include <mbarrier.h>
250 #define ECB_MEMORY_FENCE __machine_rw_barrier () 282 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
251 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () 283 #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
252 #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 ()
253 #elif __xlC__ 286 #elif __xlC__
254 #define ECB_MEMORY_FENCE __sync () 287 #define ECB_MEMORY_FENCE __sync ()
255 #endif 288 #endif
256#endif 289#endif
257 290
258#ifndef ECB_MEMORY_FENCE 291#ifndef ECB_MEMORY_FENCE
259 #if ECB_C11 && !defined __STDC_NO_ATOMICS__ 292 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
260 /* 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, */
261 /* not just C11 atomics and atomic accesses */ 294 /* not just C11 atomics and atomic accesses */
262 #include <stdatomic.h> 295 #include <stdatomic.h>
263 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
264 /* any fence other than seq_cst, which isn't very efficient for us. */
265 /* Why that is, we don't know - either the C11 memory model is quite useless */
266 /* for most usages, or gcc and clang have a bug */
267 /* I *currently* lean towards the latter, and inefficiently implement */
268 /* all three of ecb's fences as a seq_cst fence */
269 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
270 /* for all __atomic_thread_fence's except seq_cst */
271 #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)
272 #endif 299 #endif
273#endif 300#endif
274 301
275#ifndef ECB_MEMORY_FENCE 302#ifndef ECB_MEMORY_FENCE
276 #if !ECB_AVOID_PTHREADS 303 #if !ECB_AVOID_PTHREADS
… …
296 323
297#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE 324#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
298 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 325 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
299#endif 326#endif
300 327
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
301/*****************************************************************************/ 332/*****************************************************************************/
302 333
303#if ECB_CPP 334#if ECB_CPP
304 #define ecb_inline static inline 335 #define ecb_inline static inline
305#elif ECB_GCC_VERSION(2,5) 336#elif ECB_GCC_VERSION(2,5)
… …
586ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); } 617ecb_inline ecb_const uint16_t ecb_rotr16 (uint16_t x, unsigned int count) { return (x << (16 - count)) | (x >> count); }
587ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 618ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
588ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 619ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
589ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 620ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
590ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 621ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
622
623#if ECB_CPP
624
625inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
626inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
627inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
628inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
629
630inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
631inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
632inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
633inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
634
635inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
636inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
637inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
638inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
639
640inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
641inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
642inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
643inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
644
645inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
646inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
647inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
648
649inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
650inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
651inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
652inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
653
654inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
655inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
656inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
657inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
658
659#endif
591 660
592#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) 661#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
593 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) 662 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
594 #define ecb_bswap16(x) __builtin_bswap16 (x) 663 #define ecb_bswap16(x) __builtin_bswap16 (x)
595 #else 664 #else
… …
667ecb_inline ecb_const ecb_bool ecb_big_endian (void); 736ecb_inline ecb_const ecb_bool ecb_big_endian (void);
668ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; } 737ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
669ecb_inline ecb_const ecb_bool ecb_little_endian (void); 738ecb_inline ecb_const ecb_bool ecb_little_endian (void);
670ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; } 739ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
671 740
741/*****************************************************************************/
742/* unaligned load/store */
743
744ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
745ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
746ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
747
748ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
749ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
750ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
751
752ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
753ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
754ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
755
756ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
757ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
758ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
759
760ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
761ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
762ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
763
764ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
765ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
766ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
767
768ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
769ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
770ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
771
772ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
773ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
774ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
775
776ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
777ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
778ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
779
780ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
781ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
782ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
783
784#if ECB_CPP
785
786inline uint8_t ecb_bswap (uint8_t v) { return v; }
787inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
788inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
789inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
790
791template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
792template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
793template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
794template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
795template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
796template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
797template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
798template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
799
800template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
801template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
802template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
803template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
804template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
805template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
806template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
807template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
808
809#endif
810
811/*****************************************************************************/
812/* division */
813
672#if ECB_GCC_VERSION(3,0) || ECB_C99 814#if ECB_GCC_VERSION(3,0) || ECB_C99
815 /* C99 tightened the definition of %, so we can use a more efficient version */
673 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 816 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
674#else 817#else
675 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 818 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
676#endif 819#endif
677 820
… …
688 } 831 }
689#else 832#else
690 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) 833 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
691 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) 834 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
692#endif 835#endif
836
837/*****************************************************************************/
838/* array length */
693 839
694#if ecb_cplusplus_does_not_suck 840#if ecb_cplusplus_does_not_suck
695 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 841 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
696 template<typename T, int N> 842 template<typename T, int N>
697 static inline int ecb_array_length (const T (&arr)[N]) 843 static inline int ecb_array_length (const T (&arr)[N])
… …
699 return N; 845 return N;
700 } 846 }
701#else 847#else
702 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 848 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
703#endif 849#endif
850
851/*****************************************************************************/
852/* IEEE 754-2008 half float conversions */
704 853
705ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x); 854ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
706ecb_function_ ecb_const uint32_t 855ecb_function_ ecb_const uint32_t
707ecb_binary16_to_binary32 (uint32_t x) 856ecb_binary16_to_binary32 (uint32_t x)
708{ 857{
… …
737ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); 886ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
738ecb_function_ ecb_const uint16_t 887ecb_function_ ecb_const uint16_t
739ecb_binary32_to_binary16 (uint32_t x) 888ecb_binary32_to_binary16 (uint32_t x)
740{ 889{
741 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ 890 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
742 unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ 891 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
743 unsigned int m = x & 0x007fffff; 892 unsigned int m = x & 0x007fffff;
744 893
745 x &= 0x7fffffff; 894 x &= 0x7fffffff;
746 895
747 /* if it's within range of binary16 normals, use fast path */ 896 /* if it's within range of binary16 normals, use fast path */
… …
794 943
795 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */ 944 /* handle NaNs, preserve leftmost nan bits, but make sure we don't turn them into infinities */
796 m >>= 13; 945 m >>= 13;
797 946
798 return s | 0x7c00 | m | !m; 947 return s | 0x7c00 | m | !m;
948}
949
950/*******************************************************************************/
951/* fast integer to ascii */
952
953/*
954 * This code is pretty complicated because it is general. The idea behind it,
955 * however, is pretty simple: first, the number is multiplied with a scaling
956 * factor (2**bits / 10**(digits-1)) to convert the integer into a fixed-point
957 * number with the first digit in the upper bits.
958 * Then this digit is converted to text and masked out. The resulting number
959 * is then multiplied by 10, by multiplying the fixed point representation
960 * by 5 and shifting the (binary) decimal point one to the right, so a 4.28
961 * format becomes 5.27, 6.26 and so on.
962 * The rest involves only advancing the pointer if we already generated a
963 * non-zero digit, so leading zeroes are overwritten.
964 */
965
966// simply return a mask with "bits" bits set
967#define ecb_i2a_mask(type,bits) ((((type)1) << (bits)) - 1)
968
969// oputput a single digit. maskvalue is 10**digitidx
970#define ecb_i2a_digit(type,bits,digitmask,maskvalue,digitidx) \
971 if (digitmask >= maskvalue) /* constant, used to decide how many digits to generate */ \
972 { \
973 char digit = x >> (bits - digitidx); /* calculate the topmost digit */ \
974 *ptr = digit + '0'; /* output it */ \
975 nz = (digitmask == maskvalue) || nz || digit; /* first term == always output last digit */ \
976 ptr += nz; /* output digit only if non-zero digit seen */ \
977 x = (x & ecb_i2a_mask (type, bits - digitidx)) * 5; /* *10, but shift decimal point right */ \
978 }
979
980// convert integer to fixed point format and multiply out digits, highest first
981// requires magic constants: max. digits and number of bits after the decimal point
982#define ecb_i2a_def(suffix,ptr,v,type,bits,digitmask,lz) \
983ecb_inline char *ecb_i2a_ ## suffix (char *ptr, uint32_t u) \
984{ \
985 char nz = lz; /* non-zero digit seen? */ \
986 /* convert to x.bits fixed-point */ \
987 type x = u * ((ecb_i2a_mask (type, bits) + digitmask) / digitmask); \
988 /* output up to 10 digits */ \
989 ecb_i2a_digit (type,bits,digitmask, 1, 0); \
990 ecb_i2a_digit (type,bits,digitmask, 10, 1); \
991 ecb_i2a_digit (type,bits,digitmask, 100, 2); \
992 ecb_i2a_digit (type,bits,digitmask, 1000, 3); \
993 ecb_i2a_digit (type,bits,digitmask, 10000, 4); \
994 ecb_i2a_digit (type,bits,digitmask, 100000, 5); \
995 ecb_i2a_digit (type,bits,digitmask, 1000000, 6); \
996 ecb_i2a_digit (type,bits,digitmask, 10000000, 7); \
997 ecb_i2a_digit (type,bits,digitmask, 100000000, 8); \
998 ecb_i2a_digit (type,bits,digitmask, 1000000000, 9); \
999 return ptr; \
1000}
1001
1002// predefined versions of the above, for various digits
1003// ecb_i2a_xN = almost N digits, limit defined by macro
1004// ecb_i2a_N = up to N digits, leading zeroes suppressed
1005// ecb_i2a_0N = exactly N digits, including leading zeroes
1006
1007// non-leading-zero versions, limited range
1008#define ECB_I2A_MAX_X5 59074 // limit for ecb_i2a_x5
1009#define ECB_I2A_MAX_X10 2932500665 // limit for ecb_i2a_x10
1010ecb_i2a_def ( x5, ptr, v, uint32_t, 26, 10000, 0)
1011ecb_i2a_def (x10, ptr, v, uint64_t, 60, 1000000000, 0)
1012
1013// non-leading zero versions, all digits, 4 and 9 are optimal for 32/64 bit
1014ecb_i2a_def ( 2, ptr, v, uint32_t, 10, 10, 0)
1015ecb_i2a_def ( 3, ptr, v, uint32_t, 12, 100, 0)
1016ecb_i2a_def ( 4, ptr, v, uint32_t, 26, 1000, 0)
1017ecb_i2a_def ( 5, ptr, v, uint64_t, 30, 10000, 0)
1018ecb_i2a_def ( 6, ptr, v, uint64_t, 36, 100000, 0)
1019ecb_i2a_def ( 7, ptr, v, uint64_t, 44, 1000000, 0)
1020ecb_i2a_def ( 8, ptr, v, uint64_t, 50, 10000000, 0)
1021ecb_i2a_def ( 9, ptr, v, uint64_t, 56, 100000000, 0)
1022
1023// leading-zero versions, all digits, 04 and 09 are optimal for 32/64 bit
1024ecb_i2a_def (02, ptr, v, uint32_t, 10, 10, 1)
1025ecb_i2a_def (03, ptr, v, uint32_t, 12, 100, 1)
1026ecb_i2a_def (04, ptr, v, uint32_t, 26, 1000, 1)
1027ecb_i2a_def (05, ptr, v, uint64_t, 30, 10000, 1)
1028ecb_i2a_def (06, ptr, v, uint64_t, 36, 100000, 1)
1029ecb_i2a_def (07, ptr, v, uint64_t, 44, 1000000, 1)
1030ecb_i2a_def (08, ptr, v, uint64_t, 50, 10000000, 1)
1031ecb_i2a_def (09, ptr, v, uint64_t, 56, 100000000, 1)
1032
1033#define ECB_I2A_I32_DIGITS 11
1034#define ECB_I2A_U32_DIGITS 10
1035#define ECB_I2A_I64_DIGITS 20
1036#define ECB_I2A_U64_DIGITS 21
1037#define ECB_I2A_MAX_DIGITS 21
1038
1039ecb_inline char *
1040ecb_i2a_u32 (char *ptr, uint32_t u)
1041{
1042 #if ECB_64BIT_NATIVE
1043 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1044 ptr = ecb_i2a_x10 (ptr, u);
1045 else // x10 almost, but not fully, covers 32 bit
1046 {
1047 uint32_t u1 = u % 1000000000;
1048 uint32_t u2 = u / 1000000000;
1049
1050 *ptr++ = u2 + '0';
1051 ptr = ecb_i2a_09 (ptr, u1);
1052 }
1053 #else
1054 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1055 ecb_i2a_x5 (ptr, u);
1056 else if (ecb_expect_true (u <= ECB_I2A_MAX_X5 * 10000))
1057 {
1058 uint32_t u1 = u % 10000;
1059 uint32_t u2 = u / 10000;
1060
1061 ptr = ecb_i2a_x5 (ptr, u2);
1062 ptr = ecb_i2a_04 (ptr, u1);
1063 }
1064 else
1065 {
1066 uint32_t u1 = u % 10000;
1067 uint32_t ua = u / 10000;
1068 uint32_t u2 = ua % 10000;
1069 uint32_t u3 = ua / 10000;
1070
1071 ptr = ecb_i2a_2 (ptr, u3);
1072 ptr = ecb_i2a_04 (ptr, u2);
1073 ptr = ecb_i2a_04 (ptr, u1);
1074 }
1075 #endif
1076
1077 return ptr;
1078}
1079
1080ecb_inline char *
1081ecb_i2a_i32 (char *ptr, int32_t v)
1082{
1083 *ptr = '-'; ptr += v < 0;
1084 uint32_t u = v < 0 ? -(uint32_t)v : v;
1085
1086 #if ECB_64BIT_NATIVE
1087 ptr = ecb_i2a_x10 (ptr, u); // x10 fully covers 31 bit
1088 #else
1089 ptr = ecb_i2a_u32 (ptr, u);
1090 #endif
1091
1092 return ptr;
1093}
1094
1095ecb_inline char *
1096ecb_i2a_u64 (char *ptr, uint64_t u)
1097{
1098 #if ECB_64BIT_NATIVE
1099 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1100 ptr = ecb_i2a_x10 (ptr, u);
1101 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1102 {
1103 uint64_t u1 = u % 1000000000;
1104 uint64_t u2 = u / 1000000000;
1105
1106 ptr = ecb_i2a_x10 (ptr, u2);
1107 ptr = ecb_i2a_09 (ptr, u1);
1108 }
1109 else
1110 {
1111 uint64_t u1 = u % 1000000000;
1112 uint64_t ua = u / 1000000000;
1113 uint64_t u2 = ua % 1000000000;
1114 uint64_t u3 = ua / 1000000000;
1115
1116 ptr = ecb_i2a_2 (ptr, u3);
1117 ptr = ecb_i2a_09 (ptr, u2);
1118 ptr = ecb_i2a_09 (ptr, u1);
1119 }
1120 #else
1121 if (ecb_expect_true (u <= ECB_I2A_MAX_X5))
1122 ptr = ecb_i2a_x5 (ptr, u);
1123 else
1124 {
1125 uint64_t u1 = u % 10000;
1126 uint64_t u2 = u / 10000;
1127
1128 ptr = ecb_i2a_u64 (ptr, u2);
1129 ptr = ecb_i2a_04 (ptr, u1);
1130 }
1131 #endif
1132
1133 return ptr;
1134}
1135
1136ecb_inline char *
1137ecb_i2a_i64 (char *ptr, int64_t v)
1138{
1139 *ptr = '-'; ptr += v < 0;
1140 uint64_t u = v < 0 ? -(uint64_t)v : v;
1141
1142 #if ECB_64BIT_NATIVE
1143 if (ecb_expect_true (u <= ECB_I2A_MAX_X10))
1144 ptr = ecb_i2a_x10 (ptr, u);
1145 else if (ecb_expect_false (u <= ECB_I2A_MAX_X10 * 1000000000))
1146 {
1147 uint64_t u1 = u % 1000000000;
1148 uint64_t u2 = u / 1000000000;
1149
1150 ptr = ecb_i2a_x10 (ptr, u2);
1151 ptr = ecb_i2a_09 (ptr, u1);
1152 }
1153 else
1154 {
1155 uint64_t u1 = u % 1000000000;
1156 uint64_t ua = u / 1000000000;
1157 uint64_t u2 = ua % 1000000000;
1158 uint64_t u3 = ua / 1000000000;
1159
1160 // 2**31 is 19 digits, so the top is exactly one digit
1161 *ptr++ = u3 + '0';
1162 ptr = ecb_i2a_09 (ptr, u2);
1163 ptr = ecb_i2a_09 (ptr, u1);
1164 }
1165 #else
1166 ptr = ecb_i2a_u64 (ptr, u);
1167 #endif
1168
1169 return ptr;
799} 1170}
800 1171
801/*******************************************************************************/ 1172/*******************************************************************************/
802/* floating point stuff, can be disabled by defining ECB_NO_LIBM */ 1173/* floating point stuff, can be disabled by defining ECB_NO_LIBM */
803 1174
… …
817 || defined __sh__ \ 1188 || defined __sh__ \
818 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ 1189 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
819 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ 1190 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
820 || defined __aarch64__ 1191 || defined __aarch64__
821 #define ECB_STDFP 1 1192 #define ECB_STDFP 1
822 #include <string.h> /* for memcpy */
823#else 1193#else
824 #define ECB_STDFP 0 1194 #define ECB_STDFP 0
825#endif 1195#endif
826 1196
827#ifndef ECB_NO_LIBM 1197#ifndef ECB_NO_LIBM

Diff Legend

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