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

Comparing libecb/ecb.h (file contents):
Revision 1.167 by root, Sat Nov 21 16:53:50 2015 UTC vs.
Revision 1.189 by root, Mon Jun 21 21:26:48 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
… …
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 }
… …
155/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */ 182/* http://www-01.ibm.com/support/knowledgecenter/SSGH3R_13.1.0/com.ibm.xlcpp131.aix.doc/compiler_ref/compiler_builtins.html */
156#if __xlC__ && ECB_CPP 183#if __xlC__ && ECB_CPP
157 #include <builtins.h> 184 #include <builtins.h>
158#endif 185#endif
159 186
187#if 1400 <= _MSC_VER
188 #include <intrin.h> /* fence functions _ReadBarrier, also bit search functions _BitScanReverse */
189#endif
190
160#ifndef ECB_MEMORY_FENCE 191#ifndef ECB_MEMORY_FENCE
161 #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")
162 #if __i386 || __i386__ 194 #if __i386 || __i386__
163 #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")
164 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 196 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
165 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 197 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
166 #elif ECB_GCC_AMD64 198 #elif ECB_GCC_AMD64
167 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory") 199 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("mfence" : : : "memory")
168 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory") 200 #define ECB_MEMORY_FENCE_ACQUIRE __asm__ __volatile__ ("" : : : "memory")
169 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("") 201 #define ECB_MEMORY_FENCE_RELEASE __asm__ __volatile__ ("" : : : "memory")
170 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__ 202 #elif __powerpc__ || __ppc__ || __powerpc64__ || __ppc64__
171 #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 */
172 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \ 211 #elif defined __ARM_ARCH_6__ || defined __ARM_ARCH_6J__ \
173 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ 212 || defined __ARM_ARCH_6K__ || defined __ARM_ARCH_6ZK__ \
213 || defined __ARM_ARCH_6T2__
174 #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")
175 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \ 215 #elif defined __ARM_ARCH_7__ || defined __ARM_ARCH_7A__ \
176 || defined __ARM_ARCH_7M__ || defined __ARM_ARCH_7R__ 216 || defined __ARM_ARCH_7R__ || defined __ARM_ARCH_7M__
177 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory") 217 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb" : : : "memory")
178 #elif __aarch64__ 218 #elif __aarch64__
179 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory") 219 #define ECB_MEMORY_FENCE __asm__ __volatile__ ("dmb ish" : : : "memory")
180 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8) 220 #elif (__sparc || __sparc__) && !(__sparc_v8__ || defined __sparcv8)
181 #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")
… …
208 #if ECB_GCC_VERSION(4,7) 248 #if ECB_GCC_VERSION(4,7)
209 /* see comment below (stdatomic.h) about the C11 memory model. */ 249 /* see comment below (stdatomic.h) about the C11 memory model. */
210 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST) 250 #define ECB_MEMORY_FENCE __atomic_thread_fence (__ATOMIC_SEQ_CST)
211 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE) 251 #define ECB_MEMORY_FENCE_ACQUIRE __atomic_thread_fence (__ATOMIC_ACQUIRE)
212 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE) 252 #define ECB_MEMORY_FENCE_RELEASE __atomic_thread_fence (__ATOMIC_RELEASE)
253 #define ECB_MEMORY_FENCE_RELAXED __atomic_thread_fence (__ATOMIC_RELAXED)
213 254
214 #elif ECB_CLANG_EXTENSION(c_atomic) 255 #elif ECB_CLANG_EXTENSION(c_atomic)
215 /* see comment below (stdatomic.h) about the C11 memory model. */ 256 /* see comment below (stdatomic.h) about the C11 memory model. */
216 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST) 257 #define ECB_MEMORY_FENCE __c11_atomic_thread_fence (__ATOMIC_SEQ_CST)
217 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE) 258 #define ECB_MEMORY_FENCE_ACQUIRE __c11_atomic_thread_fence (__ATOMIC_ACQUIRE)
218 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE) 259 #define ECB_MEMORY_FENCE_RELEASE __c11_atomic_thread_fence (__ATOMIC_RELEASE)
260 #define ECB_MEMORY_FENCE_RELAXED __c11_atomic_thread_fence (__ATOMIC_RELAXED)
219 261
220 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__ 262 #elif ECB_GCC_VERSION(4,4) || defined __INTEL_COMPILER || defined __clang__
221 #define ECB_MEMORY_FENCE __sync_synchronize () 263 #define ECB_MEMORY_FENCE __sync_synchronize ()
222 #elif _MSC_VER >= 1500 /* VC++ 2008 */ 264 #elif _MSC_VER >= 1500 /* VC++ 2008 */
223 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */ 265 /* apparently, microsoft broke all the memory barrier stuff in Visual Studio 2008... */
… …
233 #elif defined _WIN32 275 #elif defined _WIN32
234 #include <WinNT.h> 276 #include <WinNT.h>
235 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */ 277 #define ECB_MEMORY_FENCE MemoryBarrier () /* actually just xchg on x86... scary */
236 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110 278 #elif __SUNPRO_C >= 0x5110 || __SUNPRO_CC >= 0x5110
237 #include <mbarrier.h> 279 #include <mbarrier.h>
238 #define ECB_MEMORY_FENCE __machine_rw_barrier () 280 #define ECB_MEMORY_FENCE __machine_rw_barrier ()
239 #define ECB_MEMORY_FENCE_ACQUIRE __machine_r_barrier () 281 #define ECB_MEMORY_FENCE_ACQUIRE __machine_acq_barrier ()
240 #define ECB_MEMORY_FENCE_RELEASE __machine_w_barrier () 282 #define ECB_MEMORY_FENCE_RELEASE __machine_rel_barrier ()
283 #define ECB_MEMORY_FENCE_RELAXED __compiler_barrier ()
241 #elif __xlC__ 284 #elif __xlC__
242 #define ECB_MEMORY_FENCE __sync () 285 #define ECB_MEMORY_FENCE __sync ()
243 #endif 286 #endif
244#endif 287#endif
245 288
246#ifndef ECB_MEMORY_FENCE 289#ifndef ECB_MEMORY_FENCE
247 #if ECB_C11 && !defined __STDC_NO_ATOMICS__ 290 #if ECB_C11 && !defined __STDC_NO_ATOMICS__
248 /* we assume that these memory fences work on all variables/all memory accesses, */ 291 /* we assume that these memory fences work on all variables/all memory accesses, */
249 /* not just C11 atomics and atomic accesses */ 292 /* not just C11 atomics and atomic accesses */
250 #include <stdatomic.h> 293 #include <stdatomic.h>
251 /* Unfortunately, neither gcc 4.7 nor clang 3.1 generate any instructions for */
252 /* any fence other than seq_cst, which isn't very efficient for us. */
253 /* Why that is, we don't know - either the C11 memory model is quite useless */
254 /* for most usages, or gcc and clang have a bug */
255 /* I *currently* lean towards the latter, and inefficiently implement */
256 /* all three of ecb's fences as a seq_cst fence */
257 /* Update, gcc-4.8 generates mfence for all c++ fences, but nothing */
258 /* for all __atomic_thread_fence's except seq_cst */
259 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst) 294 #define ECB_MEMORY_FENCE atomic_thread_fence (memory_order_seq_cst)
295 #define ECB_MEMORY_FENCE_ACQUIRE atomic_thread_fence (memory_order_acquire)
296 #define ECB_MEMORY_FENCE_RELEASE atomic_thread_fence (memory_order_release)
260 #endif 297 #endif
261#endif 298#endif
262 299
263#ifndef ECB_MEMORY_FENCE 300#ifndef ECB_MEMORY_FENCE
264 #if !ECB_AVOID_PTHREADS 301 #if !ECB_AVOID_PTHREADS
… …
284 321
285#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE 322#if !defined ECB_MEMORY_FENCE_RELEASE && defined ECB_MEMORY_FENCE
286 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE 323 #define ECB_MEMORY_FENCE_RELEASE ECB_MEMORY_FENCE
287#endif 324#endif
288 325
326#if !defined ECB_MEMORY_FENCE_RELAXED && defined ECB_MEMORY_FENCE
327 #define ECB_MEMORY_FENCE_RELAXED ECB_MEMORY_FENCE /* very heavy-handed */
328#endif
329
289/*****************************************************************************/ 330/*****************************************************************************/
290 331
291#if ECB_CPP 332#if ECB_CPP
292 #define ecb_inline static inline 333 #define ecb_inline static inline
293#elif ECB_GCC_VERSION(2,5) 334#elif ECB_GCC_VERSION(2,5)
… …
423#else 464#else
424 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x); 465 ecb_function_ ecb_const int ecb_ctz32 (uint32_t x);
425 ecb_function_ ecb_const int 466 ecb_function_ ecb_const int
426 ecb_ctz32 (uint32_t x) 467 ecb_ctz32 (uint32_t x)
427 { 468 {
469#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
470 unsigned long r;
471 _BitScanForward (&r, x);
472 return (int)r;
473#else
428 int r = 0; 474 int r = 0;
429 475
430 x &= ~x + 1; /* this isolates the lowest bit */ 476 x &= ~x + 1; /* this isolates the lowest bit */
431 477
432#if ECB_branchless_on_i386 478#if ECB_branchless_on_i386
… …
442 if (x & 0xff00ff00) r += 8; 488 if (x & 0xff00ff00) r += 8;
443 if (x & 0xffff0000) r += 16; 489 if (x & 0xffff0000) r += 16;
444#endif 490#endif
445 491
446 return r; 492 return r;
493#endif
447 } 494 }
448 495
449 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x); 496 ecb_function_ ecb_const int ecb_ctz64 (uint64_t x);
450 ecb_function_ ecb_const int 497 ecb_function_ ecb_const int
451 ecb_ctz64 (uint64_t x) 498 ecb_ctz64 (uint64_t x)
452 { 499 {
500#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
501 unsigned long r;
502 _BitScanForward64 (&r, x);
503 return (int)r;
504#else
453 int shift = x & 0xffffffffU ? 0 : 32; 505 int shift = x & 0xffffffff ? 0 : 32;
454 return ecb_ctz32 (x >> shift) + shift; 506 return ecb_ctz32 (x >> shift) + shift;
507#endif
455 } 508 }
456 509
457 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x); 510 ecb_function_ ecb_const int ecb_popcount32 (uint32_t x);
458 ecb_function_ ecb_const int 511 ecb_function_ ecb_const int
459 ecb_popcount32 (uint32_t x) 512 ecb_popcount32 (uint32_t x)
… …
467 } 520 }
468 521
469 ecb_function_ ecb_const int ecb_ld32 (uint32_t x); 522 ecb_function_ ecb_const int ecb_ld32 (uint32_t x);
470 ecb_function_ ecb_const int ecb_ld32 (uint32_t x) 523 ecb_function_ ecb_const int ecb_ld32 (uint32_t x)
471 { 524 {
525#if 1400 <= _MSC_VER && (_M_IX86 || _M_X64 || _M_IA64 || _M_ARM)
526 unsigned long r;
527 _BitScanReverse (&r, x);
528 return (int)r;
529#else
472 int r = 0; 530 int r = 0;
473 531
474 if (x >> 16) { x >>= 16; r += 16; } 532 if (x >> 16) { x >>= 16; r += 16; }
475 if (x >> 8) { x >>= 8; r += 8; } 533 if (x >> 8) { x >>= 8; r += 8; }
476 if (x >> 4) { x >>= 4; r += 4; } 534 if (x >> 4) { x >>= 4; r += 4; }
477 if (x >> 2) { x >>= 2; r += 2; } 535 if (x >> 2) { x >>= 2; r += 2; }
478 if (x >> 1) { r += 1; } 536 if (x >> 1) { r += 1; }
479 537
480 return r; 538 return r;
539#endif
481 } 540 }
482 541
483 ecb_function_ ecb_const int ecb_ld64 (uint64_t x); 542 ecb_function_ ecb_const int ecb_ld64 (uint64_t x);
484 ecb_function_ ecb_const int ecb_ld64 (uint64_t x) 543 ecb_function_ ecb_const int ecb_ld64 (uint64_t x)
485 { 544 {
545#if 1400 <= _MSC_VER && (_M_X64 || _M_IA64 || _M_ARM)
546 unsigned long r;
547 _BitScanReverse64 (&r, x);
548 return (int)r;
549#else
486 int r = 0; 550 int r = 0;
487 551
488 if (x >> 32) { x >>= 32; r += 32; } 552 if (x >> 32) { x >>= 32; r += 32; }
489 553
490 return r + ecb_ld32 (x); 554 return r + ecb_ld32 (x);
555#endif
491 } 556 }
492#endif 557#endif
493 558
494ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x); 559ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x);
495ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); } 560ecb_function_ ecb_const ecb_bool ecb_is_pot32 (uint32_t x) { return !(x & (x - 1)); }
… …
551ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); } 616ecb_inline ecb_const uint32_t ecb_rotl32 (uint32_t x, unsigned int count) { return (x >> (32 - count)) | (x << count); }
552ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); } 617ecb_inline ecb_const uint32_t ecb_rotr32 (uint32_t x, unsigned int count) { return (x << (32 - count)) | (x >> count); }
553ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); } 618ecb_inline ecb_const uint64_t ecb_rotl64 (uint64_t x, unsigned int count) { return (x >> (64 - count)) | (x << count); }
554ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); } 619ecb_inline ecb_const uint64_t ecb_rotr64 (uint64_t x, unsigned int count) { return (x << (64 - count)) | (x >> count); }
555 620
621#if ECB_CPP
622
623inline uint8_t ecb_ctz (uint8_t v) { return ecb_ctz32 (v); }
624inline uint16_t ecb_ctz (uint16_t v) { return ecb_ctz32 (v); }
625inline uint32_t ecb_ctz (uint32_t v) { return ecb_ctz32 (v); }
626inline uint64_t ecb_ctz (uint64_t v) { return ecb_ctz64 (v); }
627
628inline bool ecb_is_pot (uint8_t v) { return ecb_is_pot32 (v); }
629inline bool ecb_is_pot (uint16_t v) { return ecb_is_pot32 (v); }
630inline bool ecb_is_pot (uint32_t v) { return ecb_is_pot32 (v); }
631inline bool ecb_is_pot (uint64_t v) { return ecb_is_pot64 (v); }
632
633inline int ecb_ld (uint8_t v) { return ecb_ld32 (v); }
634inline int ecb_ld (uint16_t v) { return ecb_ld32 (v); }
635inline int ecb_ld (uint32_t v) { return ecb_ld32 (v); }
636inline int ecb_ld (uint64_t v) { return ecb_ld64 (v); }
637
638inline int ecb_popcount (uint8_t v) { return ecb_popcount32 (v); }
639inline int ecb_popcount (uint16_t v) { return ecb_popcount32 (v); }
640inline int ecb_popcount (uint32_t v) { return ecb_popcount32 (v); }
641inline int ecb_popcount (uint64_t v) { return ecb_popcount64 (v); }
642
643inline uint8_t ecb_bitrev (uint8_t v) { return ecb_bitrev8 (v); }
644inline uint16_t ecb_bitrev (uint16_t v) { return ecb_bitrev16 (v); }
645inline uint32_t ecb_bitrev (uint32_t v) { return ecb_bitrev32 (v); }
646
647inline uint8_t ecb_rotl (uint8_t v, unsigned int count) { return ecb_rotl8 (v, count); }
648inline uint16_t ecb_rotl (uint16_t v, unsigned int count) { return ecb_rotl16 (v, count); }
649inline uint32_t ecb_rotl (uint32_t v, unsigned int count) { return ecb_rotl32 (v, count); }
650inline uint64_t ecb_rotl (uint64_t v, unsigned int count) { return ecb_rotl64 (v, count); }
651
652inline uint8_t ecb_rotr (uint8_t v, unsigned int count) { return ecb_rotr8 (v, count); }
653inline uint16_t ecb_rotr (uint16_t v, unsigned int count) { return ecb_rotr16 (v, count); }
654inline uint32_t ecb_rotr (uint32_t v, unsigned int count) { return ecb_rotr32 (v, count); }
655inline uint64_t ecb_rotr (uint64_t v, unsigned int count) { return ecb_rotr64 (v, count); }
656
657#endif
658
556#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64)) 659#if ECB_GCC_VERSION(4,3) || (ECB_CLANG_BUILTIN(__builtin_bswap32) && ECB_CLANG_BUILTIN(__builtin_bswap64))
557 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16) 660 #if ECB_GCC_VERSION(4,8) || ECB_CLANG_BUILTIN(__builtin_bswap16)
558 #define ecb_bswap16(x) __builtin_bswap16 (x) 661 #define ecb_bswap16(x) __builtin_bswap16 (x)
559 #else 662 #else
560 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16) 663 #define ecb_bswap16(x) (__builtin_bswap32 (x) >> 16)
… …
598#endif 701#endif
599 702
600/* try to tell the compiler that some condition is definitely true */ 703/* try to tell the compiler that some condition is definitely true */
601#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0 704#define ecb_assume(cond) if (!(cond)) ecb_unreachable (); else 0
602 705
603ecb_inline ecb_const unsigned char ecb_byteorder_helper (void); 706ecb_inline ecb_const uint32_t ecb_byteorder_helper (void);
604ecb_inline ecb_const unsigned char 707ecb_inline ecb_const uint32_t
605ecb_byteorder_helper (void) 708ecb_byteorder_helper (void)
606{ 709{
607 /* the union code still generates code under pressure in gcc, */ 710 /* the union code still generates code under pressure in gcc, */
608 /* but less than using pointers, and always seems to */ 711 /* but less than using pointers, and always seems to */
609 /* successfully return a constant. */ 712 /* successfully return a constant. */
610 /* the reason why we have this horrible preprocessor mess */ 713 /* the reason why we have this horrible preprocessor mess */
611 /* is to avoid it in all cases, at least on common architectures */ 714 /* is to avoid it in all cases, at least on common architectures */
612 /* or when using a recent enough gcc version (>= 4.6) */ 715 /* or when using a recent enough gcc version (>= 4.6) */
613#if ((__i386 || __i386__) && !__VOS__) || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64
614 return 0x44;
615#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__ 716#if (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
717 || ((__i386 || __i386__ || _M_IX86 || ECB_GCC_AMD64 || ECB_MSVC_AMD64) && !__VOS__)
718 #define ECB_LITTLE_ENDIAN 1
616 return 0x44; 719 return 0x44332211;
617#elif __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ 720#elif (defined __BYTE_ORDER__ && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__) \
721 || ((__AARCH64EB__ || __MIPSEB__ || __ARMEB__) && !__VOS__)
722 #define ECB_BIG_ENDIAN 1
618 return 0x11; 723 return 0x11223344;
619#else 724#else
620 union 725 union
621 { 726 {
727 uint8_t c[4];
622 uint32_t i; 728 uint32_t u;
623 uint8_t c;
624 } u = { 0x11223344 }; 729 } u = { 0x11, 0x22, 0x33, 0x44 };
625 return u.c; 730 return u.u;
626#endif 731#endif
627} 732}
628 733
629ecb_inline ecb_const ecb_bool ecb_big_endian (void); 734ecb_inline ecb_const ecb_bool ecb_big_endian (void);
630ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11; } 735ecb_inline ecb_const ecb_bool ecb_big_endian (void) { return ecb_byteorder_helper () == 0x11223344; }
631ecb_inline ecb_const ecb_bool ecb_little_endian (void); 736ecb_inline ecb_const ecb_bool ecb_little_endian (void);
632ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44; } 737ecb_inline ecb_const ecb_bool ecb_little_endian (void) { return ecb_byteorder_helper () == 0x44332211; }
738
739/*****************************************************************************/
740/* unaligned load/store */
741
742ecb_inline uint_fast16_t ecb_be_u16_to_host (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
743ecb_inline uint_fast32_t ecb_be_u32_to_host (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
744ecb_inline uint_fast64_t ecb_be_u64_to_host (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
745
746ecb_inline uint_fast16_t ecb_le_u16_to_host (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
747ecb_inline uint_fast32_t ecb_le_u32_to_host (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
748ecb_inline uint_fast64_t ecb_le_u64_to_host (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
749
750ecb_inline uint_fast16_t ecb_peek_u16_u (const void *ptr) { uint16_t v; memcpy (&v, ptr, sizeof (v)); return v; }
751ecb_inline uint_fast32_t ecb_peek_u32_u (const void *ptr) { uint32_t v; memcpy (&v, ptr, sizeof (v)); return v; }
752ecb_inline uint_fast64_t ecb_peek_u64_u (const void *ptr) { uint64_t v; memcpy (&v, ptr, sizeof (v)); return v; }
753
754ecb_inline uint_fast16_t ecb_peek_be_u16_u (const void *ptr) { return ecb_be_u16_to_host (ecb_peek_u16_u (ptr)); }
755ecb_inline uint_fast32_t ecb_peek_be_u32_u (const void *ptr) { return ecb_be_u32_to_host (ecb_peek_u32_u (ptr)); }
756ecb_inline uint_fast64_t ecb_peek_be_u64_u (const void *ptr) { return ecb_be_u64_to_host (ecb_peek_u64_u (ptr)); }
757
758ecb_inline uint_fast16_t ecb_peek_le_u16_u (const void *ptr) { return ecb_le_u16_to_host (ecb_peek_u16_u (ptr)); }
759ecb_inline uint_fast32_t ecb_peek_le_u32_u (const void *ptr) { return ecb_le_u32_to_host (ecb_peek_u32_u (ptr)); }
760ecb_inline uint_fast64_t ecb_peek_le_u64_u (const void *ptr) { return ecb_le_u64_to_host (ecb_peek_u64_u (ptr)); }
761
762ecb_inline uint_fast16_t ecb_host_to_be_u16 (uint_fast16_t v) { return ecb_little_endian () ? ecb_bswap16 (v) : v; }
763ecb_inline uint_fast32_t ecb_host_to_be_u32 (uint_fast32_t v) { return ecb_little_endian () ? ecb_bswap32 (v) : v; }
764ecb_inline uint_fast64_t ecb_host_to_be_u64 (uint_fast64_t v) { return ecb_little_endian () ? ecb_bswap64 (v) : v; }
765
766ecb_inline uint_fast16_t ecb_host_to_le_u16 (uint_fast16_t v) { return ecb_big_endian () ? ecb_bswap16 (v) : v; }
767ecb_inline uint_fast32_t ecb_host_to_le_u32 (uint_fast32_t v) { return ecb_big_endian () ? ecb_bswap32 (v) : v; }
768ecb_inline uint_fast64_t ecb_host_to_le_u64 (uint_fast64_t v) { return ecb_big_endian () ? ecb_bswap64 (v) : v; }
769
770ecb_inline void ecb_poke_u16_u (void *ptr, uint16_t v) { memcpy (ptr, &v, sizeof (v)); }
771ecb_inline void ecb_poke_u32_u (void *ptr, uint32_t v) { memcpy (ptr, &v, sizeof (v)); }
772ecb_inline void ecb_poke_u64_u (void *ptr, uint64_t v) { memcpy (ptr, &v, sizeof (v)); }
773
774ecb_inline void ecb_poke_be_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_be_u16 (v)); }
775ecb_inline void ecb_poke_be_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_be_u32 (v)); }
776ecb_inline void ecb_poke_be_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_be_u64 (v)); }
777
778ecb_inline void ecb_poke_le_u16_u (void *ptr, uint_fast16_t v) { ecb_poke_u16_u (ptr, ecb_host_to_le_u16 (v)); }
779ecb_inline void ecb_poke_le_u32_u (void *ptr, uint_fast32_t v) { ecb_poke_u32_u (ptr, ecb_host_to_le_u32 (v)); }
780ecb_inline void ecb_poke_le_u64_u (void *ptr, uint_fast64_t v) { ecb_poke_u64_u (ptr, ecb_host_to_le_u64 (v)); }
781
782#if ECB_CPP
783
784inline uint8_t ecb_bswap (uint8_t v) { return v; }
785inline uint16_t ecb_bswap (uint16_t v) { return ecb_bswap16 (v); }
786inline uint32_t ecb_bswap (uint32_t v) { return ecb_bswap32 (v); }
787inline uint64_t ecb_bswap (uint64_t v) { return ecb_bswap64 (v); }
788
789template<typename T> inline T ecb_be_to_host (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
790template<typename T> inline T ecb_le_to_host (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
791template<typename T> inline T ecb_peek (const void *ptr) { return *(const T *)ptr; }
792template<typename T> inline T ecb_peek_be (const void *ptr) { return ecb_be_to_host (ecb_peek <T> (ptr)); }
793template<typename T> inline T ecb_peek_le (const void *ptr) { return ecb_le_to_host (ecb_peek <T> (ptr)); }
794template<typename T> inline T ecb_peek_u (const void *ptr) { T v; memcpy (&v, ptr, sizeof (v)); return v; }
795template<typename T> inline T ecb_peek_be_u (const void *ptr) { return ecb_be_to_host (ecb_peek_u<T> (ptr)); }
796template<typename T> inline T ecb_peek_le_u (const void *ptr) { return ecb_le_to_host (ecb_peek_u<T> (ptr)); }
797
798template<typename T> inline T ecb_host_to_be (T v) { return ecb_little_endian () ? ecb_bswap (v) : v; }
799template<typename T> inline T ecb_host_to_le (T v) { return ecb_big_endian () ? ecb_bswap (v) : v; }
800template<typename T> inline void ecb_poke (void *ptr, T v) { *(T *)ptr = v; }
801template<typename T> inline void ecb_poke_be (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_be (v)); }
802template<typename T> inline void ecb_poke_le (void *ptr, T v) { return ecb_poke <T> (ptr, ecb_host_to_le (v)); }
803template<typename T> inline void ecb_poke_u (void *ptr, T v) { memcpy (ptr, &v, sizeof (v)); }
804template<typename T> inline void ecb_poke_be_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_be (v)); }
805template<typename T> inline void ecb_poke_le_u (void *ptr, T v) { return ecb_poke_u<T> (ptr, ecb_host_to_le (v)); }
806
807#endif
808
809/*****************************************************************************/
810/* division */
633 811
634#if ECB_GCC_VERSION(3,0) || ECB_C99 812#if ECB_GCC_VERSION(3,0) || ECB_C99
813 /* C99 tightened the definition of %, so we can use a more efficient version */
635 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0)) 814 #define ecb_mod(m,n) ((m) % (n) + ((m) % (n) < 0 ? (n) : 0))
636#else 815#else
637 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n))) 816 #define ecb_mod(m,n) ((m) < 0 ? ((n) - 1 - ((-1 - (m)) % (n))) : ((m) % (n)))
638#endif 817#endif
639 818
… …
650 } 829 }
651#else 830#else
652 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div)) 831 #define ecb_div_rd(val,div) ((val) < 0 ? - ((-(val) + (div) - 1) / (div)) : ((val) ) / (div))
653 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div)) 832 #define ecb_div_ru(val,div) ((val) < 0 ? - ((-(val) ) / (div)) : ((val) + (div) - 1) / (div))
654#endif 833#endif
834
835/*****************************************************************************/
836/* array length */
655 837
656#if ecb_cplusplus_does_not_suck 838#if ecb_cplusplus_does_not_suck
657 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */ 839 /* does not work for local types (http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm) */
658 template<typename T, int N> 840 template<typename T, int N>
659 static inline int ecb_array_length (const T (&arr)[N]) 841 static inline int ecb_array_length (const T (&arr)[N])
… …
662 } 844 }
663#else 845#else
664 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0])) 846 #define ecb_array_length(name) (sizeof (name) / sizeof (name [0]))
665#endif 847#endif
666 848
849/*****************************************************************************/
850/* IEEE 754-2008 half float conversions */
851
667ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint16_t x); 852ecb_function_ ecb_const uint32_t ecb_binary16_to_binary32 (uint32_t x);
668ecb_function_ ecb_const uint32_t 853ecb_function_ ecb_const uint32_t
669ecb_binary16_to_binary32 (uint16_t x) 854ecb_binary16_to_binary32 (uint32_t x)
670{ 855{
671 unsigned int s = (x & 0x8000) << (31 - 15); 856 unsigned int s = (x & 0x8000) << (31 - 15);
672 int e = (x >> 10) & 0x001f; 857 int e = (x >> 10) & 0x001f;
673 unsigned int m = x & 0x03ff; 858 unsigned int m = x & 0x03ff;
674 859
… …
699ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x); 884ecb_function_ ecb_const uint16_t ecb_binary32_to_binary16 (uint32_t x);
700ecb_function_ ecb_const uint16_t 885ecb_function_ ecb_const uint16_t
701ecb_binary32_to_binary16 (uint32_t x) 886ecb_binary32_to_binary16 (uint32_t x)
702{ 887{
703 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */ 888 unsigned int s = (x >> 16) & 0x00008000; /* sign bit, the easy part */
704 unsigned int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */ 889 int e = ((x >> 23) & 0x000000ff) - (127 - 15); /* the desired exponent */
705 unsigned int m = x & 0x007fffff; 890 unsigned int m = x & 0x007fffff;
706 891
707 x &= 0x7fffffff; 892 x &= 0x7fffffff;
708 893
709 /* if it's within range of binary16 normals, use fast path */ 894 /* if it's within range of binary16 normals, use fast path */
… …
724 909
725 /* handle large numbers and infinity */ 910 /* handle large numbers and infinity */
726 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000)) 911 if (ecb_expect_true (0x477fefff < x && x <= 0x7f800000))
727 return s | 0x7c00; 912 return s | 0x7c00;
728 913
729 /* handle zero and subnormals */ 914 /* handle zero, subnormals and small numbers */
730 if (ecb_expect_true (x < 0x38800000)) 915 if (ecb_expect_true (x < 0x38800000))
731 { 916 {
732 /* zero */ 917 /* zero */
733 if (ecb_expect_true (!x)) 918 if (ecb_expect_true (!x))
734 return s; 919 return s;
735 920
736 /* handle subnormals */ 921 /* handle subnormals */
922
923 /* too small, will be zero */
924 if (e < (14 - 24)) /* might not be sharp, but is good enough */
925 return s;
737 926
738 m |= 0x00800000; /* make implicit bit explicit */ 927 m |= 0x00800000; /* make implicit bit explicit */
739 928
740 /* very tricky - we need to round to the nearest e (+10) bit value */ 929 /* very tricky - we need to round to the nearest e (+10) bit value */
741 { 930 {
… …
775 || defined __sh__ \ 964 || defined __sh__ \
776 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \ 965 || defined _M_IX86 || defined ECB_MSVC_AMD64 || defined _M_IA64 \
777 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \ 966 || (defined __arm__ && (defined __ARM_EABI__ || defined __EABI__ || defined __VFP_FP__ || defined _WIN32_WCE || defined __ANDROID__)) \
778 || defined __aarch64__ 967 || defined __aarch64__
779 #define ECB_STDFP 1 968 #define ECB_STDFP 1
780 #include <string.h> /* for memcpy */
781#else 969#else
782 #define ECB_STDFP 0 970 #define ECB_STDFP 0
783#endif 971#endif
784 972
785#ifndef ECB_NO_LIBM 973#ifndef ECB_NO_LIBM

Diff Legend

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