1 /* 2 * Copyright (c) 2016-present, Yann Collet, Facebook, Inc. 3 * All rights reserved. 4 * 5 * This source code is licensed under both the BSD-style license (found in the 6 * LICENSE file in the root directory of this source tree) and the GPLv2 (found 7 * in the COPYING file in the root directory of this source tree). 8 * You may select, at your option, one of the above-listed licenses. 9 */ 10 11 #ifndef MEM_H_MODULE 12 #define MEM_H_MODULE 13 14 #if defined (__cplusplus) 15 extern "C" { 16 #endif 17 18 /*-**************************************** 19 * Dependencies 20 ******************************************/ 21 #include <stddef.h> /* size_t, ptrdiff_t */ 22 #include <string.h> /* memcpy */ 23 24 25 /*-**************************************** 26 * Compiler specifics 27 ******************************************/ 28 #if defined(_MSC_VER) /* Visual Studio */ 29 # include <stdlib.h> /* _byteswap_ulong */ 30 # include <intrin.h> /* _byteswap_* */ 31 #endif 32 #if defined(__GNUC__) 33 # define MEM_STATIC static __inline __attribute__((unused)) 34 #elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) 35 # define MEM_STATIC static inline 36 #elif defined(_MSC_VER) 37 # define MEM_STATIC static __inline 38 #else 39 # define MEM_STATIC static /* this version may generate warnings for unused static functions; disable the relevant warning */ 40 #endif 41 42 #ifndef __has_builtin 43 # define __has_builtin(x) 0 /* compat. with non-clang compilers */ 44 #endif 45 46 /* code only tested on 32 and 64 bits systems */ 47 #define MEM_STATIC_ASSERT(c) { enum { MEM_static_assert = 1/(int)(!!(c)) }; } 48 MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); } 49 50 51 /*-************************************************************** 52 * Basic Types 53 *****************************************************************/ 54 #if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) ) 55 # include <stdint.h> 56 typedef uint8_t BYTE; 57 typedef uint16_t U16; 58 typedef int16_t S16; 59 typedef uint32_t U32; 60 typedef int32_t S32; 61 typedef uint64_t U64; 62 typedef int64_t S64; 63 #else 64 # include <limits.h> 65 #if CHAR_BIT != 8 66 # error "this implementation requires char to be exactly 8-bit type" 67 #endif 68 typedef unsigned char BYTE; 69 #if USHRT_MAX != 65535 70 # error "this implementation requires short to be exactly 16-bit type" 71 #endif 72 typedef unsigned short U16; 73 typedef signed short S16; 74 #if UINT_MAX != 4294967295 75 # error "this implementation requires int to be exactly 32-bit type" 76 #endif 77 typedef unsigned int U32; 78 typedef signed int S32; 79 /* note : there are no limits defined for long long type in C90. 80 * limits exist in C99, however, in such case, <stdint.h> is preferred */ 81 typedef unsigned long long U64; 82 typedef signed long long S64; 83 #endif 84 85 86 /*-************************************************************** 87 * Memory I/O 88 *****************************************************************/ 89 /* MEM_FORCE_MEMORY_ACCESS : 90 * By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable. 91 * Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal. 92 * The below switch allow to select different access method for improved performance. 93 * Method 0 (default) : use `memcpy()`. Safe and portable. 94 * Method 1 : `__packed` statement. It depends on compiler extension (i.e., not portable). 95 * This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`. 96 * Method 2 : direct access. This method is portable but violate C standard. 97 * It can generate buggy code on targets depending on alignment. 98 * In some circumstances, it's the only known way to get the most performance (i.e. GCC + ARMv6) 99 * See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details. 100 * Prefer these methods in priority order (0 > 1 > 2) 101 */ 102 #ifndef MEM_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */ 103 # if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) ) 104 # define MEM_FORCE_MEMORY_ACCESS 2 105 # elif defined(__INTEL_COMPILER) || defined(__GNUC__) 106 # define MEM_FORCE_MEMORY_ACCESS 1 107 # endif 108 #endif 109 110 MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; } 111 MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; } 112 113 MEM_STATIC unsigned MEM_isLittleEndian(void) 114 { 115 const union { U32 u; BYTE c[4]; } one = { 1 }; /* don't use static : performance detrimental */ 116 return one.c[0]; 117 } 118 119 #if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2) 120 121 /* violates C standard, by lying on structure alignment. 122 Only use if no other choice to achieve best performance on target platform */ 123 MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; } 124 MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; } 125 MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; } 126 MEM_STATIC size_t MEM_readST(const void* memPtr) { return *(const size_t*) memPtr; } 127 128 MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; } 129 MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; } 130 MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; } 131 132 #elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1) 133 134 /* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */ 135 /* currently only defined for gcc and icc */ 136 #if defined(_MSC_VER) || (defined(__INTEL_COMPILER) && defined(WIN32)) 137 __pragma( pack(push, 1) ) 138 typedef struct { U16 v; } unalign16; 139 typedef struct { U32 v; } unalign32; 140 typedef struct { U64 v; } unalign64; 141 typedef struct { size_t v; } unalignArch; 142 __pragma( pack(pop) ) 143 #else 144 typedef struct { U16 v; } __attribute__((packed)) unalign16; 145 typedef struct { U32 v; } __attribute__((packed)) unalign32; 146 typedef struct { U64 v; } __attribute__((packed)) unalign64; 147 typedef struct { size_t v; } __attribute__((packed)) unalignArch; 148 #endif 149 150 MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign16*)ptr)->v; } 151 MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign32*)ptr)->v; } 152 MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign64*)ptr)->v; } 153 MEM_STATIC size_t MEM_readST(const void* ptr) { return ((const unalignArch*)ptr)->v; } 154 155 MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign16*)memPtr)->v = value; } 156 MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign32*)memPtr)->v = value; } 157 MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign64*)memPtr)->v = value; } 158 159 #else 160 161 /* default method, safe and standard. 162 can sometimes prove slower */ 163 164 MEM_STATIC U16 MEM_read16(const void* memPtr) 165 { 166 U16 val; memcpy(&val, memPtr, sizeof(val)); return val; 167 } 168 169 MEM_STATIC U32 MEM_read32(const void* memPtr) 170 { 171 U32 val; memcpy(&val, memPtr, sizeof(val)); return val; 172 } 173 174 MEM_STATIC U64 MEM_read64(const void* memPtr) 175 { 176 U64 val; memcpy(&val, memPtr, sizeof(val)); return val; 177 } 178 179 MEM_STATIC size_t MEM_readST(const void* memPtr) 180 { 181 size_t val; memcpy(&val, memPtr, sizeof(val)); return val; 182 } 183 184 MEM_STATIC void MEM_write16(void* memPtr, U16 value) 185 { 186 memcpy(memPtr, &value, sizeof(value)); 187 } 188 189 MEM_STATIC void MEM_write32(void* memPtr, U32 value) 190 { 191 memcpy(memPtr, &value, sizeof(value)); 192 } 193 194 MEM_STATIC void MEM_write64(void* memPtr, U64 value) 195 { 196 memcpy(memPtr, &value, sizeof(value)); 197 } 198 199 #endif /* MEM_FORCE_MEMORY_ACCESS */ 200 201 MEM_STATIC U32 MEM_swap32(U32 in) 202 { 203 #if (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \ 204 || (defined(__clang__) && __has_builtin(__builtin_bswap32)) 205 return __builtin_bswap32(in); 206 #else 207 return ((in << 24) & 0xff000000 ) | 208 ((in << 8) & 0x00ff0000 ) | 209 ((in >> 8) & 0x0000ff00 ) | 210 ((in >> 24) & 0x000000ff ); 211 #endif 212 } 213 214 MEM_STATIC U64 MEM_swap64(U64 in) 215 { 216 #if (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \ 217 || (defined(__clang__) && __has_builtin(__builtin_bswap64)) 218 return __builtin_bswap64(in); 219 #else 220 return ((in << 56) & 0xff00000000000000ULL) | 221 ((in << 40) & 0x00ff000000000000ULL) | 222 ((in << 24) & 0x0000ff0000000000ULL) | 223 ((in << 8) & 0x000000ff00000000ULL) | 224 ((in >> 8) & 0x00000000ff000000ULL) | 225 ((in >> 24) & 0x0000000000ff0000ULL) | 226 ((in >> 40) & 0x000000000000ff00ULL) | 227 ((in >> 56) & 0x00000000000000ffULL); 228 #endif 229 } 230 231 MEM_STATIC size_t MEM_swapST(size_t in) 232 { 233 if (MEM_32bits()) 234 return (size_t)MEM_swap32((U32)in); 235 else 236 return (size_t)MEM_swap64((U64)in); 237 } 238 239 /*=== Little endian r/w ===*/ 240 241 MEM_STATIC U16 MEM_readLE16(const void* memPtr) 242 { 243 if (MEM_isLittleEndian()) 244 return MEM_read16(memPtr); 245 else { 246 const BYTE* p = (const BYTE*)memPtr; 247 return (U16)(p[0] + (p[1]<<8)); 248 } 249 } 250 251 MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val) 252 { 253 if (MEM_isLittleEndian()) { 254 MEM_write16(memPtr, val); 255 } else { 256 BYTE* p = (BYTE*)memPtr; 257 p[0] = (BYTE)val; 258 p[1] = (BYTE)(val>>8); 259 } 260 } 261 262 MEM_STATIC U32 MEM_readLE24(const void* memPtr) 263 { 264 return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16); 265 } 266 267 MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val) 268 { 269 MEM_writeLE16(memPtr, (U16)val); 270 ((BYTE*)memPtr)[2] = (BYTE)(val>>16); 271 } 272 273 MEM_STATIC U32 MEM_readLE32(const void* memPtr) 274 { 275 if (MEM_isLittleEndian()) 276 return MEM_read32(memPtr); 277 else 278 return MEM_swap32(MEM_read32(memPtr)); 279 } 280 281 MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32) 282 { 283 if (MEM_isLittleEndian()) 284 MEM_write32(memPtr, val32); 285 else 286 MEM_write32(memPtr, MEM_swap32(val32)); 287 } 288 289 MEM_STATIC U64 MEM_readLE64(const void* memPtr) 290 { 291 if (MEM_isLittleEndian()) 292 return MEM_read64(memPtr); 293 else 294 return MEM_swap64(MEM_read64(memPtr)); 295 } 296 297 MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64) 298 { 299 if (MEM_isLittleEndian()) 300 MEM_write64(memPtr, val64); 301 else 302 MEM_write64(memPtr, MEM_swap64(val64)); 303 } 304 305 MEM_STATIC size_t MEM_readLEST(const void* memPtr) 306 { 307 if (MEM_32bits()) 308 return (size_t)MEM_readLE32(memPtr); 309 else 310 return (size_t)MEM_readLE64(memPtr); 311 } 312 313 MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val) 314 { 315 if (MEM_32bits()) 316 MEM_writeLE32(memPtr, (U32)val); 317 else 318 MEM_writeLE64(memPtr, (U64)val); 319 } 320 321 /*=== Big endian r/w ===*/ 322 323 MEM_STATIC U32 MEM_readBE32(const void* memPtr) 324 { 325 if (MEM_isLittleEndian()) 326 return MEM_swap32(MEM_read32(memPtr)); 327 else 328 return MEM_read32(memPtr); 329 } 330 331 MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32) 332 { 333 if (MEM_isLittleEndian()) 334 MEM_write32(memPtr, MEM_swap32(val32)); 335 else 336 MEM_write32(memPtr, val32); 337 } 338 339 MEM_STATIC U64 MEM_readBE64(const void* memPtr) 340 { 341 if (MEM_isLittleEndian()) 342 return MEM_swap64(MEM_read64(memPtr)); 343 else 344 return MEM_read64(memPtr); 345 } 346 347 MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64) 348 { 349 if (MEM_isLittleEndian()) 350 MEM_write64(memPtr, MEM_swap64(val64)); 351 else 352 MEM_write64(memPtr, val64); 353 } 354 355 MEM_STATIC size_t MEM_readBEST(const void* memPtr) 356 { 357 if (MEM_32bits()) 358 return (size_t)MEM_readBE32(memPtr); 359 else 360 return (size_t)MEM_readBE64(memPtr); 361 } 362 363 MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val) 364 { 365 if (MEM_32bits()) 366 MEM_writeBE32(memPtr, (U32)val); 367 else 368 MEM_writeBE64(memPtr, (U64)val); 369 } 370 371 372 #if defined (__cplusplus) 373 } 374 #endif 375 376 #endif /* MEM_H_MODULE */ 377