1 /* 2 * Copyright (c) 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 ZSTD_COMPILER_H 12 #define ZSTD_COMPILER_H 13 14 #include "portability_macros.h" 15 16 /*-******************************************************* 17 * Compiler specifics 18 *********************************************************/ 19 /* force inlining */ 20 21 #if !defined(ZSTD_NO_INLINE) 22 #if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */ 23 # define INLINE_KEYWORD inline 24 #else 25 # define INLINE_KEYWORD 26 #endif 27 28 #if defined(__GNUC__) || defined(__ICCARM__) 29 # define FORCE_INLINE_ATTR __attribute__((always_inline)) 30 #elif defined(_MSC_VER) 31 # define FORCE_INLINE_ATTR __forceinline 32 #else 33 # define FORCE_INLINE_ATTR 34 #endif 35 36 #else 37 38 #define INLINE_KEYWORD 39 #define FORCE_INLINE_ATTR 40 41 #endif 42 43 /** 44 On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC). 45 This explicitly marks such functions as __cdecl so that the code will still compile 46 if a CC other than __cdecl has been made the default. 47 */ 48 #if defined(_MSC_VER) 49 # define WIN_CDECL __cdecl 50 #else 51 # define WIN_CDECL 52 #endif 53 54 /** 55 * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant 56 * parameters. They must be inlined for the compiler to eliminate the constant 57 * branches. 58 */ 59 #define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR 60 /** 61 * HINT_INLINE is used to help the compiler generate better code. It is *not* 62 * used for "templates", so it can be tweaked based on the compilers 63 * performance. 64 * 65 * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the 66 * always_inline attribute. 67 * 68 * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline 69 * attribute. 70 */ 71 #if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5 72 # define HINT_INLINE static INLINE_KEYWORD 73 #else 74 # define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR 75 #endif 76 77 /* UNUSED_ATTR tells the compiler it is okay if the function is unused. */ 78 #if defined(__GNUC__) 79 # define UNUSED_ATTR __attribute__((unused)) 80 #else 81 # define UNUSED_ATTR 82 #endif 83 84 /* force no inlining */ 85 #ifdef _MSC_VER 86 # define FORCE_NOINLINE static __declspec(noinline) 87 #else 88 # if defined(__GNUC__) || defined(__ICCARM__) 89 # define FORCE_NOINLINE static __attribute__((__noinline__)) 90 # else 91 # define FORCE_NOINLINE static 92 # endif 93 #endif 94 95 96 /* target attribute */ 97 #if defined(__GNUC__) || defined(__ICCARM__) 98 # define TARGET_ATTRIBUTE(target) __attribute__((__target__(target))) 99 #else 100 # define TARGET_ATTRIBUTE(target) 101 #endif 102 103 /* Target attribute for BMI2 dynamic dispatch. 104 * Enable lzcnt, bmi, and bmi2. 105 * We test for bmi1 & bmi2. lzcnt is included in bmi1. 106 */ 107 #define BMI2_TARGET_ATTRIBUTE TARGET_ATTRIBUTE("lzcnt,bmi,bmi2") 108 109 /* prefetch 110 * can be disabled, by declaring NO_PREFETCH build macro */ 111 #if defined(NO_PREFETCH) 112 # define PREFETCH_L1(ptr) (void)(ptr) /* disabled */ 113 # define PREFETCH_L2(ptr) (void)(ptr) /* disabled */ 114 #else 115 # if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86)) /* _mm_prefetch() is not defined outside of x86/x64 */ 116 # include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */ 117 # define PREFETCH_L1(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0) 118 # define PREFETCH_L2(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T1) 119 # elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) ) 120 # define PREFETCH_L1(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */) 121 # define PREFETCH_L2(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */) 122 # elif defined(__aarch64__) 123 # define PREFETCH_L1(ptr) __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr))) 124 # define PREFETCH_L2(ptr) __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr))) 125 # else 126 # define PREFETCH_L1(ptr) (void)(ptr) /* disabled */ 127 # define PREFETCH_L2(ptr) (void)(ptr) /* disabled */ 128 # endif 129 #endif /* NO_PREFETCH */ 130 131 #define CACHELINE_SIZE 64 132 133 #define PREFETCH_AREA(p, s) { \ 134 const char* const _ptr = (const char*)(p); \ 135 size_t const _size = (size_t)(s); \ 136 size_t _pos; \ 137 for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) { \ 138 PREFETCH_L2(_ptr + _pos); \ 139 } \ 140 } 141 142 /* vectorization 143 * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax, 144 * and some compilers, like Intel ICC and MCST LCC, do not support it at all. */ 145 #if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__) && !defined(__LCC__) 146 # if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5) 147 # define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize"))) 148 # else 149 # define DONT_VECTORIZE _Pragma("GCC optimize(\"no-tree-vectorize\")") 150 # endif 151 #else 152 # define DONT_VECTORIZE 153 #endif 154 155 /* Tell the compiler that a branch is likely or unlikely. 156 * Only use these macros if it causes the compiler to generate better code. 157 * If you can remove a LIKELY/UNLIKELY annotation without speed changes in gcc 158 * and clang, please do. 159 */ 160 #if defined(__GNUC__) 161 #define LIKELY(x) (__builtin_expect((x), 1)) 162 #define UNLIKELY(x) (__builtin_expect((x), 0)) 163 #else 164 #define LIKELY(x) (x) 165 #define UNLIKELY(x) (x) 166 #endif 167 168 /* disable warnings */ 169 #ifdef _MSC_VER /* Visual Studio */ 170 # include <intrin.h> /* For Visual 2005 */ 171 # pragma warning(disable : 4100) /* disable: C4100: unreferenced formal parameter */ 172 # pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ 173 # pragma warning(disable : 4204) /* disable: C4204: non-constant aggregate initializer */ 174 # pragma warning(disable : 4214) /* disable: C4214: non-int bitfields */ 175 # pragma warning(disable : 4324) /* disable: C4324: padded structure */ 176 #endif 177 178 /*Like DYNAMIC_BMI2 but for compile time determination of BMI2 support*/ 179 #ifndef STATIC_BMI2 180 # if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86)) 181 # ifdef __AVX2__ //MSVC does not have a BMI2 specific flag, but every CPU that supports AVX2 also supports BMI2 182 # define STATIC_BMI2 1 183 # endif 184 # endif 185 #endif 186 187 #ifndef STATIC_BMI2 188 #define STATIC_BMI2 0 189 #endif 190 191 /* compile time determination of SIMD support */ 192 #if !defined(ZSTD_NO_INTRINSICS) 193 # if defined(__SSE2__) || defined(_M_AMD64) || (defined (_M_IX86) && defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) 194 # define ZSTD_ARCH_X86_SSE2 195 # endif 196 # if defined(__ARM_NEON) || defined(_M_ARM64) 197 # define ZSTD_ARCH_ARM_NEON 198 # endif 199 # 200 # if defined(ZSTD_ARCH_X86_SSE2) 201 # include <emmintrin.h> 202 # elif defined(ZSTD_ARCH_ARM_NEON) 203 # include <arm_neon.h> 204 # endif 205 #endif 206 207 /* C-language Attributes are added in C23. */ 208 #if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute) 209 # define ZSTD_HAS_C_ATTRIBUTE(x) __has_c_attribute(x) 210 #else 211 # define ZSTD_HAS_C_ATTRIBUTE(x) 0 212 #endif 213 214 /* Only use C++ attributes in C++. Some compilers report support for C++ 215 * attributes when compiling with C. 216 */ 217 #if defined(__cplusplus) && defined(__has_cpp_attribute) 218 # define ZSTD_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x) 219 #else 220 # define ZSTD_HAS_CPP_ATTRIBUTE(x) 0 221 #endif 222 223 /* Define ZSTD_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute. 224 * - C23: https://en.cppreference.com/w/c/language/attributes/fallthrough 225 * - CPP17: https://en.cppreference.com/w/cpp/language/attributes/fallthrough 226 * - Else: __attribute__((__fallthrough__)) 227 */ 228 #ifndef ZSTD_FALLTHROUGH 229 # if ZSTD_HAS_C_ATTRIBUTE(fallthrough) 230 # define ZSTD_FALLTHROUGH [[fallthrough]] 231 # elif ZSTD_HAS_CPP_ATTRIBUTE(fallthrough) 232 # define ZSTD_FALLTHROUGH [[fallthrough]] 233 # elif __has_attribute(__fallthrough__) 234 /* Leading semicolon is to satisfy gcc-11 with -pedantic. Without the semicolon 235 * gcc complains about: a label can only be part of a statement and a declaration is not a statement. 236 */ 237 # define ZSTD_FALLTHROUGH ; __attribute__((__fallthrough__)) 238 # else 239 # define ZSTD_FALLTHROUGH 240 # endif 241 #endif 242 243 /*-************************************************************** 244 * Alignment check 245 *****************************************************************/ 246 247 /* this test was initially positioned in mem.h, 248 * but this file is removed (or replaced) for linux kernel 249 * so it's now hosted in compiler.h, 250 * which remains valid for both user & kernel spaces. 251 */ 252 253 #ifndef ZSTD_ALIGNOF 254 # if defined(__GNUC__) || defined(_MSC_VER) 255 /* covers gcc, clang & MSVC */ 256 /* note : this section must come first, before C11, 257 * due to a limitation in the kernel source generator */ 258 # define ZSTD_ALIGNOF(T) __alignof(T) 259 260 # elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) 261 /* C11 support */ 262 # include <stdalign.h> 263 # define ZSTD_ALIGNOF(T) alignof(T) 264 265 # else 266 /* No known support for alignof() - imperfect backup */ 267 # define ZSTD_ALIGNOF(T) (sizeof(void*) < sizeof(T) ? sizeof(void*) : sizeof(T)) 268 269 # endif 270 #endif /* ZSTD_ALIGNOF */ 271 272 /*-************************************************************** 273 * Sanitizer 274 *****************************************************************/ 275 276 #if ZSTD_MEMORY_SANITIZER 277 /* Not all platforms that support msan provide sanitizers/msan_interface.h. 278 * We therefore declare the functions we need ourselves, rather than trying to 279 * include the header file... */ 280 #include <stddef.h> /* size_t */ 281 #define ZSTD_DEPS_NEED_STDINT 282 #include "zstd_deps.h" /* intptr_t */ 283 284 /* Make memory region fully initialized (without changing its contents). */ 285 void __msan_unpoison(const volatile void *a, size_t size); 286 287 /* Make memory region fully uninitialized (without changing its contents). 288 This is a legacy interface that does not update origin information. Use 289 __msan_allocated_memory() instead. */ 290 void __msan_poison(const volatile void *a, size_t size); 291 292 /* Returns the offset of the first (at least partially) poisoned byte in the 293 memory range, or -1 if the whole range is good. */ 294 intptr_t __msan_test_shadow(const volatile void *x, size_t size); 295 #endif 296 297 #if ZSTD_ADDRESS_SANITIZER 298 /* Not all platforms that support asan provide sanitizers/asan_interface.h. 299 * We therefore declare the functions we need ourselves, rather than trying to 300 * include the header file... */ 301 #include <stddef.h> /* size_t */ 302 303 /** 304 * Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable. 305 * 306 * This memory must be previously allocated by your program. Instrumented 307 * code is forbidden from accessing addresses in this region until it is 308 * unpoisoned. This function is not guaranteed to poison the entire region - 309 * it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan 310 * alignment restrictions. 311 * 312 * \note This function is not thread-safe because no two threads can poison or 313 * unpoison memory in the same memory region simultaneously. 314 * 315 * \param addr Start of memory region. 316 * \param size Size of memory region. */ 317 void __asan_poison_memory_region(void const volatile *addr, size_t size); 318 319 /** 320 * Marks a memory region (<c>[addr, addr+size)</c>) as addressable. 321 * 322 * This memory must be previously allocated by your program. Accessing 323 * addresses in this region is allowed until this region is poisoned again. 324 * This function could unpoison a super-region of <c>[addr, addr+size)</c> due 325 * to ASan alignment restrictions. 326 * 327 * \note This function is not thread-safe because no two threads can 328 * poison or unpoison memory in the same memory region simultaneously. 329 * 330 * \param addr Start of memory region. 331 * \param size Size of memory region. */ 332 void __asan_unpoison_memory_region(void const volatile *addr, size_t size); 333 #endif 334 335 #endif /* ZSTD_COMPILER_H */ 336