1/* 2 * Copyright (c) 2016-2020, 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/*-******************************************************* 15* Compiler specifics 16*********************************************************/ 17/* force inlining */ 18 19#if !defined(ZSTD_NO_INLINE) 20#if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */ 21# define INLINE_KEYWORD inline 22#else 23# define INLINE_KEYWORD 24#endif 25 26#if defined(__GNUC__) || defined(__ICCARM__) 27# define FORCE_INLINE_ATTR __attribute__((always_inline)) 28#elif defined(_MSC_VER) 29# define FORCE_INLINE_ATTR __forceinline 30#else 31# define FORCE_INLINE_ATTR 32#endif 33 34#else 35 36#define INLINE_KEYWORD 37#define FORCE_INLINE_ATTR 38 39#endif 40 41/** 42 On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC). 43 This explictly marks such functions as __cdecl so that the code will still compile 44 if a CC other than __cdecl has been made the default. 45*/ 46#if defined(_MSC_VER) 47# define WIN_CDECL __cdecl 48#else 49# define WIN_CDECL 50#endif 51 52/** 53 * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant 54 * parameters. They must be inlined for the compiler to eliminate the constant 55 * branches. 56 */ 57#define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR 58/** 59 * HINT_INLINE is used to help the compiler generate better code. It is *not* 60 * used for "templates", so it can be tweaked based on the compilers 61 * performance. 62 * 63 * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the 64 * always_inline attribute. 65 * 66 * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline 67 * attribute. 68 */ 69#if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5 70# define HINT_INLINE static INLINE_KEYWORD 71#else 72# define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR 73#endif 74 75/* UNUSED_ATTR tells the compiler it is okay if the function is unused. */ 76#if defined(__GNUC__) 77# define UNUSED_ATTR __attribute__((unused)) 78#else 79# define UNUSED_ATTR 80#endif 81 82/* force no inlining */ 83#ifdef _MSC_VER 84# define FORCE_NOINLINE static __declspec(noinline) 85#else 86# if defined(__GNUC__) || defined(__ICCARM__) 87# define FORCE_NOINLINE static __attribute__((__noinline__)) 88# else 89# define FORCE_NOINLINE static 90# endif 91#endif 92 93/* target attribute */ 94#ifndef __has_attribute 95 #define __has_attribute(x) 0 /* Compatibility with non-clang compilers. */ 96#endif 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/* Enable runtime BMI2 dispatch based on the CPU. 104 * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default. 105 */ 106#ifndef DYNAMIC_BMI2 107 #if ((defined(__clang__) && __has_attribute(__target__)) \ 108 || (defined(__GNUC__) \ 109 && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \ 110 && (defined(__x86_64__) || defined(_M_X86)) \ 111 && !defined(__BMI2__) 112 # define DYNAMIC_BMI2 1 113 #else 114 # define DYNAMIC_BMI2 0 115 #endif 116#endif 117 118/* prefetch 119 * can be disabled, by declaring NO_PREFETCH build macro */ 120#if defined(NO_PREFETCH) 121# define PREFETCH_L1(ptr) (void)(ptr) /* disabled */ 122# define PREFETCH_L2(ptr) (void)(ptr) /* disabled */ 123#else 124# if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86)) /* _mm_prefetch() is not defined outside of x86/x64 */ 125# include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */ 126# define PREFETCH_L1(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0) 127# define PREFETCH_L2(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T1) 128# elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) ) 129# define PREFETCH_L1(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */) 130# define PREFETCH_L2(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */) 131# elif defined(__aarch64__) 132# define PREFETCH_L1(ptr) __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr))) 133# define PREFETCH_L2(ptr) __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr))) 134# else 135# define PREFETCH_L1(ptr) (void)(ptr) /* disabled */ 136# define PREFETCH_L2(ptr) (void)(ptr) /* disabled */ 137# endif 138#endif /* NO_PREFETCH */ 139 140#define CACHELINE_SIZE 64 141 142#define PREFETCH_AREA(p, s) { \ 143 const char* const _ptr = (const char*)(p); \ 144 size_t const _size = (size_t)(s); \ 145 size_t _pos; \ 146 for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) { \ 147 PREFETCH_L2(_ptr + _pos); \ 148 } \ 149} 150 151/* vectorization 152 * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax */ 153#if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__) 154# if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5) 155# define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize"))) 156# else 157# define DONT_VECTORIZE _Pragma("GCC optimize(\"no-tree-vectorize\")") 158# endif 159#else 160# define DONT_VECTORIZE 161#endif 162 163/* Tell the compiler that a branch is likely or unlikely. 164 * Only use these macros if it causes the compiler to generate better code. 165 * If you can remove a LIKELY/UNLIKELY annotation without speed changes in gcc 166 * and clang, please do. 167 */ 168#if defined(__GNUC__) 169#define LIKELY(x) (__builtin_expect((x), 1)) 170#define UNLIKELY(x) (__builtin_expect((x), 0)) 171#else 172#define LIKELY(x) (x) 173#define UNLIKELY(x) (x) 174#endif 175 176/* disable warnings */ 177#ifdef _MSC_VER /* Visual Studio */ 178# include <intrin.h> /* For Visual 2005 */ 179# pragma warning(disable : 4100) /* disable: C4100: unreferenced formal parameter */ 180# pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ 181# pragma warning(disable : 4204) /* disable: C4204: non-constant aggregate initializer */ 182# pragma warning(disable : 4214) /* disable: C4214: non-int bitfields */ 183# pragma warning(disable : 4324) /* disable: C4324: padded structure */ 184#endif 185 186/*Like DYNAMIC_BMI2 but for compile time determination of BMI2 support*/ 187#ifndef STATIC_BMI2 188# if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86)) 189# ifdef __AVX2__ //MSVC does not have a BMI2 specific flag, but every CPU that supports AVX2 also supports BMI2 190# define STATIC_BMI2 1 191# endif 192# endif 193#endif 194 195#ifndef STATIC_BMI2 196 #define STATIC_BMI2 0 197#endif 198 199/* compat. with non-clang compilers */ 200#ifndef __has_builtin 201# define __has_builtin(x) 0 202#endif 203 204/* compat. with non-clang compilers */ 205#ifndef __has_feature 206# define __has_feature(x) 0 207#endif 208 209/* detects whether we are being compiled under msan */ 210#ifndef ZSTD_MEMORY_SANITIZER 211# if __has_feature(memory_sanitizer) 212# define ZSTD_MEMORY_SANITIZER 1 213# else 214# define ZSTD_MEMORY_SANITIZER 0 215# endif 216#endif 217 218#if ZSTD_MEMORY_SANITIZER 219/* Not all platforms that support msan provide sanitizers/msan_interface.h. 220 * We therefore declare the functions we need ourselves, rather than trying to 221 * include the header file... */ 222#include <stddef.h> /* size_t */ 223#define ZSTD_DEPS_NEED_STDINT 224#include "zstd_deps.h" /* intptr_t */ 225 226/* Make memory region fully initialized (without changing its contents). */ 227void __msan_unpoison(const volatile void *a, size_t size); 228 229/* Make memory region fully uninitialized (without changing its contents). 230 This is a legacy interface that does not update origin information. Use 231 __msan_allocated_memory() instead. */ 232void __msan_poison(const volatile void *a, size_t size); 233 234/* Returns the offset of the first (at least partially) poisoned byte in the 235 memory range, or -1 if the whole range is good. */ 236intptr_t __msan_test_shadow(const volatile void *x, size_t size); 237#endif 238 239/* detects whether we are being compiled under asan */ 240#ifndef ZSTD_ADDRESS_SANITIZER 241# if __has_feature(address_sanitizer) 242# define ZSTD_ADDRESS_SANITIZER 1 243# elif defined(__SANITIZE_ADDRESS__) 244# define ZSTD_ADDRESS_SANITIZER 1 245# else 246# define ZSTD_ADDRESS_SANITIZER 0 247# endif 248#endif 249 250#if ZSTD_ADDRESS_SANITIZER 251/* Not all platforms that support asan provide sanitizers/asan_interface.h. 252 * We therefore declare the functions we need ourselves, rather than trying to 253 * include the header file... */ 254#include <stddef.h> /* size_t */ 255 256/** 257 * Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable. 258 * 259 * This memory must be previously allocated by your program. Instrumented 260 * code is forbidden from accessing addresses in this region until it is 261 * unpoisoned. This function is not guaranteed to poison the entire region - 262 * it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan 263 * alignment restrictions. 264 * 265 * \note This function is not thread-safe because no two threads can poison or 266 * unpoison memory in the same memory region simultaneously. 267 * 268 * \param addr Start of memory region. 269 * \param size Size of memory region. */ 270void __asan_poison_memory_region(void const volatile *addr, size_t size); 271 272/** 273 * Marks a memory region (<c>[addr, addr+size)</c>) as addressable. 274 * 275 * This memory must be previously allocated by your program. Accessing 276 * addresses in this region is allowed until this region is poisoned again. 277 * This function could unpoison a super-region of <c>[addr, addr+size)</c> due 278 * to ASan alignment restrictions. 279 * 280 * \note This function is not thread-safe because no two threads can 281 * poison or unpoison memory in the same memory region simultaneously. 282 * 283 * \param addr Start of memory region. 284 * \param size Size of memory region. */ 285void __asan_unpoison_memory_region(void const volatile *addr, size_t size); 286#endif 287 288#endif /* ZSTD_COMPILER_H */ 289