1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0-only 2 /* 3 * Copyright (c) 2016-2020, Yann Collet, Facebook, Inc. 4 * All rights reserved. 5 * 6 * This source code is licensed under both the BSD-style license (found in the 7 * LICENSE file in the root directory of this source tree) and the GPLv2 (found 8 * in the COPYING file in the root directory of this source tree). 9 * You may select, at your option, one of the above-listed licenses. 10 */ 11 12 #ifndef ZSTD_COMPILER_H 13 #define ZSTD_COMPILER_H 14 15 /*-******************************************************* 16 * Compiler specifics 17 *********************************************************/ 18 /* force inlining */ 19 20 #if !defined(ZSTD_NO_INLINE) 21 #if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */ 22 # define INLINE_KEYWORD inline 23 #else 24 # define INLINE_KEYWORD 25 #endif 26 27 #if defined(__GNUC__) || defined(__ICCARM__) 28 # define FORCE_INLINE_ATTR __attribute__((always_inline)) 29 #elif defined(_MSC_VER) 30 # define FORCE_INLINE_ATTR __forceinline 31 #else 32 # define FORCE_INLINE_ATTR 33 #endif 34 35 #else 36 37 #define INLINE_KEYWORD 38 #define FORCE_INLINE_ATTR 39 40 #endif 41 42 /** 43 * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant 44 * parameters. They must be inlined for the compiler to eliminate the constant 45 * branches. 46 */ 47 #define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR 48 /** 49 * HINT_INLINE is used to help the compiler generate better code. It is *not* 50 * used for "templates", so it can be tweaked based on the compilers 51 * performance. 52 * 53 * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the 54 * always_inline attribute. 55 * 56 * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline 57 * attribute. 58 */ 59 #if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5 60 # define HINT_INLINE static INLINE_KEYWORD 61 #else 62 # define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR 63 #endif 64 65 /* UNUSED_ATTR tells the compiler it is okay if the function is unused. */ 66 #if defined(__GNUC__) 67 # define UNUSED_ATTR __attribute__((unused)) 68 #else 69 # define UNUSED_ATTR 70 #endif 71 72 /* force no inlining */ 73 #ifdef _MSC_VER 74 # define FORCE_NOINLINE static __declspec(noinline) 75 #else 76 # if defined(__GNUC__) || defined(__ICCARM__) 77 # define FORCE_NOINLINE static __attribute__((__noinline__)) 78 # else 79 # define FORCE_NOINLINE static 80 # endif 81 #endif 82 83 /* target attribute */ 84 #ifndef __has_attribute 85 #define __has_attribute(x) 0 /* Compatibility with non-clang compilers. */ 86 #endif 87 #if defined(__GNUC__) || defined(__ICCARM__) 88 # define TARGET_ATTRIBUTE(target) __attribute__((__target__(target))) 89 #else 90 # define TARGET_ATTRIBUTE(target) 91 #endif 92 93 /* Enable runtime BMI2 dispatch based on the CPU. 94 * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default. 95 */ 96 #ifndef DYNAMIC_BMI2 97 #if ((defined(__clang__) && __has_attribute(__target__)) \ 98 || (defined(__GNUC__) \ 99 && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \ 100 && (defined(__x86_64__) || defined(_M_X86)) \ 101 && !defined(__BMI2__) 102 # define DYNAMIC_BMI2 1 103 #else 104 # define DYNAMIC_BMI2 0 105 #endif 106 #endif 107 108 /* prefetch 109 * can be disabled, by declaring NO_PREFETCH build macro */ 110 #if defined(NO_PREFETCH) 111 # define PREFETCH_L1(ptr) (void)(ptr) /* disabled */ 112 # define PREFETCH_L2(ptr) (void)(ptr) /* disabled */ 113 #else 114 # if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86)) /* _mm_prefetch() is not defined outside of x86/x64 */ 115 # include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */ 116 # define PREFETCH_L1(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0) 117 # define PREFETCH_L2(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T1) 118 # elif defined(__aarch64__) 119 # define PREFETCH_L1(ptr) __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr))) 120 # define PREFETCH_L2(ptr) __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr))) 121 # elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) ) 122 # define PREFETCH_L1(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */) 123 # define PREFETCH_L2(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */) 124 # else 125 # define PREFETCH_L1(ptr) (void)(ptr) /* disabled */ 126 # define PREFETCH_L2(ptr) (void)(ptr) /* disabled */ 127 # endif 128 #endif /* NO_PREFETCH */ 129 130 #define CACHELINE_SIZE 64 131 132 #define PREFETCH_AREA(p, s) { \ 133 const char* const _ptr = (const char*)(p); \ 134 size_t const _size = (size_t)(s); \ 135 size_t _pos; \ 136 for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) { \ 137 PREFETCH_L2(_ptr + _pos); \ 138 } \ 139 } 140 141 /* vectorization 142 * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax */ 143 #if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__) 144 # if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5) 145 # define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize"))) 146 # else 147 # define DONT_VECTORIZE _Pragma("GCC optimize(\"no-tree-vectorize\")") 148 # endif 149 #else 150 # define DONT_VECTORIZE 151 #endif 152 153 /* Tell the compiler that a branch is likely or unlikely. 154 * Only use these macros if it causes the compiler to generate better code. 155 * If you can remove a LIKELY/UNLIKELY annotation without speed changes in gcc 156 * and clang, please do. 157 */ 158 #if defined(__GNUC__) 159 #define LIKELY(x) (__builtin_expect((x), 1)) 160 #define UNLIKELY(x) (__builtin_expect((x), 0)) 161 #else 162 #define LIKELY(x) (x) 163 #define UNLIKELY(x) (x) 164 #endif 165 166 /* disable warnings */ 167 #ifdef _MSC_VER /* Visual Studio */ 168 # include <intrin.h> /* For Visual 2005 */ 169 # pragma warning(disable : 4100) /* disable: C4100: unreferenced formal parameter */ 170 # pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ 171 # pragma warning(disable : 4204) /* disable: C4204: non-constant aggregate initializer */ 172 # pragma warning(disable : 4214) /* disable: C4214: non-int bitfields */ 173 # pragma warning(disable : 4324) /* disable: C4324: padded structure */ 174 #endif 175 176 #endif /* ZSTD_COMPILER_H */ 177