10c16b537SWarner Losh /* 20c16b537SWarner Losh * Copyright (c) 2016-present, Yann Collet, Facebook, Inc. 30c16b537SWarner Losh * All rights reserved. 40c16b537SWarner Losh * 50c16b537SWarner Losh * This source code is licensed under both the BSD-style license (found in the 60c16b537SWarner Losh * LICENSE file in the root directory of this source tree) and the GPLv2 (found 70c16b537SWarner Losh * in the COPYING file in the root directory of this source tree). 80c16b537SWarner Losh * You may select, at your option, one of the above-listed licenses. 90c16b537SWarner Losh */ 100c16b537SWarner Losh 110c16b537SWarner Losh #ifndef ZSTD_COMPILER_H 120c16b537SWarner Losh #define ZSTD_COMPILER_H 130c16b537SWarner Losh 140c16b537SWarner Losh /*-******************************************************* 150c16b537SWarner Losh * Compiler specifics 160c16b537SWarner Losh *********************************************************/ 170c16b537SWarner Losh /* force inlining */ 180c16b537SWarner Losh #if defined (__GNUC__) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */ 190c16b537SWarner Losh # define INLINE_KEYWORD inline 200c16b537SWarner Losh #else 210c16b537SWarner Losh # define INLINE_KEYWORD 220c16b537SWarner Losh #endif 230c16b537SWarner Losh 240c16b537SWarner Losh #if defined(__GNUC__) 250c16b537SWarner Losh # define FORCE_INLINE_ATTR __attribute__((always_inline)) 260c16b537SWarner Losh #elif defined(_MSC_VER) 270c16b537SWarner Losh # define FORCE_INLINE_ATTR __forceinline 280c16b537SWarner Losh #else 290c16b537SWarner Losh # define FORCE_INLINE_ATTR 300c16b537SWarner Losh #endif 310c16b537SWarner Losh 320c16b537SWarner Losh /** 330c16b537SWarner Losh * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant 340c16b537SWarner Losh * parameters. They must be inlined for the compiler to elimininate the constant 350c16b537SWarner Losh * branches. 360c16b537SWarner Losh */ 370c16b537SWarner Losh #define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR 380c16b537SWarner Losh /** 390c16b537SWarner Losh * HINT_INLINE is used to help the compiler generate better code. It is *not* 400c16b537SWarner Losh * used for "templates", so it can be tweaked based on the compilers 410c16b537SWarner Losh * performance. 420c16b537SWarner Losh * 430c16b537SWarner Losh * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the 440c16b537SWarner Losh * always_inline attribute. 450c16b537SWarner Losh * 460c16b537SWarner Losh * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline 470c16b537SWarner Losh * attribute. 480c16b537SWarner Losh */ 490c16b537SWarner Losh #if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5 500c16b537SWarner Losh # define HINT_INLINE static INLINE_KEYWORD 510c16b537SWarner Losh #else 520c16b537SWarner Losh # define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR 530c16b537SWarner Losh #endif 540c16b537SWarner Losh 550c16b537SWarner Losh /* force no inlining */ 560c16b537SWarner Losh #ifdef _MSC_VER 570c16b537SWarner Losh # define FORCE_NOINLINE static __declspec(noinline) 580c16b537SWarner Losh #else 590c16b537SWarner Losh # ifdef __GNUC__ 600c16b537SWarner Losh # define FORCE_NOINLINE static __attribute__((__noinline__)) 610c16b537SWarner Losh # else 620c16b537SWarner Losh # define FORCE_NOINLINE static 630c16b537SWarner Losh # endif 640c16b537SWarner Losh #endif 650c16b537SWarner Losh 66*19fcbaf1SConrad Meyer /* target attribute */ 67*19fcbaf1SConrad Meyer #ifndef __has_attribute 68*19fcbaf1SConrad Meyer #define __has_attribute(x) 0 /* Compatibility with non-clang compilers. */ 69*19fcbaf1SConrad Meyer #endif 70*19fcbaf1SConrad Meyer #if defined(__GNUC__) 71*19fcbaf1SConrad Meyer # define TARGET_ATTRIBUTE(target) __attribute__((__target__(target))) 72*19fcbaf1SConrad Meyer #else 73*19fcbaf1SConrad Meyer # define TARGET_ATTRIBUTE(target) 74*19fcbaf1SConrad Meyer #endif 75*19fcbaf1SConrad Meyer 76*19fcbaf1SConrad Meyer /* Enable runtime BMI2 dispatch based on the CPU. 77*19fcbaf1SConrad Meyer * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default. 78*19fcbaf1SConrad Meyer */ 79*19fcbaf1SConrad Meyer #ifndef DYNAMIC_BMI2 80*19fcbaf1SConrad Meyer #if (defined(__clang__) && __has_attribute(__target__)) \ 81*19fcbaf1SConrad Meyer || (defined(__GNUC__) \ 82*19fcbaf1SConrad Meyer && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8))) \ 83*19fcbaf1SConrad Meyer && (defined(__x86_64__) || defined(_M_X86)) \ 84*19fcbaf1SConrad Meyer && !defined(__BMI2__) 85*19fcbaf1SConrad Meyer # define DYNAMIC_BMI2 1 86*19fcbaf1SConrad Meyer #else 87*19fcbaf1SConrad Meyer # define DYNAMIC_BMI2 0 88*19fcbaf1SConrad Meyer #endif 89*19fcbaf1SConrad Meyer #endif 90*19fcbaf1SConrad Meyer 910c16b537SWarner Losh /* prefetch */ 920c16b537SWarner Losh #if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86)) /* _mm_prefetch() is not defined outside of x86/x64 */ 930c16b537SWarner Losh # include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */ 940c16b537SWarner Losh # define PREFETCH(ptr) _mm_prefetch((const char*)ptr, _MM_HINT_T0) 950c16b537SWarner Losh #elif defined(__GNUC__) 960c16b537SWarner Losh # define PREFETCH(ptr) __builtin_prefetch(ptr, 0, 0) 970c16b537SWarner Losh #else 980c16b537SWarner Losh # define PREFETCH(ptr) /* disabled */ 990c16b537SWarner Losh #endif 1000c16b537SWarner Losh 1010c16b537SWarner Losh /* disable warnings */ 1020c16b537SWarner Losh #ifdef _MSC_VER /* Visual Studio */ 1030c16b537SWarner Losh # include <intrin.h> /* For Visual 2005 */ 1040c16b537SWarner Losh # pragma warning(disable : 4100) /* disable: C4100: unreferenced formal parameter */ 1050c16b537SWarner Losh # pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ 1060c16b537SWarner Losh # pragma warning(disable : 4204) /* disable: C4204: non-constant aggregate initializer */ 1070c16b537SWarner Losh # pragma warning(disable : 4214) /* disable: C4214: non-int bitfields */ 1080c16b537SWarner Losh # pragma warning(disable : 4324) /* disable: C4324: padded structure */ 1090c16b537SWarner Losh #endif 1100c16b537SWarner Losh 1110c16b537SWarner Losh #endif /* ZSTD_COMPILER_H */ 112