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 */ 18a0483764SConrad Meyer 19a0483764SConrad Meyer #if !defined(ZSTD_NO_INLINE) 200c16b537SWarner Losh #if defined (__GNUC__) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */ 210c16b537SWarner Losh # define INLINE_KEYWORD inline 220c16b537SWarner Losh #else 230c16b537SWarner Losh # define INLINE_KEYWORD 240c16b537SWarner Losh #endif 250c16b537SWarner Losh 260c16b537SWarner Losh #if defined(__GNUC__) 270c16b537SWarner Losh # define FORCE_INLINE_ATTR __attribute__((always_inline)) 280c16b537SWarner Losh #elif defined(_MSC_VER) 290c16b537SWarner Losh # define FORCE_INLINE_ATTR __forceinline 300c16b537SWarner Losh #else 310c16b537SWarner Losh # define FORCE_INLINE_ATTR 320c16b537SWarner Losh #endif 330c16b537SWarner Losh 34a0483764SConrad Meyer #else 35a0483764SConrad Meyer 36a0483764SConrad Meyer #define INLINE_KEYWORD 37a0483764SConrad Meyer #define FORCE_INLINE_ATTR 38a0483764SConrad Meyer 39a0483764SConrad Meyer #endif 40a0483764SConrad Meyer 410c16b537SWarner Losh /** 420c16b537SWarner Losh * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant 43*2b9c00cbSConrad Meyer * parameters. They must be inlined for the compiler to eliminate the constant 440c16b537SWarner Losh * branches. 450c16b537SWarner Losh */ 460c16b537SWarner Losh #define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR 470c16b537SWarner Losh /** 480c16b537SWarner Losh * HINT_INLINE is used to help the compiler generate better code. It is *not* 490c16b537SWarner Losh * used for "templates", so it can be tweaked based on the compilers 500c16b537SWarner Losh * performance. 510c16b537SWarner Losh * 520c16b537SWarner Losh * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the 530c16b537SWarner Losh * always_inline attribute. 540c16b537SWarner Losh * 550c16b537SWarner Losh * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline 560c16b537SWarner Losh * attribute. 570c16b537SWarner Losh */ 580c16b537SWarner Losh #if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5 590c16b537SWarner Losh # define HINT_INLINE static INLINE_KEYWORD 600c16b537SWarner Losh #else 610c16b537SWarner Losh # define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR 620c16b537SWarner Losh #endif 630c16b537SWarner Losh 640c16b537SWarner Losh /* force no inlining */ 650c16b537SWarner Losh #ifdef _MSC_VER 660c16b537SWarner Losh # define FORCE_NOINLINE static __declspec(noinline) 670c16b537SWarner Losh #else 680c16b537SWarner Losh # ifdef __GNUC__ 690c16b537SWarner Losh # define FORCE_NOINLINE static __attribute__((__noinline__)) 700c16b537SWarner Losh # else 710c16b537SWarner Losh # define FORCE_NOINLINE static 720c16b537SWarner Losh # endif 730c16b537SWarner Losh #endif 740c16b537SWarner Losh 7519fcbaf1SConrad Meyer /* target attribute */ 7619fcbaf1SConrad Meyer #ifndef __has_attribute 7719fcbaf1SConrad Meyer #define __has_attribute(x) 0 /* Compatibility with non-clang compilers. */ 7819fcbaf1SConrad Meyer #endif 7919fcbaf1SConrad Meyer #if defined(__GNUC__) 8019fcbaf1SConrad Meyer # define TARGET_ATTRIBUTE(target) __attribute__((__target__(target))) 8119fcbaf1SConrad Meyer #else 8219fcbaf1SConrad Meyer # define TARGET_ATTRIBUTE(target) 8319fcbaf1SConrad Meyer #endif 8419fcbaf1SConrad Meyer 8519fcbaf1SConrad Meyer /* Enable runtime BMI2 dispatch based on the CPU. 8619fcbaf1SConrad Meyer * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default. 8719fcbaf1SConrad Meyer */ 8819fcbaf1SConrad Meyer #ifndef DYNAMIC_BMI2 890f743729SConrad Meyer #if ((defined(__clang__) && __has_attribute(__target__)) \ 9019fcbaf1SConrad Meyer || (defined(__GNUC__) \ 910f743729SConrad Meyer && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \ 9219fcbaf1SConrad Meyer && (defined(__x86_64__) || defined(_M_X86)) \ 9319fcbaf1SConrad Meyer && !defined(__BMI2__) 9419fcbaf1SConrad Meyer # define DYNAMIC_BMI2 1 9519fcbaf1SConrad Meyer #else 9619fcbaf1SConrad Meyer # define DYNAMIC_BMI2 0 9719fcbaf1SConrad Meyer #endif 9819fcbaf1SConrad Meyer #endif 9919fcbaf1SConrad Meyer 1000f743729SConrad Meyer /* prefetch 101a0483764SConrad Meyer * can be disabled, by declaring NO_PREFETCH build macro */ 1020f743729SConrad Meyer #if defined(NO_PREFETCH) 103a0483764SConrad Meyer # define PREFETCH_L1(ptr) (void)(ptr) /* disabled */ 104a0483764SConrad Meyer # define PREFETCH_L2(ptr) (void)(ptr) /* disabled */ 1050f743729SConrad Meyer #else 1060c16b537SWarner Losh # if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86)) /* _mm_prefetch() is not defined outside of x86/x64 */ 1070c16b537SWarner Losh # include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */ 108a0483764SConrad Meyer # define PREFETCH_L1(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0) 109a0483764SConrad Meyer # define PREFETCH_L2(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T1) 1100f743729SConrad Meyer # elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) ) 111a0483764SConrad Meyer # define PREFETCH_L1(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */) 112a0483764SConrad Meyer # define PREFETCH_L2(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */) 1130c16b537SWarner Losh # else 114a0483764SConrad Meyer # define PREFETCH_L1(ptr) (void)(ptr) /* disabled */ 115a0483764SConrad Meyer # define PREFETCH_L2(ptr) (void)(ptr) /* disabled */ 1160c16b537SWarner Losh # endif 1170f743729SConrad Meyer #endif /* NO_PREFETCH */ 1180f743729SConrad Meyer 1190f743729SConrad Meyer #define CACHELINE_SIZE 64 1200f743729SConrad Meyer 1210f743729SConrad Meyer #define PREFETCH_AREA(p, s) { \ 1220f743729SConrad Meyer const char* const _ptr = (const char*)(p); \ 1230f743729SConrad Meyer size_t const _size = (size_t)(s); \ 1240f743729SConrad Meyer size_t _pos; \ 1250f743729SConrad Meyer for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) { \ 126a0483764SConrad Meyer PREFETCH_L2(_ptr + _pos); \ 1270f743729SConrad Meyer } \ 1280f743729SConrad Meyer } 1290c16b537SWarner Losh 1300c16b537SWarner Losh /* disable warnings */ 1310c16b537SWarner Losh #ifdef _MSC_VER /* Visual Studio */ 1320c16b537SWarner Losh # include <intrin.h> /* For Visual 2005 */ 1330c16b537SWarner Losh # pragma warning(disable : 4100) /* disable: C4100: unreferenced formal parameter */ 1340c16b537SWarner Losh # pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ 1350c16b537SWarner Losh # pragma warning(disable : 4204) /* disable: C4204: non-constant aggregate initializer */ 1360c16b537SWarner Losh # pragma warning(disable : 4214) /* disable: C4214: non-int bitfields */ 1370c16b537SWarner Losh # pragma warning(disable : 4324) /* disable: C4324: padded structure */ 1380c16b537SWarner Losh #endif 1390c16b537SWarner Losh 1400c16b537SWarner Losh #endif /* ZSTD_COMPILER_H */ 141