1 /* 1 /*
2 * Copyright (c) Yann Collet, Facebook, Inc. 2 * Copyright (c) Yann Collet, Facebook, Inc.
3 * All rights reserved. 3 * All rights reserved.
4 * 4 *
5 * This source code is licensed under both the 5 * This source code is licensed under both the BSD-style license (found in the
6 * LICENSE file in the root directory of this 6 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7 * in the COPYING file in the root directory o 7 * in the COPYING file in the root directory of this source tree).
8 * You may select, at your option, one of the 8 * You may select, at your option, one of the above-listed licenses.
9 */ 9 */
10 10
11 #ifndef ZSTD_COMPILER_H 11 #ifndef ZSTD_COMPILER_H
12 #define ZSTD_COMPILER_H 12 #define ZSTD_COMPILER_H
13 13
14 #include "portability_macros.h" <<
15 <<
16 /*-******************************************* 14 /*-*******************************************************
17 * Compiler specifics 15 * Compiler specifics
18 ********************************************** 16 *********************************************************/
19 /* force inlining */ 17 /* force inlining */
20 18
21 #if !defined(ZSTD_NO_INLINE) 19 #if !defined(ZSTD_NO_INLINE)
22 #if (defined(__GNUC__) && !defined(__STRICT_AN 20 #if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */
23 # define INLINE_KEYWORD inline 21 # define INLINE_KEYWORD inline
24 #else 22 #else
25 # define INLINE_KEYWORD 23 # define INLINE_KEYWORD
26 #endif 24 #endif
27 25
28 #define FORCE_INLINE_ATTR __attribute__((alway 26 #define FORCE_INLINE_ATTR __attribute__((always_inline))
29 27
30 #else 28 #else
31 29
32 #define INLINE_KEYWORD 30 #define INLINE_KEYWORD
33 #define FORCE_INLINE_ATTR 31 #define FORCE_INLINE_ATTR
34 32
35 #endif 33 #endif
36 34
37 /* 35 /*
38 On MSVC qsort requires that functions passed 36 On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC).
39 This explicitly marks such functions as __cd !! 37 This explictly marks such functions as __cdecl so that the code will still compile
40 if a CC other than __cdecl has been made the 38 if a CC other than __cdecl has been made the default.
41 */ 39 */
42 #define WIN_CDECL 40 #define WIN_CDECL
43 41
44 /* 42 /*
45 * FORCE_INLINE_TEMPLATE is used to define C " 43 * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
46 * parameters. They must be inlined for the co 44 * parameters. They must be inlined for the compiler to eliminate the constant
47 * branches. 45 * branches.
48 */ 46 */
49 #define FORCE_INLINE_TEMPLATE static INLINE_KE 47 #define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR
50 /* 48 /*
51 * HINT_INLINE is used to help the compiler ge 49 * HINT_INLINE is used to help the compiler generate better code. It is *not*
52 * used for "templates", so it can be tweaked 50 * used for "templates", so it can be tweaked based on the compilers
53 * performance. 51 * performance.
54 * 52 *
55 * gcc-4.8 and gcc-4.9 have been shown to bene 53 * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the
56 * always_inline attribute. 54 * always_inline attribute.
57 * 55 *
58 * clang up to 5.0.0 (trunk) benefit tremendou 56 * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline
59 * attribute. 57 * attribute.
60 */ 58 */
61 #if !defined(__clang__) && defined(__GNUC__) & 59 #if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5
62 # define HINT_INLINE static INLINE_KEYWORD 60 # define HINT_INLINE static INLINE_KEYWORD
63 #else 61 #else
64 # define HINT_INLINE static INLINE_KEYWORD FO 62 # define HINT_INLINE static INLINE_KEYWORD FORCE_INLINE_ATTR
65 #endif 63 #endif
66 64
67 /* UNUSED_ATTR tells the compiler it is okay i 65 /* UNUSED_ATTR tells the compiler it is okay if the function is unused. */
68 #define UNUSED_ATTR __attribute__((unused)) 66 #define UNUSED_ATTR __attribute__((unused))
69 67
70 /* force no inlining */ 68 /* force no inlining */
71 #define FORCE_NOINLINE static __attribute__((_ 69 #define FORCE_NOINLINE static __attribute__((__noinline__))
72 70
73 71
74 /* target attribute */ 72 /* target attribute */
>> 73 #ifndef __has_attribute
>> 74 #define __has_attribute(x) 0 /* Compatibility with non-clang compilers. */
>> 75 #endif
75 #define TARGET_ATTRIBUTE(target) __attribute__ 76 #define TARGET_ATTRIBUTE(target) __attribute__((__target__(target)))
76 77
77 /* Target attribute for BMI2 dynamic dispatch. !! 78 /* Enable runtime BMI2 dispatch based on the CPU.
78 * Enable lzcnt, bmi, and bmi2. !! 79 * Enabled for clang & gcc >=4.8 on x86 when BMI2 isn't enabled by default.
79 * We test for bmi1 & bmi2. lzcnt is included <<
80 */ 80 */
81 #define BMI2_TARGET_ATTRIBUTE TARGET_ATTRIBUTE !! 81 #ifndef DYNAMIC_BMI2
>> 82 #if ((defined(__clang__) && __has_attribute(__target__)) \
>> 83 || (defined(__GNUC__) \
>> 84 && (__GNUC__ >= 5 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)))) \
>> 85 && (defined(__x86_64__) || defined(_M_X86)) \
>> 86 && !defined(__BMI2__)
>> 87 # define DYNAMIC_BMI2 1
>> 88 #else
>> 89 # define DYNAMIC_BMI2 0
>> 90 #endif
>> 91 #endif
82 92
83 /* prefetch 93 /* prefetch
84 * can be disabled, by declaring NO_PREFETCH b 94 * can be disabled, by declaring NO_PREFETCH build macro */
85 #if ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && 95 #if ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
86 # define PREFETCH_L1(ptr) __builtin_prefetch 96 # define PREFETCH_L1(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
87 # define PREFETCH_L2(ptr) __builtin_prefetch 97 # define PREFETCH_L2(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
88 #elif defined(__aarch64__) 98 #elif defined(__aarch64__)
89 # define PREFETCH_L1(ptr) __asm__ __volatile 99 # define PREFETCH_L1(ptr) __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr)))
90 # define PREFETCH_L2(ptr) __asm__ __volatile 100 # define PREFETCH_L2(ptr) __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr)))
91 #else 101 #else
92 # define PREFETCH_L1(ptr) (void)(ptr) /* dis 102 # define PREFETCH_L1(ptr) (void)(ptr) /* disabled */
93 # define PREFETCH_L2(ptr) (void)(ptr) /* dis 103 # define PREFETCH_L2(ptr) (void)(ptr) /* disabled */
94 #endif /* NO_PREFETCH */ 104 #endif /* NO_PREFETCH */
95 105
96 #define CACHELINE_SIZE 64 106 #define CACHELINE_SIZE 64
97 107
98 #define PREFETCH_AREA(p, s) { \ 108 #define PREFETCH_AREA(p, s) { \
99 const char* const _ptr = (const char*)(p); 109 const char* const _ptr = (const char*)(p); \
100 size_t const _size = (size_t)(s); \ 110 size_t const _size = (size_t)(s); \
101 size_t _pos; \ 111 size_t _pos; \
102 for (_pos=0; _pos<_size; _pos+=CACHELINE_S 112 for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) { \
103 PREFETCH_L2(_ptr + _pos); \ 113 PREFETCH_L2(_ptr + _pos); \
104 } \ 114 } \
105 } 115 }
106 116
107 /* vectorization 117 /* vectorization
108 * older GCC (pre gcc-4.3 picked as the cutoff !! 118 * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax */
109 * and some compilers, like Intel ICC and MCST !! 119 #if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__)
110 #if !defined(__INTEL_COMPILER) && !defined(__c <<
111 # if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || 120 # if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5)
112 # define DONT_VECTORIZE __attribute__((opti 121 # define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize")))
113 # else 122 # else
114 # define DONT_VECTORIZE _Pragma("GCC optimi 123 # define DONT_VECTORIZE _Pragma("GCC optimize(\"no-tree-vectorize\")")
115 # endif 124 # endif
116 #else 125 #else
117 # define DONT_VECTORIZE 126 # define DONT_VECTORIZE
118 #endif 127 #endif
119 128
120 /* Tell the compiler that a branch is likely o 129 /* Tell the compiler that a branch is likely or unlikely.
121 * Only use these macros if it causes the comp 130 * Only use these macros if it causes the compiler to generate better code.
122 * If you can remove a LIKELY/UNLIKELY annotat 131 * If you can remove a LIKELY/UNLIKELY annotation without speed changes in gcc
123 * and clang, please do. 132 * and clang, please do.
124 */ 133 */
125 #define LIKELY(x) (__builtin_expect((x), 1)) 134 #define LIKELY(x) (__builtin_expect((x), 1))
126 #define UNLIKELY(x) (__builtin_expect((x), 0)) 135 #define UNLIKELY(x) (__builtin_expect((x), 0))
127 136
128 #if __has_builtin(__builtin_unreachable) || (d <<
129 # define ZSTD_UNREACHABLE { assert(0), __buil <<
130 #else <<
131 # define ZSTD_UNREACHABLE { assert(0); } <<
132 #endif <<
133 <<
134 /* disable warnings */ 137 /* disable warnings */
135 138
136 /*Like DYNAMIC_BMI2 but for compile time deter 139 /*Like DYNAMIC_BMI2 but for compile time determination of BMI2 support*/
137 140
138 141
139 /* compile time determination of SIMD support !! 142 /* compat. with non-clang compilers */
>> 143 #ifndef __has_builtin
>> 144 # define __has_builtin(x) 0
>> 145 #endif
>> 146
>> 147 /* compat. with non-clang compilers */
>> 148 #ifndef __has_feature
>> 149 # define __has_feature(x) 0
>> 150 #endif
140 151
141 /* C-language Attributes are added in C23. */ 152 /* C-language Attributes are added in C23. */
142 #if defined(__STDC_VERSION__) && (__STDC_VERSI 153 #if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
143 # define ZSTD_HAS_C_ATTRIBUTE(x) __has_c_attri 154 # define ZSTD_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
144 #else 155 #else
145 # define ZSTD_HAS_C_ATTRIBUTE(x) 0 156 # define ZSTD_HAS_C_ATTRIBUTE(x) 0
146 #endif 157 #endif
147 158
148 /* Only use C++ attributes in C++. Some compil 159 /* Only use C++ attributes in C++. Some compilers report support for C++
149 * attributes when compiling with C. 160 * attributes when compiling with C.
150 */ 161 */
151 #define ZSTD_HAS_CPP_ATTRIBUTE(x) 0 162 #define ZSTD_HAS_CPP_ATTRIBUTE(x) 0
152 163
153 /* Define ZSTD_FALLTHROUGH macro for annotatin 164 /* Define ZSTD_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute.
154 * - C23: https://en.cppreference.com/w/c/lang 165 * - C23: https://en.cppreference.com/w/c/language/attributes/fallthrough
155 * - CPP17: https://en.cppreference.com/w/cpp/ 166 * - CPP17: https://en.cppreference.com/w/cpp/language/attributes/fallthrough
156 * - Else: __attribute__((__fallthrough__)) 167 * - Else: __attribute__((__fallthrough__))
157 */ 168 */
158 #define ZSTD_FALLTHROUGH fallthrough 169 #define ZSTD_FALLTHROUGH fallthrough
159 170
160 /*-******************************************* !! 171 /* detects whether we are being compiled under msan */
161 * Alignment check !! 172
162 ********************************************** <<
163 <<
164 /* this test was initially positioned in mem.h <<
165 * but this file is removed (or replaced) for <<
166 * so it's now hosted in compiler.h, <<
167 * which remains valid for both user & kernel <<
168 */ <<
169 <<
170 #ifndef ZSTD_ALIGNOF <<
171 /* covers gcc, clang & MSVC */ <<
172 /* note : this section must come first, before <<
173 * due to a limitation in the kernel source ge <<
174 # define ZSTD_ALIGNOF(T) __alignof(T) <<
175 <<
176 #endif /* ZSTD_ALIGNOF */ <<
177 <<
178 /*-******************************************* <<
179 * Sanitizer <<
180 ********************************************** <<
181 173
>> 174 /* detects whether we are being compiled under asan */
182 175
183 176
184 #endif /* ZSTD_COMPILER_H */ 177 #endif /* ZSTD_COMPILER_H */
185 178
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.