1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0-only 2 /* 3 * Copyright (c) Meta Platforms, Inc. and affiliates. 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 #include <stddef.h> 16 17 #include "portability_macros.h" 18 19 /*-******************************************************* 20 * Compiler specifics 21 *********************************************************/ 22 /* force inlining */ 23 24 #if !defined(ZSTD_NO_INLINE) 25 #if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */ 26 # define INLINE_KEYWORD inline 27 #else 28 # define INLINE_KEYWORD 29 #endif 30 31 #if defined(__GNUC__) || defined(__IAR_SYSTEMS_ICC__) 32 # define FORCE_INLINE_ATTR __attribute__((always_inline)) 33 #elif defined(_MSC_VER) 34 # define FORCE_INLINE_ATTR __forceinline 35 #else 36 # define FORCE_INLINE_ATTR 37 #endif 38 39 #else 40 41 #define INLINE_KEYWORD 42 #define FORCE_INLINE_ATTR 43 44 #endif 45 46 /** 47 On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC). 48 This explicitly marks such functions as __cdecl so that the code will still compile 49 if a CC other than __cdecl has been made the default. 50 */ 51 #if defined(_MSC_VER) 52 # define WIN_CDECL __cdecl 53 #else 54 # define WIN_CDECL 55 #endif 56 57 /* UNUSED_ATTR tells the compiler it is okay if the function is unused. */ 58 #if defined(__GNUC__) || defined(__IAR_SYSTEMS_ICC__) 59 # define UNUSED_ATTR __attribute__((unused)) 60 #else 61 # define UNUSED_ATTR 62 #endif 63 64 /** 65 * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant 66 * parameters. They must be inlined for the compiler to eliminate the constant 67 * branches. 68 */ 69 #define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR UNUSED_ATTR 70 /** 71 * HINT_INLINE is used to help the compiler generate better code. It is *not* 72 * used for "templates", so it can be tweaked based on the compilers 73 * performance. 74 * 75 * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the 76 * always_inline attribute. 77 * 78 * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline 79 * attribute. 80 */ 81 #if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5 82 # define HINT_INLINE static INLINE_KEYWORD 83 #else 84 # define HINT_INLINE FORCE_INLINE_TEMPLATE 85 #endif 86 87 /* "soft" inline : 88 * The compiler is free to select if it's a good idea to inline or not. 89 * The main objective is to silence compiler warnings 90 * when a defined function in included but not used. 91 * 92 * Note : this macro is prefixed `MEM_` because it used to be provided by `mem.h` unit. 93 * Updating the prefix is probably preferable, but requires a fairly large codemod, 94 * since this name is used everywhere. 95 */ 96 #ifndef MEM_STATIC /* already defined in Linux Kernel mem.h */ 97 #if defined(__GNUC__) 98 # define MEM_STATIC static __inline UNUSED_ATTR 99 #elif defined(__IAR_SYSTEMS_ICC__) 100 # define MEM_STATIC static inline UNUSED_ATTR 101 #elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) 102 # define MEM_STATIC static inline 103 #elif defined(_MSC_VER) 104 # define MEM_STATIC static __inline 105 #else 106 # define MEM_STATIC static /* this version may generate warnings for unused static functions; disable the relevant warning */ 107 #endif 108 #endif 109 110 /* force no inlining */ 111 #ifdef _MSC_VER 112 # define FORCE_NOINLINE static __declspec(noinline) 113 #else 114 # if defined(__GNUC__) || defined(__IAR_SYSTEMS_ICC__) 115 # define FORCE_NOINLINE static __attribute__((__noinline__)) 116 # else 117 # define FORCE_NOINLINE static 118 # endif 119 #endif 120 121 122 /* target attribute */ 123 #if defined(__GNUC__) || defined(__IAR_SYSTEMS_ICC__) 124 # define TARGET_ATTRIBUTE(target) __attribute__((__target__(target))) 125 #else 126 # define TARGET_ATTRIBUTE(target) 127 #endif 128 129 /* Target attribute for BMI2 dynamic dispatch. 130 * Enable lzcnt, bmi, and bmi2. 131 * We test for bmi1 & bmi2. lzcnt is included in bmi1. 132 */ 133 #define BMI2_TARGET_ATTRIBUTE TARGET_ATTRIBUTE("lzcnt,bmi,bmi2") 134 135 /* prefetch 136 * can be disabled, by declaring NO_PREFETCH build macro */ 137 #if defined(NO_PREFETCH) 138 # define PREFETCH_L1(ptr) do { (void)(ptr); } while (0) /* disabled */ 139 # define PREFETCH_L2(ptr) do { (void)(ptr); } while (0) /* disabled */ 140 #else 141 # if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86)) && !defined(_M_ARM64EC) /* _mm_prefetch() is not defined outside of x86/x64 */ 142 # include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */ 143 # define PREFETCH_L1(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0) 144 # define PREFETCH_L2(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T1) 145 # elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) ) 146 # define PREFETCH_L1(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */) 147 # define PREFETCH_L2(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */) 148 # elif defined(__aarch64__) 149 # define PREFETCH_L1(ptr) do { __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr))); } while (0) 150 # define PREFETCH_L2(ptr) do { __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr))); } while (0) 151 # else 152 # define PREFETCH_L1(ptr) do { (void)(ptr); } while (0) /* disabled */ 153 # define PREFETCH_L2(ptr) do { (void)(ptr); } while (0) /* disabled */ 154 # endif 155 #endif /* NO_PREFETCH */ 156 157 #define CACHELINE_SIZE 64 158 159 #define PREFETCH_AREA(p, s) \ 160 do { \ 161 const char* const _ptr = (const char*)(p); \ 162 size_t const _size = (size_t)(s); \ 163 size_t _pos; \ 164 for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) { \ 165 PREFETCH_L2(_ptr + _pos); \ 166 } \ 167 } while (0) 168 169 /* vectorization 170 * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax, 171 * and some compilers, like Intel ICC and MCST LCC, do not support it at all. */ 172 #if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__) && !defined(__LCC__) 173 # if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5) 174 # define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize"))) 175 # else 176 # define DONT_VECTORIZE _Pragma("GCC optimize(\"no-tree-vectorize\")") 177 # endif 178 #else 179 # define DONT_VECTORIZE 180 #endif 181 182 /* Tell the compiler that a branch is likely or unlikely. 183 * Only use these macros if it causes the compiler to generate better code. 184 * If you can remove a LIKELY/UNLIKELY annotation without speed changes in gcc 185 * and clang, please do. 186 */ 187 #if defined(__GNUC__) 188 #define LIKELY(x) (__builtin_expect((x), 1)) 189 #define UNLIKELY(x) (__builtin_expect((x), 0)) 190 #else 191 #define LIKELY(x) (x) 192 #define UNLIKELY(x) (x) 193 #endif 194 195 #if __has_builtin(__builtin_unreachable) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 5))) 196 # define ZSTD_UNREACHABLE do { assert(0), __builtin_unreachable(); } while (0) 197 #else 198 # define ZSTD_UNREACHABLE do { assert(0); } while (0) 199 #endif 200 201 /* disable warnings */ 202 #ifdef _MSC_VER /* Visual Studio */ 203 # include <intrin.h> /* For Visual 2005 */ 204 # pragma warning(disable : 4100) /* disable: C4100: unreferenced formal parameter */ 205 # pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */ 206 # pragma warning(disable : 4204) /* disable: C4204: non-constant aggregate initializer */ 207 # pragma warning(disable : 4214) /* disable: C4214: non-int bitfields */ 208 # pragma warning(disable : 4324) /* disable: C4324: padded structure */ 209 #endif 210 211 /* compile time determination of SIMD support */ 212 #if !defined(ZSTD_NO_INTRINSICS) 213 # if defined(__AVX2__) 214 # define ZSTD_ARCH_X86_AVX2 215 # endif 216 # if defined(__SSE2__) || defined(_M_X64) || (defined (_M_IX86) && defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) 217 # define ZSTD_ARCH_X86_SSE2 218 # endif 219 # if defined(__ARM_NEON) || defined(_M_ARM64) 220 # define ZSTD_ARCH_ARM_NEON 221 # endif 222 # 223 # if defined(ZSTD_ARCH_X86_AVX2) 224 # include <immintrin.h> 225 # endif 226 # if defined(ZSTD_ARCH_X86_SSE2) 227 # include <emmintrin.h> 228 # elif defined(ZSTD_ARCH_ARM_NEON) 229 # include <arm_neon.h> 230 # endif 231 #endif 232 233 /* C-language Attributes are added in C23. */ 234 #if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute) 235 # define ZSTD_HAS_C_ATTRIBUTE(x) __has_c_attribute(x) 236 #else 237 # define ZSTD_HAS_C_ATTRIBUTE(x) 0 238 #endif 239 240 /* Only use C++ attributes in C++. Some compilers report support for C++ 241 * attributes when compiling with C. 242 */ 243 #if defined(__cplusplus) && defined(__has_cpp_attribute) 244 # define ZSTD_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x) 245 #else 246 # define ZSTD_HAS_CPP_ATTRIBUTE(x) 0 247 #endif 248 249 /* Define ZSTD_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute. 250 * - C23: https://en.cppreference.com/w/c/language/attributes/fallthrough 251 * - CPP17: https://en.cppreference.com/w/cpp/language/attributes/fallthrough 252 * - Else: __attribute__((__fallthrough__)) 253 */ 254 #ifndef ZSTD_FALLTHROUGH 255 # if ZSTD_HAS_C_ATTRIBUTE(fallthrough) 256 # define ZSTD_FALLTHROUGH [[fallthrough]] 257 # elif ZSTD_HAS_CPP_ATTRIBUTE(fallthrough) 258 # define ZSTD_FALLTHROUGH [[fallthrough]] 259 # elif __has_attribute(__fallthrough__) 260 /* Leading semicolon is to satisfy gcc-11 with -pedantic. Without the semicolon 261 * gcc complains about: a label can only be part of a statement and a declaration is not a statement. 262 */ 263 # define ZSTD_FALLTHROUGH ; __attribute__((__fallthrough__)) 264 # else 265 # define ZSTD_FALLTHROUGH 266 # endif 267 #endif 268 269 /*-************************************************************** 270 * Alignment 271 *****************************************************************/ 272 273 /* @return 1 if @u is a 2^n value, 0 otherwise 274 * useful to check a value is valid for alignment restrictions */ 275 MEM_STATIC int ZSTD_isPower2(size_t u) { 276 return (u & (u-1)) == 0; 277 } 278 279 /* this test was initially positioned in mem.h, 280 * but this file is removed (or replaced) for linux kernel 281 * so it's now hosted in compiler.h, 282 * which remains valid for both user & kernel spaces. 283 */ 284 285 #ifndef ZSTD_ALIGNOF 286 # if defined(__GNUC__) || defined(_MSC_VER) 287 /* covers gcc, clang & MSVC */ 288 /* note : this section must come first, before C11, 289 * due to a limitation in the kernel source generator */ 290 # define ZSTD_ALIGNOF(T) __alignof(T) 291 292 # elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) 293 /* C11 support */ 294 # include <stdalign.h> 295 # define ZSTD_ALIGNOF(T) alignof(T) 296 297 # else 298 /* No known support for alignof() - imperfect backup */ 299 # define ZSTD_ALIGNOF(T) (sizeof(void*) < sizeof(T) ? sizeof(void*) : sizeof(T)) 300 301 # endif 302 #endif /* ZSTD_ALIGNOF */ 303 304 #ifndef ZSTD_ALIGNED 305 /* C90-compatible alignment macro (GCC/Clang). Adjust for other compilers if needed. */ 306 # if defined(__GNUC__) || defined(__clang__) 307 # define ZSTD_ALIGNED(a) __attribute__((aligned(a))) 308 # elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* C11 */ 309 # define ZSTD_ALIGNED(a) _Alignas(a) 310 #elif defined(_MSC_VER) 311 # define ZSTD_ALIGNED(n) __declspec(align(n)) 312 # else 313 /* this compiler will require its own alignment instruction */ 314 # define ZSTD_ALIGNED(...) 315 # endif 316 #endif /* ZSTD_ALIGNED */ 317 318 319 /*-************************************************************** 320 * Sanitizer 321 *****************************************************************/ 322 323 /** 324 * Zstd relies on pointer overflow in its decompressor. 325 * We add this attribute to functions that rely on pointer overflow. 326 */ 327 #ifndef ZSTD_ALLOW_POINTER_OVERFLOW_ATTR 328 # if __has_attribute(no_sanitize) 329 # if !defined(__clang__) && defined(__GNUC__) && __GNUC__ < 8 330 /* gcc < 8 only has signed-integer-overlow which triggers on pointer overflow */ 331 # define ZSTD_ALLOW_POINTER_OVERFLOW_ATTR __attribute__((no_sanitize("signed-integer-overflow"))) 332 # else 333 /* older versions of clang [3.7, 5.0) will warn that pointer-overflow is ignored. */ 334 # define ZSTD_ALLOW_POINTER_OVERFLOW_ATTR __attribute__((no_sanitize("pointer-overflow"))) 335 # endif 336 # else 337 # define ZSTD_ALLOW_POINTER_OVERFLOW_ATTR 338 # endif 339 #endif 340 341 /** 342 * Helper function to perform a wrapped pointer difference without triggering 343 * UBSAN. 344 * 345 * @returns lhs - rhs with wrapping 346 */ 347 MEM_STATIC 348 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR 349 ptrdiff_t ZSTD_wrappedPtrDiff(unsigned char const* lhs, unsigned char const* rhs) 350 { 351 return lhs - rhs; 352 } 353 354 /** 355 * Helper function to perform a wrapped pointer add without triggering UBSAN. 356 * 357 * @return ptr + add with wrapping 358 */ 359 MEM_STATIC 360 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR 361 unsigned char const* ZSTD_wrappedPtrAdd(unsigned char const* ptr, ptrdiff_t add) 362 { 363 return ptr + add; 364 } 365 366 /** 367 * Helper function to perform a wrapped pointer subtraction without triggering 368 * UBSAN. 369 * 370 * @return ptr - sub with wrapping 371 */ 372 MEM_STATIC 373 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR 374 unsigned char const* ZSTD_wrappedPtrSub(unsigned char const* ptr, ptrdiff_t sub) 375 { 376 return ptr - sub; 377 } 378 379 /** 380 * Helper function to add to a pointer that works around C's undefined behavior 381 * of adding 0 to NULL. 382 * 383 * @returns `ptr + add` except it defines `NULL + 0 == NULL`. 384 */ 385 MEM_STATIC 386 unsigned char* ZSTD_maybeNullPtrAdd(unsigned char* ptr, ptrdiff_t add) 387 { 388 return add > 0 ? ptr + add : ptr; 389 } 390 391 /* Issue #3240 reports an ASAN failure on an llvm-mingw build. Out of an 392 * abundance of caution, disable our custom poisoning on mingw. */ 393 #ifdef __MINGW32__ 394 #ifndef ZSTD_ASAN_DONT_POISON_WORKSPACE 395 #define ZSTD_ASAN_DONT_POISON_WORKSPACE 1 396 #endif 397 #ifndef ZSTD_MSAN_DONT_POISON_WORKSPACE 398 #define ZSTD_MSAN_DONT_POISON_WORKSPACE 1 399 #endif 400 #endif 401 402 #if ZSTD_MEMORY_SANITIZER && !defined(ZSTD_MSAN_DONT_POISON_WORKSPACE) 403 /* Not all platforms that support msan provide sanitizers/msan_interface.h. 404 * We therefore declare the functions we need ourselves, rather than trying to 405 * include the header file... */ 406 #include <stddef.h> /* size_t */ 407 #define ZSTD_DEPS_NEED_STDINT 408 #include "zstd_deps.h" /* intptr_t */ 409 410 /* Make memory region fully initialized (without changing its contents). */ 411 void __msan_unpoison(const volatile void *a, size_t size); 412 413 /* Make memory region fully uninitialized (without changing its contents). 414 This is a legacy interface that does not update origin information. Use 415 __msan_allocated_memory() instead. */ 416 void __msan_poison(const volatile void *a, size_t size); 417 418 /* Returns the offset of the first (at least partially) poisoned byte in the 419 memory range, or -1 if the whole range is good. */ 420 intptr_t __msan_test_shadow(const volatile void *x, size_t size); 421 422 /* Print shadow and origin for the memory range to stderr in a human-readable 423 format. */ 424 void __msan_print_shadow(const volatile void *x, size_t size); 425 #endif 426 427 #if ZSTD_ADDRESS_SANITIZER && !defined(ZSTD_ASAN_DONT_POISON_WORKSPACE) 428 /* Not all platforms that support asan provide sanitizers/asan_interface.h. 429 * We therefore declare the functions we need ourselves, rather than trying to 430 * include the header file... */ 431 #include <stddef.h> /* size_t */ 432 433 /** 434 * Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable. 435 * 436 * This memory must be previously allocated by your program. Instrumented 437 * code is forbidden from accessing addresses in this region until it is 438 * unpoisoned. This function is not guaranteed to poison the entire region - 439 * it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan 440 * alignment restrictions. 441 * 442 * \note This function is not thread-safe because no two threads can poison or 443 * unpoison memory in the same memory region simultaneously. 444 * 445 * \param addr Start of memory region. 446 * \param size Size of memory region. */ 447 void __asan_poison_memory_region(void const volatile *addr, size_t size); 448 449 /** 450 * Marks a memory region (<c>[addr, addr+size)</c>) as addressable. 451 * 452 * This memory must be previously allocated by your program. Accessing 453 * addresses in this region is allowed until this region is poisoned again. 454 * This function could unpoison a super-region of <c>[addr, addr+size)</c> due 455 * to ASan alignment restrictions. 456 * 457 * \note This function is not thread-safe because no two threads can 458 * poison or unpoison memory in the same memory region simultaneously. 459 * 460 * \param addr Start of memory region. 461 * \param size Size of memory region. */ 462 void __asan_unpoison_memory_region(void const volatile *addr, size_t size); 463 #endif 464 465 #endif /* ZSTD_COMPILER_H */ 466