1 /*
2 * Copyright (c) Meta Platforms, Inc. and affiliates.
3 * All rights reserved.
4 *
5 * This source code is licensed under both the BSD-style license (found in the
6 * LICENSE file in the root directory of this source tree) and the GPLv2 (found
7 * in the COPYING file in the root directory of this source tree).
8 * You may select, at your option, one of the above-listed licenses.
9 */
10
11 #ifndef ZSTD_COMPILER_H
12 #define ZSTD_COMPILER_H
13
14 #include <stddef.h>
15
16 #include "portability_macros.h"
17
18 /*-*******************************************************
19 * Compiler specifics
20 *********************************************************/
21 /* force inlining */
22
23 #if !defined(ZSTD_NO_INLINE)
24 #if (defined(__GNUC__) && !defined(__STRICT_ANSI__)) || defined(__cplusplus) || defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L /* C99 */
25 # define INLINE_KEYWORD inline
26 #else
27 # define INLINE_KEYWORD
28 #endif
29
30 #if defined(__GNUC__) || defined(__IAR_SYSTEMS_ICC__)
31 # define FORCE_INLINE_ATTR __attribute__((always_inline))
32 #elif defined(_MSC_VER)
33 # define FORCE_INLINE_ATTR __forceinline
34 #else
35 # define FORCE_INLINE_ATTR
36 #endif
37
38 #else
39
40 #define INLINE_KEYWORD
41 #define FORCE_INLINE_ATTR
42
43 #endif
44
45 /**
46 On MSVC qsort requires that functions passed into it use the __cdecl calling conversion(CC).
47 This explicitly marks such functions as __cdecl so that the code will still compile
48 if a CC other than __cdecl has been made the default.
49 */
50 #if defined(_MSC_VER)
51 # define WIN_CDECL __cdecl
52 #else
53 # define WIN_CDECL
54 #endif
55
56 /* UNUSED_ATTR tells the compiler it is okay if the function is unused. */
57 #if defined(__GNUC__) || defined(__IAR_SYSTEMS_ICC__)
58 # define UNUSED_ATTR __attribute__((unused))
59 #else
60 # define UNUSED_ATTR
61 #endif
62
63 /**
64 * FORCE_INLINE_TEMPLATE is used to define C "templates", which take constant
65 * parameters. They must be inlined for the compiler to eliminate the constant
66 * branches.
67 */
68 #define FORCE_INLINE_TEMPLATE static INLINE_KEYWORD FORCE_INLINE_ATTR UNUSED_ATTR
69 /**
70 * HINT_INLINE is used to help the compiler generate better code. It is *not*
71 * used for "templates", so it can be tweaked based on the compilers
72 * performance.
73 *
74 * gcc-4.8 and gcc-4.9 have been shown to benefit from leaving off the
75 * always_inline attribute.
76 *
77 * clang up to 5.0.0 (trunk) benefit tremendously from the always_inline
78 * attribute.
79 */
80 #if !defined(__clang__) && defined(__GNUC__) && __GNUC__ >= 4 && __GNUC_MINOR__ >= 8 && __GNUC__ < 5
81 # define HINT_INLINE static INLINE_KEYWORD
82 #else
83 # define HINT_INLINE FORCE_INLINE_TEMPLATE
84 #endif
85
86 /* "soft" inline :
87 * The compiler is free to select if it's a good idea to inline or not.
88 * The main objective is to silence compiler warnings
89 * when a defined function in included but not used.
90 *
91 * Note : this macro is prefixed `MEM_` because it used to be provided by `mem.h` unit.
92 * Updating the prefix is probably preferable, but requires a fairly large codemod,
93 * since this name is used everywhere.
94 */
95 #ifndef MEM_STATIC /* already defined in Linux Kernel mem.h */
96 #if defined(__GNUC__)
97 # define MEM_STATIC static __inline UNUSED_ATTR
98 #elif defined(__IAR_SYSTEMS_ICC__)
99 # define MEM_STATIC static inline UNUSED_ATTR
100 #elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
101 # define MEM_STATIC static inline
102 #elif defined(_MSC_VER)
103 # define MEM_STATIC static __inline
104 #else
105 # define MEM_STATIC static /* this version may generate warnings for unused static functions; disable the relevant warning */
106 #endif
107 #endif
108
109 /* force no inlining */
110 #ifdef _MSC_VER
111 # define FORCE_NOINLINE static __declspec(noinline)
112 #else
113 # if defined(__GNUC__) || defined(__IAR_SYSTEMS_ICC__)
114 # define FORCE_NOINLINE static __attribute__((__noinline__))
115 # else
116 # define FORCE_NOINLINE static
117 # endif
118 #endif
119
120
121 /* target attribute */
122 #if defined(__GNUC__) || defined(__IAR_SYSTEMS_ICC__)
123 # define TARGET_ATTRIBUTE(target) __attribute__((__target__(target)))
124 #else
125 # define TARGET_ATTRIBUTE(target)
126 #endif
127
128 /* Target attribute for BMI2 dynamic dispatch.
129 * Enable lzcnt, bmi, and bmi2.
130 * We test for bmi1 & bmi2. lzcnt is included in bmi1.
131 */
132 #define BMI2_TARGET_ATTRIBUTE TARGET_ATTRIBUTE("lzcnt,bmi,bmi2")
133
134 /* prefetch
135 * can be disabled, by declaring NO_PREFETCH build macro */
136 #if defined(NO_PREFETCH)
137 # define PREFETCH_L1(ptr) do { (void)(ptr); } while (0) /* disabled */
138 # define PREFETCH_L2(ptr) do { (void)(ptr); } while (0) /* disabled */
139 #else
140 # if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_I86)) && !defined(_M_ARM64EC) /* _mm_prefetch() is not defined outside of x86/x64 */
141 # include <mmintrin.h> /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
142 # define PREFETCH_L1(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
143 # define PREFETCH_L2(ptr) _mm_prefetch((const char*)(ptr), _MM_HINT_T1)
144 # elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
145 # define PREFETCH_L1(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
146 # define PREFETCH_L2(ptr) __builtin_prefetch((ptr), 0 /* rw==read */, 2 /* locality */)
147 # elif defined(__aarch64__)
148 # define PREFETCH_L1(ptr) do { __asm__ __volatile__("prfm pldl1keep, %0" ::"Q"(*(ptr))); } while (0)
149 # define PREFETCH_L2(ptr) do { __asm__ __volatile__("prfm pldl2keep, %0" ::"Q"(*(ptr))); } while (0)
150 # else
151 # define PREFETCH_L1(ptr) do { (void)(ptr); } while (0) /* disabled */
152 # define PREFETCH_L2(ptr) do { (void)(ptr); } while (0) /* disabled */
153 # endif
154 #endif /* NO_PREFETCH */
155
156 #define CACHELINE_SIZE 64
157
158 #define PREFETCH_AREA(p, s) \
159 do { \
160 const char* const _ptr = (const char*)(p); \
161 size_t const _size = (size_t)(s); \
162 size_t _pos; \
163 for (_pos=0; _pos<_size; _pos+=CACHELINE_SIZE) { \
164 PREFETCH_L2(_ptr + _pos); \
165 } \
166 } while (0)
167
168 /* vectorization
169 * older GCC (pre gcc-4.3 picked as the cutoff) uses a different syntax,
170 * and some compilers, like Intel ICC and MCST LCC, do not support it at all. */
171 #if !defined(__INTEL_COMPILER) && !defined(__clang__) && defined(__GNUC__) && !defined(__LCC__)
172 # if (__GNUC__ == 4 && __GNUC_MINOR__ > 3) || (__GNUC__ >= 5)
173 # define DONT_VECTORIZE __attribute__((optimize("no-tree-vectorize")))
174 # else
175 # define DONT_VECTORIZE _Pragma("GCC optimize(\"no-tree-vectorize\")")
176 # endif
177 #else
178 # define DONT_VECTORIZE
179 #endif
180
181 /* Tell the compiler that a branch is likely or unlikely.
182 * Only use these macros if it causes the compiler to generate better code.
183 * If you can remove a LIKELY/UNLIKELY annotation without speed changes in gcc
184 * and clang, please do.
185 */
186 #if defined(__GNUC__)
187 #define LIKELY(x) (__builtin_expect((x), 1))
188 #define UNLIKELY(x) (__builtin_expect((x), 0))
189 #else
190 #define LIKELY(x) (x)
191 #define UNLIKELY(x) (x)
192 #endif
193
194 #if __has_builtin(__builtin_unreachable) || (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 5)))
195 # define ZSTD_UNREACHABLE do { assert(0), __builtin_unreachable(); } while (0)
196 #else
197 # define ZSTD_UNREACHABLE do { assert(0); } while (0)
198 #endif
199
200 /* disable warnings */
201 #ifdef _MSC_VER /* Visual Studio */
202 # include <intrin.h> /* For Visual 2005 */
203 # pragma warning(disable : 4100) /* disable: C4100: unreferenced formal parameter */
204 # pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
205 # pragma warning(disable : 4204) /* disable: C4204: non-constant aggregate initializer */
206 # pragma warning(disable : 4214) /* disable: C4214: non-int bitfields */
207 # pragma warning(disable : 4324) /* disable: C4324: padded structure */
208 #endif
209
210 /* compile time determination of SIMD support */
211 #if !defined(ZSTD_NO_INTRINSICS)
212 # if defined(__AVX2__)
213 # define ZSTD_ARCH_X86_AVX2
214 # endif
215 # if defined(__SSE2__) || defined(_M_X64) || (defined (_M_IX86) && defined(_M_IX86_FP) && (_M_IX86_FP >= 2))
216 # define ZSTD_ARCH_X86_SSE2
217 # endif
218 # if defined(__ARM_NEON) || defined(_M_ARM64)
219 # define ZSTD_ARCH_ARM_NEON
220 # endif
221 #
222 # if defined(ZSTD_ARCH_X86_AVX2)
223 # include <immintrin.h>
224 # endif
225 # if defined(ZSTD_ARCH_X86_SSE2)
226 # include <emmintrin.h>
227 # elif defined(ZSTD_ARCH_ARM_NEON)
228 # include <arm_neon.h>
229 # endif
230 #endif
231
232 /* C-language Attributes are added in C23. */
233 #if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
234 # define ZSTD_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
235 #else
236 # define ZSTD_HAS_C_ATTRIBUTE(x) 0
237 #endif
238
239 /* Only use C++ attributes in C++. Some compilers report support for C++
240 * attributes when compiling with C.
241 */
242 #if defined(__cplusplus) && defined(__has_cpp_attribute)
243 # define ZSTD_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
244 #else
245 # define ZSTD_HAS_CPP_ATTRIBUTE(x) 0
246 #endif
247
248 /* Define ZSTD_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute.
249 * - C23: https://en.cppreference.com/w/c/language/attributes/fallthrough
250 * - CPP17: https://en.cppreference.com/w/cpp/language/attributes/fallthrough
251 * - Else: __attribute__((__fallthrough__))
252 */
253 #ifndef ZSTD_FALLTHROUGH
254 # if ZSTD_HAS_C_ATTRIBUTE(fallthrough)
255 # define ZSTD_FALLTHROUGH [[fallthrough]]
256 # elif ZSTD_HAS_CPP_ATTRIBUTE(fallthrough)
257 # define ZSTD_FALLTHROUGH [[fallthrough]]
258 # elif __has_attribute(__fallthrough__)
259 /* Leading semicolon is to satisfy gcc-11 with -pedantic. Without the semicolon
260 * gcc complains about: a label can only be part of a statement and a declaration is not a statement.
261 */
262 # define ZSTD_FALLTHROUGH ; __attribute__((__fallthrough__))
263 # else
264 # define ZSTD_FALLTHROUGH
265 # endif
266 #endif
267
268 /*-**************************************************************
269 * Alignment
270 *****************************************************************/
271
272 /* @return 1 if @u is a 2^n value, 0 otherwise
273 * useful to check a value is valid for alignment restrictions */
ZSTD_isPower2(size_t u)274 MEM_STATIC int ZSTD_isPower2(size_t u) {
275 return (u & (u-1)) == 0;
276 }
277
278 /* this test was initially positioned in mem.h,
279 * but this file is removed (or replaced) for linux kernel
280 * so it's now hosted in compiler.h,
281 * which remains valid for both user & kernel spaces.
282 */
283
284 #ifndef ZSTD_ALIGNOF
285 # if defined(__GNUC__) || defined(_MSC_VER)
286 /* covers gcc, clang & MSVC */
287 /* note : this section must come first, before C11,
288 * due to a limitation in the kernel source generator */
289 # define ZSTD_ALIGNOF(T) __alignof(T)
290
291 # elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)
292 /* C11 support */
293 # include <stdalign.h>
294 # define ZSTD_ALIGNOF(T) alignof(T)
295
296 # else
297 /* No known support for alignof() - imperfect backup */
298 # define ZSTD_ALIGNOF(T) (sizeof(void*) < sizeof(T) ? sizeof(void*) : sizeof(T))
299
300 # endif
301 #endif /* ZSTD_ALIGNOF */
302
303 #ifndef ZSTD_ALIGNED
304 /* C90-compatible alignment macro (GCC/Clang). Adjust for other compilers if needed. */
305 # if defined(__GNUC__) || defined(__clang__)
306 # define ZSTD_ALIGNED(a) __attribute__((aligned(a)))
307 # elif defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* C11 */
308 # define ZSTD_ALIGNED(a) _Alignas(a)
309 #elif defined(_MSC_VER)
310 # define ZSTD_ALIGNED(n) __declspec(align(n))
311 # else
312 /* this compiler will require its own alignment instruction */
313 # define ZSTD_ALIGNED(...)
314 # endif
315 #endif /* ZSTD_ALIGNED */
316
317
318 /*-**************************************************************
319 * Sanitizer
320 *****************************************************************/
321
322 /**
323 * Zstd relies on pointer overflow in its decompressor.
324 * We add this attribute to functions that rely on pointer overflow.
325 */
326 #ifndef ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
327 # if __has_attribute(no_sanitize)
328 # if !defined(__clang__) && defined(__GNUC__) && __GNUC__ < 8
329 /* gcc < 8 only has signed-integer-overlow which triggers on pointer overflow */
330 # define ZSTD_ALLOW_POINTER_OVERFLOW_ATTR __attribute__((no_sanitize("signed-integer-overflow")))
331 # else
332 /* older versions of clang [3.7, 5.0) will warn that pointer-overflow is ignored. */
333 # define ZSTD_ALLOW_POINTER_OVERFLOW_ATTR __attribute__((no_sanitize("pointer-overflow")))
334 # endif
335 # else
336 # define ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
337 # endif
338 #endif
339
340 /**
341 * Helper function to perform a wrapped pointer difference without triggering
342 * UBSAN.
343 *
344 * @returns lhs - rhs with wrapping
345 */
346 MEM_STATIC
347 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_wrappedPtrDiff(unsigned char const * lhs,unsigned char const * rhs)348 ptrdiff_t ZSTD_wrappedPtrDiff(unsigned char const* lhs, unsigned char const* rhs)
349 {
350 return lhs - rhs;
351 }
352
353 /**
354 * Helper function to perform a wrapped pointer add without triggering UBSAN.
355 *
356 * @return ptr + add with wrapping
357 */
358 MEM_STATIC
359 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_wrappedPtrAdd(unsigned char const * ptr,ptrdiff_t add)360 unsigned char const* ZSTD_wrappedPtrAdd(unsigned char const* ptr, ptrdiff_t add)
361 {
362 return ptr + add;
363 }
364
365 /**
366 * Helper function to perform a wrapped pointer subtraction without triggering
367 * UBSAN.
368 *
369 * @return ptr - sub with wrapping
370 */
371 MEM_STATIC
372 ZSTD_ALLOW_POINTER_OVERFLOW_ATTR
ZSTD_wrappedPtrSub(unsigned char const * ptr,ptrdiff_t sub)373 unsigned char const* ZSTD_wrappedPtrSub(unsigned char const* ptr, ptrdiff_t sub)
374 {
375 return ptr - sub;
376 }
377
378 /**
379 * Helper function to add to a pointer that works around C's undefined behavior
380 * of adding 0 to NULL.
381 *
382 * @returns `ptr + add` except it defines `NULL + 0 == NULL`.
383 */
384 MEM_STATIC
ZSTD_maybeNullPtrAdd(unsigned char * ptr,ptrdiff_t add)385 unsigned char* ZSTD_maybeNullPtrAdd(unsigned char* ptr, ptrdiff_t add)
386 {
387 return add > 0 ? ptr + add : ptr;
388 }
389
390 /* Issue #3240 reports an ASAN failure on an llvm-mingw build. Out of an
391 * abundance of caution, disable our custom poisoning on mingw. */
392 #ifdef __MINGW32__
393 #ifndef ZSTD_ASAN_DONT_POISON_WORKSPACE
394 #define ZSTD_ASAN_DONT_POISON_WORKSPACE 1
395 #endif
396 #ifndef ZSTD_MSAN_DONT_POISON_WORKSPACE
397 #define ZSTD_MSAN_DONT_POISON_WORKSPACE 1
398 #endif
399 #endif
400
401 #if ZSTD_MEMORY_SANITIZER && !defined(ZSTD_MSAN_DONT_POISON_WORKSPACE)
402 /* Not all platforms that support msan provide sanitizers/msan_interface.h.
403 * We therefore declare the functions we need ourselves, rather than trying to
404 * include the header file... */
405 #include <stddef.h> /* size_t */
406 #define ZSTD_DEPS_NEED_STDINT
407 #include "zstd_deps.h" /* intptr_t */
408
409 /* Make memory region fully initialized (without changing its contents). */
410 void __msan_unpoison(const volatile void *a, size_t size);
411
412 /* Make memory region fully uninitialized (without changing its contents).
413 This is a legacy interface that does not update origin information. Use
414 __msan_allocated_memory() instead. */
415 void __msan_poison(const volatile void *a, size_t size);
416
417 /* Returns the offset of the first (at least partially) poisoned byte in the
418 memory range, or -1 if the whole range is good. */
419 intptr_t __msan_test_shadow(const volatile void *x, size_t size);
420
421 /* Print shadow and origin for the memory range to stderr in a human-readable
422 format. */
423 void __msan_print_shadow(const volatile void *x, size_t size);
424 #endif
425
426 #if ZSTD_ADDRESS_SANITIZER && !defined(ZSTD_ASAN_DONT_POISON_WORKSPACE)
427 /* Not all platforms that support asan provide sanitizers/asan_interface.h.
428 * We therefore declare the functions we need ourselves, rather than trying to
429 * include the header file... */
430 #include <stddef.h> /* size_t */
431
432 /**
433 * Marks a memory region (<c>[addr, addr+size)</c>) as unaddressable.
434 *
435 * This memory must be previously allocated by your program. Instrumented
436 * code is forbidden from accessing addresses in this region until it is
437 * unpoisoned. This function is not guaranteed to poison the entire region -
438 * it could poison only a subregion of <c>[addr, addr+size)</c> due to ASan
439 * alignment restrictions.
440 *
441 * \note This function is not thread-safe because no two threads can poison or
442 * unpoison memory in the same memory region simultaneously.
443 *
444 * \param addr Start of memory region.
445 * \param size Size of memory region. */
446 void __asan_poison_memory_region(void const volatile *addr, size_t size);
447
448 /**
449 * Marks a memory region (<c>[addr, addr+size)</c>) as addressable.
450 *
451 * This memory must be previously allocated by your program. Accessing
452 * addresses in this region is allowed until this region is poisoned again.
453 * This function could unpoison a super-region of <c>[addr, addr+size)</c> due
454 * to ASan alignment restrictions.
455 *
456 * \note This function is not thread-safe because no two threads can
457 * poison or unpoison memory in the same memory region simultaneously.
458 *
459 * \param addr Start of memory region.
460 * \param size Size of memory region. */
461 void __asan_unpoison_memory_region(void const volatile *addr, size_t size);
462 #endif
463
464 #endif /* ZSTD_COMPILER_H */
465