zstd/lib/common/mem.h

408 lines
13 KiB
C
Raw Normal View History

/*
2016-08-30 17:04:33 +00:00
* Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
2016-08-30 17:04:33 +00:00
*/
2015-10-18 21:18:32 +00:00
#ifndef MEM_H_MODULE
#define MEM_H_MODULE
#if defined (__cplusplus)
extern "C" {
#endif
2016-02-03 01:46:46 +00:00
/*-****************************************
* Dependencies
2015-10-18 21:18:32 +00:00
******************************************/
#include <stddef.h> /* size_t, ptrdiff_t */
#include <string.h> /* memcpy */
2015-10-18 21:18:32 +00:00
2016-02-03 01:46:46 +00:00
/*-****************************************
* Compiler specifics
2015-10-18 21:18:32 +00:00
******************************************/
#if defined(_MSC_VER) /* Visual Studio */
# include <stdlib.h> /* _byteswap_ulong */
# include <intrin.h> /* _byteswap_* */
2016-05-17 12:30:19 +00:00
#endif
#if defined(__GNUC__)
# define MEM_STATIC static __inline __attribute__((unused))
#elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
2015-10-18 21:18:32 +00:00
# define MEM_STATIC static inline
#elif defined(_MSC_VER)
# define MEM_STATIC static __inline
#else
# define MEM_STATIC static /* this version may generate warnings for unused static functions; disable the relevant warning */
#endif
#ifndef __has_builtin
# define __has_builtin(x) 0 /* compat. with non-clang compilers */
#endif
/* code only tested on 32 and 64 bits systems */
#define MEM_STATIC_ASSERT(c) { enum { MEM_static_assert = 1/(int)(!!(c)) }; }
MEM_STATIC void MEM_check(void) { MEM_STATIC_ASSERT((sizeof(size_t)==4) || (sizeof(size_t)==8)); }
/* detects whether we are being compiled under msan */
#if defined (__has_feature)
# if __has_feature(memory_sanitizer)
# define MEMORY_SANITIZER 1
# endif
#endif
#if defined (MEMORY_SANITIZER)
/* Not all platforms that support msan provide sanitizers/msan_interface.h.
* We therefore declare the functions we need ourselves, rather than trying to
* include the header file... */
#include <stdint.h> /* intptr_t */
/* Make memory region fully initialized (without changing its contents). */
void __msan_unpoison(const volatile void *a, size_t size);
/* Make memory region fully uninitialized (without changing its contents).
This is a legacy interface that does not update origin information. Use
__msan_allocated_memory() instead. */
void __msan_poison(const volatile void *a, size_t size);
/* Returns the offset of the first (at least partially) poisoned byte in the
memory range, or -1 if the whole range is good. */
intptr_t __msan_test_shadow(const volatile void *x, size_t size);
#endif
2015-10-18 21:18:32 +00:00
2016-02-03 01:46:46 +00:00
/*-**************************************************************
2015-10-18 21:18:32 +00:00
* Basic Types
*****************************************************************/
#if !defined (__VMS) && (defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
2015-10-18 21:18:32 +00:00
# include <stdint.h>
typedef uint8_t BYTE;
typedef uint16_t U16;
typedef int16_t S16;
typedef uint32_t U32;
typedef int32_t S32;
typedef uint64_t U64;
typedef int64_t S64;
2015-10-18 21:18:32 +00:00
#else
# include <limits.h>
#if CHAR_BIT != 8
# error "this implementation requires char to be exactly 8-bit type"
#endif
typedef unsigned char BYTE;
#if USHRT_MAX != 65535
# error "this implementation requires short to be exactly 16-bit type"
#endif
2015-10-18 21:18:32 +00:00
typedef unsigned short U16;
typedef signed short S16;
#if UINT_MAX != 4294967295
# error "this implementation requires int to be exactly 32-bit type"
#endif
2015-10-18 21:18:32 +00:00
typedef unsigned int U32;
typedef signed int S32;
/* note : there are no limits defined for long long type in C90.
* limits exist in C99, however, in such case, <stdint.h> is preferred */
2015-10-18 21:18:32 +00:00
typedef unsigned long long U64;
typedef signed long long S64;
#endif
2016-02-03 01:46:46 +00:00
/*-**************************************************************
2015-10-18 21:18:32 +00:00
* Memory I/O
*****************************************************************/
2016-02-11 03:17:50 +00:00
/* MEM_FORCE_MEMORY_ACCESS :
2015-10-18 21:18:32 +00:00
* By default, access to unaligned memory is controlled by `memcpy()`, which is safe and portable.
* Unfortunately, on some target/compiler combinations, the generated assembly is sub-optimal.
* The below switch allow to select different access method for improved performance.
* Method 0 (default) : use `memcpy()`. Safe and portable.
* Method 1 : `__packed` statement. It depends on compiler extension (i.e., not portable).
2015-10-18 21:18:32 +00:00
* This method is safe if your compiler supports it, and *generally* as fast or faster than `memcpy`.
* Method 2 : direct access. This method is portable but violate C standard.
2016-02-03 01:46:46 +00:00
* It can generate buggy code on targets depending on alignment.
* In some circumstances, it's the only known way to get the most performance (i.e. GCC + ARMv6)
2015-10-18 21:18:32 +00:00
* See http://fastcompression.blogspot.fr/2015/08/accessing-unaligned-memory.html for details.
* Prefer these methods in priority order (0 > 1 > 2)
*/
#ifndef MEM_FORCE_MEMORY_ACCESS /* can be defined externally, on command line for example */
# if defined(__GNUC__) && ( defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) )
# define MEM_FORCE_MEMORY_ACCESS 2
# elif defined(__INTEL_COMPILER) || defined(__GNUC__) || defined(__ICCARM__)
2015-10-18 21:18:32 +00:00
# define MEM_FORCE_MEMORY_ACCESS 1
# endif
#endif
MEM_STATIC unsigned MEM_32bits(void) { return sizeof(size_t)==4; }
MEM_STATIC unsigned MEM_64bits(void) { return sizeof(size_t)==8; }
2015-10-18 21:18:32 +00:00
MEM_STATIC unsigned MEM_isLittleEndian(void)
{
const union { U32 u; BYTE c[4]; } one = { 1 }; /* don't use static : performance detrimental */
return one.c[0];
}
#if defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==2)
2016-02-11 23:07:30 +00:00
/* violates C standard, by lying on structure alignment.
2015-10-18 21:18:32 +00:00
Only use if no other choice to achieve best performance on target platform */
MEM_STATIC U16 MEM_read16(const void* memPtr) { return *(const U16*) memPtr; }
MEM_STATIC U32 MEM_read32(const void* memPtr) { return *(const U32*) memPtr; }
MEM_STATIC U64 MEM_read64(const void* memPtr) { return *(const U64*) memPtr; }
2017-07-08 00:13:12 +00:00
MEM_STATIC size_t MEM_readST(const void* memPtr) { return *(const size_t*) memPtr; }
2015-10-18 21:18:32 +00:00
MEM_STATIC void MEM_write16(void* memPtr, U16 value) { *(U16*)memPtr = value; }
MEM_STATIC void MEM_write32(void* memPtr, U32 value) { *(U32*)memPtr = value; }
MEM_STATIC void MEM_write64(void* memPtr, U64 value) { *(U64*)memPtr = value; }
#elif defined(MEM_FORCE_MEMORY_ACCESS) && (MEM_FORCE_MEMORY_ACCESS==1)
/* __pack instructions are safer, but compiler specific, hence potentially problematic for some compilers */
/* currently only defined for gcc and icc */
#if defined(_MSC_VER) || (defined(__INTEL_COMPILER) && defined(WIN32))
2017-03-30 01:51:58 +00:00
__pragma( pack(push, 1) )
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
typedef struct { U16 v; } unalign16;
typedef struct { U32 v; } unalign32;
typedef struct { U64 v; } unalign64;
typedef struct { size_t v; } unalignArch;
__pragma( pack(pop) )
#else
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
typedef struct { U16 v; } __attribute__((packed)) unalign16;
typedef struct { U32 v; } __attribute__((packed)) unalign32;
typedef struct { U64 v; } __attribute__((packed)) unalign64;
typedef struct { size_t v; } __attribute__((packed)) unalignArch;
#endif
2015-10-18 21:18:32 +00:00
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
MEM_STATIC U16 MEM_read16(const void* ptr) { return ((const unalign16*)ptr)->v; }
MEM_STATIC U32 MEM_read32(const void* ptr) { return ((const unalign32*)ptr)->v; }
MEM_STATIC U64 MEM_read64(const void* ptr) { return ((const unalign64*)ptr)->v; }
MEM_STATIC size_t MEM_readST(const void* ptr) { return ((const unalignArch*)ptr)->v; }
2015-10-18 21:18:32 +00:00
Fixed Btree update ZSTD_updateTree() expected to be followed by a Bt match finder, which would update zc->nextToUpdate. With the new optimal match finder, it's not necessarily the case : a match might be found during repcode or hash3, and stops there because it reaches sufficient_len, without even entering the binary tree. Previous policy was to nonetheless update zc->nextToUpdate, but the current position would not be inserted, creating "holes" in the btree, aka positions that will no longer be searched. Now, when current position is not inserted, zc->nextToUpdate is not update, expecting ZSTD_updateTree() to fill the tree later on. Solution selected is that ZSTD_updateTree() takes care of properly setting zc->nextToUpdate, so that it no longer depends on a future function to do this job. It took time to get there, as the issue started with a memory sanitizer error. The pb would have been easier to spot with a proper `assert()`. So this patch add a few of them. Additionnally, I discovered that `make test` does not enable `assert()` during CLI tests. This patch enables them. Unfortunately, these `assert()` triggered other (unrelated) bugs during CLI tests, mostly within zstdmt. So this patch also fixes them. - Changed packed structure for gcc memory access : memory sanitizer would complain that a read "might" reach out-of-bound position on the ground that the `union` is larger than the type accessed. Now, to avoid this issue, each type is independent. - ZSTD_CCtxParams_setParameter() : @return provides the value of parameter, clamped/fixed appropriately. - ZSTDMT : changed constant name to ZSTDMT_JOBSIZE_MIN - ZSTDMT : multithreading is automatically disabled when srcSize <= ZSTDMT_JOBSIZE_MIN, since only one thread will be used in this case (saves memory and runtime). - ZSTDMT : nbThreads is automatically clamped on setting the value.
2017-11-16 20:18:56 +00:00
MEM_STATIC void MEM_write16(void* memPtr, U16 value) { ((unalign16*)memPtr)->v = value; }
MEM_STATIC void MEM_write32(void* memPtr, U32 value) { ((unalign32*)memPtr)->v = value; }
MEM_STATIC void MEM_write64(void* memPtr, U64 value) { ((unalign64*)memPtr)->v = value; }
2015-10-18 21:18:32 +00:00
#else
/* default method, safe and standard.
can sometimes prove slower */
MEM_STATIC U16 MEM_read16(const void* memPtr)
{
U16 val; memcpy(&val, memPtr, sizeof(val)); return val;
}
MEM_STATIC U32 MEM_read32(const void* memPtr)
{
U32 val; memcpy(&val, memPtr, sizeof(val)); return val;
}
MEM_STATIC U64 MEM_read64(const void* memPtr)
{
U64 val; memcpy(&val, memPtr, sizeof(val)); return val;
}
2016-02-11 23:07:30 +00:00
MEM_STATIC size_t MEM_readST(const void* memPtr)
{
size_t val; memcpy(&val, memPtr, sizeof(val)); return val;
}
2015-10-18 21:18:32 +00:00
MEM_STATIC void MEM_write16(void* memPtr, U16 value)
{
memcpy(memPtr, &value, sizeof(value));
}
MEM_STATIC void MEM_write32(void* memPtr, U32 value)
{
memcpy(memPtr, &value, sizeof(value));
}
MEM_STATIC void MEM_write64(void* memPtr, U64 value)
{
memcpy(memPtr, &value, sizeof(value));
}
2016-02-11 03:17:50 +00:00
#endif /* MEM_FORCE_MEMORY_ACCESS */
2015-10-18 21:18:32 +00:00
MEM_STATIC U32 MEM_swap32(U32 in)
{
#if defined(_MSC_VER) /* Visual Studio */
return _byteswap_ulong(in);
#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
|| (defined(__clang__) && __has_builtin(__builtin_bswap32))
return __builtin_bswap32(in);
#else
return ((in << 24) & 0xff000000 ) |
((in << 8) & 0x00ff0000 ) |
((in >> 8) & 0x0000ff00 ) |
((in >> 24) & 0x000000ff );
#endif
}
MEM_STATIC U64 MEM_swap64(U64 in)
{
#if defined(_MSC_VER) /* Visual Studio */
return _byteswap_uint64(in);
#elif (defined (__GNUC__) && (__GNUC__ * 100 + __GNUC_MINOR__ >= 403)) \
|| (defined(__clang__) && __has_builtin(__builtin_bswap64))
return __builtin_bswap64(in);
#else
return ((in << 56) & 0xff00000000000000ULL) |
((in << 40) & 0x00ff000000000000ULL) |
((in << 24) & 0x0000ff0000000000ULL) |
((in << 8) & 0x000000ff00000000ULL) |
((in >> 8) & 0x00000000ff000000ULL) |
((in >> 24) & 0x0000000000ff0000ULL) |
((in >> 40) & 0x000000000000ff00ULL) |
((in >> 56) & 0x00000000000000ffULL);
#endif
}
2016-05-11 16:30:24 +00:00
MEM_STATIC size_t MEM_swapST(size_t in)
{
if (MEM_32bits())
2016-05-11 16:55:27 +00:00
return (size_t)MEM_swap32((U32)in);
2016-05-11 16:30:24 +00:00
else
2016-05-11 16:55:27 +00:00
return (size_t)MEM_swap64((U64)in);
2016-05-11 16:30:24 +00:00
}
/*=== Little endian r/w ===*/
2015-10-18 21:18:32 +00:00
MEM_STATIC U16 MEM_readLE16(const void* memPtr)
{
if (MEM_isLittleEndian())
return MEM_read16(memPtr);
2016-02-03 01:46:46 +00:00
else {
2015-10-18 21:18:32 +00:00
const BYTE* p = (const BYTE*)memPtr;
return (U16)(p[0] + (p[1]<<8));
}
}
MEM_STATIC void MEM_writeLE16(void* memPtr, U16 val)
{
2016-02-03 01:46:46 +00:00
if (MEM_isLittleEndian()) {
2015-10-18 21:18:32 +00:00
MEM_write16(memPtr, val);
2016-02-03 01:46:46 +00:00
} else {
2015-10-18 21:18:32 +00:00
BYTE* p = (BYTE*)memPtr;
p[0] = (BYTE)val;
p[1] = (BYTE)(val>>8);
}
}
MEM_STATIC U32 MEM_readLE24(const void* memPtr)
{
return MEM_readLE16(memPtr) + (((const BYTE*)memPtr)[2] << 16);
}
MEM_STATIC void MEM_writeLE24(void* memPtr, U32 val)
{
MEM_writeLE16(memPtr, (U16)val);
((BYTE*)memPtr)[2] = (BYTE)(val>>16);
}
2015-10-18 21:18:32 +00:00
MEM_STATIC U32 MEM_readLE32(const void* memPtr)
{
if (MEM_isLittleEndian())
return MEM_read32(memPtr);
else
return MEM_swap32(MEM_read32(memPtr));
2015-10-18 21:18:32 +00:00
}
MEM_STATIC void MEM_writeLE32(void* memPtr, U32 val32)
{
if (MEM_isLittleEndian())
2015-10-18 21:18:32 +00:00
MEM_write32(memPtr, val32);
else
MEM_write32(memPtr, MEM_swap32(val32));
2015-10-18 21:18:32 +00:00
}
MEM_STATIC U64 MEM_readLE64(const void* memPtr)
{
if (MEM_isLittleEndian())
return MEM_read64(memPtr);
else
return MEM_swap64(MEM_read64(memPtr));
2015-10-18 21:18:32 +00:00
}
MEM_STATIC void MEM_writeLE64(void* memPtr, U64 val64)
{
if (MEM_isLittleEndian())
2015-10-18 21:18:32 +00:00
MEM_write64(memPtr, val64);
else
MEM_write64(memPtr, MEM_swap64(val64));
2015-10-18 21:18:32 +00:00
}
MEM_STATIC size_t MEM_readLEST(const void* memPtr)
{
if (MEM_32bits())
return (size_t)MEM_readLE32(memPtr);
else
return (size_t)MEM_readLE64(memPtr);
}
MEM_STATIC void MEM_writeLEST(void* memPtr, size_t val)
{
if (MEM_32bits())
MEM_writeLE32(memPtr, (U32)val);
else
MEM_writeLE64(memPtr, (U64)val);
}
/*=== Big endian r/w ===*/
MEM_STATIC U32 MEM_readBE32(const void* memPtr)
{
if (MEM_isLittleEndian())
return MEM_swap32(MEM_read32(memPtr));
else
return MEM_read32(memPtr);
}
MEM_STATIC void MEM_writeBE32(void* memPtr, U32 val32)
{
if (MEM_isLittleEndian())
MEM_write32(memPtr, MEM_swap32(val32));
else
MEM_write32(memPtr, val32);
}
MEM_STATIC U64 MEM_readBE64(const void* memPtr)
{
if (MEM_isLittleEndian())
return MEM_swap64(MEM_read64(memPtr));
else
return MEM_read64(memPtr);
}
MEM_STATIC void MEM_writeBE64(void* memPtr, U64 val64)
{
if (MEM_isLittleEndian())
MEM_write64(memPtr, MEM_swap64(val64));
else
MEM_write64(memPtr, val64);
}
MEM_STATIC size_t MEM_readBEST(const void* memPtr)
{
if (MEM_32bits())
return (size_t)MEM_readBE32(memPtr);
else
return (size_t)MEM_readBE64(memPtr);
}
MEM_STATIC void MEM_writeBEST(void* memPtr, size_t val)
{
if (MEM_32bits())
MEM_writeBE32(memPtr, (U32)val);
else
MEM_writeBE64(memPtr, (U64)val);
}
2015-10-18 21:18:32 +00:00
#if defined (__cplusplus)
}
#endif
#endif /* MEM_H_MODULE */