From 336da838ce43edae40dd708dd465f1f43a2c312d Mon Sep 17 00:00:00 2001 From: Wiimm Date: Fri, 22 Sep 2023 17:19:28 +0200 Subject: SZS 2.38a --- project/src/liblzma/7zTypes.h | 597 ++++++++ project/src/liblzma/Compiler.h | 159 ++ project/src/liblzma/CpuArch.c | 823 +++++++++++ project/src/liblzma/CpuArch.h | 523 +++++++ project/src/liblzma/LzFind.c | 1717 ++++++++++++++++++++++ project/src/liblzma/LzFind.h | 159 ++ project/src/liblzma/LzFindMt.h | 109 ++ project/src/liblzma/LzHash.h | 34 + project/src/liblzma/LzmaDec.c | 1363 +++++++++++++++++ project/src/liblzma/LzmaDec.h | 237 +++ project/src/liblzma/LzmaEnc.c | 3144 ++++++++++++++++++++++++++++++++++++++++ project/src/liblzma/LzmaEnc.h | 83 ++ project/src/liblzma/Precomp.h | 10 + project/src/liblzma/Threads.h | 240 +++ 14 files changed, 9198 insertions(+) create mode 100644 project/src/liblzma/7zTypes.h create mode 100644 project/src/liblzma/Compiler.h create mode 100644 project/src/liblzma/CpuArch.c create mode 100644 project/src/liblzma/CpuArch.h create mode 100644 project/src/liblzma/LzFind.c create mode 100644 project/src/liblzma/LzFind.h create mode 100644 project/src/liblzma/LzFindMt.h create mode 100644 project/src/liblzma/LzHash.h create mode 100644 project/src/liblzma/LzmaDec.c create mode 100644 project/src/liblzma/LzmaDec.h create mode 100644 project/src/liblzma/LzmaEnc.c create mode 100644 project/src/liblzma/LzmaEnc.h create mode 100644 project/src/liblzma/Precomp.h create mode 100644 project/src/liblzma/Threads.h (limited to 'project/src/liblzma') diff --git a/project/src/liblzma/7zTypes.h b/project/src/liblzma/7zTypes.h new file mode 100644 index 0000000..0eb9513 --- /dev/null +++ b/project/src/liblzma/7zTypes.h @@ -0,0 +1,597 @@ +/* 7zTypes.h -- Basic types +2023-04-02 : Igor Pavlov : Public domain */ + +#ifndef ZIP7_7Z_TYPES_H +#define ZIP7_7Z_TYPES_H + +#ifdef _WIN32 +/* #include */ +#else +#include +#endif + +#include + +#ifndef EXTERN_C_BEGIN +#ifdef __cplusplus +#define EXTERN_C_BEGIN extern "C" { +#define EXTERN_C_END } +#else +#define EXTERN_C_BEGIN +#define EXTERN_C_END +#endif +#endif + +EXTERN_C_BEGIN + +#define SZ_OK 0 + +#define SZ_ERROR_DATA 1 +#define SZ_ERROR_MEM 2 +#define SZ_ERROR_CRC 3 +#define SZ_ERROR_UNSUPPORTED 4 +#define SZ_ERROR_PARAM 5 +#define SZ_ERROR_INPUT_EOF 6 +#define SZ_ERROR_OUTPUT_EOF 7 +#define SZ_ERROR_READ 8 +#define SZ_ERROR_WRITE 9 +#define SZ_ERROR_PROGRESS 10 +#define SZ_ERROR_FAIL 11 +#define SZ_ERROR_THREAD 12 + +#define SZ_ERROR_ARCHIVE 16 +#define SZ_ERROR_NO_ARCHIVE 17 + +typedef int SRes; + + +#ifdef _MSC_VER + #if _MSC_VER > 1200 + #define MY_ALIGN(n) __declspec(align(n)) + #else + #define MY_ALIGN(n) + #endif +#else + /* + // C11/C++11: + #include + #define MY_ALIGN(n) alignas(n) + */ + #define MY_ALIGN(n) __attribute__ ((aligned(n))) +#endif + + +#ifdef _WIN32 + +/* typedef DWORD WRes; */ +typedef unsigned WRes; +#define MY_SRes_HRESULT_FROM_WRes(x) HRESULT_FROM_WIN32(x) + +// #define MY_HRES_ERROR_INTERNAL_ERROR MY_SRes_HRESULT_FROM_WRes(ERROR_INTERNAL_ERROR) + +#else // _WIN32 + +// #define ENV_HAVE_LSTAT +typedef int WRes; + +// (FACILITY_ERRNO = 0x800) is 7zip's FACILITY constant to represent (errno) errors in HRESULT +#define MY_FACILITY_ERRNO 0x800 +#define MY_FACILITY_WIN32 7 +#define MY_FACILITY_WRes MY_FACILITY_ERRNO + +#define MY_HRESULT_FROM_errno_CONST_ERROR(x) ((HRESULT)( \ + ( (HRESULT)(x) & 0x0000FFFF) \ + | (MY_FACILITY_WRes << 16) \ + | (HRESULT)0x80000000 )) + +#define MY_SRes_HRESULT_FROM_WRes(x) \ + ((HRESULT)(x) <= 0 ? ((HRESULT)(x)) : MY_HRESULT_FROM_errno_CONST_ERROR(x)) + +// we call macro HRESULT_FROM_WIN32 for system errors (WRes) that are (errno) +#define HRESULT_FROM_WIN32(x) MY_SRes_HRESULT_FROM_WRes(x) + +/* +#define ERROR_FILE_NOT_FOUND 2L +#define ERROR_ACCESS_DENIED 5L +#define ERROR_NO_MORE_FILES 18L +#define ERROR_LOCK_VIOLATION 33L +#define ERROR_FILE_EXISTS 80L +#define ERROR_DISK_FULL 112L +#define ERROR_NEGATIVE_SEEK 131L +#define ERROR_ALREADY_EXISTS 183L +#define ERROR_DIRECTORY 267L +#define ERROR_TOO_MANY_POSTS 298L + +#define ERROR_INTERNAL_ERROR 1359L +#define ERROR_INVALID_REPARSE_DATA 4392L +#define ERROR_REPARSE_TAG_INVALID 4393L +#define ERROR_REPARSE_TAG_MISMATCH 4394L +*/ + +// we use errno equivalents for some WIN32 errors: + +#define ERROR_INVALID_PARAMETER EINVAL +#define ERROR_INVALID_FUNCTION EINVAL +#define ERROR_ALREADY_EXISTS EEXIST +#define ERROR_FILE_EXISTS EEXIST +#define ERROR_PATH_NOT_FOUND ENOENT +#define ERROR_FILE_NOT_FOUND ENOENT +#define ERROR_DISK_FULL ENOSPC +// #define ERROR_INVALID_HANDLE EBADF + +// we use FACILITY_WIN32 for errors that has no errno equivalent +// Too many posts were made to a semaphore. +#define ERROR_TOO_MANY_POSTS ((HRESULT)0x8007012AL) +#define ERROR_INVALID_REPARSE_DATA ((HRESULT)0x80071128L) +#define ERROR_REPARSE_TAG_INVALID ((HRESULT)0x80071129L) + +// if (MY_FACILITY_WRes != FACILITY_WIN32), +// we use FACILITY_WIN32 for COM errors: +#define E_OUTOFMEMORY ((HRESULT)0x8007000EL) +#define E_INVALIDARG ((HRESULT)0x80070057L) +#define MY_E_ERROR_NEGATIVE_SEEK ((HRESULT)0x80070083L) + +/* +// we can use FACILITY_ERRNO for some COM errors, that have errno equivalents: +#define E_OUTOFMEMORY MY_HRESULT_FROM_errno_CONST_ERROR(ENOMEM) +#define E_INVALIDARG MY_HRESULT_FROM_errno_CONST_ERROR(EINVAL) +#define MY_E_ERROR_NEGATIVE_SEEK MY_HRESULT_FROM_errno_CONST_ERROR(EINVAL) +*/ + +#define TEXT(quote) quote + +#define FILE_ATTRIBUTE_READONLY 0x0001 +#define FILE_ATTRIBUTE_HIDDEN 0x0002 +#define FILE_ATTRIBUTE_SYSTEM 0x0004 +#define FILE_ATTRIBUTE_DIRECTORY 0x0010 +#define FILE_ATTRIBUTE_ARCHIVE 0x0020 +#define FILE_ATTRIBUTE_DEVICE 0x0040 +#define FILE_ATTRIBUTE_NORMAL 0x0080 +#define FILE_ATTRIBUTE_TEMPORARY 0x0100 +#define FILE_ATTRIBUTE_SPARSE_FILE 0x0200 +#define FILE_ATTRIBUTE_REPARSE_POINT 0x0400 +#define FILE_ATTRIBUTE_COMPRESSED 0x0800 +#define FILE_ATTRIBUTE_OFFLINE 0x1000 +#define FILE_ATTRIBUTE_NOT_CONTENT_INDEXED 0x2000 +#define FILE_ATTRIBUTE_ENCRYPTED 0x4000 + +#define FILE_ATTRIBUTE_UNIX_EXTENSION 0x8000 /* trick for Unix */ + +#endif + + +#ifndef RINOK +#define RINOK(x) { const int _result_ = (x); if (_result_ != 0) return _result_; } +#endif + +#ifndef RINOK_WRes +#define RINOK_WRes(x) { const WRes _result_ = (x); if (_result_ != 0) return _result_; } +#endif + +typedef unsigned char Byte; +typedef short Int16; +typedef unsigned short UInt16; + +#ifdef Z7_DECL_Int32_AS_long +typedef long Int32; +typedef unsigned long UInt32; +#else +typedef int Int32; +typedef unsigned int UInt32; +#endif + + +#ifndef _WIN32 + +typedef int INT; +typedef Int32 INT32; +typedef unsigned int UINT; +typedef UInt32 UINT32; +typedef INT32 LONG; // LONG, ULONG and DWORD must be 32-bit for _WIN32 compatibility +typedef UINT32 ULONG; + +#undef DWORD +typedef UINT32 DWORD; + +#define VOID void + +#define HRESULT LONG + +typedef void *LPVOID; +// typedef void VOID; +// typedef ULONG_PTR DWORD_PTR, *PDWORD_PTR; +// gcc / clang on Unix : sizeof(long==sizeof(void*) in 32 or 64 bits) +typedef long INT_PTR; +typedef unsigned long UINT_PTR; +typedef long LONG_PTR; +typedef unsigned long DWORD_PTR; + +typedef size_t SIZE_T; + +#endif // _WIN32 + + +#define MY_HRES_ERROR_INTERNAL_ERROR ((HRESULT)0x8007054FL) + + +#ifdef Z7_DECL_Int64_AS_long + +typedef long Int64; +typedef unsigned long UInt64; + +#else + +#if (defined(_MSC_VER) || defined(__BORLANDC__)) && !defined(__clang__) +typedef __int64 Int64; +typedef unsigned __int64 UInt64; +#else +#if defined(__clang__) || defined(__GNUC__) +#include +typedef int64_t Int64; +typedef uint64_t UInt64; +#else +typedef long long int Int64; +typedef unsigned long long int UInt64; +// #define UINT64_CONST(n) n ## ULL +#endif +#endif + +#endif + +#define UINT64_CONST(n) n + + +#ifdef Z7_DECL_SizeT_AS_unsigned_int +typedef unsigned int SizeT; +#else +typedef size_t SizeT; +#endif + +/* +#if (defined(_MSC_VER) && _MSC_VER <= 1200) +typedef size_t MY_uintptr_t; +#else +#include +typedef uintptr_t MY_uintptr_t; +#endif +*/ + +typedef int BoolInt; +/* typedef BoolInt Bool; */ +#define True 1 +#define False 0 + + +#ifdef _WIN32 +#define Z7_STDCALL __stdcall +#else +#define Z7_STDCALL +#endif + +#ifdef _MSC_VER + +#if _MSC_VER >= 1300 +#define Z7_NO_INLINE __declspec(noinline) +#else +#define Z7_NO_INLINE +#endif + +#define Z7_FORCE_INLINE __forceinline + +#define Z7_CDECL __cdecl +#define Z7_FASTCALL __fastcall + +#else // _MSC_VER + +#if (defined(__GNUC__) && (__GNUC__ >= 4)) \ + || (defined(__clang__) && (__clang_major__ >= 4)) \ + || defined(__INTEL_COMPILER) \ + || defined(__xlC__) +#define Z7_NO_INLINE __attribute__((noinline)) +#define Z7_FORCE_INLINE __attribute__((always_inline)) inline +#else +#define Z7_NO_INLINE +#define Z7_FORCE_INLINE +#endif + +#define Z7_CDECL + +#if defined(_M_IX86) \ + || defined(__i386__) +// #define Z7_FASTCALL __attribute__((fastcall)) +// #define Z7_FASTCALL __attribute__((cdecl)) +#define Z7_FASTCALL +#elif defined(MY_CPU_AMD64) +// #define Z7_FASTCALL __attribute__((ms_abi)) +#define Z7_FASTCALL +#else +#define Z7_FASTCALL +#endif + +#endif // _MSC_VER + + +/* The following interfaces use first parameter as pointer to structure */ + +// #define Z7_C_IFACE_CONST_QUAL +#define Z7_C_IFACE_CONST_QUAL const + +#define Z7_C_IFACE_DECL(a) \ + struct a ## _; \ + typedef Z7_C_IFACE_CONST_QUAL struct a ## _ * a ## Ptr; \ + typedef struct a ## _ a; \ + struct a ## _ + + +Z7_C_IFACE_DECL (IByteIn) +{ + Byte (*Read)(IByteInPtr p); /* reads one byte, returns 0 in case of EOF or error */ +}; +#define IByteIn_Read(p) (p)->Read(p) + + +Z7_C_IFACE_DECL (IByteOut) +{ + void (*Write)(IByteOutPtr p, Byte b); +}; +#define IByteOut_Write(p, b) (p)->Write(p, b) + + +Z7_C_IFACE_DECL (ISeqInStream) +{ + SRes (*Read)(ISeqInStreamPtr p, void *buf, size_t *size); + /* if (input(*size) != 0 && output(*size) == 0) means end_of_stream. + (output(*size) < input(*size)) is allowed */ +}; +#define ISeqInStream_Read(p, buf, size) (p)->Read(p, buf, size) + +/* try to read as much as avail in stream and limited by (*processedSize) */ +SRes SeqInStream_ReadMax(ISeqInStreamPtr stream, void *buf, size_t *processedSize); +/* it can return SZ_ERROR_INPUT_EOF */ +// SRes SeqInStream_Read(ISeqInStreamPtr stream, void *buf, size_t size); +// SRes SeqInStream_Read2(ISeqInStreamPtr stream, void *buf, size_t size, SRes errorType); +SRes SeqInStream_ReadByte(ISeqInStreamPtr stream, Byte *buf); + + +Z7_C_IFACE_DECL (ISeqOutStream) +{ + size_t (*Write)(ISeqOutStreamPtr p, const void *buf, size_t size); + /* Returns: result - the number of actually written bytes. + (result < size) means error */ +}; +#define ISeqOutStream_Write(p, buf, size) (p)->Write(p, buf, size) + +typedef enum +{ + SZ_SEEK_SET = 0, + SZ_SEEK_CUR = 1, + SZ_SEEK_END = 2 +} ESzSeek; + + +Z7_C_IFACE_DECL (ISeekInStream) +{ + SRes (*Read)(ISeekInStreamPtr p, void *buf, size_t *size); /* same as ISeqInStream::Read */ + SRes (*Seek)(ISeekInStreamPtr p, Int64 *pos, ESzSeek origin); +}; +#define ISeekInStream_Read(p, buf, size) (p)->Read(p, buf, size) +#define ISeekInStream_Seek(p, pos, origin) (p)->Seek(p, pos, origin) + + +Z7_C_IFACE_DECL (ILookInStream) +{ + SRes (*Look)(ILookInStreamPtr p, const void **buf, size_t *size); + /* if (input(*size) != 0 && output(*size) == 0) means end_of_stream. + (output(*size) > input(*size)) is not allowed + (output(*size) < input(*size)) is allowed */ + SRes (*Skip)(ILookInStreamPtr p, size_t offset); + /* offset must be <= output(*size) of Look */ + SRes (*Read)(ILookInStreamPtr p, void *buf, size_t *size); + /* reads directly (without buffer). It's same as ISeqInStream::Read */ + SRes (*Seek)(ILookInStreamPtr p, Int64 *pos, ESzSeek origin); +}; + +#define ILookInStream_Look(p, buf, size) (p)->Look(p, buf, size) +#define ILookInStream_Skip(p, offset) (p)->Skip(p, offset) +#define ILookInStream_Read(p, buf, size) (p)->Read(p, buf, size) +#define ILookInStream_Seek(p, pos, origin) (p)->Seek(p, pos, origin) + + +SRes LookInStream_LookRead(ILookInStreamPtr stream, void *buf, size_t *size); +SRes LookInStream_SeekTo(ILookInStreamPtr stream, UInt64 offset); + +/* reads via ILookInStream::Read */ +SRes LookInStream_Read2(ILookInStreamPtr stream, void *buf, size_t size, SRes errorType); +SRes LookInStream_Read(ILookInStreamPtr stream, void *buf, size_t size); + + +typedef struct +{ + ILookInStream vt; + ISeekInStreamPtr realStream; + + size_t pos; + size_t size; /* it's data size */ + + /* the following variables must be set outside */ + Byte *buf; + size_t bufSize; +} CLookToRead2; + +void LookToRead2_CreateVTable(CLookToRead2 *p, int lookahead); + +#define LookToRead2_INIT(p) { (p)->pos = (p)->size = 0; } + + +typedef struct +{ + ISeqInStream vt; + ILookInStreamPtr realStream; +} CSecToLook; + +void SecToLook_CreateVTable(CSecToLook *p); + + + +typedef struct +{ + ISeqInStream vt; + ILookInStreamPtr realStream; +} CSecToRead; + +void SecToRead_CreateVTable(CSecToRead *p); + + +Z7_C_IFACE_DECL (ICompressProgress) +{ + SRes (*Progress)(ICompressProgressPtr p, UInt64 inSize, UInt64 outSize); + /* Returns: result. (result != SZ_OK) means break. + Value (UInt64)(Int64)-1 for size means unknown value. */ +}; + +#define ICompressProgress_Progress(p, inSize, outSize) (p)->Progress(p, inSize, outSize) + + + +typedef struct ISzAlloc ISzAlloc; +typedef const ISzAlloc * ISzAllocPtr; + +struct ISzAlloc +{ + void *(*Alloc)(ISzAllocPtr p, size_t size); + void (*Free)(ISzAllocPtr p, void *address); /* address can be 0 */ +}; + +#define ISzAlloc_Alloc(p, size) (p)->Alloc(p, size) +#define ISzAlloc_Free(p, a) (p)->Free(p, a) + +/* deprecated */ +#define IAlloc_Alloc(p, size) ISzAlloc_Alloc(p, size) +#define IAlloc_Free(p, a) ISzAlloc_Free(p, a) + + + + + +#ifndef MY_offsetof + #ifdef offsetof + #define MY_offsetof(type, m) offsetof(type, m) + /* + #define MY_offsetof(type, m) FIELD_OFFSET(type, m) + */ + #else + #define MY_offsetof(type, m) ((size_t)&(((type *)0)->m)) + #endif +#endif + + + +#ifndef Z7_container_of + +/* +#define Z7_container_of(ptr, type, m) container_of(ptr, type, m) +#define Z7_container_of(ptr, type, m) CONTAINING_RECORD(ptr, type, m) +#define Z7_container_of(ptr, type, m) ((type *)((char *)(ptr) - offsetof(type, m))) +#define Z7_container_of(ptr, type, m) (&((type *)0)->m == (ptr), ((type *)(((char *)(ptr)) - MY_offsetof(type, m)))) +*/ + +/* + GCC shows warning: "perhaps the 'offsetof' macro was used incorrectly" + GCC 3.4.4 : classes with constructor + GCC 4.8.1 : classes with non-public variable members" +*/ + +#define Z7_container_of(ptr, type, m) \ + ((type *)(void *)((char *)(void *) \ + (1 ? (ptr) : &((type *)NULL)->m) - MY_offsetof(type, m))) + +#define Z7_container_of_CONST(ptr, type, m) \ + ((const type *)(const void *)((const char *)(const void *) \ + (1 ? (ptr) : &((type *)NULL)->m) - MY_offsetof(type, m))) + +/* +#define Z7_container_of_NON_CONST_FROM_CONST(ptr, type, m) \ + ((type *)(void *)(const void *)((const char *)(const void *) \ + (1 ? (ptr) : &((type *)NULL)->m) - MY_offsetof(type, m))) +*/ + +#endif + +#define Z7_CONTAINER_FROM_VTBL_SIMPLE(ptr, type, m) ((type *)(void *)(ptr)) + +// #define Z7_CONTAINER_FROM_VTBL(ptr, type, m) Z7_CONTAINER_FROM_VTBL_SIMPLE(ptr, type, m) +#define Z7_CONTAINER_FROM_VTBL(ptr, type, m) Z7_container_of(ptr, type, m) +// #define Z7_CONTAINER_FROM_VTBL(ptr, type, m) Z7_container_of_NON_CONST_FROM_CONST(ptr, type, m) + +#define Z7_CONTAINER_FROM_VTBL_CONST(ptr, type, m) Z7_container_of_CONST(ptr, type, m) + +#define Z7_CONTAINER_FROM_VTBL_CLS(ptr, type, m) Z7_CONTAINER_FROM_VTBL_SIMPLE(ptr, type, m) +/* +#define Z7_CONTAINER_FROM_VTBL_CLS(ptr, type, m) Z7_CONTAINER_FROM_VTBL(ptr, type, m) +*/ +#if defined (__clang__) || defined(__GNUC__) +#define Z7_DIAGNOSCTIC_IGNORE_BEGIN_CAST_QUAL \ + _Pragma("GCC diagnostic push") \ + _Pragma("GCC diagnostic ignored \"-Wcast-qual\"") +#define Z7_DIAGNOSCTIC_IGNORE_END_CAST_QUAL \ + _Pragma("GCC diagnostic pop") +#else +#define Z7_DIAGNOSCTIC_IGNORE_BEGIN_CAST_QUAL +#define Z7_DIAGNOSCTIC_IGNORE_END_CAST_QUAL +#endif + +#define Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR(ptr, type, m, p) \ + Z7_DIAGNOSCTIC_IGNORE_BEGIN_CAST_QUAL \ + type *p = Z7_CONTAINER_FROM_VTBL(ptr, type, m); \ + Z7_DIAGNOSCTIC_IGNORE_END_CAST_QUAL + +#define Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(type) \ + Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR(pp, type, vt, p) + + +// #define ZIP7_DECLARE_HANDLE(name) typedef void *name; +#define Z7_DECLARE_HANDLE(name) struct name##_dummy{int unused;}; typedef struct name##_dummy *name; + + +#define Z7_memset_0_ARRAY(a) memset((a), 0, sizeof(a)) + +#ifndef Z7_ARRAY_SIZE +#define Z7_ARRAY_SIZE(a) (sizeof(a) / sizeof((a)[0])) +#endif + + +#ifdef _WIN32 + +#define CHAR_PATH_SEPARATOR '\\' +#define WCHAR_PATH_SEPARATOR L'\\' +#define STRING_PATH_SEPARATOR "\\" +#define WSTRING_PATH_SEPARATOR L"\\" + +#else + +#define CHAR_PATH_SEPARATOR '/' +#define WCHAR_PATH_SEPARATOR L'/' +#define STRING_PATH_SEPARATOR "/" +#define WSTRING_PATH_SEPARATOR L"/" + +#endif + +#define k_PropVar_TimePrec_0 0 +#define k_PropVar_TimePrec_Unix 1 +#define k_PropVar_TimePrec_DOS 2 +#define k_PropVar_TimePrec_HighPrec 3 +#define k_PropVar_TimePrec_Base 16 +#define k_PropVar_TimePrec_100ns (k_PropVar_TimePrec_Base + 7) +#define k_PropVar_TimePrec_1ns (k_PropVar_TimePrec_Base + 9) + +EXTERN_C_END + +#endif + +/* +#ifndef Z7_ST +#ifdef _7ZIP_ST +#define Z7_ST +#endif +#endif +*/ diff --git a/project/src/liblzma/Compiler.h b/project/src/liblzma/Compiler.h new file mode 100644 index 0000000..ca4618b --- /dev/null +++ b/project/src/liblzma/Compiler.h @@ -0,0 +1,159 @@ +/* Compiler.h : Compiler specific defines and pragmas +2023-04-02 : Igor Pavlov : Public domain */ + +#ifndef ZIP7_INC_COMPILER_H +#define ZIP7_INC_COMPILER_H + +#if defined(__clang__) +# define Z7_CLANG_VERSION (__clang_major__ * 10000 + __clang_minor__ * 100 + __clang_patchlevel__) +#endif +#if defined(__clang__) && defined(__apple_build_version__) +# define Z7_APPLE_CLANG_VERSION Z7_CLANG_VERSION +#elif defined(__clang__) +# define Z7_LLVM_CLANG_VERSION Z7_CLANG_VERSION +#elif defined(__GNUC__) +# define Z7_GCC_VERSION (__GNUC__ * 10000 + __GNUC_MINOR__ * 100 + __GNUC_PATCHLEVEL__) +#endif + +#ifdef _MSC_VER +#if !defined(__clang__) && !defined(__GNUC__) +#define Z7_MSC_VER_ORIGINAL _MSC_VER +#endif +#endif + +#if defined(__MINGW32__) || defined(__MINGW64__) +#define Z7_MINGW +#endif + +// #pragma GCC diagnostic ignored "-Wunknown-pragmas" + +#ifdef __clang__ +// padding size of '' with 4 bytes to alignment boundary +#pragma GCC diagnostic ignored "-Wpadded" +#endif + + +#ifdef _MSC_VER + + #ifdef UNDER_CE + #define RPC_NO_WINDOWS_H + /* #pragma warning(disable : 4115) // '_RPC_ASYNC_STATE' : named type definition in parentheses */ + #pragma warning(disable : 4201) // nonstandard extension used : nameless struct/union + #pragma warning(disable : 4214) // nonstandard extension used : bit field types other than int + #endif + +#if defined(_MSC_VER) && _MSC_VER >= 1800 +#pragma warning(disable : 4464) // relative include path contains '..' +#endif + +// == 1200 : -O1 : for __forceinline +// >= 1900 : -O1 : for printf +#pragma warning(disable : 4710) // function not inlined + +#if _MSC_VER < 1900 +// winnt.h: 'Int64ShllMod32' +#pragma warning(disable : 4514) // unreferenced inline function has been removed +#endif + +#if _MSC_VER < 1300 +// #pragma warning(disable : 4702) // unreachable code +// Bra.c : -O1: +#pragma warning(disable : 4714) // function marked as __forceinline not inlined +#endif + +/* +#if _MSC_VER > 1400 && _MSC_VER <= 1900 +// strcat: This function or variable may be unsafe +// sysinfoapi.h: kit10: GetVersion was declared deprecated +#pragma warning(disable : 4996) +#endif +*/ + +#if _MSC_VER > 1200 +// -Wall warnings + +#pragma warning(disable : 4711) // function selected for automatic inline expansion +#pragma warning(disable : 4820) // '2' bytes padding added after data member + +#if _MSC_VER >= 1400 && _MSC_VER < 1920 +// 1400: string.h: _DBG_MEMCPY_INLINE_ +// 1600 - 191x : smmintrin.h __cplusplus' +// is not defined as a preprocessor macro, replacing with '0' for '#if/#elif' +#pragma warning(disable : 4668) + +// 1400 - 1600 : WinDef.h : 'FARPROC' : +// 1900 - 191x : immintrin.h: _readfsbase_u32 +// no function prototype given : converting '()' to '(void)' +#pragma warning(disable : 4255) +#endif + +#if _MSC_VER >= 1914 +// Compiler will insert Spectre mitigation for memory load if /Qspectre switch specified +#pragma warning(disable : 5045) +#endif + +#endif // _MSC_VER > 1200 +#endif // _MSC_VER + + +#if defined(__clang__) && (__clang_major__ >= 4) + #define Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE \ + _Pragma("clang loop unroll(disable)") \ + _Pragma("clang loop vectorize(disable)") + #define Z7_ATTRIB_NO_VECTORIZE +#elif defined(__GNUC__) && (__GNUC__ >= 5) + #define Z7_ATTRIB_NO_VECTORIZE __attribute__((optimize("no-tree-vectorize"))) + // __attribute__((optimize("no-unroll-loops"))); + #define Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE +#elif defined(_MSC_VER) && (_MSC_VER >= 1920) + #define Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE \ + _Pragma("loop( no_vector )") + #define Z7_ATTRIB_NO_VECTORIZE +#else + #define Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE + #define Z7_ATTRIB_NO_VECTORIZE +#endif + +#if defined(MY_CPU_X86_OR_AMD64) && ( \ + defined(__clang__) && (__clang_major__ >= 4) \ + || defined(__GNUC__) && (__GNUC__ >= 5)) + #define Z7_ATTRIB_NO_SSE __attribute__((__target__("no-sse"))) +#else + #define Z7_ATTRIB_NO_SSE +#endif + +#define Z7_ATTRIB_NO_VECTOR \ + Z7_ATTRIB_NO_VECTORIZE \ + Z7_ATTRIB_NO_SSE + + +#if defined(__clang__) && (__clang_major__ >= 8) \ + || defined(__GNUC__) && (__GNUC__ >= 1000) \ + /* || defined(_MSC_VER) && (_MSC_VER >= 1920) */ + // GCC is not good for __builtin_expect() + #define Z7_LIKELY(x) (__builtin_expect((x), 1)) + #define Z7_UNLIKELY(x) (__builtin_expect((x), 0)) + // #define Z7_unlikely [[unlikely]] + // #define Z7_likely [[likely]] +#else + #define Z7_LIKELY(x) (x) + #define Z7_UNLIKELY(x) (x) + // #define Z7_likely +#endif + + +#if (defined(Z7_CLANG_VERSION) && (Z7_CLANG_VERSION >= 36000)) +#define Z7_DIAGNOSCTIC_IGNORE_BEGIN_RESERVED_MACRO_IDENTIFIER \ + _Pragma("GCC diagnostic push") \ + _Pragma("GCC diagnostic ignored \"-Wreserved-macro-identifier\"") +#define Z7_DIAGNOSCTIC_IGNORE_END_RESERVED_MACRO_IDENTIFIER \ + _Pragma("GCC diagnostic pop") +#else +#define Z7_DIAGNOSCTIC_IGNORE_BEGIN_RESERVED_MACRO_IDENTIFIER +#define Z7_DIAGNOSCTIC_IGNORE_END_RESERVED_MACRO_IDENTIFIER +#endif + +#define UNUSED_VAR(x) (void)x; +/* #define UNUSED_VAR(x) x=x; */ + +#endif diff --git a/project/src/liblzma/CpuArch.c b/project/src/liblzma/CpuArch.c new file mode 100644 index 0000000..5092d38 --- /dev/null +++ b/project/src/liblzma/CpuArch.c @@ -0,0 +1,823 @@ +/* CpuArch.c -- CPU specific code +2023-05-18 : Igor Pavlov : Public domain */ + +#include "Precomp.h" + +// #include + +#include "CpuArch.h" + +#ifdef MY_CPU_X86_OR_AMD64 + +#undef NEED_CHECK_FOR_CPUID +#if !defined(MY_CPU_AMD64) +#define NEED_CHECK_FOR_CPUID +#endif + +/* + cpuid instruction supports (subFunction) parameter in ECX, + that is used only with some specific (function) parameter values. + But we always use only (subFunction==0). +*/ +/* + __cpuid(): MSVC and GCC/CLANG use same function/macro name + but parameters are different. + We use MSVC __cpuid() parameters style for our z7_x86_cpuid() function. +*/ + +#if defined(__GNUC__) /* && (__GNUC__ >= 10) */ \ + || defined(__clang__) /* && (__clang_major__ >= 10) */ + +/* there was some CLANG/GCC compilers that have issues with + rbx(ebx) handling in asm blocks in -fPIC mode (__PIC__ is defined). + compiler's contains the macro __cpuid() that is similar to our code. + The history of __cpuid() changes in CLANG/GCC: + GCC: + 2007: it preserved ebx for (__PIC__ && __i386__) + 2013: it preserved rbx and ebx for __PIC__ + 2014: it doesn't preserves rbx and ebx anymore + we suppose that (__GNUC__ >= 5) fixed that __PIC__ ebx/rbx problem. + CLANG: + 2014+: it preserves rbx, but only for 64-bit code. No __PIC__ check. + Why CLANG cares about 64-bit mode only, and doesn't care about ebx (in 32-bit)? + Do we need __PIC__ test for CLANG or we must care about rbx even if + __PIC__ is not defined? +*/ + +#define ASM_LN "\n" + +#if defined(MY_CPU_AMD64) && defined(__PIC__) \ + && ((defined (__GNUC__) && (__GNUC__ < 5)) || defined(__clang__)) + +#define x86_cpuid_MACRO(p, func) { \ + __asm__ __volatile__ ( \ + ASM_LN "mov %%rbx, %q1" \ + ASM_LN "cpuid" \ + ASM_LN "xchg %%rbx, %q1" \ + : "=a" ((p)[0]), "=&r" ((p)[1]), "=c" ((p)[2]), "=d" ((p)[3]) : "0" (func), "2"(0)); } + + /* "=&r" selects free register. It can select even rbx, if that register is free. + "=&D" for (RDI) also works, but the code can be larger with "=&D" + "2"(0) means (subFunction = 0), + 2 is (zero-based) index in the output constraint list "=c" (ECX). */ + +#elif defined(MY_CPU_X86) && defined(__PIC__) \ + && ((defined (__GNUC__) && (__GNUC__ < 5)) || defined(__clang__)) + +#define x86_cpuid_MACRO(p, func) { \ + __asm__ __volatile__ ( \ + ASM_LN "mov %%ebx, %k1" \ + ASM_LN "cpuid" \ + ASM_LN "xchg %%ebx, %k1" \ + : "=a" ((p)[0]), "=&r" ((p)[1]), "=c" ((p)[2]), "=d" ((p)[3]) : "0" (func), "2"(0)); } + +#else + +#define x86_cpuid_MACRO(p, func) { \ + __asm__ __volatile__ ( \ + ASM_LN "cpuid" \ + : "=a" ((p)[0]), "=b" ((p)[1]), "=c" ((p)[2]), "=d" ((p)[3]) : "0" (func), "2"(0)); } + +#endif + + +void Z7_FASTCALL z7_x86_cpuid(UInt32 p[4], UInt32 func) +{ + x86_cpuid_MACRO(p, func) +} + + +Z7_NO_INLINE +UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void) +{ + #if defined(NEED_CHECK_FOR_CPUID) + #define EFALGS_CPUID_BIT 21 + UInt32 a; + __asm__ __volatile__ ( + ASM_LN "pushf" + ASM_LN "pushf" + ASM_LN "pop %0" + // ASM_LN "movl %0, %1" + // ASM_LN "xorl $0x200000, %0" + ASM_LN "btc %1, %0" + ASM_LN "push %0" + ASM_LN "popf" + ASM_LN "pushf" + ASM_LN "pop %0" + ASM_LN "xorl (%%esp), %0" + + ASM_LN "popf" + ASM_LN + : "=&r" (a) // "=a" + : "i" (EFALGS_CPUID_BIT) + ); + if ((a & (1 << EFALGS_CPUID_BIT)) == 0) + return 0; + #endif + { + UInt32 p[4]; + x86_cpuid_MACRO(p, 0) + return p[0]; + } +} + +#undef ASM_LN + +#elif !defined(_MSC_VER) + +/* +// for gcc/clang and other: we can try to use __cpuid macro: +#include +void Z7_FASTCALL z7_x86_cpuid(UInt32 p[4], UInt32 func) +{ + __cpuid(func, p[0], p[1], p[2], p[3]); +} +UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void) +{ + return (UInt32)__get_cpuid_max(0, NULL); +} +*/ +// for unsupported cpuid: +void Z7_FASTCALL z7_x86_cpuid(UInt32 p[4], UInt32 func) +{ + UNUSED_VAR(func) + p[0] = p[1] = p[2] = p[3] = 0; +} +UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void) +{ + return 0; +} + +#else // _MSC_VER + +#if !defined(MY_CPU_AMD64) + +UInt32 __declspec(naked) Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void) +{ + #if defined(NEED_CHECK_FOR_CPUID) + #define EFALGS_CPUID_BIT 21 + __asm pushfd + __asm pushfd + /* + __asm pop eax + // __asm mov edx, eax + __asm btc eax, EFALGS_CPUID_BIT + __asm push eax + */ + __asm btc dword ptr [esp], EFALGS_CPUID_BIT + __asm popfd + __asm pushfd + __asm pop eax + // __asm xor eax, edx + __asm xor eax, [esp] + // __asm push edx + __asm popfd + __asm and eax, (1 shl EFALGS_CPUID_BIT) + __asm jz end_func + #endif + __asm push ebx + __asm xor eax, eax // func + __asm xor ecx, ecx // subFunction (optional) for (func == 0) + __asm cpuid + __asm pop ebx + #if defined(NEED_CHECK_FOR_CPUID) + end_func: + #endif + __asm ret 0 +} + +void __declspec(naked) Z7_FASTCALL z7_x86_cpuid(UInt32 p[4], UInt32 func) +{ + UNUSED_VAR(p) + UNUSED_VAR(func) + __asm push ebx + __asm push edi + __asm mov edi, ecx // p + __asm mov eax, edx // func + __asm xor ecx, ecx // subfunction (optional) for (func == 0) + __asm cpuid + __asm mov [edi ], eax + __asm mov [edi + 4], ebx + __asm mov [edi + 8], ecx + __asm mov [edi + 12], edx + __asm pop edi + __asm pop ebx + __asm ret 0 +} + +#else // MY_CPU_AMD64 + + #if _MSC_VER >= 1600 + #include + #define MY_cpuidex __cpuidex + #else +/* + __cpuid (func == (0 or 7)) requires subfunction number in ECX. + MSDN: The __cpuid intrinsic clears the ECX register before calling the cpuid instruction. + __cpuid() in new MSVC clears ECX. + __cpuid() in old MSVC (14.00) x64 doesn't clear ECX + We still can use __cpuid for low (func) values that don't require ECX, + but __cpuid() in old MSVC will be incorrect for some func values: (func == 7). + So here we use the hack for old MSVC to send (subFunction) in ECX register to cpuid instruction, + where ECX value is first parameter for FASTCALL / NO_INLINE func, + So the caller of MY_cpuidex_HACK() sets ECX as subFunction, and + old MSVC for __cpuid() doesn't change ECX and cpuid instruction gets (subFunction) value. + +DON'T remove Z7_NO_INLINE and Z7_FASTCALL for MY_cpuidex_HACK(): !!! +*/ +static +Z7_NO_INLINE void Z7_FASTCALL MY_cpuidex_HACK(UInt32 subFunction, UInt32 func, int *CPUInfo) +{ + UNUSED_VAR(subFunction) + __cpuid(CPUInfo, func); +} + #define MY_cpuidex(info, func, func2) MY_cpuidex_HACK(func2, func, info) + #pragma message("======== MY_cpuidex_HACK WAS USED ========") + #endif // _MSC_VER >= 1600 + +#if !defined(MY_CPU_AMD64) +/* inlining for __cpuid() in MSVC x86 (32-bit) produces big ineffective code, + so we disable inlining here */ +Z7_NO_INLINE +#endif +void Z7_FASTCALL z7_x86_cpuid(UInt32 p[4], UInt32 func) +{ + MY_cpuidex((int *)p, (int)func, 0); +} + +Z7_NO_INLINE +UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void) +{ + int a[4]; + MY_cpuidex(a, 0, 0); + return a[0]; +} + +#endif // MY_CPU_AMD64 +#endif // _MSC_VER + +#if defined(NEED_CHECK_FOR_CPUID) +#define CHECK_CPUID_IS_SUPPORTED { if (z7_x86_cpuid_GetMaxFunc() == 0) return 0; } +#else +#define CHECK_CPUID_IS_SUPPORTED +#endif +#undef NEED_CHECK_FOR_CPUID + + +static +BoolInt x86cpuid_Func_1(UInt32 *p) +{ + CHECK_CPUID_IS_SUPPORTED + z7_x86_cpuid(p, 1); + return True; +} + +/* +static const UInt32 kVendors[][1] = +{ + { 0x756E6547 }, // , 0x49656E69, 0x6C65746E }, + { 0x68747541 }, // , 0x69746E65, 0x444D4163 }, + { 0x746E6543 } // , 0x48727561, 0x736C7561 } +}; +*/ + +/* +typedef struct +{ + UInt32 maxFunc; + UInt32 vendor[3]; + UInt32 ver; + UInt32 b; + UInt32 c; + UInt32 d; +} Cx86cpuid; + +enum +{ + CPU_FIRM_INTEL, + CPU_FIRM_AMD, + CPU_FIRM_VIA +}; +int x86cpuid_GetFirm(const Cx86cpuid *p); +#define x86cpuid_ver_GetFamily(ver) (((ver >> 16) & 0xff0) | ((ver >> 8) & 0xf)) +#define x86cpuid_ver_GetModel(ver) (((ver >> 12) & 0xf0) | ((ver >> 4) & 0xf)) +#define x86cpuid_ver_GetStepping(ver) (ver & 0xf) + +int x86cpuid_GetFirm(const Cx86cpuid *p) +{ + unsigned i; + for (i = 0; i < sizeof(kVendors) / sizeof(kVendors[0]); i++) + { + const UInt32 *v = kVendors[i]; + if (v[0] == p->vendor[0] + // && v[1] == p->vendor[1] + // && v[2] == p->vendor[2] + ) + return (int)i; + } + return -1; +} + +BoolInt CPU_Is_InOrder() +{ + Cx86cpuid p; + UInt32 family, model; + if (!x86cpuid_CheckAndRead(&p)) + return True; + + family = x86cpuid_ver_GetFamily(p.ver); + model = x86cpuid_ver_GetModel(p.ver); + + switch (x86cpuid_GetFirm(&p)) + { + case CPU_FIRM_INTEL: return (family < 6 || (family == 6 && ( + // In-Order Atom CPU + model == 0x1C // 45 nm, N4xx, D4xx, N5xx, D5xx, 230, 330 + || model == 0x26 // 45 nm, Z6xx + || model == 0x27 // 32 nm, Z2460 + || model == 0x35 // 32 nm, Z2760 + || model == 0x36 // 32 nm, N2xxx, D2xxx + ))); + case CPU_FIRM_AMD: return (family < 5 || (family == 5 && (model < 6 || model == 0xA))); + case CPU_FIRM_VIA: return (family < 6 || (family == 6 && model < 0xF)); + } + return False; // v23 : unknown processors are not In-Order +} +*/ + +#ifdef _WIN32 +#include "7zWindows.h" +#endif + +#if !defined(MY_CPU_AMD64) && defined(_WIN32) + +/* for legacy SSE ia32: there is no user-space cpu instruction to check + that OS supports SSE register storing/restoring on context switches. + So we need some OS-specific function to check that it's safe to use SSE registers. +*/ + +Z7_FORCE_INLINE +static BoolInt CPU_Sys_Is_SSE_Supported(void) +{ +#ifdef _MSC_VER + #pragma warning(push) + #pragma warning(disable : 4996) // `GetVersion': was declared deprecated +#endif + /* low byte is major version of Windows + We suppose that any Windows version since + Windows2000 (major == 5) supports SSE registers */ + return (Byte)GetVersion() >= 5; +#if defined(_MSC_VER) + #pragma warning(pop) +#endif +} +#define CHECK_SYS_SSE_SUPPORT if (!CPU_Sys_Is_SSE_Supported()) return False; +#else +#define CHECK_SYS_SSE_SUPPORT +#endif + + +#if !defined(MY_CPU_AMD64) + +BoolInt CPU_IsSupported_CMOV(void) +{ + UInt32 a[4]; + if (!x86cpuid_Func_1(&a[0])) + return 0; + return (a[3] >> 15) & 1; +} + +BoolInt CPU_IsSupported_SSE(void) +{ + UInt32 a[4]; + CHECK_SYS_SSE_SUPPORT + if (!x86cpuid_Func_1(&a[0])) + return 0; + return (a[3] >> 25) & 1; +} + +BoolInt CPU_IsSupported_SSE2(void) +{ + UInt32 a[4]; + CHECK_SYS_SSE_SUPPORT + if (!x86cpuid_Func_1(&a[0])) + return 0; + return (a[3] >> 26) & 1; +} + +#endif + + +static UInt32 x86cpuid_Func_1_ECX(void) +{ + UInt32 a[4]; + CHECK_SYS_SSE_SUPPORT + if (!x86cpuid_Func_1(&a[0])) + return 0; + return a[2]; +} + +BoolInt CPU_IsSupported_AES(void) +{ + return (x86cpuid_Func_1_ECX() >> 25) & 1; +} + +BoolInt CPU_IsSupported_SSSE3(void) +{ + return (x86cpuid_Func_1_ECX() >> 9) & 1; +} + +BoolInt CPU_IsSupported_SSE41(void) +{ + return (x86cpuid_Func_1_ECX() >> 19) & 1; +} + +BoolInt CPU_IsSupported_SHA(void) +{ + CHECK_SYS_SSE_SUPPORT + + if (z7_x86_cpuid_GetMaxFunc() < 7) + return False; + { + UInt32 d[4]; + z7_x86_cpuid(d, 7); + return (d[1] >> 29) & 1; + } +} + +/* +MSVC: _xgetbv() intrinsic is available since VS2010SP1. + MSVC also defines (_XCR_XFEATURE_ENABLED_MASK) macro in + that we can use or check. + For any 32-bit x86 we can use asm code in MSVC, + but MSVC asm code is huge after compilation. + So _xgetbv() is better + +ICC: _xgetbv() intrinsic is available (in what version of ICC?) + ICC defines (__GNUC___) and it supports gnu assembler + also ICC supports MASM style code with -use-msasm switch. + but ICC doesn't support __attribute__((__target__)) + +GCC/CLANG 9: + _xgetbv() is macro that works via __builtin_ia32_xgetbv() + and we need __attribute__((__target__("xsave")). + But with __target__("xsave") the function will be not + inlined to function that has no __target__("xsave") attribute. + If we want _xgetbv() call inlining, then we should use asm version + instead of calling _xgetbv(). + Note:intrinsic is broke before GCC 8.2: + https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85684 +*/ + +#if defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 1100) \ + || defined(_MSC_VER) && (_MSC_VER >= 1600) && (_MSC_FULL_VER >= 160040219) \ + || defined(__GNUC__) && (__GNUC__ >= 9) \ + || defined(__clang__) && (__clang_major__ >= 9) +// we define ATTRIB_XGETBV, if we want to use predefined _xgetbv() from compiler +#if defined(__INTEL_COMPILER) +#define ATTRIB_XGETBV +#elif defined(__GNUC__) || defined(__clang__) +// we don't define ATTRIB_XGETBV here, because asm version is better for inlining. +// #define ATTRIB_XGETBV __attribute__((__target__("xsave"))) +#else +#define ATTRIB_XGETBV +#endif +#endif + +#if defined(ATTRIB_XGETBV) +#include +#endif + + +// XFEATURE_ENABLED_MASK/XCR0 +#define MY_XCR_XFEATURE_ENABLED_MASK 0 + +#if defined(ATTRIB_XGETBV) +ATTRIB_XGETBV +#endif +static UInt64 x86_xgetbv_0(UInt32 num) +{ +#if defined(ATTRIB_XGETBV) + { + return + #if (defined(_MSC_VER)) + _xgetbv(num); + #else + __builtin_ia32_xgetbv( + #if !defined(__clang__) + (int) + #endif + num); + #endif + } + +#elif defined(__GNUC__) || defined(__clang__) || defined(__SUNPRO_CC) + + UInt32 a, d; + #if defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 4)) + __asm__ + ( + "xgetbv" + : "=a"(a), "=d"(d) : "c"(num) : "cc" + ); + #else // is old gcc + __asm__ + ( + ".byte 0x0f, 0x01, 0xd0" "\n\t" + : "=a"(a), "=d"(d) : "c"(num) : "cc" + ); + #endif + return ((UInt64)d << 32) | a; + // return a; + +#elif defined(_MSC_VER) && !defined(MY_CPU_AMD64) + + UInt32 a, d; + __asm { + push eax + push edx + push ecx + mov ecx, num; + // xor ecx, ecx // = MY_XCR_XFEATURE_ENABLED_MASK + _emit 0x0f + _emit 0x01 + _emit 0xd0 + mov a, eax + mov d, edx + pop ecx + pop edx + pop eax + } + return ((UInt64)d << 32) | a; + // return a; + +#else // it's unknown compiler + // #error "Need xgetbv function" + UNUSED_VAR(num) + // for MSVC-X64 we could call external function from external file. + /* Actually we had checked OSXSAVE/AVX in cpuid before. + So it's expected that OS supports at least AVX and below. */ + // if (num != MY_XCR_XFEATURE_ENABLED_MASK) return 0; // if not XCR0 + return + // (1 << 0) | // x87 + (1 << 1) // SSE + | (1 << 2); // AVX + +#endif +} + +#ifdef _WIN32 +/* + Windows versions do not know about new ISA extensions that + can be introduced. But we still can use new extensions, + even if Windows doesn't report about supporting them, + But we can use new extensions, only if Windows knows about new ISA extension + that changes the number or size of registers: SSE, AVX/XSAVE, AVX512 + So it's enough to check + MY_PF_AVX_INSTRUCTIONS_AVAILABLE + instead of + MY_PF_AVX2_INSTRUCTIONS_AVAILABLE +*/ +#define MY_PF_XSAVE_ENABLED 17 +// #define MY_PF_SSSE3_INSTRUCTIONS_AVAILABLE 36 +// #define MY_PF_SSE4_1_INSTRUCTIONS_AVAILABLE 37 +// #define MY_PF_SSE4_2_INSTRUCTIONS_AVAILABLE 38 +// #define MY_PF_AVX_INSTRUCTIONS_AVAILABLE 39 +// #define MY_PF_AVX2_INSTRUCTIONS_AVAILABLE 40 +// #define MY_PF_AVX512F_INSTRUCTIONS_AVAILABLE 41 +#endif + +BoolInt CPU_IsSupported_AVX(void) +{ + #ifdef _WIN32 + if (!IsProcessorFeaturePresent(MY_PF_XSAVE_ENABLED)) + return False; + /* PF_AVX_INSTRUCTIONS_AVAILABLE probably is supported starting from + some latest Win10 revisions. But we need AVX in older Windows also. + So we don't use the following check: */ + /* + if (!IsProcessorFeaturePresent(MY_PF_AVX_INSTRUCTIONS_AVAILABLE)) + return False; + */ + #endif + + /* + OS must use new special XSAVE/XRSTOR instructions to save + AVX registers when it required for context switching. + At OS statring: + OS sets CR4.OSXSAVE flag to signal the processor that OS supports the XSAVE extensions. + Also OS sets bitmask in XCR0 register that defines what + registers will be processed by XSAVE instruction: + XCR0.SSE[bit 0] - x87 registers and state + XCR0.SSE[bit 1] - SSE registers and state + XCR0.AVX[bit 2] - AVX registers and state + CR4.OSXSAVE is reflected to CPUID.1:ECX.OSXSAVE[bit 27]. + So we can read that bit in user-space. + XCR0 is available for reading in user-space by new XGETBV instruction. + */ + { + const UInt32 c = x86cpuid_Func_1_ECX(); + if (0 == (1 + & (c >> 28) // AVX instructions are supported by hardware + & (c >> 27))) // OSXSAVE bit: XSAVE and related instructions are enabled by OS. + return False; + } + + /* also we can check + CPUID.1:ECX.XSAVE [bit 26] : that shows that + XSAVE, XRESTOR, XSETBV, XGETBV instructions are supported by hardware. + But that check is redundant, because if OSXSAVE bit is set, then XSAVE is also set */ + + /* If OS have enabled XSAVE extension instructions (OSXSAVE == 1), + in most cases we expect that OS also will support storing/restoring + for AVX and SSE states at least. + But to be ensure for that we call user-space instruction + XGETBV(0) to get XCR0 value that contains bitmask that defines + what exact states(registers) OS have enabled for storing/restoring. + */ + + { + const UInt32 bm = (UInt32)x86_xgetbv_0(MY_XCR_XFEATURE_ENABLED_MASK); + // printf("\n=== XGetBV=%d\n", bm); + return 1 + & (bm >> 1) // SSE state is supported (set by OS) for storing/restoring + & (bm >> 2); // AVX state is supported (set by OS) for storing/restoring + } + // since Win7SP1: we can use GetEnabledXStateFeatures(); +} + + +BoolInt CPU_IsSupported_AVX2(void) +{ + if (!CPU_IsSupported_AVX()) + return False; + if (z7_x86_cpuid_GetMaxFunc() < 7) + return False; + { + UInt32 d[4]; + z7_x86_cpuid(d, 7); + // printf("\ncpuid(7): ebx=%8x ecx=%8x\n", d[1], d[2]); + return 1 + & (d[1] >> 5); // avx2 + } +} + +BoolInt CPU_IsSupported_VAES_AVX2(void) +{ + if (!CPU_IsSupported_AVX()) + return False; + if (z7_x86_cpuid_GetMaxFunc() < 7) + return False; + { + UInt32 d[4]; + z7_x86_cpuid(d, 7); + // printf("\ncpuid(7): ebx=%8x ecx=%8x\n", d[1], d[2]); + return 1 + & (d[1] >> 5) // avx2 + // & (d[1] >> 31) // avx512vl + & (d[2] >> 9); // vaes // VEX-256/EVEX + } +} + +BoolInt CPU_IsSupported_PageGB(void) +{ + CHECK_CPUID_IS_SUPPORTED + { + UInt32 d[4]; + z7_x86_cpuid(d, 0x80000000); + if (d[0] < 0x80000001) + return False; + z7_x86_cpuid(d, 0x80000001); + return (d[3] >> 26) & 1; + } +} + + +#elif defined(MY_CPU_ARM_OR_ARM64) + +#ifdef _WIN32 + +#include "7zWindows.h" + +BoolInt CPU_IsSupported_CRC32(void) { return IsProcessorFeaturePresent(PF_ARM_V8_CRC32_INSTRUCTIONS_AVAILABLE) ? 1 : 0; } +BoolInt CPU_IsSupported_CRYPTO(void) { return IsProcessorFeaturePresent(PF_ARM_V8_CRYPTO_INSTRUCTIONS_AVAILABLE) ? 1 : 0; } +BoolInt CPU_IsSupported_NEON(void) { return IsProcessorFeaturePresent(PF_ARM_NEON_INSTRUCTIONS_AVAILABLE) ? 1 : 0; } + +#else + +#if defined(__APPLE__) + +/* +#include +#include +static void Print_sysctlbyname(const char *name) +{ + size_t bufSize = 256; + char buf[256]; + int res = sysctlbyname(name, &buf, &bufSize, NULL, 0); + { + int i; + printf("\nres = %d : %s : '%s' : bufSize = %d, numeric", res, name, buf, (unsigned)bufSize); + for (i = 0; i < 20; i++) + printf(" %2x", (unsigned)(Byte)buf[i]); + + } +} +*/ +/* + Print_sysctlbyname("hw.pagesize"); + Print_sysctlbyname("machdep.cpu.brand_string"); +*/ + +static BoolInt z7_sysctlbyname_Get_BoolInt(const char *name) +{ + UInt32 val = 0; + if (z7_sysctlbyname_Get_UInt32(name, &val) == 0 && val == 1) + return 1; + return 0; +} + +BoolInt CPU_IsSupported_CRC32(void) +{ + return z7_sysctlbyname_Get_BoolInt("hw.optional.armv8_crc32"); +} + +BoolInt CPU_IsSupported_NEON(void) +{ + return z7_sysctlbyname_Get_BoolInt("hw.optional.neon"); +} + +#ifdef MY_CPU_ARM64 +#define APPLE_CRYPTO_SUPPORT_VAL 1 +#else +#define APPLE_CRYPTO_SUPPORT_VAL 0 +#endif + +BoolInt CPU_IsSupported_SHA1(void) { return APPLE_CRYPTO_SUPPORT_VAL; } +BoolInt CPU_IsSupported_SHA2(void) { return APPLE_CRYPTO_SUPPORT_VAL; } +BoolInt CPU_IsSupported_AES (void) { return APPLE_CRYPTO_SUPPORT_VAL; } + + +#else // __APPLE__ + +#include + +#define USE_HWCAP + +#ifdef USE_HWCAP + +#include + + #define MY_HWCAP_CHECK_FUNC_2(name1, name2) \ + BoolInt CPU_IsSupported_ ## name1() { return (getauxval(AT_HWCAP) & (HWCAP_ ## name2)) ? 1 : 0; } + +#ifdef MY_CPU_ARM64 + #define MY_HWCAP_CHECK_FUNC(name) \ + MY_HWCAP_CHECK_FUNC_2(name, name) + MY_HWCAP_CHECK_FUNC_2(NEON, ASIMD) +// MY_HWCAP_CHECK_FUNC (ASIMD) +#elif defined(MY_CPU_ARM) + #define MY_HWCAP_CHECK_FUNC(name) \ + BoolInt CPU_IsSupported_ ## name() { return (getauxval(AT_HWCAP2) & (HWCAP2_ ## name)) ? 1 : 0; } + MY_HWCAP_CHECK_FUNC_2(NEON, NEON) +#endif + +#else // USE_HWCAP + + #define MY_HWCAP_CHECK_FUNC(name) \ + BoolInt CPU_IsSupported_ ## name() { return 0; } + MY_HWCAP_CHECK_FUNC(NEON) + +#endif // USE_HWCAP + +MY_HWCAP_CHECK_FUNC (CRC32) +MY_HWCAP_CHECK_FUNC (SHA1) +MY_HWCAP_CHECK_FUNC (SHA2) +MY_HWCAP_CHECK_FUNC (AES) + +#endif // __APPLE__ +#endif // _WIN32 + +#endif // MY_CPU_ARM_OR_ARM64 + + + +#ifdef __APPLE__ + +#include + +int z7_sysctlbyname_Get(const char *name, void *buf, size_t *bufSize) +{ + return sysctlbyname(name, buf, bufSize, NULL, 0); +} + +int z7_sysctlbyname_Get_UInt32(const char *name, UInt32 *val) +{ + size_t bufSize = sizeof(*val); + const int res = z7_sysctlbyname_Get(name, val, &bufSize); + if (res == 0 && bufSize != sizeof(*val)) + return EFAULT; + return res; +} + +#endif diff --git a/project/src/liblzma/CpuArch.h b/project/src/liblzma/CpuArch.h new file mode 100644 index 0000000..6c1cd2a --- /dev/null +++ b/project/src/liblzma/CpuArch.h @@ -0,0 +1,523 @@ +/* CpuArch.h -- CPU specific code +2023-04-02 : Igor Pavlov : Public domain */ + +#ifndef ZIP7_INC_CPU_ARCH_H +#define ZIP7_INC_CPU_ARCH_H + +#include "7zTypes.h" + +EXTERN_C_BEGIN + +/* +MY_CPU_LE means that CPU is LITTLE ENDIAN. +MY_CPU_BE means that CPU is BIG ENDIAN. +If MY_CPU_LE and MY_CPU_BE are not defined, we don't know about ENDIANNESS of platform. + +MY_CPU_LE_UNALIGN means that CPU is LITTLE ENDIAN and CPU supports unaligned memory accesses. + +MY_CPU_64BIT means that processor can work with 64-bit registers. + MY_CPU_64BIT can be used to select fast code branch + MY_CPU_64BIT doesn't mean that (sizeof(void *) == 8) +*/ + +#if defined(_M_X64) \ + || defined(_M_AMD64) \ + || defined(__x86_64__) \ + || defined(__AMD64__) \ + || defined(__amd64__) + #define MY_CPU_AMD64 + #ifdef __ILP32__ + #define MY_CPU_NAME "x32" + #define MY_CPU_SIZEOF_POINTER 4 + #else + #define MY_CPU_NAME "x64" + #define MY_CPU_SIZEOF_POINTER 8 + #endif + #define MY_CPU_64BIT +#endif + + +#if defined(_M_IX86) \ + || defined(__i386__) + #define MY_CPU_X86 + #define MY_CPU_NAME "x86" + /* #define MY_CPU_32BIT */ + #define MY_CPU_SIZEOF_POINTER 4 +#endif + + +#if defined(_M_ARM64) \ + || defined(__AARCH64EL__) \ + || defined(__AARCH64EB__) \ + || defined(__aarch64__) + #define MY_CPU_ARM64 + #ifdef __ILP32__ + #define MY_CPU_NAME "arm64-32" + #define MY_CPU_SIZEOF_POINTER 4 + #else + #define MY_CPU_NAME "arm64" + #define MY_CPU_SIZEOF_POINTER 8 + #endif + #define MY_CPU_64BIT +#endif + + +#if defined(_M_ARM) \ + || defined(_M_ARM_NT) \ + || defined(_M_ARMT) \ + || defined(__arm__) \ + || defined(__thumb__) \ + || defined(__ARMEL__) \ + || defined(__ARMEB__) \ + || defined(__THUMBEL__) \ + || defined(__THUMBEB__) + #define MY_CPU_ARM + + #if defined(__thumb__) || defined(__THUMBEL__) || defined(_M_ARMT) + #define MY_CPU_ARMT + #define MY_CPU_NAME "armt" + #else + #define MY_CPU_ARM32 + #define MY_CPU_NAME "arm" + #endif + /* #define MY_CPU_32BIT */ + #define MY_CPU_SIZEOF_POINTER 4 +#endif + + +#if defined(_M_IA64) \ + || defined(__ia64__) + #define MY_CPU_IA64 + #define MY_CPU_NAME "ia64" + #define MY_CPU_64BIT +#endif + + +#if defined(__mips64) \ + || defined(__mips64__) \ + || (defined(__mips) && (__mips == 64 || __mips == 4 || __mips == 3)) + #define MY_CPU_NAME "mips64" + #define MY_CPU_64BIT +#elif defined(__mips__) + #define MY_CPU_NAME "mips" + /* #define MY_CPU_32BIT */ +#endif + + +#if defined(__ppc64__) \ + || defined(__powerpc64__) \ + || defined(__ppc__) \ + || defined(__powerpc__) \ + || defined(__PPC__) \ + || defined(_POWER) + +#define MY_CPU_PPC_OR_PPC64 + +#if defined(__ppc64__) \ + || defined(__powerpc64__) \ + || defined(_LP64) \ + || defined(__64BIT__) + #ifdef __ILP32__ + #define MY_CPU_NAME "ppc64-32" + #define MY_CPU_SIZEOF_POINTER 4 + #else + #define MY_CPU_NAME "ppc64" + #define MY_CPU_SIZEOF_POINTER 8 + #endif + #define MY_CPU_64BIT +#else + #define MY_CPU_NAME "ppc" + #define MY_CPU_SIZEOF_POINTER 4 + /* #define MY_CPU_32BIT */ +#endif +#endif + + +#if defined(__riscv) \ + || defined(__riscv__) + #if __riscv_xlen == 32 + #define MY_CPU_NAME "riscv32" + #elif __riscv_xlen == 64 + #define MY_CPU_NAME "riscv64" + #else + #define MY_CPU_NAME "riscv" + #endif +#endif + + +#if defined(MY_CPU_X86) || defined(MY_CPU_AMD64) +#define MY_CPU_X86_OR_AMD64 +#endif + +#if defined(MY_CPU_ARM) || defined(MY_CPU_ARM64) +#define MY_CPU_ARM_OR_ARM64 +#endif + + +#ifdef _WIN32 + + #ifdef MY_CPU_ARM + #define MY_CPU_ARM_LE + #endif + + #ifdef MY_CPU_ARM64 + #define MY_CPU_ARM64_LE + #endif + + #ifdef _M_IA64 + #define MY_CPU_IA64_LE + #endif + +#endif + + +#if defined(MY_CPU_X86_OR_AMD64) \ + || defined(MY_CPU_ARM_LE) \ + || defined(MY_CPU_ARM64_LE) \ + || defined(MY_CPU_IA64_LE) \ + || defined(__LITTLE_ENDIAN__) \ + || defined(__ARMEL__) \ + || defined(__THUMBEL__) \ + || defined(__AARCH64EL__) \ + || defined(__MIPSEL__) \ + || defined(__MIPSEL) \ + || defined(_MIPSEL) \ + || defined(__BFIN__) \ + || (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)) + #define MY_CPU_LE +#endif + +#if defined(__BIG_ENDIAN__) \ + || defined(__ARMEB__) \ + || defined(__THUMBEB__) \ + || defined(__AARCH64EB__) \ + || defined(__MIPSEB__) \ + || defined(__MIPSEB) \ + || defined(_MIPSEB) \ + || defined(__m68k__) \ + || defined(__s390__) \ + || defined(__s390x__) \ + || defined(__zarch__) \ + || (defined(__BYTE_ORDER__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)) + #define MY_CPU_BE +#endif + + +#if defined(MY_CPU_LE) && defined(MY_CPU_BE) + #error Stop_Compiling_Bad_Endian +#endif + +#if !defined(MY_CPU_LE) && !defined(MY_CPU_BE) + #error Stop_Compiling_CPU_ENDIAN_must_be_detected_at_compile_time +#endif + +#if defined(MY_CPU_32BIT) && defined(MY_CPU_64BIT) + #error Stop_Compiling_Bad_32_64_BIT +#endif + +#ifdef __SIZEOF_POINTER__ + #ifdef MY_CPU_SIZEOF_POINTER + #if MY_CPU_SIZEOF_POINTER != __SIZEOF_POINTER__ + #error Stop_Compiling_Bad_MY_CPU_PTR_SIZE + #endif + #else + #define MY_CPU_SIZEOF_POINTER __SIZEOF_POINTER__ + #endif +#endif + +#if defined(MY_CPU_SIZEOF_POINTER) && (MY_CPU_SIZEOF_POINTER == 4) +#if defined (_LP64) + #error Stop_Compiling_Bad_MY_CPU_PTR_SIZE +#endif +#endif + +#ifdef _MSC_VER + #if _MSC_VER >= 1300 + #define MY_CPU_pragma_pack_push_1 __pragma(pack(push, 1)) + #define MY_CPU_pragma_pop __pragma(pack(pop)) + #else + #define MY_CPU_pragma_pack_push_1 + #define MY_CPU_pragma_pop + #endif +#else + #ifdef __xlC__ + #define MY_CPU_pragma_pack_push_1 _Pragma("pack(1)") + #define MY_CPU_pragma_pop _Pragma("pack()") + #else + #define MY_CPU_pragma_pack_push_1 _Pragma("pack(push, 1)") + #define MY_CPU_pragma_pop _Pragma("pack(pop)") + #endif +#endif + + +#ifndef MY_CPU_NAME + #ifdef MY_CPU_LE + #define MY_CPU_NAME "LE" + #elif defined(MY_CPU_BE) + #define MY_CPU_NAME "BE" + #else + /* + #define MY_CPU_NAME "" + */ + #endif +#endif + + + + + +#ifdef __has_builtin + #define Z7_has_builtin(x) __has_builtin(x) +#else + #define Z7_has_builtin(x) 0 +#endif + + +#define Z7_BSWAP32_CONST(v) \ + ( (((UInt32)(v) << 24) ) \ + | (((UInt32)(v) << 8) & (UInt32)0xff0000) \ + | (((UInt32)(v) >> 8) & (UInt32)0xff00 ) \ + | (((UInt32)(v) >> 24) )) + + +#if defined(_MSC_VER) && (_MSC_VER >= 1300) + +#include + +/* Note: these macros will use bswap instruction (486), that is unsupported in 386 cpu */ + +#pragma intrinsic(_byteswap_ushort) +#pragma intrinsic(_byteswap_ulong) +#pragma intrinsic(_byteswap_uint64) + +#define Z7_BSWAP16(v) _byteswap_ushort(v) +#define Z7_BSWAP32(v) _byteswap_ulong (v) +#define Z7_BSWAP64(v) _byteswap_uint64(v) +#define Z7_CPU_FAST_BSWAP_SUPPORTED + +#elif (defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 3))) \ + || (defined(__clang__) && Z7_has_builtin(__builtin_bswap16)) + +#define Z7_BSWAP16(v) __builtin_bswap16(v) +#define Z7_BSWAP32(v) __builtin_bswap32(v) +#define Z7_BSWAP64(v) __builtin_bswap64(v) +#define Z7_CPU_FAST_BSWAP_SUPPORTED + +#else + +#define Z7_BSWAP16(v) ((UInt16) \ + ( ((UInt32)(v) << 8) \ + | ((UInt32)(v) >> 8) \ + )) + +#define Z7_BSWAP32(v) Z7_BSWAP32_CONST(v) + +#define Z7_BSWAP64(v) \ + ( ( ( (UInt64)(v) ) << 8 * 7 ) \ + | ( ( (UInt64)(v) & ((UInt32)0xff << 8 * 1) ) << 8 * 5 ) \ + | ( ( (UInt64)(v) & ((UInt32)0xff << 8 * 2) ) << 8 * 3 ) \ + | ( ( (UInt64)(v) & ((UInt32)0xff << 8 * 3) ) << 8 * 1 ) \ + | ( ( (UInt64)(v) >> 8 * 1 ) & ((UInt32)0xff << 8 * 3) ) \ + | ( ( (UInt64)(v) >> 8 * 3 ) & ((UInt32)0xff << 8 * 2) ) \ + | ( ( (UInt64)(v) >> 8 * 5 ) & ((UInt32)0xff << 8 * 1) ) \ + | ( ( (UInt64)(v) >> 8 * 7 ) ) \ + ) + +#endif + + + +#ifdef MY_CPU_LE + #if defined(MY_CPU_X86_OR_AMD64) \ + || defined(MY_CPU_ARM64) + #define MY_CPU_LE_UNALIGN + #define MY_CPU_LE_UNALIGN_64 + #elif defined(__ARM_FEATURE_UNALIGNED) + /* gcc9 for 32-bit arm can use LDRD instruction that requires 32-bit alignment. + So we can't use unaligned 64-bit operations. */ + #define MY_CPU_LE_UNALIGN + #endif +#endif + + +#ifdef MY_CPU_LE_UNALIGN + +#define GetUi16(p) (*(const UInt16 *)(const void *)(p)) +#define GetUi32(p) (*(const UInt32 *)(const void *)(p)) +#ifdef MY_CPU_LE_UNALIGN_64 +#define GetUi64(p) (*(const UInt64 *)(const void *)(p)) +#define SetUi64(p, v) { *(UInt64 *)(void *)(p) = (v); } +#endif + +#define SetUi16(p, v) { *(UInt16 *)(void *)(p) = (v); } +#define SetUi32(p, v) { *(UInt32 *)(void *)(p) = (v); } + +#else + +#define GetUi16(p) ( (UInt16) ( \ + ((const Byte *)(p))[0] | \ + ((UInt16)((const Byte *)(p))[1] << 8) )) + +#define GetUi32(p) ( \ + ((const Byte *)(p))[0] | \ + ((UInt32)((const Byte *)(p))[1] << 8) | \ + ((UInt32)((const Byte *)(p))[2] << 16) | \ + ((UInt32)((const Byte *)(p))[3] << 24)) + +#define SetUi16(p, v) { Byte *_ppp_ = (Byte *)(p); UInt32 _vvv_ = (v); \ + _ppp_[0] = (Byte)_vvv_; \ + _ppp_[1] = (Byte)(_vvv_ >> 8); } + +#define SetUi32(p, v) { Byte *_ppp_ = (Byte *)(p); UInt32 _vvv_ = (v); \ + _ppp_[0] = (Byte)_vvv_; \ + _ppp_[1] = (Byte)(_vvv_ >> 8); \ + _ppp_[2] = (Byte)(_vvv_ >> 16); \ + _ppp_[3] = (Byte)(_vvv_ >> 24); } + +#endif + + +#ifndef GetUi64 +#define GetUi64(p) (GetUi32(p) | ((UInt64)GetUi32(((const Byte *)(p)) + 4) << 32)) +#endif + +#ifndef SetUi64 +#define SetUi64(p, v) { Byte *_ppp2_ = (Byte *)(p); UInt64 _vvv2_ = (v); \ + SetUi32(_ppp2_ , (UInt32)_vvv2_) \ + SetUi32(_ppp2_ + 4, (UInt32)(_vvv2_ >> 32)) } +#endif + + +#if defined(MY_CPU_LE_UNALIGN) && defined(Z7_CPU_FAST_BSWAP_SUPPORTED) + +#define GetBe32(p) Z7_BSWAP32 (*(const UInt32 *)(const void *)(p)) +#define SetBe32(p, v) { (*(UInt32 *)(void *)(p)) = Z7_BSWAP32(v); } + +#if defined(MY_CPU_LE_UNALIGN_64) +#define GetBe64(p) Z7_BSWAP64 (*(const UInt64 *)(const void *)(p)) +#endif + +#else + +#define GetBe32(p) ( \ + ((UInt32)((const Byte *)(p))[0] << 24) | \ + ((UInt32)((const Byte *)(p))[1] << 16) | \ + ((UInt32)((const Byte *)(p))[2] << 8) | \ + ((const Byte *)(p))[3] ) + +#define SetBe32(p, v) { Byte *_ppp_ = (Byte *)(p); UInt32 _vvv_ = (v); \ + _ppp_[0] = (Byte)(_vvv_ >> 24); \ + _ppp_[1] = (Byte)(_vvv_ >> 16); \ + _ppp_[2] = (Byte)(_vvv_ >> 8); \ + _ppp_[3] = (Byte)_vvv_; } + +#endif + +#ifndef GetBe64 +#define GetBe64(p) (((UInt64)GetBe32(p) << 32) | GetBe32(((const Byte *)(p)) + 4)) +#endif + +#ifndef GetBe16 +#define GetBe16(p) ( (UInt16) ( \ + ((UInt16)((const Byte *)(p))[0] << 8) | \ + ((const Byte *)(p))[1] )) +#endif + + +#if defined(MY_CPU_BE) +#define Z7_CONV_BE_TO_NATIVE_CONST32(v) (v) +#define Z7_CONV_LE_TO_NATIVE_CONST32(v) Z7_BSWAP32_CONST(v) +#define Z7_CONV_NATIVE_TO_BE_32(v) (v) +#elif defined(MY_CPU_LE) +#define Z7_CONV_BE_TO_NATIVE_CONST32(v) Z7_BSWAP32_CONST(v) +#define Z7_CONV_LE_TO_NATIVE_CONST32(v) (v) +#define Z7_CONV_NATIVE_TO_BE_32(v) Z7_BSWAP32(v) +#else +#error Stop_Compiling_Unknown_Endian_CONV +#endif + + +#if defined(MY_CPU_BE) + +#define GetBe32a(p) (*(const UInt32 *)(const void *)(p)) +#define GetBe16a(p) (*(const UInt16 *)(const void *)(p)) +#define SetBe32a(p, v) { *(UInt32 *)(void *)(p) = (v); } +#define SetBe16a(p, v) { *(UInt16 *)(void *)(p) = (v); } + +#define GetUi32a(p) GetUi32(p) +#define GetUi16a(p) GetUi16(p) +#define SetUi32a(p, v) SetUi32(p, v) +#define SetUi16a(p, v) SetUi16(p, v) + +#elif defined(MY_CPU_LE) + +#define GetUi32a(p) (*(const UInt32 *)(const void *)(p)) +#define GetUi16a(p) (*(const UInt16 *)(const void *)(p)) +#define SetUi32a(p, v) { *(UInt32 *)(void *)(p) = (v); } +#define SetUi16a(p, v) { *(UInt16 *)(void *)(p) = (v); } + +#define GetBe32a(p) GetBe32(p) +#define GetBe16a(p) GetBe16(p) +#define SetBe32a(p, v) SetBe32(p, v) +#define SetBe16a(p, v) SetBe16(p, v) + +#else +#error Stop_Compiling_Unknown_Endian_CPU_a +#endif + + +#if defined(MY_CPU_X86_OR_AMD64) \ + || defined(MY_CPU_ARM_OR_ARM64) \ + || defined(MY_CPU_PPC_OR_PPC64) + #define Z7_CPU_FAST_ROTATE_SUPPORTED +#endif + + +#ifdef MY_CPU_X86_OR_AMD64 + +void Z7_FASTCALL z7_x86_cpuid(UInt32 a[4], UInt32 function); +UInt32 Z7_FASTCALL z7_x86_cpuid_GetMaxFunc(void); +#if defined(MY_CPU_AMD64) +#define Z7_IF_X86_CPUID_SUPPORTED +#else +#define Z7_IF_X86_CPUID_SUPPORTED if (z7_x86_cpuid_GetMaxFunc()) +#endif + +BoolInt CPU_IsSupported_AES(void); +BoolInt CPU_IsSupported_AVX(void); +BoolInt CPU_IsSupported_AVX2(void); +BoolInt CPU_IsSupported_VAES_AVX2(void); +BoolInt CPU_IsSupported_CMOV(void); +BoolInt CPU_IsSupported_SSE(void); +BoolInt CPU_IsSupported_SSE2(void); +BoolInt CPU_IsSupported_SSSE3(void); +BoolInt CPU_IsSupported_SSE41(void); +BoolInt CPU_IsSupported_SHA(void); +BoolInt CPU_IsSupported_PageGB(void); + +#elif defined(MY_CPU_ARM_OR_ARM64) + +BoolInt CPU_IsSupported_CRC32(void); +BoolInt CPU_IsSupported_NEON(void); + +#if defined(_WIN32) +BoolInt CPU_IsSupported_CRYPTO(void); +#define CPU_IsSupported_SHA1 CPU_IsSupported_CRYPTO +#define CPU_IsSupported_SHA2 CPU_IsSupported_CRYPTO +#define CPU_IsSupported_AES CPU_IsSupported_CRYPTO +#else +BoolInt CPU_IsSupported_SHA1(void); +BoolInt CPU_IsSupported_SHA2(void); +BoolInt CPU_IsSupported_AES(void); +#endif + +#endif + +#if defined(__APPLE__) +int z7_sysctlbyname_Get(const char *name, void *buf, size_t *bufSize); +int z7_sysctlbyname_Get_UInt32(const char *name, UInt32 *val); +#endif + +EXTERN_C_END + +#endif diff --git a/project/src/liblzma/LzFind.c b/project/src/liblzma/LzFind.c new file mode 100644 index 0000000..7589a4c --- /dev/null +++ b/project/src/liblzma/LzFind.c @@ -0,0 +1,1717 @@ +/* LzFind.c -- Match finder for LZ algorithms +2023-03-14 : Igor Pavlov : Public domain */ + +#include "Precomp.h" + +#include +// #include + +#include "CpuArch.h" +#include "LzFind.h" +#include "LzHash.h" + +#define kBlockMoveAlign (1 << 7) // alignment for memmove() +#define kBlockSizeAlign (1 << 16) // alignment for block allocation +#define kBlockSizeReserveMin (1 << 24) // it's 1/256 from 4 GB dictinary + +#define kEmptyHashValue 0 + +#define kMaxValForNormalize ((UInt32)0) +// #define kMaxValForNormalize ((UInt32)(1 << 20) + 0xfff) // for debug + +// #define kNormalizeAlign (1 << 7) // alignment for speculated accesses + +#define GET_AVAIL_BYTES(p) \ + Inline_MatchFinder_GetNumAvailableBytes(p) + + +// #define kFix5HashSize (kHash2Size + kHash3Size + kHash4Size) +#define kFix5HashSize kFix4HashSize + +/* + HASH2_CALC: + if (hv) match, then cur[0] and cur[1] also match +*/ +#define HASH2_CALC hv = GetUi16(cur); + +// (crc[0 ... 255] & 0xFF) provides one-to-one correspondence to [0 ... 255] + +/* + HASH3_CALC: + if (cur[0]) and (h2) match, then cur[1] also match + if (cur[0]) and (hv) match, then cur[1] and cur[2] also match +*/ +#define HASH3_CALC { \ + UInt32 temp = p->crc[cur[0]] ^ cur[1]; \ + h2 = temp & (kHash2Size - 1); \ + hv = (temp ^ ((UInt32)cur[2] << 8)) & p->hashMask; } + +#define HASH4_CALC { \ + UInt32 temp = p->crc[cur[0]] ^ cur[1]; \ + h2 = temp & (kHash2Size - 1); \ + temp ^= ((UInt32)cur[2] << 8); \ + h3 = temp & (kHash3Size - 1); \ + hv = (temp ^ (p->crc[cur[3]] << kLzHash_CrcShift_1)) & p->hashMask; } + +#define HASH5_CALC { \ + UInt32 temp = p->crc[cur[0]] ^ cur[1]; \ + h2 = temp & (kHash2Size - 1); \ + temp ^= ((UInt32)cur[2] << 8); \ + h3 = temp & (kHash3Size - 1); \ + temp ^= (p->crc[cur[3]] << kLzHash_CrcShift_1); \ + /* h4 = temp & p->hash4Mask; */ /* (kHash4Size - 1); */ \ + hv = (temp ^ (p->crc[cur[4]] << kLzHash_CrcShift_2)) & p->hashMask; } + +#define HASH_ZIP_CALC hv = ((cur[2] | ((UInt32)cur[0] << 8)) ^ p->crc[cur[1]]) & 0xFFFF; + + +static void LzInWindow_Free(CMatchFinder *p, ISzAllocPtr alloc) +{ + // if (!p->directInput) + { + ISzAlloc_Free(alloc, p->bufBase); + p->bufBase = NULL; + } +} + + +static int LzInWindow_Create2(CMatchFinder *p, UInt32 blockSize, ISzAllocPtr alloc) +{ + if (blockSize == 0) + return 0; + if (!p->bufBase || p->blockSize != blockSize) + { + // size_t blockSizeT; + LzInWindow_Free(p, alloc); + p->blockSize = blockSize; + // blockSizeT = blockSize; + + // printf("\nblockSize = 0x%x\n", blockSize); + /* + #if defined _WIN64 + // we can allocate 4GiB, but still use UInt32 for (p->blockSize) + // we use UInt32 type for (p->blockSize), because + // we don't want to wrap over 4 GiB, + // when we use (p->streamPos - p->pos) that is UInt32. + if (blockSize >= (UInt32)0 - (UInt32)kBlockSizeAlign) + { + blockSizeT = ((size_t)1 << 32); + printf("\nchanged to blockSizeT = 4GiB\n"); + } + #endif + */ + + p->bufBase = (Byte *)ISzAlloc_Alloc(alloc, blockSize); + // printf("\nbufferBase = %p\n", p->bufBase); + // return 0; // for debug + } + return (p->bufBase != NULL); +} + +static const Byte *MatchFinder_GetPointerToCurrentPos(CMatchFinder *p) { return p->buffer; } + +static UInt32 MatchFinder_GetNumAvailableBytes(CMatchFinder *p) { return GET_AVAIL_BYTES(p); } + + +Z7_NO_INLINE +static void MatchFinder_ReadBlock(CMatchFinder *p) +{ + if (p->streamEndWasReached || p->result != SZ_OK) + return; + + /* We use (p->streamPos - p->pos) value. + (p->streamPos < p->pos) is allowed. */ + + if (p->directInput) + { + UInt32 curSize = 0xFFFFFFFF - GET_AVAIL_BYTES(p); + if (curSize > p->directInputRem) + curSize = (UInt32)p->directInputRem; + p->streamPos += curSize; + p->directInputRem -= curSize; + if (p->directInputRem == 0) + p->streamEndWasReached = 1; + return; + } + + for (;;) + { + const Byte *dest = p->buffer + GET_AVAIL_BYTES(p); + size_t size = (size_t)(p->bufBase + p->blockSize - dest); + if (size == 0) + { + /* we call ReadBlock() after NeedMove() and MoveBlock(). + NeedMove() and MoveBlock() povide more than (keepSizeAfter) + to the end of (blockSize). + So we don't execute this branch in normal code flow. + We can go here, if we will call ReadBlock() before NeedMove(), MoveBlock(). + */ + // p->result = SZ_ERROR_FAIL; // we can show error here + return; + } + + // #define kRead 3 + // if (size > kRead) size = kRead; // for debug + + /* + // we need cast (Byte *)dest. + #ifdef __clang__ + #pragma GCC diagnostic ignored "-Wcast-qual" + #endif + */ + p->result = ISeqInStream_Read(p->stream, + p->bufBase + (dest - p->bufBase), &size); + if (p->result != SZ_OK) + return; + if (size == 0) + { + p->streamEndWasReached = 1; + return; + } + p->streamPos += (UInt32)size; + if (GET_AVAIL_BYTES(p) > p->keepSizeAfter) + return; + /* here and in another (p->keepSizeAfter) checks we keep on 1 byte more than was requested by Create() function + (GET_AVAIL_BYTES(p) >= p->keepSizeAfter) - minimal required size */ + } + + // on exit: (p->result != SZ_OK || p->streamEndWasReached || GET_AVAIL_BYTES(p) > p->keepSizeAfter) +} + + + +Z7_NO_INLINE +void MatchFinder_MoveBlock(CMatchFinder *p) +{ + const size_t offset = (size_t)(p->buffer - p->bufBase) - p->keepSizeBefore; + const size_t keepBefore = (offset & (kBlockMoveAlign - 1)) + p->keepSizeBefore; + p->buffer = p->bufBase + keepBefore; + memmove(p->bufBase, + p->bufBase + (offset & ~((size_t)kBlockMoveAlign - 1)), + keepBefore + (size_t)GET_AVAIL_BYTES(p)); +} + +/* We call MoveBlock() before ReadBlock(). + So MoveBlock() can be wasteful operation, if the whole input data + can fit in current block even without calling MoveBlock(). + in important case where (dataSize <= historySize) + condition (p->blockSize > dataSize + p->keepSizeAfter) is met + So there is no MoveBlock() in that case case. +*/ + +int MatchFinder_NeedMove(CMatchFinder *p) +{ + if (p->directInput) + return 0; + if (p->streamEndWasReached || p->result != SZ_OK) + return 0; + return ((size_t)(p->bufBase + p->blockSize - p->buffer) <= p->keepSizeAfter); +} + +void MatchFinder_ReadIfRequired(CMatchFinder *p) +{ + if (p->keepSizeAfter >= GET_AVAIL_BYTES(p)) + MatchFinder_ReadBlock(p); +} + + + +static void MatchFinder_SetDefaultSettings(CMatchFinder *p) +{ + p->cutValue = 32; + p->btMode = 1; + p->numHashBytes = 4; + p->numHashBytes_Min = 2; + p->numHashOutBits = 0; + p->bigHash = 0; +} + +#define kCrcPoly 0xEDB88320 + +void MatchFinder_Construct(CMatchFinder *p) +{ + unsigned i; + p->buffer = NULL; + p->bufBase = NULL; + p->directInput = 0; + p->stream = NULL; + p->hash = NULL; + p->expectedDataSize = (UInt64)(Int64)-1; + MatchFinder_SetDefaultSettings(p); + + for (i = 0; i < 256; i++) + { + UInt32 r = (UInt32)i; + unsigned j; + for (j = 0; j < 8; j++) + r = (r >> 1) ^ (kCrcPoly & ((UInt32)0 - (r & 1))); + p->crc[i] = r; + } +} + +#undef kCrcPoly + +static void MatchFinder_FreeThisClassMemory(CMatchFinder *p, ISzAllocPtr alloc) +{ + ISzAlloc_Free(alloc, p->hash); + p->hash = NULL; +} + +void MatchFinder_Free(CMatchFinder *p, ISzAllocPtr alloc) +{ + MatchFinder_FreeThisClassMemory(p, alloc); + LzInWindow_Free(p, alloc); +} + +static CLzRef* AllocRefs(size_t num, ISzAllocPtr alloc) +{ + const size_t sizeInBytes = (size_t)num * sizeof(CLzRef); + if (sizeInBytes / sizeof(CLzRef) != num) + return NULL; + return (CLzRef *)ISzAlloc_Alloc(alloc, sizeInBytes); +} + +#if (kBlockSizeReserveMin < kBlockSizeAlign * 2) + #error Stop_Compiling_Bad_Reserve +#endif + + + +static UInt32 GetBlockSize(CMatchFinder *p, UInt32 historySize) +{ + UInt32 blockSize = (p->keepSizeBefore + p->keepSizeAfter); + /* + if (historySize > kMaxHistorySize) + return 0; + */ + // printf("\nhistorySize == 0x%x\n", historySize); + + if (p->keepSizeBefore < historySize || blockSize < p->keepSizeBefore) // if 32-bit overflow + return 0; + + { + const UInt32 kBlockSizeMax = (UInt32)0 - (UInt32)kBlockSizeAlign; + const UInt32 rem = kBlockSizeMax - blockSize; + const UInt32 reserve = (blockSize >> (blockSize < ((UInt32)1 << 30) ? 1 : 2)) + + (1 << 12) + kBlockMoveAlign + kBlockSizeAlign; // do not overflow 32-bit here + if (blockSize >= kBlockSizeMax + || rem < kBlockSizeReserveMin) // we reject settings that will be slow + return 0; + if (reserve >= rem) + blockSize = kBlockSizeMax; + else + { + blockSize += reserve; + blockSize &= ~(UInt32)(kBlockSizeAlign - 1); + } + } + // printf("\n LzFind_blockSize = %x\n", blockSize); + // printf("\n LzFind_blockSize = %d\n", blockSize >> 20); + return blockSize; +} + + +// input is historySize +static UInt32 MatchFinder_GetHashMask2(CMatchFinder *p, UInt32 hs) +{ + if (p->numHashBytes == 2) + return (1 << 16) - 1; + if (hs != 0) + hs--; + hs |= (hs >> 1); + hs |= (hs >> 2); + hs |= (hs >> 4); + hs |= (hs >> 8); + // we propagated 16 bits in (hs). Low 16 bits must be set later + if (hs >= (1 << 24)) + { + if (p->numHashBytes == 3) + hs = (1 << 24) - 1; + /* if (bigHash) mode, GetHeads4b() in LzFindMt.c needs (hs >= ((1 << 24) - 1))) */ + } + // (hash_size >= (1 << 16)) : Required for (numHashBytes > 2) + hs |= (1 << 16) - 1; /* don't change it! */ + // bt5: we adjust the size with recommended minimum size + if (p->numHashBytes >= 5) + hs |= (256 << kLzHash_CrcShift_2) - 1; + return hs; +} + +// input is historySize +static UInt32 MatchFinder_GetHashMask(CMatchFinder *p, UInt32 hs) +{ + if (p->numHashBytes == 2) + return (1 << 16) - 1; + if (hs != 0) + hs--; + hs |= (hs >> 1); + hs |= (hs >> 2); + hs |= (hs >> 4); + hs |= (hs >> 8); + // we propagated 16 bits in (hs). Low 16 bits must be set later + hs >>= 1; + if (hs >= (1 << 24)) + { + if (p->numHashBytes == 3) + hs = (1 << 24) - 1; + else + hs >>= 1; + /* if (bigHash) mode, GetHeads4b() in LzFindMt.c needs (hs >= ((1 << 24) - 1))) */ + } + // (hash_size >= (1 << 16)) : Required for (numHashBytes > 2) + hs |= (1 << 16) - 1; /* don't change it! */ + // bt5: we adjust the size with recommended minimum size + if (p->numHashBytes >= 5) + hs |= (256 << kLzHash_CrcShift_2) - 1; + return hs; +} + + +int MatchFinder_Create(CMatchFinder *p, UInt32 historySize, + UInt32 keepAddBufferBefore, UInt32 matchMaxLen, UInt32 keepAddBufferAfter, + ISzAllocPtr alloc) +{ + /* we need one additional byte in (p->keepSizeBefore), + since we use MoveBlock() after (p->pos++) and before dictionary using */ + // keepAddBufferBefore = (UInt32)0xFFFFFFFF - (1 << 22); // for debug + p->keepSizeBefore = historySize + keepAddBufferBefore + 1; + + keepAddBufferAfter += matchMaxLen; + /* we need (p->keepSizeAfter >= p->numHashBytes) */ + if (keepAddBufferAfter < p->numHashBytes) + keepAddBufferAfter = p->numHashBytes; + // keepAddBufferAfter -= 2; // for debug + p->keepSizeAfter = keepAddBufferAfter; + + if (p->directInput) + p->blockSize = 0; + if (p->directInput || LzInWindow_Create2(p, GetBlockSize(p, historySize), alloc)) + { + size_t hashSizeSum; + { + UInt32 hs; + UInt32 hsCur; + + if (p->numHashOutBits != 0) + { + unsigned numBits = p->numHashOutBits; + const unsigned nbMax = + (p->numHashBytes == 2 ? 16 : + (p->numHashBytes == 3 ? 24 : 32)); + if (numBits > nbMax) + numBits = nbMax; + if (numBits >= 32) + hs = (UInt32)0 - 1; + else + hs = ((UInt32)1 << numBits) - 1; + // (hash_size >= (1 << 16)) : Required for (numHashBytes > 2) + hs |= (1 << 16) - 1; /* don't change it! */ + if (p->numHashBytes >= 5) + hs |= (256 << kLzHash_CrcShift_2) - 1; + { + const UInt32 hs2 = MatchFinder_GetHashMask2(p, historySize); + if (hs > hs2) + hs = hs2; + } + hsCur = hs; + if (p->expectedDataSize < historySize) + { + const UInt32 hs2 = MatchFinder_GetHashMask2(p, (UInt32)p->expectedDataSize); + if (hsCur > hs2) + hsCur = hs2; + } + } + else + { + hs = MatchFinder_GetHashMask(p, historySize); + hsCur = hs; + if (p->expectedDataSize < historySize) + { + hsCur = MatchFinder_GetHashMask(p, (UInt32)p->expectedDataSize); + if (hsCur > hs) // is it possible? + hsCur = hs; + } + } + + p->hashMask = hsCur; + + hashSizeSum = hs; + hashSizeSum++; + if (hashSizeSum < hs) + return 0; + { + UInt32 fixedHashSize = 0; + if (p->numHashBytes > 2 && p->numHashBytes_Min <= 2) fixedHashSize += kHash2Size; + if (p->numHashBytes > 3 && p->numHashBytes_Min <= 3) fixedHashSize += kHash3Size; + // if (p->numHashBytes > 4) p->fixedHashSize += hs4; // kHash4Size; + hashSizeSum += fixedHashSize; + p->fixedHashSize = fixedHashSize; + } + } + + p->matchMaxLen = matchMaxLen; + + { + size_t newSize; + size_t numSons; + const UInt32 newCyclicBufferSize = historySize + 1; // do not change it + p->historySize = historySize; + p->cyclicBufferSize = newCyclicBufferSize; // it must be = (historySize + 1) + + numSons = newCyclicBufferSize; + if (p->btMode) + numSons <<= 1; + newSize = hashSizeSum + numSons; + + if (numSons < newCyclicBufferSize || newSize < numSons) + return 0; + + // aligned size is not required here, but it can be better for some loops + #define NUM_REFS_ALIGN_MASK 0xF + newSize = (newSize + NUM_REFS_ALIGN_MASK) & ~(size_t)NUM_REFS_ALIGN_MASK; + + // 22.02: we don't reallocate buffer, if old size is enough + if (p->hash && p->numRefs >= newSize) + return 1; + + MatchFinder_FreeThisClassMemory(p, alloc); + p->numRefs = newSize; + p->hash = AllocRefs(newSize, alloc); + + if (p->hash) + { + p->son = p->hash + hashSizeSum; + return 1; + } + } + } + + MatchFinder_Free(p, alloc); + return 0; +} + + +static void MatchFinder_SetLimits(CMatchFinder *p) +{ + UInt32 k; + UInt32 n = kMaxValForNormalize - p->pos; + if (n == 0) + n = (UInt32)(Int32)-1; // we allow (pos == 0) at start even with (kMaxValForNormalize == 0) + + k = p->cyclicBufferSize - p->cyclicBufferPos; + if (k < n) + n = k; + + k = GET_AVAIL_BYTES(p); + { + const UInt32 ksa = p->keepSizeAfter; + UInt32 mm = p->matchMaxLen; + if (k > ksa) + k -= ksa; // we must limit exactly to keepSizeAfter for ReadBlock + else if (k >= mm) + { + // the limitation for (p->lenLimit) update + k -= mm; // optimization : to reduce the number of checks + k++; + // k = 1; // non-optimized version : for debug + } + else + { + mm = k; + if (k != 0) + k = 1; + } + p->lenLimit = mm; + } + if (k < n) + n = k; + + p->posLimit = p->pos + n; +} + + +void MatchFinder_Init_LowHash(CMatchFinder *p) +{ + size_t i; + CLzRef *items = p->hash; + const size_t numItems = p->fixedHashSize; + for (i = 0; i < numItems; i++) + items[i] = kEmptyHashValue; +} + + +void MatchFinder_Init_HighHash(CMatchFinder *p) +{ + size_t i; + CLzRef *items = p->hash + p->fixedHashSize; + const size_t numItems = (size_t)p->hashMask + 1; + for (i = 0; i < numItems; i++) + items[i] = kEmptyHashValue; +} + + +void MatchFinder_Init_4(CMatchFinder *p) +{ + if (!p->directInput) + p->buffer = p->bufBase; + { + /* kEmptyHashValue = 0 (Zero) is used in hash tables as NO-VALUE marker. + the code in CMatchFinderMt expects (pos = 1) */ + p->pos = + p->streamPos = + 1; // it's smallest optimal value. do not change it + // 0; // for debug + } + p->result = SZ_OK; + p->streamEndWasReached = 0; +} + + +// (CYC_TO_POS_OFFSET == 0) is expected by some optimized code +#define CYC_TO_POS_OFFSET 0 +// #define CYC_TO_POS_OFFSET 1 // for debug + +void MatchFinder_Init(CMatchFinder *p) +{ + MatchFinder_Init_HighHash(p); + MatchFinder_Init_LowHash(p); + MatchFinder_Init_4(p); + // if (readData) + MatchFinder_ReadBlock(p); + + /* if we init (cyclicBufferPos = pos), then we can use one variable + instead of both (cyclicBufferPos) and (pos) : only before (cyclicBufferPos) wrapping */ + p->cyclicBufferPos = (p->pos - CYC_TO_POS_OFFSET); // init with relation to (pos) + // p->cyclicBufferPos = 0; // smallest value + // p->son[0] = p->son[1] = 0; // unused: we can init skipped record for speculated accesses. + MatchFinder_SetLimits(p); +} + + + +#ifdef MY_CPU_X86_OR_AMD64 + #if defined(__clang__) && (__clang_major__ >= 4) \ + || defined(Z7_GCC_VERSION) && (Z7_GCC_VERSION >= 40701) + // || defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 1900) + + #define USE_LZFIND_SATUR_SUB_128 + #define USE_LZFIND_SATUR_SUB_256 + #define LZFIND_ATTRIB_SSE41 __attribute__((__target__("sse4.1"))) + #define LZFIND_ATTRIB_AVX2 __attribute__((__target__("avx2"))) + #elif defined(_MSC_VER) + #if (_MSC_VER >= 1600) + #define USE_LZFIND_SATUR_SUB_128 + #endif + #if (_MSC_VER >= 1900) + #define USE_LZFIND_SATUR_SUB_256 + #endif + #endif + +// #elif defined(MY_CPU_ARM_OR_ARM64) +#elif defined(MY_CPU_ARM64) + + #if defined(__clang__) && (__clang_major__ >= 8) \ + || defined(__GNUC__) && (__GNUC__ >= 8) + #define USE_LZFIND_SATUR_SUB_128 + #ifdef MY_CPU_ARM64 + // #define LZFIND_ATTRIB_SSE41 __attribute__((__target__(""))) + #else + // #define LZFIND_ATTRIB_SSE41 __attribute__((__target__("fpu=crypto-neon-fp-armv8"))) + #endif + + #elif defined(_MSC_VER) + #if (_MSC_VER >= 1910) + #define USE_LZFIND_SATUR_SUB_128 + #endif + #endif + + #if defined(_MSC_VER) && defined(MY_CPU_ARM64) + #include + #else + #include + #endif + +#endif + + +#ifdef USE_LZFIND_SATUR_SUB_128 + +// #define Z7_SHOW_HW_STATUS + +#ifdef Z7_SHOW_HW_STATUS +#include +#define PRF(x) x +PRF(;) +#else +#define PRF(x) +#endif + + +#ifdef MY_CPU_ARM_OR_ARM64 + +#ifdef MY_CPU_ARM64 +// #define FORCE_LZFIND_SATUR_SUB_128 +#endif +typedef uint32x4_t LzFind_v128; +#define SASUB_128_V(v, s) \ + vsubq_u32(vmaxq_u32(v, s), s) + +#else // MY_CPU_ARM_OR_ARM64 + +#include // sse4.1 + +typedef __m128i LzFind_v128; +// SSE 4.1 +#define SASUB_128_V(v, s) \ + _mm_sub_epi32(_mm_max_epu32(v, s), s) + +#endif // MY_CPU_ARM_OR_ARM64 + + +#define SASUB_128(i) \ + *( LzFind_v128 *)( void *)(items + (i) * 4) = SASUB_128_V( \ + *(const LzFind_v128 *)(const void *)(items + (i) * 4), sub2); + + +Z7_NO_INLINE +static +#ifdef LZFIND_ATTRIB_SSE41 +LZFIND_ATTRIB_SSE41 +#endif +void +Z7_FASTCALL +LzFind_SaturSub_128(UInt32 subValue, CLzRef *items, const CLzRef *lim) +{ + const LzFind_v128 sub2 = + #ifdef MY_CPU_ARM_OR_ARM64 + vdupq_n_u32(subValue); + #else + _mm_set_epi32((Int32)subValue, (Int32)subValue, (Int32)subValue, (Int32)subValue); + #endif + Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE + do + { + SASUB_128(0) SASUB_128(1) items += 2 * 4; + SASUB_128(0) SASUB_128(1) items += 2 * 4; + } + while (items != lim); +} + + + +#ifdef USE_LZFIND_SATUR_SUB_256 + +#include // avx +/* +clang :immintrin.h uses +#if !(defined(_MSC_VER) || defined(__SCE__)) || __has_feature(modules) || \ + defined(__AVX2__) +#include +#endif +so we need for clang-cl */ + +#if defined(__clang__) +#include +#include +#endif + +// AVX2: +#define SASUB_256(i) \ + *( __m256i *)( void *)(items + (i) * 8) = \ + _mm256_sub_epi32(_mm256_max_epu32( \ + *(const __m256i *)(const void *)(items + (i) * 8), sub2), sub2); + +Z7_NO_INLINE +static +#ifdef LZFIND_ATTRIB_AVX2 +LZFIND_ATTRIB_AVX2 +#endif +void +Z7_FASTCALL +LzFind_SaturSub_256(UInt32 subValue, CLzRef *items, const CLzRef *lim) +{ + const __m256i sub2 = _mm256_set_epi32( + (Int32)subValue, (Int32)subValue, (Int32)subValue, (Int32)subValue, + (Int32)subValue, (Int32)subValue, (Int32)subValue, (Int32)subValue); + Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE + do + { + SASUB_256(0) SASUB_256(1) items += 2 * 8; + SASUB_256(0) SASUB_256(1) items += 2 * 8; + } + while (items != lim); +} +#endif // USE_LZFIND_SATUR_SUB_256 + +#ifndef FORCE_LZFIND_SATUR_SUB_128 +typedef void (Z7_FASTCALL *LZFIND_SATUR_SUB_CODE_FUNC)( + UInt32 subValue, CLzRef *items, const CLzRef *lim); +static LZFIND_SATUR_SUB_CODE_FUNC g_LzFind_SaturSub; +#endif // FORCE_LZFIND_SATUR_SUB_128 + +#endif // USE_LZFIND_SATUR_SUB_128 + + +// kEmptyHashValue must be zero +// #define SASUB_32(i) { UInt32 v = items[i]; UInt32 m = v - subValue; if (v < subValue) m = kEmptyHashValue; items[i] = m; } +#define SASUB_32(i) { UInt32 v = items[i]; if (v < subValue) v = subValue; items[i] = v - subValue; } + +#ifdef FORCE_LZFIND_SATUR_SUB_128 + +#define DEFAULT_SaturSub LzFind_SaturSub_128 + +#else + +#define DEFAULT_SaturSub LzFind_SaturSub_32 + +Z7_NO_INLINE +static +void +Z7_FASTCALL +LzFind_SaturSub_32(UInt32 subValue, CLzRef *items, const CLzRef *lim) +{ + Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE + do + { + SASUB_32(0) SASUB_32(1) items += 2; + SASUB_32(0) SASUB_32(1) items += 2; + SASUB_32(0) SASUB_32(1) items += 2; + SASUB_32(0) SASUB_32(1) items += 2; + } + while (items != lim); +} + +#endif + + +Z7_NO_INLINE +void MatchFinder_Normalize3(UInt32 subValue, CLzRef *items, size_t numItems) +{ + #define LZFIND_NORM_ALIGN_BLOCK_SIZE (1 << 7) + Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE + for (; numItems != 0 && ((unsigned)(ptrdiff_t)items & (LZFIND_NORM_ALIGN_BLOCK_SIZE - 1)) != 0; numItems--) + { + SASUB_32(0) + items++; + } + { + const size_t k_Align_Mask = (LZFIND_NORM_ALIGN_BLOCK_SIZE / 4 - 1); + CLzRef *lim = items + (numItems & ~(size_t)k_Align_Mask); + numItems &= k_Align_Mask; + if (items != lim) + { + #if defined(USE_LZFIND_SATUR_SUB_128) && !defined(FORCE_LZFIND_SATUR_SUB_128) + if (g_LzFind_SaturSub) + g_LzFind_SaturSub(subValue, items, lim); + else + #endif + DEFAULT_SaturSub(subValue, items, lim); + } + items = lim; + } + Z7_PRAGMA_OPT_DISABLE_LOOP_UNROLL_VECTORIZE + for (; numItems != 0; numItems--) + { + SASUB_32(0) + items++; + } +} + + + +// call MatchFinder_CheckLimits() only after (p->pos++) update + +Z7_NO_INLINE +static void MatchFinder_CheckLimits(CMatchFinder *p) +{ + if (// !p->streamEndWasReached && p->result == SZ_OK && + p->keepSizeAfter == GET_AVAIL_BYTES(p)) + { + // we try to read only in exact state (p->keepSizeAfter == GET_AVAIL_BYTES(p)) + if (MatchFinder_NeedMove(p)) + MatchFinder_MoveBlock(p); + MatchFinder_ReadBlock(p); + } + + if (p->pos == kMaxValForNormalize) + if (GET_AVAIL_BYTES(p) >= p->numHashBytes) // optional optimization for last bytes of data. + /* + if we disable normalization for last bytes of data, and + if (data_size == 4 GiB), we don't call wastfull normalization, + but (pos) will be wrapped over Zero (0) in that case. + And we cannot resume later to normal operation + */ + { + // MatchFinder_Normalize(p); + /* after normalization we need (p->pos >= p->historySize + 1); */ + /* we can reduce subValue to aligned value, if want to keep alignment + of (p->pos) and (p->buffer) for speculated accesses. */ + const UInt32 subValue = (p->pos - p->historySize - 1) /* & ~(UInt32)(kNormalizeAlign - 1) */; + // const UInt32 subValue = (1 << 15); // for debug + // printf("\nMatchFinder_Normalize() subValue == 0x%x\n", subValue); + MatchFinder_REDUCE_OFFSETS(p, subValue) + MatchFinder_Normalize3(subValue, p->hash, (size_t)p->hashMask + 1 + p->fixedHashSize); + { + size_t numSonRefs = p->cyclicBufferSize; + if (p->btMode) + numSonRefs <<= 1; + MatchFinder_Normalize3(subValue, p->son, numSonRefs); + } + } + + if (p->cyclicBufferPos == p->cyclicBufferSize) + p->cyclicBufferPos = 0; + + MatchFinder_SetLimits(p); +} + + +/* + (lenLimit > maxLen) +*/ +Z7_FORCE_INLINE +static UInt32 * Hc_GetMatchesSpec(size_t lenLimit, UInt32 curMatch, UInt32 pos, const Byte *cur, CLzRef *son, + size_t _cyclicBufferPos, UInt32 _cyclicBufferSize, UInt32 cutValue, + UInt32 *d, unsigned maxLen) +{ + /* + son[_cyclicBufferPos] = curMatch; + for (;;) + { + UInt32 delta = pos - curMatch; + if (cutValue-- == 0 || delta >= _cyclicBufferSize) + return d; + { + const Byte *pb = cur - delta; + curMatch = son[_cyclicBufferPos - delta + ((delta > _cyclicBufferPos) ? _cyclicBufferSize : 0)]; + if (pb[maxLen] == cur[maxLen] && *pb == *cur) + { + UInt32 len = 0; + while (++len != lenLimit) + if (pb[len] != cur[len]) + break; + if (maxLen < len) + { + maxLen = len; + *d++ = len; + *d++ = delta - 1; + if (len == lenLimit) + return d; + } + } + } + } + */ + + const Byte *lim = cur + lenLimit; + son[_cyclicBufferPos] = curMatch; + + do + { + UInt32 delta; + + if (curMatch == 0) + break; + // if (curMatch2 >= curMatch) return NULL; + delta = pos - curMatch; + if (delta >= _cyclicBufferSize) + break; + { + ptrdiff_t diff; + curMatch = son[_cyclicBufferPos - delta + ((delta > _cyclicBufferPos) ? _cyclicBufferSize : 0)]; + diff = (ptrdiff_t)0 - (ptrdiff_t)delta; + if (cur[maxLen] == cur[(ptrdiff_t)maxLen + diff]) + { + const Byte *c = cur; + while (*c == c[diff]) + { + if (++c == lim) + { + d[0] = (UInt32)(lim - cur); + d[1] = delta - 1; + return d + 2; + } + } + { + const unsigned len = (unsigned)(c - cur); + if (maxLen < len) + { + maxLen = len; + d[0] = (UInt32)len; + d[1] = delta - 1; + d += 2; + } + } + } + } + } + while (--cutValue); + + return d; +} + + +Z7_FORCE_INLINE +UInt32 * GetMatchesSpec1(UInt32 lenLimit, UInt32 curMatch, UInt32 pos, const Byte *cur, CLzRef *son, + size_t _cyclicBufferPos, UInt32 _cyclicBufferSize, UInt32 cutValue, + UInt32 *d, UInt32 maxLen) +{ + CLzRef *ptr0 = son + ((size_t)_cyclicBufferPos << 1) + 1; + CLzRef *ptr1 = son + ((size_t)_cyclicBufferPos << 1); + unsigned len0 = 0, len1 = 0; + + UInt32 cmCheck; + + // if (curMatch >= pos) { *ptr0 = *ptr1 = kEmptyHashValue; return NULL; } + + cmCheck = (UInt32)(pos - _cyclicBufferSize); + if ((UInt32)pos <= _cyclicBufferSize) + cmCheck = 0; + + if (cmCheck < curMatch) + do + { + const UInt32 delta = pos - curMatch; + { + CLzRef *pair = son + ((size_t)(_cyclicBufferPos - delta + ((delta > _cyclicBufferPos) ? _cyclicBufferSize : 0)) << 1); + const Byte *pb = cur - delta; + unsigned len = (len0 < len1 ? len0 : len1); + const UInt32 pair0 = pair[0]; + if (pb[len] == cur[len]) + { + if (++len != lenLimit && pb[len] == cur[len]) + while (++len != lenLimit) + if (pb[len] != cur[len]) + break; + if (maxLen < len) + { + maxLen = (UInt32)len; + *d++ = (UInt32)len; + *d++ = delta - 1; + if (len == lenLimit) + { + *ptr1 = pair0; + *ptr0 = pair[1]; + return d; + } + } + } + if (pb[len] < cur[len]) + { + *ptr1 = curMatch; + // const UInt32 curMatch2 = pair[1]; + // if (curMatch2 >= curMatch) { *ptr0 = *ptr1 = kEmptyHashValue; return NULL; } + // curMatch = curMatch2; + curMatch = pair[1]; + ptr1 = pair + 1; + len1 = len; + } + else + { + *ptr0 = curMatch; + curMatch = pair[0]; + ptr0 = pair; + len0 = len; + } + } + } + while(--cutValue && cmCheck < curMatch); + + *ptr0 = *ptr1 = kEmptyHashValue; + return d; +} + + +static void SkipMatchesSpec(UInt32 lenLimit, UInt32 curMatch, UInt32 pos, const Byte *cur, CLzRef *son, + size_t _cyclicBufferPos, UInt32 _cyclicBufferSize, UInt32 cutValue) +{ + CLzRef *ptr0 = son + ((size_t)_cyclicBufferPos << 1) + 1; + CLzRef *ptr1 = son + ((size_t)_cyclicBufferPos << 1); + unsigned len0 = 0, len1 = 0; + + UInt32 cmCheck; + + cmCheck = (UInt32)(pos - _cyclicBufferSize); + if ((UInt32)pos <= _cyclicBufferSize) + cmCheck = 0; + + if (// curMatch >= pos || // failure + cmCheck < curMatch) + do + { + const UInt32 delta = pos - curMatch; + { + CLzRef *pair = son + ((size_t)(_cyclicBufferPos - delta + ((delta > _cyclicBufferPos) ? _cyclicBufferSize : 0)) << 1); + const Byte *pb = cur - delta; + unsigned len = (len0 < len1 ? len0 : len1); + if (pb[len] == cur[len]) + { + while (++len != lenLimit) + if (pb[len] != cur[len]) + break; + { + if (len == lenLimit) + { + *ptr1 = pair[0]; + *ptr0 = pair[1]; + return; + } + } + } + if (pb[len] < cur[len]) + { + *ptr1 = curMatch; + curMatch = pair[1]; + ptr1 = pair + 1; + len1 = len; + } + else + { + *ptr0 = curMatch; + curMatch = pair[0]; + ptr0 = pair; + len0 = len; + } + } + } + while(--cutValue && cmCheck < curMatch); + + *ptr0 = *ptr1 = kEmptyHashValue; + return; +} + + +#define MOVE_POS \ + ++p->cyclicBufferPos; \ + p->buffer++; \ + { const UInt32 pos1 = p->pos + 1; p->pos = pos1; if (pos1 == p->posLimit) MatchFinder_CheckLimits(p); } + +#define MOVE_POS_RET MOVE_POS return distances; + +Z7_NO_INLINE +static void MatchFinder_MovePos(CMatchFinder *p) +{ + /* we go here at the end of stream data, when (avail < num_hash_bytes) + We don't update sons[cyclicBufferPos << btMode]. + So (sons) record will contain junk. And we cannot resume match searching + to normal operation, even if we will provide more input data in buffer. + p->sons[p->cyclicBufferPos << p->btMode] = 0; // kEmptyHashValue + if (p->btMode) + p->sons[(p->cyclicBufferPos << p->btMode) + 1] = 0; // kEmptyHashValue + */ + MOVE_POS +} + +#define GET_MATCHES_HEADER2(minLen, ret_op) \ + unsigned lenLimit; UInt32 hv; const Byte *cur; UInt32 curMatch; \ + lenLimit = (unsigned)p->lenLimit; { if (lenLimit < minLen) { MatchFinder_MovePos(p); ret_op; }} \ + cur = p->buffer; + +#define GET_MATCHES_HEADER(minLen) GET_MATCHES_HEADER2(minLen, return distances) +#define SKIP_HEADER(minLen) do { GET_MATCHES_HEADER2(minLen, continue) + +#define MF_PARAMS(p) lenLimit, curMatch, p->pos, p->buffer, p->son, p->cyclicBufferPos, p->cyclicBufferSize, p->cutValue + +#define SKIP_FOOTER SkipMatchesSpec(MF_PARAMS(p)); MOVE_POS } while (--num); + +#define GET_MATCHES_FOOTER_BASE(_maxLen_, func) \ + distances = func(MF_PARAMS(p), \ + distances, (UInt32)_maxLen_); MOVE_POS_RET + +#define GET_MATCHES_FOOTER_BT(_maxLen_) \ + GET_MATCHES_FOOTER_BASE(_maxLen_, GetMatchesSpec1) + +#define GET_MATCHES_FOOTER_HC(_maxLen_) \ + GET_MATCHES_FOOTER_BASE(_maxLen_, Hc_GetMatchesSpec) + + + +#define UPDATE_maxLen { \ + const ptrdiff_t diff = (ptrdiff_t)0 - (ptrdiff_t)d2; \ + const Byte *c = cur + maxLen; \ + const Byte *lim = cur + lenLimit; \ + for (; c != lim; c++) if (*(c + diff) != *c) break; \ + maxLen = (unsigned)(c - cur); } + +static UInt32* Bt2_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances) +{ + GET_MATCHES_HEADER(2) + HASH2_CALC + curMatch = p->hash[hv]; + p->hash[hv] = p->pos; + GET_MATCHES_FOOTER_BT(1) +} + +UInt32* Bt3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances) +{ + GET_MATCHES_HEADER(3) + HASH_ZIP_CALC + curMatch = p->hash[hv]; + p->hash[hv] = p->pos; + GET_MATCHES_FOOTER_BT(2) +} + + +#define SET_mmm \ + mmm = p->cyclicBufferSize; \ + if (pos < mmm) \ + mmm = pos; + + +static UInt32* Bt3_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances) +{ + UInt32 mmm; + UInt32 h2, d2, pos; + unsigned maxLen; + UInt32 *hash; + GET_MATCHES_HEADER(3) + + HASH3_CALC + + hash = p->hash; + pos = p->pos; + + d2 = pos - hash[h2]; + + curMatch = (hash + kFix3HashSize)[hv]; + + hash[h2] = pos; + (hash + kFix3HashSize)[hv] = pos; + + SET_mmm + + maxLen = 2; + + if (d2 < mmm && *(cur - d2) == *cur) + { + UPDATE_maxLen + distances[0] = (UInt32)maxLen; + distances[1] = d2 - 1; + distances += 2; + if (maxLen == lenLimit) + { + SkipMatchesSpec(MF_PARAMS(p)); + MOVE_POS_RET + } + } + + GET_MATCHES_FOOTER_BT(maxLen) +} + + +static UInt32* Bt4_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances) +{ + UInt32 mmm; + UInt32 h2, h3, d2, d3, pos; + unsigned maxLen; + UInt32 *hash; + GET_MATCHES_HEADER(4) + + HASH4_CALC + + hash = p->hash; + pos = p->pos; + + d2 = pos - hash [h2]; + d3 = pos - (hash + kFix3HashSize)[h3]; + curMatch = (hash + kFix4HashSize)[hv]; + + hash [h2] = pos; + (hash + kFix3HashSize)[h3] = pos; + (hash + kFix4HashSize)[hv] = pos; + + SET_mmm + + maxLen = 3; + + for (;;) + { + if (d2 < mmm && *(cur - d2) == *cur) + { + distances[0] = 2; + distances[1] = d2 - 1; + distances += 2; + if (*(cur - d2 + 2) == cur[2]) + { + // distances[-2] = 3; + } + else if (d3 < mmm && *(cur - d3) == *cur) + { + d2 = d3; + distances[1] = d3 - 1; + distances += 2; + } + else + break; + } + else if (d3 < mmm && *(cur - d3) == *cur) + { + d2 = d3; + distances[1] = d3 - 1; + distances += 2; + } + else + break; + + UPDATE_maxLen + distances[-2] = (UInt32)maxLen; + if (maxLen == lenLimit) + { + SkipMatchesSpec(MF_PARAMS(p)); + MOVE_POS_RET + } + break; + } + + GET_MATCHES_FOOTER_BT(maxLen) +} + + +static UInt32* Bt5_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances) +{ + UInt32 mmm; + UInt32 h2, h3, d2, d3, maxLen, pos; + UInt32 *hash; + GET_MATCHES_HEADER(5) + + HASH5_CALC + + hash = p->hash; + pos = p->pos; + + d2 = pos - hash [h2]; + d3 = pos - (hash + kFix3HashSize)[h3]; + // d4 = pos - (hash + kFix4HashSize)[h4]; + + curMatch = (hash + kFix5HashSize)[hv]; + + hash [h2] = pos; + (hash + kFix3HashSize)[h3] = pos; + // (hash + kFix4HashSize)[h4] = pos; + (hash + kFix5HashSize)[hv] = pos; + + SET_mmm + + maxLen = 4; + + for (;;) + { + if (d2 < mmm && *(cur - d2) == *cur) + { + distances[0] = 2; + distances[1] = d2 - 1; + distances += 2; + if (*(cur - d2 + 2) == cur[2]) + { + } + else if (d3 < mmm && *(cur - d3) == *cur) + { + distances[1] = d3 - 1; + distances += 2; + d2 = d3; + } + else + break; + } + else if (d3 < mmm && *(cur - d3) == *cur) + { + distances[1] = d3 - 1; + distances += 2; + d2 = d3; + } + else + break; + + distances[-2] = 3; + if (*(cur - d2 + 3) != cur[3]) + break; + UPDATE_maxLen + distances[-2] = (UInt32)maxLen; + if (maxLen == lenLimit) + { + SkipMatchesSpec(MF_PARAMS(p)); + MOVE_POS_RET + } + break; + } + + GET_MATCHES_FOOTER_BT(maxLen) +} + + +static UInt32* Hc4_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances) +{ + UInt32 mmm; + UInt32 h2, h3, d2, d3, pos; + unsigned maxLen; + UInt32 *hash; + GET_MATCHES_HEADER(4) + + HASH4_CALC + + hash = p->hash; + pos = p->pos; + + d2 = pos - hash [h2]; + d3 = pos - (hash + kFix3HashSize)[h3]; + curMatch = (hash + kFix4HashSize)[hv]; + + hash [h2] = pos; + (hash + kFix3HashSize)[h3] = pos; + (hash + kFix4HashSize)[hv] = pos; + + SET_mmm + + maxLen = 3; + + for (;;) + { + if (d2 < mmm && *(cur - d2) == *cur) + { + distances[0] = 2; + distances[1] = d2 - 1; + distances += 2; + if (*(cur - d2 + 2) == cur[2]) + { + // distances[-2] = 3; + } + else if (d3 < mmm && *(cur - d3) == *cur) + { + d2 = d3; + distances[1] = d3 - 1; + distances += 2; + } + else + break; + } + else if (d3 < mmm && *(cur - d3) == *cur) + { + d2 = d3; + distances[1] = d3 - 1; + distances += 2; + } + else + break; + + UPDATE_maxLen + distances[-2] = (UInt32)maxLen; + if (maxLen == lenLimit) + { + p->son[p->cyclicBufferPos] = curMatch; + MOVE_POS_RET + } + break; + } + + GET_MATCHES_FOOTER_HC(maxLen) +} + + +static UInt32 * Hc5_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances) +{ + UInt32 mmm; + UInt32 h2, h3, d2, d3, maxLen, pos; + UInt32 *hash; + GET_MATCHES_HEADER(5) + + HASH5_CALC + + hash = p->hash; + pos = p->pos; + + d2 = pos - hash [h2]; + d3 = pos - (hash + kFix3HashSize)[h3]; + // d4 = pos - (hash + kFix4HashSize)[h4]; + + curMatch = (hash + kFix5HashSize)[hv]; + + hash [h2] = pos; + (hash + kFix3HashSize)[h3] = pos; + // (hash + kFix4HashSize)[h4] = pos; + (hash + kFix5HashSize)[hv] = pos; + + SET_mmm + + maxLen = 4; + + for (;;) + { + if (d2 < mmm && *(cur - d2) == *cur) + { + distances[0] = 2; + distances[1] = d2 - 1; + distances += 2; + if (*(cur - d2 + 2) == cur[2]) + { + } + else if (d3 < mmm && *(cur - d3) == *cur) + { + distances[1] = d3 - 1; + distances += 2; + d2 = d3; + } + else + break; + } + else if (d3 < mmm && *(cur - d3) == *cur) + { + distances[1] = d3 - 1; + distances += 2; + d2 = d3; + } + else + break; + + distances[-2] = 3; + if (*(cur - d2 + 3) != cur[3]) + break; + UPDATE_maxLen + distances[-2] = maxLen; + if (maxLen == lenLimit) + { + p->son[p->cyclicBufferPos] = curMatch; + MOVE_POS_RET + } + break; + } + + GET_MATCHES_FOOTER_HC(maxLen) +} + + +UInt32* Hc3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances) +{ + GET_MATCHES_HEADER(3) + HASH_ZIP_CALC + curMatch = p->hash[hv]; + p->hash[hv] = p->pos; + GET_MATCHES_FOOTER_HC(2) +} + + +static void Bt2_MatchFinder_Skip(CMatchFinder *p, UInt32 num) +{ + SKIP_HEADER(2) + { + HASH2_CALC + curMatch = p->hash[hv]; + p->hash[hv] = p->pos; + } + SKIP_FOOTER +} + +void Bt3Zip_MatchFinder_Skip(CMatchFinder *p, UInt32 num) +{ + SKIP_HEADER(3) + { + HASH_ZIP_CALC + curMatch = p->hash[hv]; + p->hash[hv] = p->pos; + } + SKIP_FOOTER +} + +static void Bt3_MatchFinder_Skip(CMatchFinder *p, UInt32 num) +{ + SKIP_HEADER(3) + { + UInt32 h2; + UInt32 *hash; + HASH3_CALC + hash = p->hash; + curMatch = (hash + kFix3HashSize)[hv]; + hash[h2] = + (hash + kFix3HashSize)[hv] = p->pos; + } + SKIP_FOOTER +} + +static void Bt4_MatchFinder_Skip(CMatchFinder *p, UInt32 num) +{ + SKIP_HEADER(4) + { + UInt32 h2, h3; + UInt32 *hash; + HASH4_CALC + hash = p->hash; + curMatch = (hash + kFix4HashSize)[hv]; + hash [h2] = + (hash + kFix3HashSize)[h3] = + (hash + kFix4HashSize)[hv] = p->pos; + } + SKIP_FOOTER +} + +static void Bt5_MatchFinder_Skip(CMatchFinder *p, UInt32 num) +{ + SKIP_HEADER(5) + { + UInt32 h2, h3; + UInt32 *hash; + HASH5_CALC + hash = p->hash; + curMatch = (hash + kFix5HashSize)[hv]; + hash [h2] = + (hash + kFix3HashSize)[h3] = + // (hash + kFix4HashSize)[h4] = + (hash + kFix5HashSize)[hv] = p->pos; + } + SKIP_FOOTER +} + + +#define HC_SKIP_HEADER(minLen) \ + do { if (p->lenLimit < minLen) { MatchFinder_MovePos(p); num--; continue; } { \ + const Byte *cur; \ + UInt32 *hash; \ + UInt32 *son; \ + UInt32 pos = p->pos; \ + UInt32 num2 = num; \ + /* (p->pos == p->posLimit) is not allowed here !!! */ \ + { const UInt32 rem = p->posLimit - pos; if (num2 > rem) num2 = rem; } \ + num -= num2; \ + { const UInt32 cycPos = p->cyclicBufferPos; \ + son = p->son + cycPos; \ + p->cyclicBufferPos = cycPos + num2; } \ + cur = p->buffer; \ + hash = p->hash; \ + do { \ + UInt32 curMatch; \ + UInt32 hv; + + +#define HC_SKIP_FOOTER \ + cur++; pos++; *son++ = curMatch; \ + } while (--num2); \ + p->buffer = cur; \ + p->pos = pos; \ + if (pos == p->posLimit) MatchFinder_CheckLimits(p); \ + }} while(num); \ + + +static void Hc4_MatchFinder_Skip(CMatchFinder *p, UInt32 num) +{ + HC_SKIP_HEADER(4) + + UInt32 h2, h3; + HASH4_CALC + curMatch = (hash + kFix4HashSize)[hv]; + hash [h2] = + (hash + kFix3HashSize)[h3] = + (hash + kFix4HashSize)[hv] = pos; + + HC_SKIP_FOOTER +} + + +static void Hc5_MatchFinder_Skip(CMatchFinder *p, UInt32 num) +{ + HC_SKIP_HEADER(5) + + UInt32 h2, h3; + HASH5_CALC + curMatch = (hash + kFix5HashSize)[hv]; + hash [h2] = + (hash + kFix3HashSize)[h3] = + // (hash + kFix4HashSize)[h4] = + (hash + kFix5HashSize)[hv] = pos; + + HC_SKIP_FOOTER +} + + +void Hc3Zip_MatchFinder_Skip(CMatchFinder *p, UInt32 num) +{ + HC_SKIP_HEADER(3) + + HASH_ZIP_CALC + curMatch = hash[hv]; + hash[hv] = pos; + + HC_SKIP_FOOTER +} + + +void MatchFinder_CreateVTable(CMatchFinder *p, IMatchFinder2 *vTable) +{ + vTable->Init = (Mf_Init_Func)MatchFinder_Init; + vTable->GetNumAvailableBytes = (Mf_GetNumAvailableBytes_Func)MatchFinder_GetNumAvailableBytes; + vTable->GetPointerToCurrentPos = (Mf_GetPointerToCurrentPos_Func)MatchFinder_GetPointerToCurrentPos; + if (!p->btMode) + { + if (p->numHashBytes <= 4) + { + vTable->GetMatches = (Mf_GetMatches_Func)Hc4_MatchFinder_GetMatches; + vTable->Skip = (Mf_Skip_Func)Hc4_MatchFinder_Skip; + } + else + { + vTable->GetMatches = (Mf_GetMatches_Func)Hc5_MatchFinder_GetMatches; + vTable->Skip = (Mf_Skip_Func)Hc5_MatchFinder_Skip; + } + } + else if (p->numHashBytes == 2) + { + vTable->GetMatches = (Mf_GetMatches_Func)Bt2_MatchFinder_GetMatches; + vTable->Skip = (Mf_Skip_Func)Bt2_MatchFinder_Skip; + } + else if (p->numHashBytes == 3) + { + vTable->GetMatches = (Mf_GetMatches_Func)Bt3_MatchFinder_GetMatches; + vTable->Skip = (Mf_Skip_Func)Bt3_MatchFinder_Skip; + } + else if (p->numHashBytes == 4) + { + vTable->GetMatches = (Mf_GetMatches_Func)Bt4_MatchFinder_GetMatches; + vTable->Skip = (Mf_Skip_Func)Bt4_MatchFinder_Skip; + } + else + { + vTable->GetMatches = (Mf_GetMatches_Func)Bt5_MatchFinder_GetMatches; + vTable->Skip = (Mf_Skip_Func)Bt5_MatchFinder_Skip; + } +} + + + +void LzFindPrepare(void) +{ + #ifndef FORCE_LZFIND_SATUR_SUB_128 + #ifdef USE_LZFIND_SATUR_SUB_128 + LZFIND_SATUR_SUB_CODE_FUNC f = NULL; + #ifdef MY_CPU_ARM_OR_ARM64 + { + if (CPU_IsSupported_NEON()) + { + // #pragma message ("=== LzFind NEON") + PRF(printf("\n=== LzFind NEON\n")); + f = LzFind_SaturSub_128; + } + // f = 0; // for debug + } + #else // MY_CPU_ARM_OR_ARM64 + if (CPU_IsSupported_SSE41()) + { + // #pragma message ("=== LzFind SSE41") + PRF(printf("\n=== LzFind SSE41\n")); + f = LzFind_SaturSub_128; + + #ifdef USE_LZFIND_SATUR_SUB_256 + if (CPU_IsSupported_AVX2()) + { + // #pragma message ("=== LzFind AVX2") + PRF(printf("\n=== LzFind AVX2\n")); + f = LzFind_SaturSub_256; + } + #endif + } + #endif // MY_CPU_ARM_OR_ARM64 + g_LzFind_SaturSub = f; + #endif // USE_LZFIND_SATUR_SUB_128 + #endif // FORCE_LZFIND_SATUR_SUB_128 +} + + +#undef MOVE_POS +#undef MOVE_POS_RET +#undef PRF diff --git a/project/src/liblzma/LzFind.h b/project/src/liblzma/LzFind.h new file mode 100644 index 0000000..5acdb98 --- /dev/null +++ b/project/src/liblzma/LzFind.h @@ -0,0 +1,159 @@ +/* LzFind.h -- Match finder for LZ algorithms +2023-03-04 : Igor Pavlov : Public domain */ + +#ifndef ZIP7_INC_LZ_FIND_H +#define ZIP7_INC_LZ_FIND_H + +#include "7zTypes.h" + +EXTERN_C_BEGIN + +typedef UInt32 CLzRef; + +typedef struct +{ + const Byte *buffer; + UInt32 pos; + UInt32 posLimit; + UInt32 streamPos; /* wrap over Zero is allowed (streamPos < pos). Use (UInt32)(streamPos - pos) */ + UInt32 lenLimit; + + UInt32 cyclicBufferPos; + UInt32 cyclicBufferSize; /* it must be = (historySize + 1) */ + + Byte streamEndWasReached; + Byte btMode; + Byte bigHash; + Byte directInput; + + UInt32 matchMaxLen; + CLzRef *hash; + CLzRef *son; + UInt32 hashMask; + UInt32 cutValue; + + Byte *bufBase; + ISeqInStreamPtr stream; + + UInt32 blockSize; + UInt32 keepSizeBefore; + UInt32 keepSizeAfter; + + UInt32 numHashBytes; + size_t directInputRem; + UInt32 historySize; + UInt32 fixedHashSize; + Byte numHashBytes_Min; + Byte numHashOutBits; + Byte _pad2_[2]; + SRes result; + UInt32 crc[256]; + size_t numRefs; + + UInt64 expectedDataSize; +} CMatchFinder; + +#define Inline_MatchFinder_GetPointerToCurrentPos(p) ((const Byte *)(p)->buffer) + +#define Inline_MatchFinder_GetNumAvailableBytes(p) ((UInt32)((p)->streamPos - (p)->pos)) + +/* +#define Inline_MatchFinder_IsFinishedOK(p) \ + ((p)->streamEndWasReached \ + && (p)->streamPos == (p)->pos \ + && (!(p)->directInput || (p)->directInputRem == 0)) +*/ + +int MatchFinder_NeedMove(CMatchFinder *p); +/* Byte *MatchFinder_GetPointerToCurrentPos(CMatchFinder *p); */ +void MatchFinder_MoveBlock(CMatchFinder *p); +void MatchFinder_ReadIfRequired(CMatchFinder *p); + +void MatchFinder_Construct(CMatchFinder *p); + +/* (directInput = 0) is default value. + It's required to provide correct (directInput) value + before calling MatchFinder_Create(). + You can set (directInput) by any of the following calls: + - MatchFinder_SET_DIRECT_INPUT_BUF() + - MatchFinder_SET_STREAM() + - MatchFinder_SET_STREAM_MODE() +*/ + +#define MatchFinder_SET_DIRECT_INPUT_BUF(p, _src_, _srcLen_) { \ + (p)->stream = NULL; \ + (p)->directInput = 1; \ + (p)->buffer = (_src_); \ + (p)->directInputRem = (_srcLen_); } + +/* +#define MatchFinder_SET_STREAM_MODE(p) { \ + (p)->directInput = 0; } +*/ + +#define MatchFinder_SET_STREAM(p, _stream_) { \ + (p)->stream = _stream_; \ + (p)->directInput = 0; } + + +int MatchFinder_Create(CMatchFinder *p, UInt32 historySize, + UInt32 keepAddBufferBefore, UInt32 matchMaxLen, UInt32 keepAddBufferAfter, + ISzAllocPtr alloc); +void MatchFinder_Free(CMatchFinder *p, ISzAllocPtr alloc); +void MatchFinder_Normalize3(UInt32 subValue, CLzRef *items, size_t numItems); + +/* +#define MatchFinder_INIT_POS(p, val) \ + (p)->pos = (val); \ + (p)->streamPos = (val); +*/ + +// void MatchFinder_ReduceOffsets(CMatchFinder *p, UInt32 subValue); +#define MatchFinder_REDUCE_OFFSETS(p, subValue) \ + (p)->pos -= (subValue); \ + (p)->streamPos -= (subValue); + + +UInt32 * GetMatchesSpec1(UInt32 lenLimit, UInt32 curMatch, UInt32 pos, const Byte *buffer, CLzRef *son, + size_t _cyclicBufferPos, UInt32 _cyclicBufferSize, UInt32 _cutValue, + UInt32 *distances, UInt32 maxLen); + +/* +Conditions: + Mf_GetNumAvailableBytes_Func must be called before each Mf_GetMatchLen_Func. + Mf_GetPointerToCurrentPos_Func's result must be used only before any other function +*/ + +typedef void (*Mf_Init_Func)(void *object); +typedef UInt32 (*Mf_GetNumAvailableBytes_Func)(void *object); +typedef const Byte * (*Mf_GetPointerToCurrentPos_Func)(void *object); +typedef UInt32 * (*Mf_GetMatches_Func)(void *object, UInt32 *distances); +typedef void (*Mf_Skip_Func)(void *object, UInt32); + +typedef struct +{ + Mf_Init_Func Init; + Mf_GetNumAvailableBytes_Func GetNumAvailableBytes; + Mf_GetPointerToCurrentPos_Func GetPointerToCurrentPos; + Mf_GetMatches_Func GetMatches; + Mf_Skip_Func Skip; +} IMatchFinder2; + +void MatchFinder_CreateVTable(CMatchFinder *p, IMatchFinder2 *vTable); + +void MatchFinder_Init_LowHash(CMatchFinder *p); +void MatchFinder_Init_HighHash(CMatchFinder *p); +void MatchFinder_Init_4(CMatchFinder *p); +void MatchFinder_Init(CMatchFinder *p); + +UInt32* Bt3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances); +UInt32* Hc3Zip_MatchFinder_GetMatches(CMatchFinder *p, UInt32 *distances); + +void Bt3Zip_MatchFinder_Skip(CMatchFinder *p, UInt32 num); +void Hc3Zip_MatchFinder_Skip(CMatchFinder *p, UInt32 num); + +void LzFindPrepare(void); + +EXTERN_C_END + +#endif diff --git a/project/src/liblzma/LzFindMt.h b/project/src/liblzma/LzFindMt.h new file mode 100644 index 0000000..5002d0c --- /dev/null +++ b/project/src/liblzma/LzFindMt.h @@ -0,0 +1,109 @@ +/* LzFindMt.h -- multithreaded Match finder for LZ algorithms +2023-03-05 : Igor Pavlov : Public domain */ + +#ifndef ZIP7_INC_LZ_FIND_MT_H +#define ZIP7_INC_LZ_FIND_MT_H + +#include "LzFind.h" +#include "Threads.h" + +EXTERN_C_BEGIN + +typedef struct +{ + UInt32 numProcessedBlocks; + CThread thread; + UInt64 affinity; + + BoolInt wasCreated; + BoolInt needStart; + BoolInt csWasInitialized; + BoolInt csWasEntered; + + BoolInt exit; + BoolInt stopWriting; + + CAutoResetEvent canStart; + CAutoResetEvent wasStopped; + CSemaphore freeSemaphore; + CSemaphore filledSemaphore; + CCriticalSection cs; + // UInt32 numBlocks_Sent; +} CMtSync; + +typedef UInt32 * (*Mf_Mix_Matches)(void *p, UInt32 matchMinPos, UInt32 *distances); + +/* kMtCacheLineDummy must be >= size_of_CPU_cache_line */ +#define kMtCacheLineDummy 128 + +typedef void (*Mf_GetHeads)(const Byte *buffer, UInt32 pos, + UInt32 *hash, UInt32 hashMask, UInt32 *heads, UInt32 numHeads, const UInt32 *crc); + +typedef struct +{ + /* LZ */ + const Byte *pointerToCurPos; + UInt32 *btBuf; + const UInt32 *btBufPos; + const UInt32 *btBufPosLimit; + UInt32 lzPos; + UInt32 btNumAvailBytes; + + UInt32 *hash; + UInt32 fixedHashSize; + // UInt32 hash4Mask; + UInt32 historySize; + const UInt32 *crc; + + Mf_Mix_Matches MixMatchesFunc; + UInt32 failure_LZ_BT; // failure in BT transfered to LZ + // UInt32 failure_LZ_LZ; // failure in LZ tables + UInt32 failureBuf[1]; + // UInt32 crc[256]; + + /* LZ + BT */ + CMtSync btSync; + Byte btDummy[kMtCacheLineDummy]; + + /* BT */ + UInt32 *hashBuf; + UInt32 hashBufPos; + UInt32 hashBufPosLimit; + UInt32 hashNumAvail; + UInt32 failure_BT; + + + CLzRef *son; + UInt32 matchMaxLen; + UInt32 numHashBytes; + UInt32 pos; + const Byte *buffer; + UInt32 cyclicBufferPos; + UInt32 cyclicBufferSize; /* it must be = (historySize + 1) */ + UInt32 cutValue; + + /* BT + Hash */ + CMtSync hashSync; + /* Byte hashDummy[kMtCacheLineDummy]; */ + + /* Hash */ + Mf_GetHeads GetHeadsFunc; + CMatchFinder *MatchFinder; + // CMatchFinder MatchFinder; +} CMatchFinderMt; + +// only for Mt part +void MatchFinderMt_Construct(CMatchFinderMt *p); +void MatchFinderMt_Destruct(CMatchFinderMt *p, ISzAllocPtr alloc); + +SRes MatchFinderMt_Create(CMatchFinderMt *p, UInt32 historySize, UInt32 keepAddBufferBefore, + UInt32 matchMaxLen, UInt32 keepAddBufferAfter, ISzAllocPtr alloc); +void MatchFinderMt_CreateVTable(CMatchFinderMt *p, IMatchFinder2 *vTable); + +/* call MatchFinderMt_InitMt() before IMatchFinder::Init() */ +SRes MatchFinderMt_InitMt(CMatchFinderMt *p); +void MatchFinderMt_ReleaseStream(CMatchFinderMt *p); + +EXTERN_C_END + +#endif diff --git a/project/src/liblzma/LzHash.h b/project/src/liblzma/LzHash.h new file mode 100644 index 0000000..3b172eb --- /dev/null +++ b/project/src/liblzma/LzHash.h @@ -0,0 +1,34 @@ +/* LzHash.h -- HASH constants for LZ algorithms +2023-03-05 : Igor Pavlov : Public domain */ + +#ifndef ZIP7_INC_LZ_HASH_H +#define ZIP7_INC_LZ_HASH_H + +/* + (kHash2Size >= (1 << 8)) : Required + (kHash3Size >= (1 << 16)) : Required +*/ + +#define kHash2Size (1 << 10) +#define kHash3Size (1 << 16) +// #define kHash4Size (1 << 20) + +#define kFix3HashSize (kHash2Size) +#define kFix4HashSize (kHash2Size + kHash3Size) +// #define kFix5HashSize (kHash2Size + kHash3Size + kHash4Size) + +/* + We use up to 3 crc values for hash: + crc0 + crc1 << Shift_1 + crc2 << Shift_2 + (Shift_1 = 5) and (Shift_2 = 10) is good tradeoff. + Small values for Shift are not good for collision rate. + Big value for Shift_2 increases the minimum size + of hash table, that will be slow for small files. +*/ + +#define kLzHash_CrcShift_1 5 +#define kLzHash_CrcShift_2 10 + +#endif diff --git a/project/src/liblzma/LzmaDec.c b/project/src/liblzma/LzmaDec.c new file mode 100644 index 0000000..b6bcd50 --- /dev/null +++ b/project/src/liblzma/LzmaDec.c @@ -0,0 +1,1363 @@ +/* LzmaDec.c -- LZMA Decoder +2023-04-07 : Igor Pavlov : Public domain */ + +#include "Precomp.h" + +#include + +/* #include "CpuArch.h" */ +#include "LzmaDec.h" + +// #define kNumTopBits 24 +#define kTopValue ((UInt32)1 << 24) + +#define kNumBitModelTotalBits 11 +#define kBitModelTotal (1 << kNumBitModelTotalBits) + +#define RC_INIT_SIZE 5 + +#ifndef Z7_LZMA_DEC_OPT + +#define kNumMoveBits 5 +#define NORMALIZE if (range < kTopValue) { range <<= 8; code = (code << 8) | (*buf++); } + +#define IF_BIT_0(p) ttt = *(p); NORMALIZE; bound = (range >> kNumBitModelTotalBits) * (UInt32)ttt; if (code < bound) +#define UPDATE_0(p) range = bound; *(p) = (CLzmaProb)(ttt + ((kBitModelTotal - ttt) >> kNumMoveBits)); +#define UPDATE_1(p) range -= bound; code -= bound; *(p) = (CLzmaProb)(ttt - (ttt >> kNumMoveBits)); +#define GET_BIT2(p, i, A0, A1) IF_BIT_0(p) \ + { UPDATE_0(p) i = (i + i); A0; } else \ + { UPDATE_1(p) i = (i + i) + 1; A1; } + +#define TREE_GET_BIT(probs, i) { GET_BIT2(probs + i, i, ;, ;); } + +#define REV_BIT(p, i, A0, A1) IF_BIT_0(p + i) \ + { UPDATE_0(p + i) A0; } else \ + { UPDATE_1(p + i) A1; } +#define REV_BIT_VAR( p, i, m) REV_BIT(p, i, i += m; m += m, m += m; i += m; ) +#define REV_BIT_CONST(p, i, m) REV_BIT(p, i, i += m; , i += m * 2; ) +#define REV_BIT_LAST( p, i, m) REV_BIT(p, i, i -= m , ; ) + +#define TREE_DECODE(probs, limit, i) \ + { i = 1; do { TREE_GET_BIT(probs, i); } while (i < limit); i -= limit; } + +/* #define Z7_LZMA_SIZE_OPT */ + +#ifdef Z7_LZMA_SIZE_OPT +#define TREE_6_DECODE(probs, i) TREE_DECODE(probs, (1 << 6), i) +#else +#define TREE_6_DECODE(probs, i) \ + { i = 1; \ + TREE_GET_BIT(probs, i) \ + TREE_GET_BIT(probs, i) \ + TREE_GET_BIT(probs, i) \ + TREE_GET_BIT(probs, i) \ + TREE_GET_BIT(probs, i) \ + TREE_GET_BIT(probs, i) \ + i -= 0x40; } +#endif + +#define NORMAL_LITER_DEC TREE_GET_BIT(prob, symbol) +#define MATCHED_LITER_DEC \ + matchByte += matchByte; \ + bit = offs; \ + offs &= matchByte; \ + probLit = prob + (offs + bit + symbol); \ + GET_BIT2(probLit, symbol, offs ^= bit; , ;) + +#endif // Z7_LZMA_DEC_OPT + + +#define NORMALIZE_CHECK if (range < kTopValue) { if (buf >= bufLimit) return DUMMY_INPUT_EOF; range <<= 8; code = (code << 8) | (*buf++); } + +#define IF_BIT_0_CHECK(p) ttt = *(p); NORMALIZE_CHECK bound = (range >> kNumBitModelTotalBits) * (UInt32)ttt; if (code < bound) +#define UPDATE_0_CHECK range = bound; +#define UPDATE_1_CHECK range -= bound; code -= bound; +#define GET_BIT2_CHECK(p, i, A0, A1) IF_BIT_0_CHECK(p) \ + { UPDATE_0_CHECK i = (i + i); A0; } else \ + { UPDATE_1_CHECK i = (i + i) + 1; A1; } +#define GET_BIT_CHECK(p, i) GET_BIT2_CHECK(p, i, ; , ;) +#define TREE_DECODE_CHECK(probs, limit, i) \ + { i = 1; do { GET_BIT_CHECK(probs + i, i) } while (i < limit); i -= limit; } + + +#define REV_BIT_CHECK(p, i, m) IF_BIT_0_CHECK(p + i) \ + { UPDATE_0_CHECK i += m; m += m; } else \ + { UPDATE_1_CHECK m += m; i += m; } + + +#define kNumPosBitsMax 4 +#define kNumPosStatesMax (1 << kNumPosBitsMax) + +#define kLenNumLowBits 3 +#define kLenNumLowSymbols (1 << kLenNumLowBits) +#define kLenNumHighBits 8 +#define kLenNumHighSymbols (1 << kLenNumHighBits) + +#define LenLow 0 +#define LenHigh (LenLow + 2 * (kNumPosStatesMax << kLenNumLowBits)) +#define kNumLenProbs (LenHigh + kLenNumHighSymbols) + +#define LenChoice LenLow +#define LenChoice2 (LenLow + (1 << kLenNumLowBits)) + +#define kNumStates 12 +#define kNumStates2 16 +#define kNumLitStates 7 + +#define kStartPosModelIndex 4 +#define kEndPosModelIndex 14 +#define kNumFullDistances (1 << (kEndPosModelIndex >> 1)) + +#define kNumPosSlotBits 6 +#define kNumLenToPosStates 4 + +#define kNumAlignBits 4 +#define kAlignTableSize (1 << kNumAlignBits) + +#define kMatchMinLen 2 +#define kMatchSpecLenStart (kMatchMinLen + kLenNumLowSymbols * 2 + kLenNumHighSymbols) + +#define kMatchSpecLen_Error_Data (1 << 9) +#define kMatchSpecLen_Error_Fail (kMatchSpecLen_Error_Data - 1) + +/* External ASM code needs same CLzmaProb array layout. So don't change it. */ + +/* (probs_1664) is faster and better for code size at some platforms */ +/* +#ifdef MY_CPU_X86_OR_AMD64 +*/ +#define kStartOffset 1664 +#define GET_PROBS p->probs_1664 +/* +#define GET_PROBS p->probs + kStartOffset +#else +#define kStartOffset 0 +#define GET_PROBS p->probs +#endif +*/ + +#define SpecPos (-kStartOffset) +#define IsRep0Long (SpecPos + kNumFullDistances) +#define RepLenCoder (IsRep0Long + (kNumStates2 << kNumPosBitsMax)) +#define LenCoder (RepLenCoder + kNumLenProbs) +#define IsMatch (LenCoder + kNumLenProbs) +#define Align (IsMatch + (kNumStates2 << kNumPosBitsMax)) +#define IsRep (Align + kAlignTableSize) +#define IsRepG0 (IsRep + kNumStates) +#define IsRepG1 (IsRepG0 + kNumStates) +#define IsRepG2 (IsRepG1 + kNumStates) +#define PosSlot (IsRepG2 + kNumStates) +#define Literal (PosSlot + (kNumLenToPosStates << kNumPosSlotBits)) +#define NUM_BASE_PROBS (Literal + kStartOffset) + +#if Align != 0 && kStartOffset != 0 + #error Stop_Compiling_Bad_LZMA_kAlign +#endif + +#if NUM_BASE_PROBS != 1984 + #error Stop_Compiling_Bad_LZMA_PROBS +#endif + + +#define LZMA_LIT_SIZE 0x300 + +#define LzmaProps_GetNumProbs(p) (NUM_BASE_PROBS + ((UInt32)LZMA_LIT_SIZE << ((p)->lc + (p)->lp))) + + +#define CALC_POS_STATE(processedPos, pbMask) (((processedPos) & (pbMask)) << 4) +#define COMBINED_PS_STATE (posState + state) +#define GET_LEN_STATE (posState) + +#define LZMA_DIC_MIN (1 << 12) + +/* +p->remainLen : shows status of LZMA decoder: + < kMatchSpecLenStart : the number of bytes to be copied with (p->rep0) offset + = kMatchSpecLenStart : the LZMA stream was finished with end mark + = kMatchSpecLenStart + 1 : need init range coder + = kMatchSpecLenStart + 2 : need init range coder and state + = kMatchSpecLen_Error_Fail : Internal Code Failure + = kMatchSpecLen_Error_Data + [0 ... 273] : LZMA Data Error +*/ + +/* ---------- LZMA_DECODE_REAL ---------- */ +/* +LzmaDec_DecodeReal_3() can be implemented in external ASM file. +3 - is the code compatibility version of that function for check at link time. +*/ + +#define LZMA_DECODE_REAL LzmaDec_DecodeReal_3 + +/* +LZMA_DECODE_REAL() +In: + RangeCoder is normalized + if (p->dicPos == limit) + { + LzmaDec_TryDummy() was called before to exclude LITERAL and MATCH-REP cases. + So first symbol can be only MATCH-NON-REP. And if that MATCH-NON-REP symbol + is not END_OF_PAYALOAD_MARKER, then the function doesn't write any byte to dictionary, + the function returns SZ_OK, and the caller can use (p->remainLen) and (p->reps[0]) later. + } + +Processing: + The first LZMA symbol will be decoded in any case. + All main checks for limits are at the end of main loop, + It decodes additional LZMA-symbols while (p->buf < bufLimit && dicPos < limit), + RangeCoder is still without last normalization when (p->buf < bufLimit) is being checked. + But if (p->buf < bufLimit), the caller provided at least (LZMA_REQUIRED_INPUT_MAX + 1) bytes for + next iteration before limit (bufLimit + LZMA_REQUIRED_INPUT_MAX), + that is enough for worst case LZMA symbol with one additional RangeCoder normalization for one bit. + So that function never reads bufLimit [LZMA_REQUIRED_INPUT_MAX] byte. + +Out: + RangeCoder is normalized + Result: + SZ_OK - OK + p->remainLen: + < kMatchSpecLenStart : the number of bytes to be copied with (p->reps[0]) offset + = kMatchSpecLenStart : the LZMA stream was finished with end mark + + SZ_ERROR_DATA - error, when the MATCH-Symbol refers out of dictionary + p->remainLen : undefined + p->reps[*] : undefined +*/ + + +#ifdef Z7_LZMA_DEC_OPT + +int Z7_FASTCALL LZMA_DECODE_REAL(CLzmaDec *p, SizeT limit, const Byte *bufLimit); + +#else + +static +int Z7_FASTCALL LZMA_DECODE_REAL(CLzmaDec *p, SizeT limit, const Byte *bufLimit) +{ + CLzmaProb *probs = GET_PROBS; + unsigned state = (unsigned)p->state; + UInt32 rep0 = p->reps[0], rep1 = p->reps[1], rep2 = p->reps[2], rep3 = p->reps[3]; + unsigned pbMask = ((unsigned)1 << (p->prop.pb)) - 1; + unsigned lc = p->prop.lc; + unsigned lpMask = ((unsigned)0x100 << p->prop.lp) - ((unsigned)0x100 >> lc); + + Byte *dic = p->dic; + SizeT dicBufSize = p->dicBufSize; + SizeT dicPos = p->dicPos; + + UInt32 processedPos = p->processedPos; + UInt32 checkDicSize = p->checkDicSize; + unsigned len = 0; + + const Byte *buf = p->buf; + UInt32 range = p->range; + UInt32 code = p->code; + + do + { + CLzmaProb *prob; + UInt32 bound; + unsigned ttt; + unsigned posState = CALC_POS_STATE(processedPos, pbMask); + + prob = probs + IsMatch + COMBINED_PS_STATE; + IF_BIT_0(prob) + { + unsigned symbol; + UPDATE_0(prob) + prob = probs + Literal; + if (processedPos != 0 || checkDicSize != 0) + prob += (UInt32)3 * ((((processedPos << 8) + dic[(dicPos == 0 ? dicBufSize : dicPos) - 1]) & lpMask) << lc); + processedPos++; + + if (state < kNumLitStates) + { + state -= (state < 4) ? state : 3; + symbol = 1; + #ifdef Z7_LZMA_SIZE_OPT + do { NORMAL_LITER_DEC } while (symbol < 0x100); + #else + NORMAL_LITER_DEC + NORMAL_LITER_DEC + NORMAL_LITER_DEC + NORMAL_LITER_DEC + NORMAL_LITER_DEC + NORMAL_LITER_DEC + NORMAL_LITER_DEC + NORMAL_LITER_DEC + #endif + } + else + { + unsigned matchByte = dic[dicPos - rep0 + (dicPos < rep0 ? dicBufSize : 0)]; + unsigned offs = 0x100; + state -= (state < 10) ? 3 : 6; + symbol = 1; + #ifdef Z7_LZMA_SIZE_OPT + do + { + unsigned bit; + CLzmaProb *probLit; + MATCHED_LITER_DEC + } + while (symbol < 0x100); + #else + { + unsigned bit; + CLzmaProb *probLit; + MATCHED_LITER_DEC + MATCHED_LITER_DEC + MATCHED_LITER_DEC + MATCHED_LITER_DEC + MATCHED_LITER_DEC + MATCHED_LITER_DEC + MATCHED_LITER_DEC + MATCHED_LITER_DEC + } + #endif + } + + dic[dicPos++] = (Byte)symbol; + continue; + } + + { + UPDATE_1(prob) + prob = probs + IsRep + state; + IF_BIT_0(prob) + { + UPDATE_0(prob) + state += kNumStates; + prob = probs + LenCoder; + } + else + { + UPDATE_1(prob) + prob = probs + IsRepG0 + state; + IF_BIT_0(prob) + { + UPDATE_0(prob) + prob = probs + IsRep0Long + COMBINED_PS_STATE; + IF_BIT_0(prob) + { + UPDATE_0(prob) + + // that case was checked before with kBadRepCode + // if (checkDicSize == 0 && processedPos == 0) { len = kMatchSpecLen_Error_Data + 1; break; } + // The caller doesn't allow (dicPos == limit) case here + // so we don't need the following check: + // if (dicPos == limit) { state = state < kNumLitStates ? 9 : 11; len = 1; break; } + + dic[dicPos] = dic[dicPos - rep0 + (dicPos < rep0 ? dicBufSize : 0)]; + dicPos++; + processedPos++; + state = state < kNumLitStates ? 9 : 11; + continue; + } + UPDATE_1(prob) + } + else + { + UInt32 distance; + UPDATE_1(prob) + prob = probs + IsRepG1 + state; + IF_BIT_0(prob) + { + UPDATE_0(prob) + distance = rep1; + } + else + { + UPDATE_1(prob) + prob = probs + IsRepG2 + state; + IF_BIT_0(prob) + { + UPDATE_0(prob) + distance = rep2; + } + else + { + UPDATE_1(prob) + distance = rep3; + rep3 = rep2; + } + rep2 = rep1; + } + rep1 = rep0; + rep0 = distance; + } + state = state < kNumLitStates ? 8 : 11; + prob = probs + RepLenCoder; + } + + #ifdef Z7_LZMA_SIZE_OPT + { + unsigned lim, offset; + CLzmaProb *probLen = prob + LenChoice; + IF_BIT_0(probLen) + { + UPDATE_0(probLen) + probLen = prob + LenLow + GET_LEN_STATE; + offset = 0; + lim = (1 << kLenNumLowBits); + } + else + { + UPDATE_1(probLen) + probLen = prob + LenChoice2; + IF_BIT_0(probLen) + { + UPDATE_0(probLen) + probLen = prob + LenLow + GET_LEN_STATE + (1 << kLenNumLowBits); + offset = kLenNumLowSymbols; + lim = (1 << kLenNumLowBits); + } + else + { + UPDATE_1(probLen) + probLen = prob + LenHigh; + offset = kLenNumLowSymbols * 2; + lim = (1 << kLenNumHighBits); + } + } + TREE_DECODE(probLen, lim, len) + len += offset; + } + #else + { + CLzmaProb *probLen = prob + LenChoice; + IF_BIT_0(probLen) + { + UPDATE_0(probLen) + probLen = prob + LenLow + GET_LEN_STATE; + len = 1; + TREE_GET_BIT(probLen, len) + TREE_GET_BIT(probLen, len) + TREE_GET_BIT(probLen, len) + len -= 8; + } + else + { + UPDATE_1(probLen) + probLen = prob + LenChoice2; + IF_BIT_0(probLen) + { + UPDATE_0(probLen) + probLen = prob + LenLow + GET_LEN_STATE + (1 << kLenNumLowBits); + len = 1; + TREE_GET_BIT(probLen, len) + TREE_GET_BIT(probLen, len) + TREE_GET_BIT(probLen, len) + } + else + { + UPDATE_1(probLen) + probLen = prob + LenHigh; + TREE_DECODE(probLen, (1 << kLenNumHighBits), len) + len += kLenNumLowSymbols * 2; + } + } + } + #endif + + if (state >= kNumStates) + { + UInt32 distance; + prob = probs + PosSlot + + ((len < kNumLenToPosStates ? len : kNumLenToPosStates - 1) << kNumPosSlotBits); + TREE_6_DECODE(prob, distance) + if (distance >= kStartPosModelIndex) + { + unsigned posSlot = (unsigned)distance; + unsigned numDirectBits = (unsigned)(((distance >> 1) - 1)); + distance = (2 | (distance & 1)); + if (posSlot < kEndPosModelIndex) + { + distance <<= numDirectBits; + prob = probs + SpecPos; + { + UInt32 m = 1; + distance++; + do + { + REV_BIT_VAR(prob, distance, m) + } + while (--numDirectBits); + distance -= m; + } + } + else + { + numDirectBits -= kNumAlignBits; + do + { + NORMALIZE + range >>= 1; + + { + UInt32 t; + code -= range; + t = (0 - ((UInt32)code >> 31)); /* (UInt32)((Int32)code >> 31) */ + distance = (distance << 1) + (t + 1); + code += range & t; + } + /* + distance <<= 1; + if (code >= range) + { + code -= range; + distance |= 1; + } + */ + } + while (--numDirectBits); + prob = probs + Align; + distance <<= kNumAlignBits; + { + unsigned i = 1; + REV_BIT_CONST(prob, i, 1) + REV_BIT_CONST(prob, i, 2) + REV_BIT_CONST(prob, i, 4) + REV_BIT_LAST (prob, i, 8) + distance |= i; + } + if (distance == (UInt32)0xFFFFFFFF) + { + len = kMatchSpecLenStart; + state -= kNumStates; + break; + } + } + } + + rep3 = rep2; + rep2 = rep1; + rep1 = rep0; + rep0 = distance + 1; + state = (state < kNumStates + kNumLitStates) ? kNumLitStates : kNumLitStates + 3; + if (distance >= (checkDicSize == 0 ? processedPos: checkDicSize)) + { + len += kMatchSpecLen_Error_Data + kMatchMinLen; + // len = kMatchSpecLen_Error_Data; + // len += kMatchMinLen; + break; + } + } + + len += kMatchMinLen; + + { + SizeT rem; + unsigned curLen; + SizeT pos; + + if ((rem = limit - dicPos) == 0) + { + /* + We stop decoding and return SZ_OK, and we can resume decoding later. + Any error conditions can be tested later in caller code. + For more strict mode we can stop decoding with error + // len += kMatchSpecLen_Error_Data; + */ + break; + } + + curLen = ((rem < len) ? (unsigned)rem : len); + pos = dicPos - rep0 + (dicPos < rep0 ? dicBufSize : 0); + + processedPos += (UInt32)curLen; + + len -= curLen; + if (curLen <= dicBufSize - pos) + { + Byte *dest = dic + dicPos; + ptrdiff_t src = (ptrdiff_t)pos - (ptrdiff_t)dicPos; + const Byte *lim = dest + curLen; + dicPos += (SizeT)curLen; + do + *(dest) = (Byte)*(dest + src); + while (++dest != lim); + } + else + { + do + { + dic[dicPos++] = dic[pos]; + if (++pos == dicBufSize) + pos = 0; + } + while (--curLen != 0); + } + } + } + } + while (dicPos < limit && buf < bufLimit); + + NORMALIZE + + p->buf = buf; + p->range = range; + p->code = code; + p->remainLen = (UInt32)len; // & (kMatchSpecLen_Error_Data - 1); // we can write real length for error matches too. + p->dicPos = dicPos; + p->processedPos = processedPos; + p->reps[0] = rep0; + p->reps[1] = rep1; + p->reps[2] = rep2; + p->reps[3] = rep3; + p->state = (UInt32)state; + if (len >= kMatchSpecLen_Error_Data) + return SZ_ERROR_DATA; + return SZ_OK; +} +#endif + + + +static void Z7_FASTCALL LzmaDec_WriteRem(CLzmaDec *p, SizeT limit) +{ + unsigned len = (unsigned)p->remainLen; + if (len == 0 /* || len >= kMatchSpecLenStart */) + return; + { + SizeT dicPos = p->dicPos; + Byte *dic; + SizeT dicBufSize; + SizeT rep0; /* we use SizeT to avoid the BUG of VC14 for AMD64 */ + { + SizeT rem = limit - dicPos; + if (rem < len) + { + len = (unsigned)(rem); + if (len == 0) + return; + } + } + + if (p->checkDicSize == 0 && p->prop.dicSize - p->processedPos <= len) + p->checkDicSize = p->prop.dicSize; + + p->processedPos += (UInt32)len; + p->remainLen -= (UInt32)len; + dic = p->dic; + rep0 = p->reps[0]; + dicBufSize = p->dicBufSize; + do + { + dic[dicPos] = dic[dicPos - rep0 + (dicPos < rep0 ? dicBufSize : 0)]; + dicPos++; + } + while (--len); + p->dicPos = dicPos; + } +} + + +/* +At staring of new stream we have one of the following symbols: + - Literal - is allowed + - Non-Rep-Match - is allowed only if it's end marker symbol + - Rep-Match - is not allowed +We use early check of (RangeCoder:Code) over kBadRepCode to simplify main decoding code +*/ + +#define kRange0 0xFFFFFFFF +#define kBound0 ((kRange0 >> kNumBitModelTotalBits) << (kNumBitModelTotalBits - 1)) +#define kBadRepCode (kBound0 + (((kRange0 - kBound0) >> kNumBitModelTotalBits) << (kNumBitModelTotalBits - 1))) +#if kBadRepCode != (0xC0000000 - 0x400) + #error Stop_Compiling_Bad_LZMA_Check +#endif + + +/* +LzmaDec_DecodeReal2(): + It calls LZMA_DECODE_REAL() and it adjusts limit according (p->checkDicSize). + +We correct (p->checkDicSize) after LZMA_DECODE_REAL() and in LzmaDec_WriteRem(), +and we support the following state of (p->checkDicSize): + if (total_processed < p->prop.dicSize) then + { + (total_processed == p->processedPos) + (p->checkDicSize == 0) + } + else + (p->checkDicSize == p->prop.dicSize) +*/ + +static int Z7_FASTCALL LzmaDec_DecodeReal2(CLzmaDec *p, SizeT limit, const Byte *bufLimit) +{ + if (p->checkDicSize == 0) + { + UInt32 rem = p->prop.dicSize - p->processedPos; + if (limit - p->dicPos > rem) + limit = p->dicPos + rem; + } + { + int res = LZMA_DECODE_REAL(p, limit, bufLimit); + if (p->checkDicSize == 0 && p->processedPos >= p->prop.dicSize) + p->checkDicSize = p->prop.dicSize; + return res; + } +} + + + +typedef enum +{ + DUMMY_INPUT_EOF, /* need more input data */ + DUMMY_LIT, + DUMMY_MATCH, + DUMMY_REP +} ELzmaDummy; + + +#define IS_DUMMY_END_MARKER_POSSIBLE(dummyRes) ((dummyRes) == DUMMY_MATCH) + +static ELzmaDummy LzmaDec_TryDummy(const CLzmaDec *p, const Byte *buf, const Byte **bufOut) +{ + UInt32 range = p->range; + UInt32 code = p->code; + const Byte *bufLimit = *bufOut; + const CLzmaProb *probs = GET_PROBS; + unsigned state = (unsigned)p->state; + ELzmaDummy res; + + for (;;) + { + const CLzmaProb *prob; + UInt32 bound; + unsigned ttt; + unsigned posState = CALC_POS_STATE(p->processedPos, ((unsigned)1 << p->prop.pb) - 1); + + prob = probs + IsMatch + COMBINED_PS_STATE; + IF_BIT_0_CHECK(prob) + { + UPDATE_0_CHECK + + prob = probs + Literal; + if (p->checkDicSize != 0 || p->processedPos != 0) + prob += ((UInt32)LZMA_LIT_SIZE * + ((((p->processedPos) & (((unsigned)1 << (p->prop.lp)) - 1)) << p->prop.lc) + + ((unsigned)p->dic[(p->dicPos == 0 ? p->dicBufSize : p->dicPos) - 1] >> (8 - p->prop.lc)))); + + if (state < kNumLitStates) + { + unsigned symbol = 1; + do { GET_BIT_CHECK(prob + symbol, symbol) } while (symbol < 0x100); + } + else + { + unsigned matchByte = p->dic[p->dicPos - p->reps[0] + + (p->dicPos < p->reps[0] ? p->dicBufSize : 0)]; + unsigned offs = 0x100; + unsigned symbol = 1; + do + { + unsigned bit; + const CLzmaProb *probLit; + matchByte += matchByte; + bit = offs; + offs &= matchByte; + probLit = prob + (offs + bit + symbol); + GET_BIT2_CHECK(probLit, symbol, offs ^= bit; , ; ) + } + while (symbol < 0x100); + } + res = DUMMY_LIT; + } + else + { + unsigned len; + UPDATE_1_CHECK + + prob = probs + IsRep + state; + IF_BIT_0_CHECK(prob) + { + UPDATE_0_CHECK + state = 0; + prob = probs + LenCoder; + res = DUMMY_MATCH; + } + else + { + UPDATE_1_CHECK + res = DUMMY_REP; + prob = probs + IsRepG0 + state; + IF_BIT_0_CHECK(prob) + { + UPDATE_0_CHECK + prob = probs + IsRep0Long + COMBINED_PS_STATE; + IF_BIT_0_CHECK(prob) + { + UPDATE_0_CHECK + break; + } + else + { + UPDATE_1_CHECK + } + } + else + { + UPDATE_1_CHECK + prob = probs + IsRepG1 + state; + IF_BIT_0_CHECK(prob) + { + UPDATE_0_CHECK + } + else + { + UPDATE_1_CHECK + prob = probs + IsRepG2 + state; + IF_BIT_0_CHECK(prob) + { + UPDATE_0_CHECK + } + else + { + UPDATE_1_CHECK + } + } + } + state = kNumStates; + prob = probs + RepLenCoder; + } + { + unsigned limit, offset; + const CLzmaProb *probLen = prob + LenChoice; + IF_BIT_0_CHECK(probLen) + { + UPDATE_0_CHECK + probLen = prob + LenLow + GET_LEN_STATE; + offset = 0; + limit = 1 << kLenNumLowBits; + } + else + { + UPDATE_1_CHECK + probLen = prob + LenChoice2; + IF_BIT_0_CHECK(probLen) + { + UPDATE_0_CHECK + probLen = prob + LenLow + GET_LEN_STATE + (1 << kLenNumLowBits); + offset = kLenNumLowSymbols; + limit = 1 << kLenNumLowBits; + } + else + { + UPDATE_1_CHECK + probLen = prob + LenHigh; + offset = kLenNumLowSymbols * 2; + limit = 1 << kLenNumHighBits; + } + } + TREE_DECODE_CHECK(probLen, limit, len) + len += offset; + } + + if (state < 4) + { + unsigned posSlot; + prob = probs + PosSlot + + ((len < kNumLenToPosStates - 1 ? len : kNumLenToPosStates - 1) << + kNumPosSlotBits); + TREE_DECODE_CHECK(prob, 1 << kNumPosSlotBits, posSlot) + if (posSlot >= kStartPosModelIndex) + { + unsigned numDirectBits = ((posSlot >> 1) - 1); + + if (posSlot < kEndPosModelIndex) + { + prob = probs + SpecPos + ((2 | (posSlot & 1)) << numDirectBits); + } + else + { + numDirectBits -= kNumAlignBits; + do + { + NORMALIZE_CHECK + range >>= 1; + code -= range & (((code - range) >> 31) - 1); + /* if (code >= range) code -= range; */ + } + while (--numDirectBits); + prob = probs + Align; + numDirectBits = kNumAlignBits; + } + { + unsigned i = 1; + unsigned m = 1; + do + { + REV_BIT_CHECK(prob, i, m) + } + while (--numDirectBits); + } + } + } + } + break; + } + NORMALIZE_CHECK + + *bufOut = buf; + return res; +} + +void LzmaDec_InitDicAndState(CLzmaDec *p, BoolInt initDic, BoolInt initState); +void LzmaDec_InitDicAndState(CLzmaDec *p, BoolInt initDic, BoolInt initState) +{ + p->remainLen = kMatchSpecLenStart + 1; + p->tempBufSize = 0; + + if (initDic) + { + p->processedPos = 0; + p->checkDicSize = 0; + p->remainLen = kMatchSpecLenStart + 2; + } + if (initState) + p->remainLen = kMatchSpecLenStart + 2; +} + +void LzmaDec_Init(CLzmaDec *p) +{ + p->dicPos = 0; + LzmaDec_InitDicAndState(p, True, True); +} + + +/* +LZMA supports optional end_marker. +So the decoder can lookahead for one additional LZMA-Symbol to check end_marker. +That additional LZMA-Symbol can require up to LZMA_REQUIRED_INPUT_MAX bytes in input stream. +When the decoder reaches dicLimit, it looks (finishMode) parameter: + if (finishMode == LZMA_FINISH_ANY), the decoder doesn't lookahead + if (finishMode != LZMA_FINISH_ANY), the decoder lookahead, if end_marker is possible for current position + +When the decoder lookahead, and the lookahead symbol is not end_marker, we have two ways: + 1) Strict mode (default) : the decoder returns SZ_ERROR_DATA. + 2) The relaxed mode (alternative mode) : we could return SZ_OK, and the caller + must check (status) value. The caller can show the error, + if the end of stream is expected, and the (status) is noit + LZMA_STATUS_FINISHED_WITH_MARK or LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK. +*/ + + +#define RETURN_NOT_FINISHED_FOR_FINISH \ + *status = LZMA_STATUS_NOT_FINISHED; \ + return SZ_ERROR_DATA; // for strict mode + // return SZ_OK; // for relaxed mode + + +SRes LzmaDec_DecodeToDic(CLzmaDec *p, SizeT dicLimit, const Byte *src, SizeT *srcLen, + ELzmaFinishMode finishMode, ELzmaStatus *status) +{ + SizeT inSize = *srcLen; + (*srcLen) = 0; + *status = LZMA_STATUS_NOT_SPECIFIED; + + if (p->remainLen > kMatchSpecLenStart) + { + if (p->remainLen > kMatchSpecLenStart + 2) + return p->remainLen == kMatchSpecLen_Error_Fail ? SZ_ERROR_FAIL : SZ_ERROR_DATA; + + for (; inSize > 0 && p->tempBufSize < RC_INIT_SIZE; (*srcLen)++, inSize--) + p->tempBuf[p->tempBufSize++] = *src++; + if (p->tempBufSize != 0 && p->tempBuf[0] != 0) + return SZ_ERROR_DATA; + if (p->tempBufSize < RC_INIT_SIZE) + { + *status = LZMA_STATUS_NEEDS_MORE_INPUT; + return SZ_OK; + } + p->code = + ((UInt32)p->tempBuf[1] << 24) + | ((UInt32)p->tempBuf[2] << 16) + | ((UInt32)p->tempBuf[3] << 8) + | ((UInt32)p->tempBuf[4]); + + if (p->checkDicSize == 0 + && p->processedPos == 0 + && p->code >= kBadRepCode) + return SZ_ERROR_DATA; + + p->range = 0xFFFFFFFF; + p->tempBufSize = 0; + + if (p->remainLen > kMatchSpecLenStart + 1) + { + SizeT numProbs = LzmaProps_GetNumProbs(&p->prop); + SizeT i; + CLzmaProb *probs = p->probs; + for (i = 0; i < numProbs; i++) + probs[i] = kBitModelTotal >> 1; + p->reps[0] = p->reps[1] = p->reps[2] = p->reps[3] = 1; + p->state = 0; + } + + p->remainLen = 0; + } + + for (;;) + { + if (p->remainLen == kMatchSpecLenStart) + { + if (p->code != 0) + return SZ_ERROR_DATA; + *status = LZMA_STATUS_FINISHED_WITH_MARK; + return SZ_OK; + } + + LzmaDec_WriteRem(p, dicLimit); + + { + // (p->remainLen == 0 || p->dicPos == dicLimit) + + int checkEndMarkNow = 0; + + if (p->dicPos >= dicLimit) + { + if (p->remainLen == 0 && p->code == 0) + { + *status = LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK; + return SZ_OK; + } + if (finishMode == LZMA_FINISH_ANY) + { + *status = LZMA_STATUS_NOT_FINISHED; + return SZ_OK; + } + if (p->remainLen != 0) + { + RETURN_NOT_FINISHED_FOR_FINISH + } + checkEndMarkNow = 1; + } + + // (p->remainLen == 0) + + if (p->tempBufSize == 0) + { + const Byte *bufLimit; + int dummyProcessed = -1; + + if (inSize < LZMA_REQUIRED_INPUT_MAX || checkEndMarkNow) + { + const Byte *bufOut = src + inSize; + + ELzmaDummy dummyRes = LzmaDec_TryDummy(p, src, &bufOut); + + if (dummyRes == DUMMY_INPUT_EOF) + { + size_t i; + if (inSize >= LZMA_REQUIRED_INPUT_MAX) + break; + (*srcLen) += inSize; + p->tempBufSize = (unsigned)inSize; + for (i = 0; i < inSize; i++) + p->tempBuf[i] = src[i]; + *status = LZMA_STATUS_NEEDS_MORE_INPUT; + return SZ_OK; + } + + dummyProcessed = (int)(bufOut - src); + if ((unsigned)dummyProcessed > LZMA_REQUIRED_INPUT_MAX) + break; + + if (checkEndMarkNow && !IS_DUMMY_END_MARKER_POSSIBLE(dummyRes)) + { + unsigned i; + (*srcLen) += (unsigned)dummyProcessed; + p->tempBufSize = (unsigned)dummyProcessed; + for (i = 0; i < (unsigned)dummyProcessed; i++) + p->tempBuf[i] = src[i]; + // p->remainLen = kMatchSpecLen_Error_Data; + RETURN_NOT_FINISHED_FOR_FINISH + } + + bufLimit = src; + // we will decode only one iteration + } + else + bufLimit = src + inSize - LZMA_REQUIRED_INPUT_MAX; + + p->buf = src; + + { + int res = LzmaDec_DecodeReal2(p, dicLimit, bufLimit); + + SizeT processed = (SizeT)(p->buf - src); + + if (dummyProcessed < 0) + { + if (processed > inSize) + break; + } + else if ((unsigned)dummyProcessed != processed) + break; + + src += processed; + inSize -= processed; + (*srcLen) += processed; + + if (res != SZ_OK) + { + p->remainLen = kMatchSpecLen_Error_Data; + return SZ_ERROR_DATA; + } + } + continue; + } + + { + // we have some data in (p->tempBuf) + // in strict mode: tempBufSize is not enough for one Symbol decoding. + // in relaxed mode: tempBufSize not larger than required for one Symbol decoding. + + unsigned rem = p->tempBufSize; + unsigned ahead = 0; + int dummyProcessed = -1; + + while (rem < LZMA_REQUIRED_INPUT_MAX && ahead < inSize) + p->tempBuf[rem++] = src[ahead++]; + + // ahead - the size of new data copied from (src) to (p->tempBuf) + // rem - the size of temp buffer including new data from (src) + + if (rem < LZMA_REQUIRED_INPUT_MAX || checkEndMarkNow) + { + const Byte *bufOut = p->tempBuf + rem; + + ELzmaDummy dummyRes = LzmaDec_TryDummy(p, p->tempBuf, &bufOut); + + if (dummyRes == DUMMY_INPUT_EOF) + { + if (rem >= LZMA_REQUIRED_INPUT_MAX) + break; + p->tempBufSize = rem; + (*srcLen) += (SizeT)ahead; + *status = LZMA_STATUS_NEEDS_MORE_INPUT; + return SZ_OK; + } + + dummyProcessed = (int)(bufOut - p->tempBuf); + + if ((unsigned)dummyProcessed < p->tempBufSize) + break; + + if (checkEndMarkNow && !IS_DUMMY_END_MARKER_POSSIBLE(dummyRes)) + { + (*srcLen) += (unsigned)dummyProcessed - p->tempBufSize; + p->tempBufSize = (unsigned)dummyProcessed; + // p->remainLen = kMatchSpecLen_Error_Data; + RETURN_NOT_FINISHED_FOR_FINISH + } + } + + p->buf = p->tempBuf; + + { + // we decode one symbol from (p->tempBuf) here, so the (bufLimit) is equal to (p->buf) + int res = LzmaDec_DecodeReal2(p, dicLimit, p->buf); + + SizeT processed = (SizeT)(p->buf - p->tempBuf); + rem = p->tempBufSize; + + if (dummyProcessed < 0) + { + if (processed > LZMA_REQUIRED_INPUT_MAX) + break; + if (processed < rem) + break; + } + else if ((unsigned)dummyProcessed != processed) + break; + + processed -= rem; + + src += processed; + inSize -= processed; + (*srcLen) += processed; + p->tempBufSize = 0; + + if (res != SZ_OK) + { + p->remainLen = kMatchSpecLen_Error_Data; + return SZ_ERROR_DATA; + } + } + } + } + } + + /* Some unexpected error: internal error of code, memory corruption or hardware failure */ + p->remainLen = kMatchSpecLen_Error_Fail; + return SZ_ERROR_FAIL; +} + + + +SRes LzmaDec_DecodeToBuf(CLzmaDec *p, Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status) +{ + SizeT outSize = *destLen; + SizeT inSize = *srcLen; + *srcLen = *destLen = 0; + for (;;) + { + SizeT inSizeCur = inSize, outSizeCur, dicPos; + ELzmaFinishMode curFinishMode; + SRes res; + if (p->dicPos == p->dicBufSize) + p->dicPos = 0; + dicPos = p->dicPos; + if (outSize > p->dicBufSize - dicPos) + { + outSizeCur = p->dicBufSize; + curFinishMode = LZMA_FINISH_ANY; + } + else + { + outSizeCur = dicPos + outSize; + curFinishMode = finishMode; + } + + res = LzmaDec_DecodeToDic(p, outSizeCur, src, &inSizeCur, curFinishMode, status); + src += inSizeCur; + inSize -= inSizeCur; + *srcLen += inSizeCur; + outSizeCur = p->dicPos - dicPos; + memcpy(dest, p->dic + dicPos, outSizeCur); + dest += outSizeCur; + outSize -= outSizeCur; + *destLen += outSizeCur; + if (res != 0) + return res; + if (outSizeCur == 0 || outSize == 0) + return SZ_OK; + } +} + +void LzmaDec_FreeProbs(CLzmaDec *p, ISzAllocPtr alloc) +{ + ISzAlloc_Free(alloc, p->probs); + p->probs = NULL; +} + +static void LzmaDec_FreeDict(CLzmaDec *p, ISzAllocPtr alloc) +{ + ISzAlloc_Free(alloc, p->dic); + p->dic = NULL; +} + +void LzmaDec_Free(CLzmaDec *p, ISzAllocPtr alloc) +{ + LzmaDec_FreeProbs(p, alloc); + LzmaDec_FreeDict(p, alloc); +} + +SRes LzmaProps_Decode(CLzmaProps *p, const Byte *data, unsigned size) +{ + UInt32 dicSize; + Byte d; + + if (size < LZMA_PROPS_SIZE) + return SZ_ERROR_UNSUPPORTED; + else + dicSize = data[1] | ((UInt32)data[2] << 8) | ((UInt32)data[3] << 16) | ((UInt32)data[4] << 24); + + if (dicSize < LZMA_DIC_MIN) + dicSize = LZMA_DIC_MIN; + p->dicSize = dicSize; + + d = data[0]; + if (d >= (9 * 5 * 5)) + return SZ_ERROR_UNSUPPORTED; + + p->lc = (Byte)(d % 9); + d /= 9; + p->pb = (Byte)(d / 5); + p->lp = (Byte)(d % 5); + + return SZ_OK; +} + +static SRes LzmaDec_AllocateProbs2(CLzmaDec *p, const CLzmaProps *propNew, ISzAllocPtr alloc) +{ + UInt32 numProbs = LzmaProps_GetNumProbs(propNew); + if (!p->probs || numProbs != p->numProbs) + { + LzmaDec_FreeProbs(p, alloc); + p->probs = (CLzmaProb *)ISzAlloc_Alloc(alloc, numProbs * sizeof(CLzmaProb)); + if (!p->probs) + return SZ_ERROR_MEM; + p->probs_1664 = p->probs + 1664; + p->numProbs = numProbs; + } + return SZ_OK; +} + +SRes LzmaDec_AllocateProbs(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAllocPtr alloc) +{ + CLzmaProps propNew; + RINOK(LzmaProps_Decode(&propNew, props, propsSize)) + RINOK(LzmaDec_AllocateProbs2(p, &propNew, alloc)) + p->prop = propNew; + return SZ_OK; +} + +SRes LzmaDec_Allocate(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAllocPtr alloc) +{ + CLzmaProps propNew; + SizeT dicBufSize; + RINOK(LzmaProps_Decode(&propNew, props, propsSize)) + RINOK(LzmaDec_AllocateProbs2(p, &propNew, alloc)) + + { + UInt32 dictSize = propNew.dicSize; + SizeT mask = ((UInt32)1 << 12) - 1; + if (dictSize >= ((UInt32)1 << 30)) mask = ((UInt32)1 << 22) - 1; + else if (dictSize >= ((UInt32)1 << 22)) mask = ((UInt32)1 << 20) - 1; + dicBufSize = ((SizeT)dictSize + mask) & ~mask; + if (dicBufSize < dictSize) + dicBufSize = dictSize; + } + + if (!p->dic || dicBufSize != p->dicBufSize) + { + LzmaDec_FreeDict(p, alloc); + p->dic = (Byte *)ISzAlloc_Alloc(alloc, dicBufSize); + if (!p->dic) + { + LzmaDec_FreeProbs(p, alloc); + return SZ_ERROR_MEM; + } + } + p->dicBufSize = dicBufSize; + p->prop = propNew; + return SZ_OK; +} + +SRes LzmaDecode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen, + const Byte *propData, unsigned propSize, ELzmaFinishMode finishMode, + ELzmaStatus *status, ISzAllocPtr alloc) +{ + CLzmaDec p; + SRes res; + SizeT outSize = *destLen, inSize = *srcLen; + *destLen = *srcLen = 0; + *status = LZMA_STATUS_NOT_SPECIFIED; + if (inSize < RC_INIT_SIZE) + return SZ_ERROR_INPUT_EOF; + LzmaDec_CONSTRUCT(&p) + RINOK(LzmaDec_AllocateProbs(&p, propData, propSize, alloc)) + p.dic = dest; + p.dicBufSize = outSize; + LzmaDec_Init(&p); + *srcLen = inSize; + res = LzmaDec_DecodeToDic(&p, outSize, src, srcLen, finishMode, status); + *destLen = p.dicPos; + if (res == SZ_OK && *status == LZMA_STATUS_NEEDS_MORE_INPUT) + res = SZ_ERROR_INPUT_EOF; + LzmaDec_FreeProbs(&p, alloc); + return res; +} diff --git a/project/src/liblzma/LzmaDec.h b/project/src/liblzma/LzmaDec.h new file mode 100644 index 0000000..6432f60 --- /dev/null +++ b/project/src/liblzma/LzmaDec.h @@ -0,0 +1,237 @@ +/* LzmaDec.h -- LZMA Decoder +2023-04-02 : Igor Pavlov : Public domain */ + +#ifndef ZIP7_INC_LZMA_DEC_H +#define ZIP7_INC_LZMA_DEC_H + +#include "7zTypes.h" + +EXTERN_C_BEGIN + +/* #define Z7_LZMA_PROB32 */ +/* Z7_LZMA_PROB32 can increase the speed on some CPUs, + but memory usage for CLzmaDec::probs will be doubled in that case */ + +typedef +#ifdef Z7_LZMA_PROB32 + UInt32 +#else + UInt16 +#endif + CLzmaProb; + + +/* ---------- LZMA Properties ---------- */ + +#define LZMA_PROPS_SIZE 5 + +typedef struct +{ + Byte lc; + Byte lp; + Byte pb; + Byte _pad_; + UInt32 dicSize; +} CLzmaProps; + +/* LzmaProps_Decode - decodes properties +Returns: + SZ_OK + SZ_ERROR_UNSUPPORTED - Unsupported properties +*/ + +SRes LzmaProps_Decode(CLzmaProps *p, const Byte *data, unsigned size); + + +/* ---------- LZMA Decoder state ---------- */ + +/* LZMA_REQUIRED_INPUT_MAX = number of required input bytes for worst case. + Num bits = log2((2^11 / 31) ^ 22) + 26 < 134 + 26 = 160; */ + +#define LZMA_REQUIRED_INPUT_MAX 20 + +typedef struct +{ + /* Don't change this structure. ASM code can use it. */ + CLzmaProps prop; + CLzmaProb *probs; + CLzmaProb *probs_1664; + Byte *dic; + SizeT dicBufSize; + SizeT dicPos; + const Byte *buf; + UInt32 range; + UInt32 code; + UInt32 processedPos; + UInt32 checkDicSize; + UInt32 reps[4]; + UInt32 state; + UInt32 remainLen; + + UInt32 numProbs; + unsigned tempBufSize; + Byte tempBuf[LZMA_REQUIRED_INPUT_MAX]; +} CLzmaDec; + +#define LzmaDec_CONSTRUCT(p) { (p)->dic = NULL; (p)->probs = NULL; } +#define LzmaDec_Construct(p) LzmaDec_CONSTRUCT(p) + +void LzmaDec_Init(CLzmaDec *p); + +/* There are two types of LZMA streams: + - Stream with end mark. That end mark adds about 6 bytes to compressed size. + - Stream without end mark. You must know exact uncompressed size to decompress such stream. */ + +typedef enum +{ + LZMA_FINISH_ANY, /* finish at any point */ + LZMA_FINISH_END /* block must be finished at the end */ +} ELzmaFinishMode; + +/* ELzmaFinishMode has meaning only if the decoding reaches output limit !!! + + You must use LZMA_FINISH_END, when you know that current output buffer + covers last bytes of block. In other cases you must use LZMA_FINISH_ANY. + + If LZMA decoder sees end marker before reaching output limit, it returns SZ_OK, + and output value of destLen will be less than output buffer size limit. + You can check status result also. + + You can use multiple checks to test data integrity after full decompression: + 1) Check Result and "status" variable. + 2) Check that output(destLen) = uncompressedSize, if you know real uncompressedSize. + 3) Check that output(srcLen) = compressedSize, if you know real compressedSize. + You must use correct finish mode in that case. */ + +typedef enum +{ + LZMA_STATUS_NOT_SPECIFIED, /* use main error code instead */ + LZMA_STATUS_FINISHED_WITH_MARK, /* stream was finished with end mark. */ + LZMA_STATUS_NOT_FINISHED, /* stream was not finished */ + LZMA_STATUS_NEEDS_MORE_INPUT, /* you must provide more input bytes */ + LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK /* there is probability that stream was finished without end mark */ +} ELzmaStatus; + +/* ELzmaStatus is used only as output value for function call */ + + +/* ---------- Interfaces ---------- */ + +/* There are 3 levels of interfaces: + 1) Dictionary Interface + 2) Buffer Interface + 3) One Call Interface + You can select any of these interfaces, but don't mix functions from different + groups for same object. */ + + +/* There are two variants to allocate state for Dictionary Interface: + 1) LzmaDec_Allocate / LzmaDec_Free + 2) LzmaDec_AllocateProbs / LzmaDec_FreeProbs + You can use variant 2, if you set dictionary buffer manually. + For Buffer Interface you must always use variant 1. + +LzmaDec_Allocate* can return: + SZ_OK + SZ_ERROR_MEM - Memory allocation error + SZ_ERROR_UNSUPPORTED - Unsupported properties +*/ + +SRes LzmaDec_AllocateProbs(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAllocPtr alloc); +void LzmaDec_FreeProbs(CLzmaDec *p, ISzAllocPtr alloc); + +SRes LzmaDec_Allocate(CLzmaDec *p, const Byte *props, unsigned propsSize, ISzAllocPtr alloc); +void LzmaDec_Free(CLzmaDec *p, ISzAllocPtr alloc); + +/* ---------- Dictionary Interface ---------- */ + +/* You can use it, if you want to eliminate the overhead for data copying from + dictionary to some other external buffer. + You must work with CLzmaDec variables directly in this interface. + + STEPS: + LzmaDec_Construct() + LzmaDec_Allocate() + for (each new stream) + { + LzmaDec_Init() + while (it needs more decompression) + { + LzmaDec_DecodeToDic() + use data from CLzmaDec::dic and update CLzmaDec::dicPos + } + } + LzmaDec_Free() +*/ + +/* LzmaDec_DecodeToDic + + The decoding to internal dictionary buffer (CLzmaDec::dic). + You must manually update CLzmaDec::dicPos, if it reaches CLzmaDec::dicBufSize !!! + +finishMode: + It has meaning only if the decoding reaches output limit (dicLimit). + LZMA_FINISH_ANY - Decode just dicLimit bytes. + LZMA_FINISH_END - Stream must be finished after dicLimit. + +Returns: + SZ_OK + status: + LZMA_STATUS_FINISHED_WITH_MARK + LZMA_STATUS_NOT_FINISHED + LZMA_STATUS_NEEDS_MORE_INPUT + LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK + SZ_ERROR_DATA - Data error + SZ_ERROR_FAIL - Some unexpected error: internal error of code, memory corruption or hardware failure +*/ + +SRes LzmaDec_DecodeToDic(CLzmaDec *p, SizeT dicLimit, + const Byte *src, SizeT *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status); + + +/* ---------- Buffer Interface ---------- */ + +/* It's zlib-like interface. + See LzmaDec_DecodeToDic description for information about STEPS and return results, + but you must use LzmaDec_DecodeToBuf instead of LzmaDec_DecodeToDic and you don't need + to work with CLzmaDec variables manually. + +finishMode: + It has meaning only if the decoding reaches output limit (*destLen). + LZMA_FINISH_ANY - Decode just destLen bytes. + LZMA_FINISH_END - Stream must be finished after (*destLen). +*/ + +SRes LzmaDec_DecodeToBuf(CLzmaDec *p, Byte *dest, SizeT *destLen, + const Byte *src, SizeT *srcLen, ELzmaFinishMode finishMode, ELzmaStatus *status); + + +/* ---------- One Call Interface ---------- */ + +/* LzmaDecode + +finishMode: + It has meaning only if the decoding reaches output limit (*destLen). + LZMA_FINISH_ANY - Decode just destLen bytes. + LZMA_FINISH_END - Stream must be finished after (*destLen). + +Returns: + SZ_OK + status: + LZMA_STATUS_FINISHED_WITH_MARK + LZMA_STATUS_NOT_FINISHED + LZMA_STATUS_MAYBE_FINISHED_WITHOUT_MARK + SZ_ERROR_DATA - Data error + SZ_ERROR_MEM - Memory allocation error + SZ_ERROR_UNSUPPORTED - Unsupported properties + SZ_ERROR_INPUT_EOF - It needs more bytes in input buffer (src). + SZ_ERROR_FAIL - Some unexpected error: internal error of code, memory corruption or hardware failure +*/ + +SRes LzmaDecode(Byte *dest, SizeT *destLen, const Byte *src, SizeT *srcLen, + const Byte *propData, unsigned propSize, ELzmaFinishMode finishMode, + ELzmaStatus *status, ISzAllocPtr alloc); + +EXTERN_C_END + +#endif diff --git a/project/src/liblzma/LzmaEnc.c b/project/src/liblzma/LzmaEnc.c new file mode 100644 index 0000000..64bb1aa --- /dev/null +++ b/project/src/liblzma/LzmaEnc.c @@ -0,0 +1,3144 @@ +/* LzmaEnc.c -- LZMA Encoder +2023-04-13: Igor Pavlov : Public domain */ + +#include "Precomp.h" + +#include + +/* #define SHOW_STAT */ +/* #define SHOW_STAT2 */ + +#if defined(SHOW_STAT) || defined(SHOW_STAT2) +#include +#endif + +#include "CpuArch.h" +#include "LzmaEnc.h" + +#include "LzFind.h" +#ifndef Z7_ST +#include "LzFindMt.h" +#endif + +/* the following LzmaEnc_* declarations is internal LZMA interface for LZMA2 encoder */ + +SRes LzmaEnc_PrepareForLzma2(CLzmaEncHandle p, ISeqInStreamPtr inStream, UInt32 keepWindowSize, + ISzAllocPtr alloc, ISzAllocPtr allocBig); +SRes LzmaEnc_MemPrepare(CLzmaEncHandle p, const Byte *src, SizeT srcLen, + UInt32 keepWindowSize, ISzAllocPtr alloc, ISzAllocPtr allocBig); +SRes LzmaEnc_CodeOneMemBlock(CLzmaEncHandle p, BoolInt reInit, + Byte *dest, size_t *destLen, UInt32 desiredPackSize, UInt32 *unpackSize); +const Byte *LzmaEnc_GetCurBuf(CLzmaEncHandle p); +void LzmaEnc_Finish(CLzmaEncHandle p); +void LzmaEnc_SaveState(CLzmaEncHandle p); +void LzmaEnc_RestoreState(CLzmaEncHandle p); + +#ifdef SHOW_STAT +static unsigned g_STAT_OFFSET = 0; +#endif + +/* for good normalization speed we still reserve 256 MB before 4 GB range */ +#define kLzmaMaxHistorySize ((UInt32)15 << 28) + +// #define kNumTopBits 24 +#define kTopValue ((UInt32)1 << 24) + +#define kNumBitModelTotalBits 11 +#define kBitModelTotal (1 << kNumBitModelTotalBits) +#define kNumMoveBits 5 +#define kProbInitValue (kBitModelTotal >> 1) + +#define kNumMoveReducingBits 4 +#define kNumBitPriceShiftBits 4 +// #define kBitPrice (1 << kNumBitPriceShiftBits) + +#define REP_LEN_COUNT 64 + +void LzmaEncProps_Init(CLzmaEncProps *p) +{ + p->level = 5; + p->dictSize = p->mc = 0; + p->reduceSize = (UInt64)(Int64)-1; + p->lc = p->lp = p->pb = p->algo = p->fb = p->btMode = p->numHashBytes = p->numThreads = -1; + p->numHashOutBits = 0; + p->writeEndMark = 0; + p->affinity = 0; +} + +void LzmaEncProps_Normalize(CLzmaEncProps *p) +{ + int level = p->level; + if (level < 0) level = 5; + p->level = level; + + if (p->dictSize == 0) + p->dictSize = + ( level <= 3 ? ((UInt32)1 << (level * 2 + 16)) : + ( level <= 6 ? ((UInt32)1 << (level + 19)) : + ( level <= 7 ? ((UInt32)1 << 25) : ((UInt32)1 << 26) + ))); + + if (p->dictSize > p->reduceSize) + { + UInt32 v = (UInt32)p->reduceSize; + const UInt32 kReduceMin = ((UInt32)1 << 12); + if (v < kReduceMin) + v = kReduceMin; + if (p->dictSize > v) + p->dictSize = v; + } + + if (p->lc < 0) p->lc = 3; + if (p->lp < 0) p->lp = 0; + if (p->pb < 0) p->pb = 2; + + if (p->algo < 0) p->algo = (level < 5 ? 0 : 1); + if (p->fb < 0) p->fb = (level < 7 ? 32 : 64); + if (p->btMode < 0) p->btMode = (p->algo == 0 ? 0 : 1); + if (p->numHashBytes < 0) p->numHashBytes = (p->btMode ? 4 : 5); + if (p->mc == 0) p->mc = (16 + ((unsigned)p->fb >> 1)) >> (p->btMode ? 0 : 1); + + if (p->numThreads < 0) + p->numThreads = + #ifndef Z7_ST + ((p->btMode && p->algo) ? 2 : 1); + #else + 1; + #endif +} + +UInt32 LzmaEncProps_GetDictSize(const CLzmaEncProps *props2) +{ + CLzmaEncProps props = *props2; + LzmaEncProps_Normalize(&props); + return props.dictSize; +} + + +/* +x86/x64: + +BSR: + IF (SRC == 0) ZF = 1, DEST is undefined; + AMD : DEST is unchanged; + IF (SRC != 0) ZF = 0; DEST is index of top non-zero bit + BSR is slow in some processors + +LZCNT: + IF (SRC == 0) CF = 1, DEST is size_in_bits_of_register(src) (32 or 64) + IF (SRC != 0) CF = 0, DEST = num_lead_zero_bits + IF (DEST == 0) ZF = 1; + +LZCNT works only in new processors starting from Haswell. +if LZCNT is not supported by processor, then it's executed as BSR. +LZCNT can be faster than BSR, if supported. +*/ + +// #define LZMA_LOG_BSR + +#if defined(MY_CPU_ARM_OR_ARM64) /* || defined(MY_CPU_X86_OR_AMD64) */ + + #if (defined(__clang__) && (__clang_major__ >= 6)) \ + || (defined(__GNUC__) && (__GNUC__ >= 6)) + #define LZMA_LOG_BSR + #elif defined(_MSC_VER) && (_MSC_VER >= 1300) + // #if defined(MY_CPU_ARM_OR_ARM64) + #define LZMA_LOG_BSR + // #endif + #endif +#endif + +// #include + +#ifdef LZMA_LOG_BSR + +#if defined(__clang__) \ + || defined(__GNUC__) + +/* + C code: : (30 - __builtin_clz(x)) + gcc9/gcc10 for x64 /x86 : 30 - (bsr(x) xor 31) + clang10 for x64 : 31 + (bsr(x) xor -32) +*/ + + #define MY_clz(x) ((unsigned)__builtin_clz(x)) + // __lzcnt32 + // __builtin_ia32_lzcnt_u32 + +#else // #if defined(_MSC_VER) + + #ifdef MY_CPU_ARM_OR_ARM64 + + #define MY_clz _CountLeadingZeros + + #else // if defined(MY_CPU_X86_OR_AMD64) + + // #define MY_clz __lzcnt // we can use lzcnt (unsupported by old CPU) + // _BitScanReverse code is not optimal for some MSVC compilers + #define BSR2_RET(pos, res) { unsigned long zz; _BitScanReverse(&zz, (pos)); zz--; \ + res = (zz + zz) + (pos >> zz); } + + #endif // MY_CPU_X86_OR_AMD64 + +#endif // _MSC_VER + + +#ifndef BSR2_RET + + #define BSR2_RET(pos, res) { unsigned zz = 30 - MY_clz(pos); \ + res = (zz + zz) + (pos >> zz); } + +#endif + + +unsigned GetPosSlot1(UInt32 pos); +unsigned GetPosSlot1(UInt32 pos) +{ + unsigned res; + BSR2_RET(pos, res); + return res; +} +#define GetPosSlot2(pos, res) { BSR2_RET(pos, res); } +#define GetPosSlot(pos, res) { if (pos < 2) res = pos; else BSR2_RET(pos, res); } + + +#else // ! LZMA_LOG_BSR + +#define kNumLogBits (11 + sizeof(size_t) / 8 * 3) + +#define kDicLogSizeMaxCompress ((kNumLogBits - 1) * 2 + 7) + +static void LzmaEnc_FastPosInit(Byte *g_FastPos) +{ + unsigned slot; + g_FastPos[0] = 0; + g_FastPos[1] = 1; + g_FastPos += 2; + + for (slot = 2; slot < kNumLogBits * 2; slot++) + { + size_t k = ((size_t)1 << ((slot >> 1) - 1)); + size_t j; + for (j = 0; j < k; j++) + g_FastPos[j] = (Byte)slot; + g_FastPos += k; + } +} + +/* we can use ((limit - pos) >> 31) only if (pos < ((UInt32)1 << 31)) */ +/* +#define BSR2_RET(pos, res) { unsigned zz = 6 + ((kNumLogBits - 1) & \ + (0 - (((((UInt32)1 << (kNumLogBits + 6)) - 1) - pos) >> 31))); \ + res = p->g_FastPos[pos >> zz] + (zz * 2); } +*/ + +/* +#define BSR2_RET(pos, res) { unsigned zz = 6 + ((kNumLogBits - 1) & \ + (0 - (((((UInt32)1 << (kNumLogBits)) - 1) - (pos >> 6)) >> 31))); \ + res = p->g_FastPos[pos >> zz] + (zz * 2); } +*/ + +#define BSR2_RET(pos, res) { unsigned zz = (pos < (1 << (kNumLogBits + 6))) ? 6 : 6 + kNumLogBits - 1; \ + res = p->g_FastPos[pos >> zz] + (zz * 2); } + +/* +#define BSR2_RET(pos, res) { res = (pos < (1 << (kNumLogBits + 6))) ? \ + p->g_FastPos[pos >> 6] + 12 : \ + p->g_FastPos[pos >> (6 + kNumLogBits - 1)] + (6 + (kNumLogBits - 1)) * 2; } +*/ + +#define GetPosSlot1(pos) p->g_FastPos[pos] +#define GetPosSlot2(pos, res) { BSR2_RET(pos, res); } +#define GetPosSlot(pos, res) { if (pos < kNumFullDistances) res = p->g_FastPos[pos & (kNumFullDistances - 1)]; else BSR2_RET(pos, res); } + +#endif // LZMA_LOG_BSR + + +#define LZMA_NUM_REPS 4 + +typedef UInt16 CState; +typedef UInt16 CExtra; + +typedef struct +{ + UInt32 price; + CState state; + CExtra extra; + // 0 : normal + // 1 : LIT : MATCH + // > 1 : MATCH (extra-1) : LIT : REP0 (len) + UInt32 len; + UInt32 dist; + UInt32 reps[LZMA_NUM_REPS]; +} COptimal; + + +// 18.06 +#define kNumOpts (1 << 11) +#define kPackReserve (kNumOpts * 8) +// #define kNumOpts (1 << 12) +// #define kPackReserve (1 + kNumOpts * 2) + +#define kNumLenToPosStates 4 +#define kNumPosSlotBits 6 +// #define kDicLogSizeMin 0 +#define kDicLogSizeMax 32 +#define kDistTableSizeMax (kDicLogSizeMax * 2) + +#define kNumAlignBits 4 +#define kAlignTableSize (1 << kNumAlignBits) +#define kAlignMask (kAlignTableSize - 1) + +#define kStartPosModelIndex 4 +#define kEndPosModelIndex 14 +#define kNumFullDistances (1 << (kEndPosModelIndex >> 1)) + +typedef +#ifdef Z7_LZMA_PROB32 + UInt32 +#else + UInt16 +#endif + CLzmaProb; + +#define LZMA_PB_MAX 4 +#define LZMA_LC_MAX 8 +#define LZMA_LP_MAX 4 + +#define LZMA_NUM_PB_STATES_MAX (1 << LZMA_PB_MAX) + +#define kLenNumLowBits 3 +#define kLenNumLowSymbols (1 << kLenNumLowBits) +#define kLenNumHighBits 8 +#define kLenNumHighSymbols (1 << kLenNumHighBits) +#define kLenNumSymbolsTotal (kLenNumLowSymbols * 2 + kLenNumHighSymbols) + +#define LZMA_MATCH_LEN_MIN 2 +#define LZMA_MATCH_LEN_MAX (LZMA_MATCH_LEN_MIN + kLenNumSymbolsTotal - 1) + +#define kNumStates 12 + + +typedef struct +{ + CLzmaProb low[LZMA_NUM_PB_STATES_MAX << (kLenNumLowBits + 1)]; + CLzmaProb high[kLenNumHighSymbols]; +} CLenEnc; + + +typedef struct +{ + unsigned tableSize; + UInt32 prices[LZMA_NUM_PB_STATES_MAX][kLenNumSymbolsTotal]; + // UInt32 prices1[LZMA_NUM_PB_STATES_MAX][kLenNumLowSymbols * 2]; + // UInt32 prices2[kLenNumSymbolsTotal]; +} CLenPriceEnc; + +#define GET_PRICE_LEN(p, posState, len) \ + ((p)->prices[posState][(size_t)(len) - LZMA_MATCH_LEN_MIN]) + +/* +#define GET_PRICE_LEN(p, posState, len) \ + ((p)->prices2[(size_t)(len) - 2] + ((p)->prices1[posState][((len) - 2) & (kLenNumLowSymbols * 2 - 1)] & (((len) - 2 - kLenNumLowSymbols * 2) >> 9))) +*/ + +typedef struct +{ + UInt32 range; + unsigned cache; + UInt64 low; + UInt64 cacheSize; + Byte *buf; + Byte *bufLim; + Byte *bufBase; + ISeqOutStreamPtr outStream; + UInt64 processed; + SRes res; +} CRangeEnc; + + +typedef struct +{ + CLzmaProb *litProbs; + + unsigned state; + UInt32 reps[LZMA_NUM_REPS]; + + CLzmaProb posAlignEncoder[1 << kNumAlignBits]; + CLzmaProb isRep[kNumStates]; + CLzmaProb isRepG0[kNumStates]; + CLzmaProb isRepG1[kNumStates]; + CLzmaProb isRepG2[kNumStates]; + CLzmaProb isMatch[kNumStates][LZMA_NUM_PB_STATES_MAX]; + CLzmaProb isRep0Long[kNumStates][LZMA_NUM_PB_STATES_MAX]; + + CLzmaProb posSlotEncoder[kNumLenToPosStates][1 << kNumPosSlotBits]; + CLzmaProb posEncoders[kNumFullDistances]; + + CLenEnc lenProbs; + CLenEnc repLenProbs; + +} CSaveState; + + +typedef UInt32 CProbPrice; + + +struct CLzmaEnc +{ + void *matchFinderObj; + IMatchFinder2 matchFinder; + + unsigned optCur; + unsigned optEnd; + + unsigned longestMatchLen; + unsigned numPairs; + UInt32 numAvail; + + unsigned state; + unsigned numFastBytes; + unsigned additionalOffset; + UInt32 reps[LZMA_NUM_REPS]; + unsigned lpMask, pbMask; + CLzmaProb *litProbs; + CRangeEnc rc; + + UInt32 backRes; + + unsigned lc, lp, pb; + unsigned lclp; + + BoolInt fastMode; + BoolInt writeEndMark; + BoolInt finished; + BoolInt multiThread; + BoolInt needInit; + // BoolInt _maxMode; + + UInt64 nowPos64; + + unsigned matchPriceCount; + // unsigned alignPriceCount; + int repLenEncCounter; + + unsigned distTableSize; + + UInt32 dictSize; + SRes result; + + #ifndef Z7_ST + BoolInt mtMode; + // begin of CMatchFinderMt is used in LZ thread + CMatchFinderMt matchFinderMt; + // end of CMatchFinderMt is used in BT and HASH threads + // #else + // CMatchFinder matchFinderBase; + #endif + CMatchFinder matchFinderBase; + + + // we suppose that we have 8-bytes alignment after CMatchFinder + + #ifndef Z7_ST + Byte pad[128]; + #endif + + // LZ thread + CProbPrice ProbPrices[kBitModelTotal >> kNumMoveReducingBits]; + + // we want {len , dist} pairs to be 8-bytes aligned in matches array + UInt32 matches[LZMA_MATCH_LEN_MAX * 2 + 2]; + + // we want 8-bytes alignment here + UInt32 alignPrices[kAlignTableSize]; + UInt32 posSlotPrices[kNumLenToPosStates][kDistTableSizeMax]; + UInt32 distancesPrices[kNumLenToPosStates][kNumFullDistances]; + + CLzmaProb posAlignEncoder[1 << kNumAlignBits]; + CLzmaProb isRep[kNumStates]; + CLzmaProb isRepG0[kNumStates]; + CLzmaProb isRepG1[kNumStates]; + CLzmaProb isRepG2[kNumStates]; + CLzmaProb isMatch[kNumStates][LZMA_NUM_PB_STATES_MAX]; + CLzmaProb isRep0Long[kNumStates][LZMA_NUM_PB_STATES_MAX]; + CLzmaProb posSlotEncoder[kNumLenToPosStates][1 << kNumPosSlotBits]; + CLzmaProb posEncoders[kNumFullDistances]; + + CLenEnc lenProbs; + CLenEnc repLenProbs; + + #ifndef LZMA_LOG_BSR + Byte g_FastPos[1 << kNumLogBits]; + #endif + + CLenPriceEnc lenEnc; + CLenPriceEnc repLenEnc; + + COptimal opt[kNumOpts]; + + CSaveState saveState; + + // BoolInt mf_Failure; + #ifndef Z7_ST + Byte pad2[128]; + #endif +}; + + +#define MFB (p->matchFinderBase) +/* +#ifndef Z7_ST +#define MFB (p->matchFinderMt.MatchFinder) +#endif +*/ + +// #define GET_CLzmaEnc_p CLzmaEnc *p = (CLzmaEnc*)(void *)p; +// #define GET_const_CLzmaEnc_p const CLzmaEnc *p = (const CLzmaEnc*)(const void *)p; + +#define COPY_ARR(dest, src, arr) memcpy((dest)->arr, (src)->arr, sizeof((src)->arr)); + +#define COPY_LZMA_ENC_STATE(d, s, p) \ + (d)->state = (s)->state; \ + COPY_ARR(d, s, reps) \ + COPY_ARR(d, s, posAlignEncoder) \ + COPY_ARR(d, s, isRep) \ + COPY_ARR(d, s, isRepG0) \ + COPY_ARR(d, s, isRepG1) \ + COPY_ARR(d, s, isRepG2) \ + COPY_ARR(d, s, isMatch) \ + COPY_ARR(d, s, isRep0Long) \ + COPY_ARR(d, s, posSlotEncoder) \ + COPY_ARR(d, s, posEncoders) \ + (d)->lenProbs = (s)->lenProbs; \ + (d)->repLenProbs = (s)->repLenProbs; \ + memcpy((d)->litProbs, (s)->litProbs, ((UInt32)0x300 << (p)->lclp) * sizeof(CLzmaProb)); + +void LzmaEnc_SaveState(CLzmaEncHandle p) +{ + // GET_CLzmaEnc_p + CSaveState *v = &p->saveState; + COPY_LZMA_ENC_STATE(v, p, p) +} + +void LzmaEnc_RestoreState(CLzmaEncHandle p) +{ + // GET_CLzmaEnc_p + const CSaveState *v = &p->saveState; + COPY_LZMA_ENC_STATE(p, v, p) +} + + +Z7_NO_INLINE +SRes LzmaEnc_SetProps(CLzmaEncHandle p, const CLzmaEncProps *props2) +{ + // GET_CLzmaEnc_p + CLzmaEncProps props = *props2; + LzmaEncProps_Normalize(&props); + + if (props.lc > LZMA_LC_MAX + || props.lp > LZMA_LP_MAX + || props.pb > LZMA_PB_MAX) + return SZ_ERROR_PARAM; + + + if (props.dictSize > kLzmaMaxHistorySize) + props.dictSize = kLzmaMaxHistorySize; + + #ifndef LZMA_LOG_BSR + { + const UInt64 dict64 = props.dictSize; + if (dict64 > ((UInt64)1 << kDicLogSizeMaxCompress)) + return SZ_ERROR_PARAM; + } + #endif + + p->dictSize = props.dictSize; + { + unsigned fb = (unsigned)props.fb; + if (fb < 5) + fb = 5; + if (fb > LZMA_MATCH_LEN_MAX) + fb = LZMA_MATCH_LEN_MAX; + p->numFastBytes = fb; + } + p->lc = (unsigned)props.lc; + p->lp = (unsigned)props.lp; + p->pb = (unsigned)props.pb; + p->fastMode = (props.algo == 0); + // p->_maxMode = True; + MFB.btMode = (Byte)(props.btMode ? 1 : 0); + // MFB.btMode = (Byte)(props.btMode); + { + unsigned numHashBytes = 4; + if (props.btMode) + { + if (props.numHashBytes < 2) numHashBytes = 2; + else if (props.numHashBytes < 4) numHashBytes = (unsigned)props.numHashBytes; + } + if (props.numHashBytes >= 5) numHashBytes = 5; + + MFB.numHashBytes = numHashBytes; + // MFB.numHashBytes_Min = 2; + MFB.numHashOutBits = (Byte)props.numHashOutBits; + } + + MFB.cutValue = props.mc; + + p->writeEndMark = (BoolInt)props.writeEndMark; + + #ifndef Z7_ST + /* + if (newMultiThread != _multiThread) + { + ReleaseMatchFinder(); + _multiThread = newMultiThread; + } + */ + p->multiThread = (props.numThreads > 1); + p->matchFinderMt.btSync.affinity = + p->matchFinderMt.hashSync.affinity = props.affinity; + #endif + + return SZ_OK; +} + + +void LzmaEnc_SetDataSize(CLzmaEncHandle p, UInt64 expectedDataSiize) +{ + // GET_CLzmaEnc_p + MFB.expectedDataSize = expectedDataSiize; +} + + +#define kState_Start 0 +#define kState_LitAfterMatch 4 +#define kState_LitAfterRep 5 +#define kState_MatchAfterLit 7 +#define kState_RepAfterLit 8 + +static const Byte kLiteralNextStates[kNumStates] = {0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 4, 5}; +static const Byte kMatchNextStates[kNumStates] = {7, 7, 7, 7, 7, 7, 7, 10, 10, 10, 10, 10}; +static const Byte kRepNextStates[kNumStates] = {8, 8, 8, 8, 8, 8, 8, 11, 11, 11, 11, 11}; +static const Byte kShortRepNextStates[kNumStates]= {9, 9, 9, 9, 9, 9, 9, 11, 11, 11, 11, 11}; + +#define IsLitState(s) ((s) < 7) +#define GetLenToPosState2(len) (((len) < kNumLenToPosStates - 1) ? (len) : kNumLenToPosStates - 1) +#define GetLenToPosState(len) (((len) < kNumLenToPosStates + 1) ? (len) - 2 : kNumLenToPosStates - 1) + +#define kInfinityPrice (1 << 30) + +static void RangeEnc_Construct(CRangeEnc *p) +{ + p->outStream = NULL; + p->bufBase = NULL; +} + +#define RangeEnc_GetProcessed(p) ( (p)->processed + (size_t)((p)->buf - (p)->bufBase) + (p)->cacheSize) +#define RangeEnc_GetProcessed_sizet(p) ((size_t)(p)->processed + (size_t)((p)->buf - (p)->bufBase) + (size_t)(p)->cacheSize) + +#define RC_BUF_SIZE (1 << 16) + +static int RangeEnc_Alloc(CRangeEnc *p, ISzAllocPtr alloc) +{ + if (!p->bufBase) + { + p->bufBase = (Byte *)ISzAlloc_Alloc(alloc, RC_BUF_SIZE); + if (!p->bufBase) + return 0; + p->bufLim = p->bufBase + RC_BUF_SIZE; + } + return 1; +} + +static void RangeEnc_Free(CRangeEnc *p, ISzAllocPtr alloc) +{ + ISzAlloc_Free(alloc, p->bufBase); + p->bufBase = NULL; +} + +static void RangeEnc_Init(CRangeEnc *p) +{ + p->range = 0xFFFFFFFF; + p->cache = 0; + p->low = 0; + p->cacheSize = 0; + + p->buf = p->bufBase; + + p->processed = 0; + p->res = SZ_OK; +} + +Z7_NO_INLINE static void RangeEnc_FlushStream(CRangeEnc *p) +{ + const size_t num = (size_t)(p->buf - p->bufBase); + if (p->res == SZ_OK) + { + if (num != ISeqOutStream_Write(p->outStream, p->bufBase, num)) + p->res = SZ_ERROR_WRITE; + } + p->processed += num; + p->buf = p->bufBase; +} + +Z7_NO_INLINE static void Z7_FASTCALL RangeEnc_ShiftLow(CRangeEnc *p) +{ + UInt32 low = (UInt32)p->low; + unsigned high = (unsigned)(p->low >> 32); + p->low = (UInt32)(low << 8); + if (low < (UInt32)0xFF000000 || high != 0) + { + { + Byte *buf = p->buf; + *buf++ = (Byte)(p->cache + high); + p->cache = (unsigned)(low >> 24); + p->buf = buf; + if (buf == p->bufLim) + RangeEnc_FlushStream(p); + if (p->cacheSize == 0) + return; + } + high += 0xFF; + for (;;) + { + Byte *buf = p->buf; + *buf++ = (Byte)(high); + p->buf = buf; + if (buf == p->bufLim) + RangeEnc_FlushStream(p); + if (--p->cacheSize == 0) + return; + } + } + p->cacheSize++; +} + +static void RangeEnc_FlushData(CRangeEnc *p) +{ + int i; + for (i = 0; i < 5; i++) + RangeEnc_ShiftLow(p); +} + +#define RC_NORM(p) if (range < kTopValue) { range <<= 8; RangeEnc_ShiftLow(p); } + +#define RC_BIT_PRE(p, prob) \ + ttt = *(prob); \ + newBound = (range >> kNumBitModelTotalBits) * ttt; + +// #define Z7_LZMA_ENC_USE_BRANCH + +#ifdef Z7_LZMA_ENC_USE_BRANCH + +#define RC_BIT(p, prob, bit) { \ + RC_BIT_PRE(p, prob) \ + if (bit == 0) { range = newBound; ttt += (kBitModelTotal - ttt) >> kNumMoveBits; } \ + else { (p)->low += newBound; range -= newBound; ttt -= ttt >> kNumMoveBits; } \ + *(prob) = (CLzmaProb)ttt; \ + RC_NORM(p) \ + } + +#else + +#define RC_BIT(p, prob, bit) { \ + UInt32 mask; \ + RC_BIT_PRE(p, prob) \ + mask = 0 - (UInt32)bit; \ + range &= mask; \ + mask &= newBound; \ + range -= mask; \ + (p)->low += mask; \ + mask = (UInt32)bit - 1; \ + range += newBound & mask; \ + mask &= (kBitModelTotal - ((1 << kNumMoveBits) - 1)); \ + mask += ((1 << kNumMoveBits) - 1); \ + ttt += (UInt32)((Int32)(mask - ttt) >> kNumMoveBits); \ + *(prob) = (CLzmaProb)ttt; \ + RC_NORM(p) \ + } + +#endif + + + + +#define RC_BIT_0_BASE(p, prob) \ + range = newBound; *(prob) = (CLzmaProb)(ttt + ((kBitModelTotal - ttt) >> kNumMoveBits)); + +#define RC_BIT_1_BASE(p, prob) \ + range -= newBound; (p)->low += newBound; *(prob) = (CLzmaProb)(ttt - (ttt >> kNumMoveBits)); \ + +#define RC_BIT_0(p, prob) \ + RC_BIT_0_BASE(p, prob) \ + RC_NORM(p) + +#define RC_BIT_1(p, prob) \ + RC_BIT_1_BASE(p, prob) \ + RC_NORM(p) + +static void RangeEnc_EncodeBit_0(CRangeEnc *p, CLzmaProb *prob) +{ + UInt32 range, ttt, newBound; + range = p->range; + RC_BIT_PRE(p, prob) + RC_BIT_0(p, prob) + p->range = range; +} + +static void LitEnc_Encode(CRangeEnc *p, CLzmaProb *probs, UInt32 sym) +{ + UInt32 range = p->range; + sym |= 0x100; + do + { + UInt32 ttt, newBound; + // RangeEnc_EncodeBit(p, probs + (sym >> 8), (sym >> 7) & 1); + CLzmaProb *prob = probs + (sym >> 8); + UInt32 bit = (sym >> 7) & 1; + sym <<= 1; + RC_BIT(p, prob, bit) + } + while (sym < 0x10000); + p->range = range; +} + +static void LitEnc_EncodeMatched(CRangeEnc *p, CLzmaProb *probs, UInt32 sym, UInt32 matchByte) +{ + UInt32 range = p->range; + UInt32 offs = 0x100; + sym |= 0x100; + do + { + UInt32 ttt, newBound; + CLzmaProb *prob; + UInt32 bit; + matchByte <<= 1; + // RangeEnc_EncodeBit(p, probs + (offs + (matchByte & offs) + (sym >> 8)), (sym >> 7) & 1); + prob = probs + (offs + (matchByte & offs) + (sym >> 8)); + bit = (sym >> 7) & 1; + sym <<= 1; + offs &= ~(matchByte ^ sym); + RC_BIT(p, prob, bit) + } + while (sym < 0x10000); + p->range = range; +} + + + +static void LzmaEnc_InitPriceTables(CProbPrice *ProbPrices) +{ + UInt32 i; + for (i = 0; i < (kBitModelTotal >> kNumMoveReducingBits); i++) + { + const unsigned kCyclesBits = kNumBitPriceShiftBits; + UInt32 w = (i << kNumMoveReducingBits) + (1 << (kNumMoveReducingBits - 1)); + unsigned bitCount = 0; + unsigned j; + for (j = 0; j < kCyclesBits; j++) + { + w = w * w; + bitCount <<= 1; + while (w >= ((UInt32)1 << 16)) + { + w >>= 1; + bitCount++; + } + } + ProbPrices[i] = (CProbPrice)(((unsigned)kNumBitModelTotalBits << kCyclesBits) - 15 - bitCount); + // printf("\n%3d: %5d", i, ProbPrices[i]); + } +} + + +#define GET_PRICE(prob, bit) \ + p->ProbPrices[((prob) ^ (unsigned)(((-(int)(bit))) & (kBitModelTotal - 1))) >> kNumMoveReducingBits] + +#define GET_PRICEa(prob, bit) \ + ProbPrices[((prob) ^ (unsigned)((-((int)(bit))) & (kBitModelTotal - 1))) >> kNumMoveReducingBits] + +#define GET_PRICE_0(prob) p->ProbPrices[(prob) >> kNumMoveReducingBits] +#define GET_PRICE_1(prob) p->ProbPrices[((prob) ^ (kBitModelTotal - 1)) >> kNumMoveReducingBits] + +#define GET_PRICEa_0(prob) ProbPrices[(prob) >> kNumMoveReducingBits] +#define GET_PRICEa_1(prob) ProbPrices[((prob) ^ (kBitModelTotal - 1)) >> kNumMoveReducingBits] + + +static UInt32 LitEnc_GetPrice(const CLzmaProb *probs, UInt32 sym, const CProbPrice *ProbPrices) +{ + UInt32 price = 0; + sym |= 0x100; + do + { + unsigned bit = sym & 1; + sym >>= 1; + price += GET_PRICEa(probs[sym], bit); + } + while (sym >= 2); + return price; +} + + +static UInt32 LitEnc_Matched_GetPrice(const CLzmaProb *probs, UInt32 sym, UInt32 matchByte, const CProbPrice *ProbPrices) +{ + UInt32 price = 0; + UInt32 offs = 0x100; + sym |= 0x100; + do + { + matchByte <<= 1; + price += GET_PRICEa(probs[offs + (matchByte & offs) + (sym >> 8)], (sym >> 7) & 1); + sym <<= 1; + offs &= ~(matchByte ^ sym); + } + while (sym < 0x10000); + return price; +} + + +static void RcTree_ReverseEncode(CRangeEnc *rc, CLzmaProb *probs, unsigned numBits, unsigned sym) +{ + UInt32 range = rc->range; + unsigned m = 1; + do + { + UInt32 ttt, newBound; + unsigned bit = sym & 1; + // RangeEnc_EncodeBit(rc, probs + m, bit); + sym >>= 1; + RC_BIT(rc, probs + m, bit) + m = (m << 1) | bit; + } + while (--numBits); + rc->range = range; +} + + + +static void LenEnc_Init(CLenEnc *p) +{ + unsigned i; + for (i = 0; i < (LZMA_NUM_PB_STATES_MAX << (kLenNumLowBits + 1)); i++) + p->low[i] = kProbInitValue; + for (i = 0; i < kLenNumHighSymbols; i++) + p->high[i] = kProbInitValue; +} + +static void LenEnc_Encode(CLenEnc *p, CRangeEnc *rc, unsigned sym, unsigned posState) +{ + UInt32 range, ttt, newBound; + CLzmaProb *probs = p->low; + range = rc->range; + RC_BIT_PRE(rc, probs) + if (sym >= kLenNumLowSymbols) + { + RC_BIT_1(rc, probs) + probs += kLenNumLowSymbols; + RC_BIT_PRE(rc, probs) + if (sym >= kLenNumLowSymbols * 2) + { + RC_BIT_1(rc, probs) + rc->range = range; + // RcTree_Encode(rc, p->high, kLenNumHighBits, sym - kLenNumLowSymbols * 2); + LitEnc_Encode(rc, p->high, sym - kLenNumLowSymbols * 2); + return; + } + sym -= kLenNumLowSymbols; + } + + // RcTree_Encode(rc, probs + (posState << kLenNumLowBits), kLenNumLowBits, sym); + { + unsigned m; + unsigned bit; + RC_BIT_0(rc, probs) + probs += (posState << (1 + kLenNumLowBits)); + bit = (sym >> 2) ; RC_BIT(rc, probs + 1, bit) m = (1 << 1) + bit; + bit = (sym >> 1) & 1; RC_BIT(rc, probs + m, bit) m = (m << 1) + bit; + bit = sym & 1; RC_BIT(rc, probs + m, bit) + rc->range = range; + } +} + +static void SetPrices_3(const CLzmaProb *probs, UInt32 startPrice, UInt32 *prices, const CProbPrice *ProbPrices) +{ + unsigned i; + for (i = 0; i < 8; i += 2) + { + UInt32 price = startPrice; + UInt32 prob; + price += GET_PRICEa(probs[1 ], (i >> 2)); + price += GET_PRICEa(probs[2 + (i >> 2)], (i >> 1) & 1); + prob = probs[4 + (i >> 1)]; + prices[i ] = price + GET_PRICEa_0(prob); + prices[i + 1] = price + GET_PRICEa_1(prob); + } +} + + +Z7_NO_INLINE static void Z7_FASTCALL LenPriceEnc_UpdateTables( + CLenPriceEnc *p, + unsigned numPosStates, + const CLenEnc *enc, + const CProbPrice *ProbPrices) +{ + UInt32 b; + + { + unsigned prob = enc->low[0]; + UInt32 a, c; + unsigned posState; + b = GET_PRICEa_1(prob); + a = GET_PRICEa_0(prob); + c = b + GET_PRICEa_0(enc->low[kLenNumLowSymbols]); + for (posState = 0; posState < numPosStates; posState++) + { + UInt32 *prices = p->prices[posState]; + const CLzmaProb *probs = enc->low + (posState << (1 + kLenNumLowBits)); + SetPrices_3(probs, a, prices, ProbPrices); + SetPrices_3(probs + kLenNumLowSymbols, c, prices + kLenNumLowSymbols, ProbPrices); + } + } + + /* + { + unsigned i; + UInt32 b; + a = GET_PRICEa_0(enc->low[0]); + for (i = 0; i < kLenNumLowSymbols; i++) + p->prices2[i] = a; + a = GET_PRICEa_1(enc->low[0]); + b = a + GET_PRICEa_0(enc->low[kLenNumLowSymbols]); + for (i = kLenNumLowSymbols; i < kLenNumLowSymbols * 2; i++) + p->prices2[i] = b; + a += GET_PRICEa_1(enc->low[kLenNumLowSymbols]); + } + */ + + // p->counter = numSymbols; + // p->counter = 64; + + { + unsigned i = p->tableSize; + + if (i > kLenNumLowSymbols * 2) + { + const CLzmaProb *probs = enc->high; + UInt32 *prices = p->prices[0] + kLenNumLowSymbols * 2; + i -= kLenNumLowSymbols * 2 - 1; + i >>= 1; + b += GET_PRICEa_1(enc->low[kLenNumLowSymbols]); + do + { + /* + p->prices2[i] = a + + // RcTree_GetPrice(enc->high, kLenNumHighBits, i - kLenNumLowSymbols * 2, ProbPrices); + LitEnc_GetPrice(probs, i - kLenNumLowSymbols * 2, ProbPrices); + */ + // UInt32 price = a + RcTree_GetPrice(probs, kLenNumHighBits - 1, sym, ProbPrices); + unsigned sym = --i + (1 << (kLenNumHighBits - 1)); + UInt32 price = b; + do + { + unsigned bit = sym & 1; + sym >>= 1; + price += GET_PRICEa(probs[sym], bit); + } + while (sym >= 2); + + { + unsigned prob = probs[(size_t)i + (1 << (kLenNumHighBits - 1))]; + prices[(size_t)i * 2 ] = price + GET_PRICEa_0(prob); + prices[(size_t)i * 2 + 1] = price + GET_PRICEa_1(prob); + } + } + while (i); + + { + unsigned posState; + size_t num = (p->tableSize - kLenNumLowSymbols * 2) * sizeof(p->prices[0][0]); + for (posState = 1; posState < numPosStates; posState++) + memcpy(p->prices[posState] + kLenNumLowSymbols * 2, p->prices[0] + kLenNumLowSymbols * 2, num); + } + } + } +} + +/* + #ifdef SHOW_STAT + g_STAT_OFFSET += num; + printf("\n MovePos %u", num); + #endif +*/ + +#define MOVE_POS(p, num) { \ + p->additionalOffset += (num); \ + p->matchFinder.Skip(p->matchFinderObj, (UInt32)(num)); } + + +static unsigned ReadMatchDistances(CLzmaEnc *p, unsigned *numPairsRes) +{ + unsigned numPairs; + + p->additionalOffset++; + p->numAvail = p->matchFinder.GetNumAvailableBytes(p->matchFinderObj); + { + const UInt32 *d = p->matchFinder.GetMatches(p->matchFinderObj, p->matches); + // if (!d) { p->mf_Failure = True; *numPairsRes = 0; return 0; } + numPairs = (unsigned)(d - p->matches); + } + *numPairsRes = numPairs; + + #ifdef SHOW_STAT + printf("\n i = %u numPairs = %u ", g_STAT_OFFSET, numPairs / 2); + g_STAT_OFFSET++; + { + unsigned i; + for (i = 0; i < numPairs; i += 2) + printf("%2u %6u | ", p->matches[i], p->matches[i + 1]); + } + #endif + + if (numPairs == 0) + return 0; + { + const unsigned len = p->matches[(size_t)numPairs - 2]; + if (len != p->numFastBytes) + return len; + { + UInt32 numAvail = p->numAvail; + if (numAvail > LZMA_MATCH_LEN_MAX) + numAvail = LZMA_MATCH_LEN_MAX; + { + const Byte *p1 = p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - 1; + const Byte *p2 = p1 + len; + const ptrdiff_t dif = (ptrdiff_t)-1 - (ptrdiff_t)p->matches[(size_t)numPairs - 1]; + const Byte *lim = p1 + numAvail; + for (; p2 != lim && *p2 == p2[dif]; p2++) + {} + return (unsigned)(p2 - p1); + } + } + } +} + +#define MARK_LIT ((UInt32)(Int32)-1) + +#define MakeAs_Lit(p) { (p)->dist = MARK_LIT; (p)->extra = 0; } +#define MakeAs_ShortRep(p) { (p)->dist = 0; (p)->extra = 0; } +#define IsShortRep(p) ((p)->dist == 0) + + +#define GetPrice_ShortRep(p, state, posState) \ + ( GET_PRICE_0(p->isRepG0[state]) + GET_PRICE_0(p->isRep0Long[state][posState])) + +#define GetPrice_Rep_0(p, state, posState) ( \ + GET_PRICE_1(p->isMatch[state][posState]) \ + + GET_PRICE_1(p->isRep0Long[state][posState])) \ + + GET_PRICE_1(p->isRep[state]) \ + + GET_PRICE_0(p->isRepG0[state]) + +Z7_FORCE_INLINE +static UInt32 GetPrice_PureRep(const CLzmaEnc *p, unsigned repIndex, size_t state, size_t posState) +{ + UInt32 price; + UInt32 prob = p->isRepG0[state]; + if (repIndex == 0) + { + price = GET_PRICE_0(prob); + price += GET_PRICE_1(p->isRep0Long[state][posState]); + } + else + { + price = GET_PRICE_1(prob); + prob = p->isRepG1[state]; + if (repIndex == 1) + price += GET_PRICE_0(prob); + else + { + price += GET_PRICE_1(prob); + price += GET_PRICE(p->isRepG2[state], repIndex - 2); + } + } + return price; +} + + +static unsigned Backward(CLzmaEnc *p, unsigned cur) +{ + unsigned wr = cur + 1; + p->optEnd = wr; + + for (;;) + { + UInt32 dist = p->opt[cur].dist; + unsigned len = (unsigned)p->opt[cur].len; + unsigned extra = (unsigned)p->opt[cur].extra; + cur -= len; + + if (extra) + { + wr--; + p->opt[wr].len = (UInt32)len; + cur -= extra; + len = extra; + if (extra == 1) + { + p->opt[wr].dist = dist; + dist = MARK_LIT; + } + else + { + p->opt[wr].dist = 0; + len--; + wr--; + p->opt[wr].dist = MARK_LIT; + p->opt[wr].len = 1; + } + } + + if (cur == 0) + { + p->backRes = dist; + p->optCur = wr; + return len; + } + + wr--; + p->opt[wr].dist = dist; + p->opt[wr].len = (UInt32)len; + } +} + + + +#define LIT_PROBS(pos, prevByte) \ + (p->litProbs + (UInt32)3 * (((((pos) << 8) + (prevByte)) & p->lpMask) << p->lc)) + + +static unsigned GetOptimum(CLzmaEnc *p, UInt32 position) +{ + unsigned last, cur; + UInt32 reps[LZMA_NUM_REPS]; + unsigned repLens[LZMA_NUM_REPS]; + UInt32 *matches; + + { + UInt32 numAvail; + unsigned numPairs, mainLen, repMaxIndex, i, posState; + UInt32 matchPrice, repMatchPrice; + const Byte *data; + Byte curByte, matchByte; + + p->optCur = p->optEnd = 0; + + if (p->additionalOffset == 0) + mainLen = ReadMatchDistances(p, &numPairs); + else + { + mainLen = p->longestMatchLen; + numPairs = p->numPairs; + } + + numAvail = p->numAvail; + if (numAvail < 2) + { + p->backRes = MARK_LIT; + return 1; + } + if (numAvail > LZMA_MATCH_LEN_MAX) + numAvail = LZMA_MATCH_LEN_MAX; + + data = p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - 1; + repMaxIndex = 0; + + for (i = 0; i < LZMA_NUM_REPS; i++) + { + unsigned len; + const Byte *data2; + reps[i] = p->reps[i]; + data2 = data - reps[i]; + if (data[0] != data2[0] || data[1] != data2[1]) + { + repLens[i] = 0; + continue; + } + for (len = 2; len < numAvail && data[len] == data2[len]; len++) + {} + repLens[i] = len; + if (len > repLens[repMaxIndex]) + repMaxIndex = i; + if (len == LZMA_MATCH_LEN_MAX) // 21.03 : optimization + break; + } + + if (repLens[repMaxIndex] >= p->numFastBytes) + { + unsigned len; + p->backRes = (UInt32)repMaxIndex; + len = repLens[repMaxIndex]; + MOVE_POS(p, len - 1) + return len; + } + + matches = p->matches; + #define MATCHES matches + // #define MATCHES p->matches + + if (mainLen >= p->numFastBytes) + { + p->backRes = MATCHES[(size_t)numPairs - 1] + LZMA_NUM_REPS; + MOVE_POS(p, mainLen - 1) + return mainLen; + } + + curByte = *data; + matchByte = *(data - reps[0]); + + last = repLens[repMaxIndex]; + if (last <= mainLen) + last = mainLen; + + if (last < 2 && curByte != matchByte) + { + p->backRes = MARK_LIT; + return 1; + } + + p->opt[0].state = (CState)p->state; + + posState = (position & p->pbMask); + + { + const CLzmaProb *probs = LIT_PROBS(position, *(data - 1)); + p->opt[1].price = GET_PRICE_0(p->isMatch[p->state][posState]) + + (!IsLitState(p->state) ? + LitEnc_Matched_GetPrice(probs, curByte, matchByte, p->ProbPrices) : + LitEnc_GetPrice(probs, curByte, p->ProbPrices)); + } + + MakeAs_Lit(&p->opt[1]) + + matchPrice = GET_PRICE_1(p->isMatch[p->state][posState]); + repMatchPrice = matchPrice + GET_PRICE_1(p->isRep[p->state]); + + // 18.06 + if (matchByte == curByte && repLens[0] == 0) + { + UInt32 shortRepPrice = repMatchPrice + GetPrice_ShortRep(p, p->state, posState); + if (shortRepPrice < p->opt[1].price) + { + p->opt[1].price = shortRepPrice; + MakeAs_ShortRep(&p->opt[1]) + } + if (last < 2) + { + p->backRes = p->opt[1].dist; + return 1; + } + } + + p->opt[1].len = 1; + + p->opt[0].reps[0] = reps[0]; + p->opt[0].reps[1] = reps[1]; + p->opt[0].reps[2] = reps[2]; + p->opt[0].reps[3] = reps[3]; + + // ---------- REP ---------- + + for (i = 0; i < LZMA_NUM_REPS; i++) + { + unsigned repLen = repLens[i]; + UInt32 price; + if (repLen < 2) + continue; + price = repMatchPrice + GetPrice_PureRep(p, i, p->state, posState); + do + { + UInt32 price2 = price + GET_PRICE_LEN(&p->repLenEnc, posState, repLen); + COptimal *opt = &p->opt[repLen]; + if (price2 < opt->price) + { + opt->price = price2; + opt->len = (UInt32)repLen; + opt->dist = (UInt32)i; + opt->extra = 0; + } + } + while (--repLen >= 2); + } + + + // ---------- MATCH ---------- + { + unsigned len = repLens[0] + 1; + if (len <= mainLen) + { + unsigned offs = 0; + UInt32 normalMatchPrice = matchPrice + GET_PRICE_0(p->isRep[p->state]); + + if (len < 2) + len = 2; + else + while (len > MATCHES[offs]) + offs += 2; + + for (; ; len++) + { + COptimal *opt; + UInt32 dist = MATCHES[(size_t)offs + 1]; + UInt32 price = normalMatchPrice + GET_PRICE_LEN(&p->lenEnc, posState, len); + unsigned lenToPosState = GetLenToPosState(len); + + if (dist < kNumFullDistances) + price += p->distancesPrices[lenToPosState][dist & (kNumFullDistances - 1)]; + else + { + unsigned slot; + GetPosSlot2(dist, slot) + price += p->alignPrices[dist & kAlignMask]; + price += p->posSlotPrices[lenToPosState][slot]; + } + + opt = &p->opt[len]; + + if (price < opt->price) + { + opt->price = price; + opt->len = (UInt32)len; + opt->dist = dist + LZMA_NUM_REPS; + opt->extra = 0; + } + + if (len == MATCHES[offs]) + { + offs += 2; + if (offs == numPairs) + break; + } + } + } + } + + + cur = 0; + + #ifdef SHOW_STAT2 + /* if (position >= 0) */ + { + unsigned i; + printf("\n pos = %4X", position); + for (i = cur; i <= last; i++) + printf("\nprice[%4X] = %u", position - cur + i, p->opt[i].price); + } + #endif + } + + + + // ---------- Optimal Parsing ---------- + + for (;;) + { + unsigned numAvail; + UInt32 numAvailFull; + unsigned newLen, numPairs, prev, state, posState, startLen; + UInt32 litPrice, matchPrice, repMatchPrice; + BoolInt nextIsLit; + Byte curByte, matchByte; + const Byte *data; + COptimal *curOpt, *nextOpt; + + if (++cur == last) + break; + + // 18.06 + if (cur >= kNumOpts - 64) + { + unsigned j, best; + UInt32 price = p->opt[cur].price; + best = cur; + for (j = cur + 1; j <= last; j++) + { + UInt32 price2 = p->opt[j].price; + if (price >= price2) + { + price = price2; + best = j; + } + } + { + unsigned delta = best - cur; + if (delta != 0) + { + MOVE_POS(p, delta) + } + } + cur = best; + break; + } + + newLen = ReadMatchDistances(p, &numPairs); + + if (newLen >= p->numFastBytes) + { + p->numPairs = numPairs; + p->longestMatchLen = newLen; + break; + } + + curOpt = &p->opt[cur]; + + position++; + + // we need that check here, if skip_items in p->opt are possible + /* + if (curOpt->price >= kInfinityPrice) + continue; + */ + + prev = cur - curOpt->len; + + if (curOpt->len == 1) + { + state = (unsigned)p->opt[prev].state; + if (IsShortRep(curOpt)) + state = kShortRepNextStates[state]; + else + state = kLiteralNextStates[state]; + } + else + { + const COptimal *prevOpt; + UInt32 b0; + UInt32 dist = curOpt->dist; + + if (curOpt->extra) + { + prev -= (unsigned)curOpt->extra; + state = kState_RepAfterLit; + if (curOpt->extra == 1) + state = (dist < LZMA_NUM_REPS ? kState_RepAfterLit : kState_MatchAfterLit); + } + else + { + state = (unsigned)p->opt[prev].state; + if (dist < LZMA_NUM_REPS) + state = kRepNextStates[state]; + else + state = kMatchNextStates[state]; + } + + prevOpt = &p->opt[prev]; + b0 = prevOpt->reps[0]; + + if (dist < LZMA_NUM_REPS) + { + if (dist == 0) + { + reps[0] = b0; + reps[1] = prevOpt->reps[1]; + reps[2] = prevOpt->reps[2]; + reps[3] = prevOpt->reps[3]; + } + else + { + reps[1] = b0; + b0 = prevOpt->reps[1]; + if (dist == 1) + { + reps[0] = b0; + reps[2] = prevOpt->reps[2]; + reps[3] = prevOpt->reps[3]; + } + else + { + reps[2] = b0; + reps[0] = prevOpt->reps[dist]; + reps[3] = prevOpt->reps[dist ^ 1]; + } + } + } + else + { + reps[0] = (dist - LZMA_NUM_REPS + 1); + reps[1] = b0; + reps[2] = prevOpt->reps[1]; + reps[3] = prevOpt->reps[2]; + } + } + + curOpt->state = (CState)state; + curOpt->reps[0] = reps[0]; + curOpt->reps[1] = reps[1]; + curOpt->reps[2] = reps[2]; + curOpt->reps[3] = reps[3]; + + data = p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - 1; + curByte = *data; + matchByte = *(data - reps[0]); + + posState = (position & p->pbMask); + + /* + The order of Price checks: + < LIT + <= SHORT_REP + < LIT : REP_0 + < REP [ : LIT : REP_0 ] + < MATCH [ : LIT : REP_0 ] + */ + + { + UInt32 curPrice = curOpt->price; + unsigned prob = p->isMatch[state][posState]; + matchPrice = curPrice + GET_PRICE_1(prob); + litPrice = curPrice + GET_PRICE_0(prob); + } + + nextOpt = &p->opt[(size_t)cur + 1]; + nextIsLit = False; + + // here we can allow skip_items in p->opt, if we don't check (nextOpt->price < kInfinityPrice) + // 18.new.06 + if ((nextOpt->price < kInfinityPrice + // && !IsLitState(state) + && matchByte == curByte) + || litPrice > nextOpt->price + ) + litPrice = 0; + else + { + const CLzmaProb *probs = LIT_PROBS(position, *(data - 1)); + litPrice += (!IsLitState(state) ? + LitEnc_Matched_GetPrice(probs, curByte, matchByte, p->ProbPrices) : + LitEnc_GetPrice(probs, curByte, p->ProbPrices)); + + if (litPrice < nextOpt->price) + { + nextOpt->price = litPrice; + nextOpt->len = 1; + MakeAs_Lit(nextOpt) + nextIsLit = True; + } + } + + repMatchPrice = matchPrice + GET_PRICE_1(p->isRep[state]); + + numAvailFull = p->numAvail; + { + unsigned temp = kNumOpts - 1 - cur; + if (numAvailFull > temp) + numAvailFull = (UInt32)temp; + } + + // 18.06 + // ---------- SHORT_REP ---------- + if (IsLitState(state)) // 18.new + if (matchByte == curByte) + if (repMatchPrice < nextOpt->price) // 18.new + // if (numAvailFull < 2 || data[1] != *(data - reps[0] + 1)) + if ( + // nextOpt->price >= kInfinityPrice || + nextOpt->len < 2 // we can check nextOpt->len, if skip items are not allowed in p->opt + || (nextOpt->dist != 0 + // && nextOpt->extra <= 1 // 17.old + ) + ) + { + UInt32 shortRepPrice = repMatchPrice + GetPrice_ShortRep(p, state, posState); + // if (shortRepPrice <= nextOpt->price) // 17.old + if (shortRepPrice < nextOpt->price) // 18.new + { + nextOpt->price = shortRepPrice; + nextOpt->len = 1; + MakeAs_ShortRep(nextOpt) + nextIsLit = False; + } + } + + if (numAvailFull < 2) + continue; + numAvail = (numAvailFull <= p->numFastBytes ? numAvailFull : p->numFastBytes); + + // numAvail <= p->numFastBytes + + // ---------- LIT : REP_0 ---------- + + if (!nextIsLit + && litPrice != 0 // 18.new + && matchByte != curByte + && numAvailFull > 2) + { + const Byte *data2 = data - reps[0]; + if (data[1] == data2[1] && data[2] == data2[2]) + { + unsigned len; + unsigned limit = p->numFastBytes + 1; + if (limit > numAvailFull) + limit = numAvailFull; + for (len = 3; len < limit && data[len] == data2[len]; len++) + {} + + { + unsigned state2 = kLiteralNextStates[state]; + unsigned posState2 = (position + 1) & p->pbMask; + UInt32 price = litPrice + GetPrice_Rep_0(p, state2, posState2); + { + unsigned offset = cur + len; + + if (last < offset) + last = offset; + + // do + { + UInt32 price2; + COptimal *opt; + len--; + // price2 = price + GetPrice_Len_Rep_0(p, len, state2, posState2); + price2 = price + GET_PRICE_LEN(&p->repLenEnc, posState2, len); + + opt = &p->opt[offset]; + // offset--; + if (price2 < opt->price) + { + opt->price = price2; + opt->len = (UInt32)len; + opt->dist = 0; + opt->extra = 1; + } + } + // while (len >= 3); + } + } + } + } + + startLen = 2; /* speed optimization */ + + { + // ---------- REP ---------- + unsigned repIndex = 0; // 17.old + // unsigned repIndex = IsLitState(state) ? 0 : 1; // 18.notused + for (; repIndex < LZMA_NUM_REPS; repIndex++) + { + unsigned len; + UInt32 price; + const Byte *data2 = data - reps[repIndex]; + if (data[0] != data2[0] || data[1] != data2[1]) + continue; + + for (len = 2; len < numAvail && data[len] == data2[len]; len++) + {} + + // if (len < startLen) continue; // 18.new: speed optimization + + { + unsigned offset = cur + len; + if (last < offset) + last = offset; + } + { + unsigned len2 = len; + price = repMatchPrice + GetPrice_PureRep(p, repIndex, state, posState); + do + { + UInt32 price2 = price + GET_PRICE_LEN(&p->repLenEnc, posState, len2); + COptimal *opt = &p->opt[cur + len2]; + if (price2 < opt->price) + { + opt->price = price2; + opt->len = (UInt32)len2; + opt->dist = (UInt32)repIndex; + opt->extra = 0; + } + } + while (--len2 >= 2); + } + + if (repIndex == 0) startLen = len + 1; // 17.old + // startLen = len + 1; // 18.new + + /* if (_maxMode) */ + { + // ---------- REP : LIT : REP_0 ---------- + // numFastBytes + 1 + numFastBytes + + unsigned len2 = len + 1; + unsigned limit = len2 + p->numFastBytes; + if (limit > numAvailFull) + limit = numAvailFull; + + len2 += 2; + if (len2 <= limit) + if (data[len2 - 2] == data2[len2 - 2]) + if (data[len2 - 1] == data2[len2 - 1]) + { + unsigned state2 = kRepNextStates[state]; + unsigned posState2 = (position + len) & p->pbMask; + price += GET_PRICE_LEN(&p->repLenEnc, posState, len) + + GET_PRICE_0(p->isMatch[state2][posState2]) + + LitEnc_Matched_GetPrice(LIT_PROBS(position + len, data[(size_t)len - 1]), + data[len], data2[len], p->ProbPrices); + + // state2 = kLiteralNextStates[state2]; + state2 = kState_LitAfterRep; + posState2 = (posState2 + 1) & p->pbMask; + + + price += GetPrice_Rep_0(p, state2, posState2); + + for (; len2 < limit && data[len2] == data2[len2]; len2++) + {} + + len2 -= len; + // if (len2 >= 3) + { + { + unsigned offset = cur + len + len2; + + if (last < offset) + last = offset; + // do + { + UInt32 price2; + COptimal *opt; + len2--; + // price2 = price + GetPrice_Len_Rep_0(p, len2, state2, posState2); + price2 = price + GET_PRICE_LEN(&p->repLenEnc, posState2, len2); + + opt = &p->opt[offset]; + // offset--; + if (price2 < opt->price) + { + opt->price = price2; + opt->len = (UInt32)len2; + opt->extra = (CExtra)(len + 1); + opt->dist = (UInt32)repIndex; + } + } + // while (len2 >= 3); + } + } + } + } + } + } + + + // ---------- MATCH ---------- + /* for (unsigned len = 2; len <= newLen; len++) */ + if (newLen > numAvail) + { + newLen = numAvail; + for (numPairs = 0; newLen > MATCHES[numPairs]; numPairs += 2); + MATCHES[numPairs] = (UInt32)newLen; + numPairs += 2; + } + + // startLen = 2; /* speed optimization */ + + if (newLen >= startLen) + { + UInt32 normalMatchPrice = matchPrice + GET_PRICE_0(p->isRep[state]); + UInt32 dist; + unsigned offs, posSlot, len; + + { + unsigned offset = cur + newLen; + if (last < offset) + last = offset; + } + + offs = 0; + while (startLen > MATCHES[offs]) + offs += 2; + dist = MATCHES[(size_t)offs + 1]; + + // if (dist >= kNumFullDistances) + GetPosSlot2(dist, posSlot) + + for (len = /*2*/ startLen; ; len++) + { + UInt32 price = normalMatchPrice + GET_PRICE_LEN(&p->lenEnc, posState, len); + { + COptimal *opt; + unsigned lenNorm = len - 2; + lenNorm = GetLenToPosState2(lenNorm); + if (dist < kNumFullDistances) + price += p->distancesPrices[lenNorm][dist & (kNumFullDistances - 1)]; + else + price += p->posSlotPrices[lenNorm][posSlot] + p->alignPrices[dist & kAlignMask]; + + opt = &p->opt[cur + len]; + if (price < opt->price) + { + opt->price = price; + opt->len = (UInt32)len; + opt->dist = dist + LZMA_NUM_REPS; + opt->extra = 0; + } + } + + if (len == MATCHES[offs]) + { + // if (p->_maxMode) { + // MATCH : LIT : REP_0 + + const Byte *data2 = data - dist - 1; + unsigned len2 = len + 1; + unsigned limit = len2 + p->numFastBytes; + if (limit > numAvailFull) + limit = numAvailFull; + + len2 += 2; + if (len2 <= limit) + if (data[len2 - 2] == data2[len2 - 2]) + if (data[len2 - 1] == data2[len2 - 1]) + { + for (; len2 < limit && data[len2] == data2[len2]; len2++) + {} + + len2 -= len; + + // if (len2 >= 3) + { + unsigned state2 = kMatchNextStates[state]; + unsigned posState2 = (position + len) & p->pbMask; + unsigned offset; + price += GET_PRICE_0(p->isMatch[state2][posState2]); + price += LitEnc_Matched_GetPrice(LIT_PROBS(position + len, data[(size_t)len - 1]), + data[len], data2[len], p->ProbPrices); + + // state2 = kLiteralNextStates[state2]; + state2 = kState_LitAfterMatch; + + posState2 = (posState2 + 1) & p->pbMask; + price += GetPrice_Rep_0(p, state2, posState2); + + offset = cur + len + len2; + + if (last < offset) + last = offset; + // do + { + UInt32 price2; + COptimal *opt; + len2--; + // price2 = price + GetPrice_Len_Rep_0(p, len2, state2, posState2); + price2 = price + GET_PRICE_LEN(&p->repLenEnc, posState2, len2); + opt = &p->opt[offset]; + // offset--; + if (price2 < opt->price) + { + opt->price = price2; + opt->len = (UInt32)len2; + opt->extra = (CExtra)(len + 1); + opt->dist = dist + LZMA_NUM_REPS; + } + } + // while (len2 >= 3); + } + + } + + offs += 2; + if (offs == numPairs) + break; + dist = MATCHES[(size_t)offs + 1]; + // if (dist >= kNumFullDistances) + GetPosSlot2(dist, posSlot) + } + } + } + } + + do + p->opt[last].price = kInfinityPrice; + while (--last); + + return Backward(p, cur); +} + + + +#define ChangePair(smallDist, bigDist) (((bigDist) >> 7) > (smallDist)) + + + +static unsigned GetOptimumFast(CLzmaEnc *p) +{ + UInt32 numAvail, mainDist; + unsigned mainLen, numPairs, repIndex, repLen, i; + const Byte *data; + + if (p->additionalOffset == 0) + mainLen = ReadMatchDistances(p, &numPairs); + else + { + mainLen = p->longestMatchLen; + numPairs = p->numPairs; + } + + numAvail = p->numAvail; + p->backRes = MARK_LIT; + if (numAvail < 2) + return 1; + // if (mainLen < 2 && p->state == 0) return 1; // 18.06.notused + if (numAvail > LZMA_MATCH_LEN_MAX) + numAvail = LZMA_MATCH_LEN_MAX; + data = p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - 1; + repLen = repIndex = 0; + + for (i = 0; i < LZMA_NUM_REPS; i++) + { + unsigned len; + const Byte *data2 = data - p->reps[i]; + if (data[0] != data2[0] || data[1] != data2[1]) + continue; + for (len = 2; len < numAvail && data[len] == data2[len]; len++) + {} + if (len >= p->numFastBytes) + { + p->backRes = (UInt32)i; + MOVE_POS(p, len - 1) + return len; + } + if (len > repLen) + { + repIndex = i; + repLen = len; + } + } + + if (mainLen >= p->numFastBytes) + { + p->backRes = p->matches[(size_t)numPairs - 1] + LZMA_NUM_REPS; + MOVE_POS(p, mainLen - 1) + return mainLen; + } + + mainDist = 0; /* for GCC */ + + if (mainLen >= 2) + { + mainDist = p->matches[(size_t)numPairs - 1]; + while (numPairs > 2) + { + UInt32 dist2; + if (mainLen != p->matches[(size_t)numPairs - 4] + 1) + break; + dist2 = p->matches[(size_t)numPairs - 3]; + if (!ChangePair(dist2, mainDist)) + break; + numPairs -= 2; + mainLen--; + mainDist = dist2; + } + if (mainLen == 2 && mainDist >= 0x80) + mainLen = 1; + } + + if (repLen >= 2) + if ( repLen + 1 >= mainLen + || (repLen + 2 >= mainLen && mainDist >= (1 << 9)) + || (repLen + 3 >= mainLen && mainDist >= (1 << 15))) + { + p->backRes = (UInt32)repIndex; + MOVE_POS(p, repLen - 1) + return repLen; + } + + if (mainLen < 2 || numAvail <= 2) + return 1; + + { + unsigned len1 = ReadMatchDistances(p, &p->numPairs); + p->longestMatchLen = len1; + + if (len1 >= 2) + { + UInt32 newDist = p->matches[(size_t)p->numPairs - 1]; + if ( (len1 >= mainLen && newDist < mainDist) + || (len1 == mainLen + 1 && !ChangePair(mainDist, newDist)) + || (len1 > mainLen + 1) + || (len1 + 1 >= mainLen && mainLen >= 3 && ChangePair(newDist, mainDist))) + return 1; + } + } + + data = p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - 1; + + for (i = 0; i < LZMA_NUM_REPS; i++) + { + unsigned len, limit; + const Byte *data2 = data - p->reps[i]; + if (data[0] != data2[0] || data[1] != data2[1]) + continue; + limit = mainLen - 1; + for (len = 2;; len++) + { + if (len >= limit) + return 1; + if (data[len] != data2[len]) + break; + } + } + + p->backRes = mainDist + LZMA_NUM_REPS; + if (mainLen != 2) + { + MOVE_POS(p, mainLen - 2) + } + return mainLen; +} + + + + +static void WriteEndMarker(CLzmaEnc *p, unsigned posState) +{ + UInt32 range; + range = p->rc.range; + { + UInt32 ttt, newBound; + CLzmaProb *prob = &p->isMatch[p->state][posState]; + RC_BIT_PRE(&p->rc, prob) + RC_BIT_1(&p->rc, prob) + prob = &p->isRep[p->state]; + RC_BIT_PRE(&p->rc, prob) + RC_BIT_0(&p->rc, prob) + } + p->state = kMatchNextStates[p->state]; + + p->rc.range = range; + LenEnc_Encode(&p->lenProbs, &p->rc, 0, posState); + range = p->rc.range; + + { + // RcTree_Encode_PosSlot(&p->rc, p->posSlotEncoder[0], (1 << kNumPosSlotBits) - 1); + CLzmaProb *probs = p->posSlotEncoder[0]; + unsigned m = 1; + do + { + UInt32 ttt, newBound; + RC_BIT_PRE(p, probs + m) + RC_BIT_1(&p->rc, probs + m) + m = (m << 1) + 1; + } + while (m < (1 << kNumPosSlotBits)); + } + { + // RangeEnc_EncodeDirectBits(&p->rc, ((UInt32)1 << (30 - kNumAlignBits)) - 1, 30 - kNumAlignBits); UInt32 range = p->range; + unsigned numBits = 30 - kNumAlignBits; + do + { + range >>= 1; + p->rc.low += range; + RC_NORM(&p->rc) + } + while (--numBits); + } + + { + // RcTree_ReverseEncode(&p->rc, p->posAlignEncoder, kNumAlignBits, kAlignMask); + CLzmaProb *probs = p->posAlignEncoder; + unsigned m = 1; + do + { + UInt32 ttt, newBound; + RC_BIT_PRE(p, probs + m) + RC_BIT_1(&p->rc, probs + m) + m = (m << 1) + 1; + } + while (m < kAlignTableSize); + } + p->rc.range = range; +} + + +static SRes CheckErrors(CLzmaEnc *p) +{ + if (p->result != SZ_OK) + return p->result; + if (p->rc.res != SZ_OK) + p->result = SZ_ERROR_WRITE; + + #ifndef Z7_ST + if ( + // p->mf_Failure || + (p->mtMode && + ( // p->matchFinderMt.failure_LZ_LZ || + p->matchFinderMt.failure_LZ_BT)) + ) + { + p->result = MY_HRES_ERROR_INTERNAL_ERROR; + // printf("\nCheckErrors p->matchFinderMt.failureLZ\n"); + } + #endif + + if (MFB.result != SZ_OK) + p->result = SZ_ERROR_READ; + + if (p->result != SZ_OK) + p->finished = True; + return p->result; +} + + +Z7_NO_INLINE static SRes Flush(CLzmaEnc *p, UInt32 nowPos) +{ + /* ReleaseMFStream(); */ + p->finished = True; + if (p->writeEndMark) + WriteEndMarker(p, nowPos & p->pbMask); + RangeEnc_FlushData(&p->rc); + RangeEnc_FlushStream(&p->rc); + return CheckErrors(p); +} + + +Z7_NO_INLINE static void FillAlignPrices(CLzmaEnc *p) +{ + unsigned i; + const CProbPrice *ProbPrices = p->ProbPrices; + const CLzmaProb *probs = p->posAlignEncoder; + // p->alignPriceCount = 0; + for (i = 0; i < kAlignTableSize / 2; i++) + { + UInt32 price = 0; + unsigned sym = i; + unsigned m = 1; + unsigned bit; + UInt32 prob; + bit = sym & 1; sym >>= 1; price += GET_PRICEa(probs[m], bit); m = (m << 1) + bit; + bit = sym & 1; sym >>= 1; price += GET_PRICEa(probs[m], bit); m = (m << 1) + bit; + bit = sym & 1; sym >>= 1; price += GET_PRICEa(probs[m], bit); m = (m << 1) + bit; + prob = probs[m]; + p->alignPrices[i ] = price + GET_PRICEa_0(prob); + p->alignPrices[i + 8] = price + GET_PRICEa_1(prob); + // p->alignPrices[i] = RcTree_ReverseGetPrice(p->posAlignEncoder, kNumAlignBits, i, p->ProbPrices); + } +} + + +Z7_NO_INLINE static void FillDistancesPrices(CLzmaEnc *p) +{ + // int y; for (y = 0; y < 100; y++) { + + UInt32 tempPrices[kNumFullDistances]; + unsigned i, lps; + + const CProbPrice *ProbPrices = p->ProbPrices; + p->matchPriceCount = 0; + + for (i = kStartPosModelIndex / 2; i < kNumFullDistances / 2; i++) + { + unsigned posSlot = GetPosSlot1(i); + unsigned footerBits = (posSlot >> 1) - 1; + unsigned base = ((2 | (posSlot & 1)) << footerBits); + const CLzmaProb *probs = p->posEncoders + (size_t)base * 2; + // tempPrices[i] = RcTree_ReverseGetPrice(p->posEncoders + base, footerBits, i - base, p->ProbPrices); + UInt32 price = 0; + unsigned m = 1; + unsigned sym = i; + unsigned offset = (unsigned)1 << footerBits; + base += i; + + if (footerBits) + do + { + unsigned bit = sym & 1; + sym >>= 1; + price += GET_PRICEa(probs[m], bit); + m = (m << 1) + bit; + } + while (--footerBits); + + { + unsigned prob = probs[m]; + tempPrices[base ] = price + GET_PRICEa_0(prob); + tempPrices[base + offset] = price + GET_PRICEa_1(prob); + } + } + + for (lps = 0; lps < kNumLenToPosStates; lps++) + { + unsigned slot; + unsigned distTableSize2 = (p->distTableSize + 1) >> 1; + UInt32 *posSlotPrices = p->posSlotPrices[lps]; + const CLzmaProb *probs = p->posSlotEncoder[lps]; + + for (slot = 0; slot < distTableSize2; slot++) + { + // posSlotPrices[slot] = RcTree_GetPrice(encoder, kNumPosSlotBits, slot, p->ProbPrices); + UInt32 price; + unsigned bit; + unsigned sym = slot + (1 << (kNumPosSlotBits - 1)); + unsigned prob; + bit = sym & 1; sym >>= 1; price = GET_PRICEa(probs[sym], bit); + bit = sym & 1; sym >>= 1; price += GET_PRICEa(probs[sym], bit); + bit = sym & 1; sym >>= 1; price += GET_PRICEa(probs[sym], bit); + bit = sym & 1; sym >>= 1; price += GET_PRICEa(probs[sym], bit); + bit = sym & 1; sym >>= 1; price += GET_PRICEa(probs[sym], bit); + prob = probs[(size_t)slot + (1 << (kNumPosSlotBits - 1))]; + posSlotPrices[(size_t)slot * 2 ] = price + GET_PRICEa_0(prob); + posSlotPrices[(size_t)slot * 2 + 1] = price + GET_PRICEa_1(prob); + } + + { + UInt32 delta = ((UInt32)((kEndPosModelIndex / 2 - 1) - kNumAlignBits) << kNumBitPriceShiftBits); + for (slot = kEndPosModelIndex / 2; slot < distTableSize2; slot++) + { + posSlotPrices[(size_t)slot * 2 ] += delta; + posSlotPrices[(size_t)slot * 2 + 1] += delta; + delta += ((UInt32)1 << kNumBitPriceShiftBits); + } + } + + { + UInt32 *dp = p->distancesPrices[lps]; + + dp[0] = posSlotPrices[0]; + dp[1] = posSlotPrices[1]; + dp[2] = posSlotPrices[2]; + dp[3] = posSlotPrices[3]; + + for (i = 4; i < kNumFullDistances; i += 2) + { + UInt32 slotPrice = posSlotPrices[GetPosSlot1(i)]; + dp[i ] = slotPrice + tempPrices[i]; + dp[i + 1] = slotPrice + tempPrices[i + 1]; + } + } + } + // } +} + + + +static void LzmaEnc_Construct(CLzmaEnc *p) +{ + RangeEnc_Construct(&p->rc); + MatchFinder_Construct(&MFB); + + #ifndef Z7_ST + p->matchFinderMt.MatchFinder = &MFB; + MatchFinderMt_Construct(&p->matchFinderMt); + #endif + + { + CLzmaEncProps props; + LzmaEncProps_Init(&props); + LzmaEnc_SetProps((CLzmaEncHandle)(void *)p, &props); + } + + #ifndef LZMA_LOG_BSR + LzmaEnc_FastPosInit(p->g_FastPos); + #endif + + LzmaEnc_InitPriceTables(p->ProbPrices); + p->litProbs = NULL; + p->saveState.litProbs = NULL; +} + +CLzmaEncHandle LzmaEnc_Create(ISzAllocPtr alloc) +{ + void *p; + p = ISzAlloc_Alloc(alloc, sizeof(CLzmaEnc)); + if (p) + LzmaEnc_Construct((CLzmaEnc *)p); + return p; +} + +static void LzmaEnc_FreeLits(CLzmaEnc *p, ISzAllocPtr alloc) +{ + ISzAlloc_Free(alloc, p->litProbs); + ISzAlloc_Free(alloc, p->saveState.litProbs); + p->litProbs = NULL; + p->saveState.litProbs = NULL; +} + +static void LzmaEnc_Destruct(CLzmaEnc *p, ISzAllocPtr alloc, ISzAllocPtr allocBig) +{ + #ifndef Z7_ST + MatchFinderMt_Destruct(&p->matchFinderMt, allocBig); + #endif + + MatchFinder_Free(&MFB, allocBig); + LzmaEnc_FreeLits(p, alloc); + RangeEnc_Free(&p->rc, alloc); +} + +void LzmaEnc_Destroy(CLzmaEncHandle p, ISzAllocPtr alloc, ISzAllocPtr allocBig) +{ + // GET_CLzmaEnc_p + LzmaEnc_Destruct(p, alloc, allocBig); + ISzAlloc_Free(alloc, p); +} + + +Z7_NO_INLINE +static SRes LzmaEnc_CodeOneBlock(CLzmaEnc *p, UInt32 maxPackSize, UInt32 maxUnpackSize) +{ + UInt32 nowPos32, startPos32; + if (p->needInit) + { + #ifndef Z7_ST + if (p->mtMode) + { + RINOK(MatchFinderMt_InitMt(&p->matchFinderMt)) + } + #endif + p->matchFinder.Init(p->matchFinderObj); + p->needInit = 0; + } + + if (p->finished) + return p->result; + RINOK(CheckErrors(p)) + + nowPos32 = (UInt32)p->nowPos64; + startPos32 = nowPos32; + + if (p->nowPos64 == 0) + { + unsigned numPairs; + Byte curByte; + if (p->matchFinder.GetNumAvailableBytes(p->matchFinderObj) == 0) + return Flush(p, nowPos32); + ReadMatchDistances(p, &numPairs); + RangeEnc_EncodeBit_0(&p->rc, &p->isMatch[kState_Start][0]); + // p->state = kLiteralNextStates[p->state]; + curByte = *(p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - p->additionalOffset); + LitEnc_Encode(&p->rc, p->litProbs, curByte); + p->additionalOffset--; + nowPos32++; + } + + if (p->matchFinder.GetNumAvailableBytes(p->matchFinderObj) != 0) + + for (;;) + { + UInt32 dist; + unsigned len, posState; + UInt32 range, ttt, newBound; + CLzmaProb *probs; + + if (p->fastMode) + len = GetOptimumFast(p); + else + { + unsigned oci = p->optCur; + if (p->optEnd == oci) + len = GetOptimum(p, nowPos32); + else + { + const COptimal *opt = &p->opt[oci]; + len = opt->len; + p->backRes = opt->dist; + p->optCur = oci + 1; + } + } + + posState = (unsigned)nowPos32 & p->pbMask; + range = p->rc.range; + probs = &p->isMatch[p->state][posState]; + + RC_BIT_PRE(&p->rc, probs) + + dist = p->backRes; + + #ifdef SHOW_STAT2 + printf("\n pos = %6X, len = %3u pos = %6u", nowPos32, len, dist); + #endif + + if (dist == MARK_LIT) + { + Byte curByte; + const Byte *data; + unsigned state; + + RC_BIT_0(&p->rc, probs) + p->rc.range = range; + data = p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - p->additionalOffset; + probs = LIT_PROBS(nowPos32, *(data - 1)); + curByte = *data; + state = p->state; + p->state = kLiteralNextStates[state]; + if (IsLitState(state)) + LitEnc_Encode(&p->rc, probs, curByte); + else + LitEnc_EncodeMatched(&p->rc, probs, curByte, *(data - p->reps[0])); + } + else + { + RC_BIT_1(&p->rc, probs) + probs = &p->isRep[p->state]; + RC_BIT_PRE(&p->rc, probs) + + if (dist < LZMA_NUM_REPS) + { + RC_BIT_1(&p->rc, probs) + probs = &p->isRepG0[p->state]; + RC_BIT_PRE(&p->rc, probs) + if (dist == 0) + { + RC_BIT_0(&p->rc, probs) + probs = &p->isRep0Long[p->state][posState]; + RC_BIT_PRE(&p->rc, probs) + if (len != 1) + { + RC_BIT_1_BASE(&p->rc, probs) + } + else + { + RC_BIT_0_BASE(&p->rc, probs) + p->state = kShortRepNextStates[p->state]; + } + } + else + { + RC_BIT_1(&p->rc, probs) + probs = &p->isRepG1[p->state]; + RC_BIT_PRE(&p->rc, probs) + if (dist == 1) + { + RC_BIT_0_BASE(&p->rc, probs) + dist = p->reps[1]; + } + else + { + RC_BIT_1(&p->rc, probs) + probs = &p->isRepG2[p->state]; + RC_BIT_PRE(&p->rc, probs) + if (dist == 2) + { + RC_BIT_0_BASE(&p->rc, probs) + dist = p->reps[2]; + } + else + { + RC_BIT_1_BASE(&p->rc, probs) + dist = p->reps[3]; + p->reps[3] = p->reps[2]; + } + p->reps[2] = p->reps[1]; + } + p->reps[1] = p->reps[0]; + p->reps[0] = dist; + } + + RC_NORM(&p->rc) + + p->rc.range = range; + + if (len != 1) + { + LenEnc_Encode(&p->repLenProbs, &p->rc, len - LZMA_MATCH_LEN_MIN, posState); + --p->repLenEncCounter; + p->state = kRepNextStates[p->state]; + } + } + else + { + unsigned posSlot; + RC_BIT_0(&p->rc, probs) + p->rc.range = range; + p->state = kMatchNextStates[p->state]; + + LenEnc_Encode(&p->lenProbs, &p->rc, len - LZMA_MATCH_LEN_MIN, posState); + // --p->lenEnc.counter; + + dist -= LZMA_NUM_REPS; + p->reps[3] = p->reps[2]; + p->reps[2] = p->reps[1]; + p->reps[1] = p->reps[0]; + p->reps[0] = dist + 1; + + p->matchPriceCount++; + GetPosSlot(dist, posSlot) + // RcTree_Encode_PosSlot(&p->rc, p->posSlotEncoder[GetLenToPosState(len)], posSlot); + { + UInt32 sym = (UInt32)posSlot + (1 << kNumPosSlotBits); + range = p->rc.range; + probs = p->posSlotEncoder[GetLenToPosState(len)]; + do + { + CLzmaProb *prob = probs + (sym >> kNumPosSlotBits); + UInt32 bit = (sym >> (kNumPosSlotBits - 1)) & 1; + sym <<= 1; + RC_BIT(&p->rc, prob, bit) + } + while (sym < (1 << kNumPosSlotBits * 2)); + p->rc.range = range; + } + + if (dist >= kStartPosModelIndex) + { + unsigned footerBits = ((posSlot >> 1) - 1); + + if (dist < kNumFullDistances) + { + unsigned base = ((2 | (posSlot & 1)) << footerBits); + RcTree_ReverseEncode(&p->rc, p->posEncoders + base, footerBits, (unsigned)(dist /* - base */)); + } + else + { + UInt32 pos2 = (dist | 0xF) << (32 - footerBits); + range = p->rc.range; + // RangeEnc_EncodeDirectBits(&p->rc, posReduced >> kNumAlignBits, footerBits - kNumAlignBits); + /* + do + { + range >>= 1; + p->rc.low += range & (0 - ((dist >> --footerBits) & 1)); + RC_NORM(&p->rc) + } + while (footerBits > kNumAlignBits); + */ + do + { + range >>= 1; + p->rc.low += range & (0 - (pos2 >> 31)); + pos2 += pos2; + RC_NORM(&p->rc) + } + while (pos2 != 0xF0000000); + + + // RcTree_ReverseEncode(&p->rc, p->posAlignEncoder, kNumAlignBits, posReduced & kAlignMask); + + { + unsigned m = 1; + unsigned bit; + bit = dist & 1; dist >>= 1; RC_BIT(&p->rc, p->posAlignEncoder + m, bit) m = (m << 1) + bit; + bit = dist & 1; dist >>= 1; RC_BIT(&p->rc, p->posAlignEncoder + m, bit) m = (m << 1) + bit; + bit = dist & 1; dist >>= 1; RC_BIT(&p->rc, p->posAlignEncoder + m, bit) m = (m << 1) + bit; + bit = dist & 1; RC_BIT(&p->rc, p->posAlignEncoder + m, bit) + p->rc.range = range; + // p->alignPriceCount++; + } + } + } + } + } + + nowPos32 += (UInt32)len; + p->additionalOffset -= len; + + if (p->additionalOffset == 0) + { + UInt32 processed; + + if (!p->fastMode) + { + /* + if (p->alignPriceCount >= 16) // kAlignTableSize + FillAlignPrices(p); + if (p->matchPriceCount >= 128) + FillDistancesPrices(p); + if (p->lenEnc.counter <= 0) + LenPriceEnc_UpdateTables(&p->lenEnc, 1 << p->pb, &p->lenProbs, p->ProbPrices); + */ + if (p->matchPriceCount >= 64) + { + FillAlignPrices(p); + // { int y; for (y = 0; y < 100; y++) { + FillDistancesPrices(p); + // }} + LenPriceEnc_UpdateTables(&p->lenEnc, (unsigned)1 << p->pb, &p->lenProbs, p->ProbPrices); + } + if (p->repLenEncCounter <= 0) + { + p->repLenEncCounter = REP_LEN_COUNT; + LenPriceEnc_UpdateTables(&p->repLenEnc, (unsigned)1 << p->pb, &p->repLenProbs, p->ProbPrices); + } + } + + if (p->matchFinder.GetNumAvailableBytes(p->matchFinderObj) == 0) + break; + processed = nowPos32 - startPos32; + + if (maxPackSize) + { + if (processed + kNumOpts + 300 >= maxUnpackSize + || RangeEnc_GetProcessed_sizet(&p->rc) + kPackReserve >= maxPackSize) + break; + } + else if (processed >= (1 << 17)) + { + p->nowPos64 += nowPos32 - startPos32; + return CheckErrors(p); + } + } + } + + p->nowPos64 += nowPos32 - startPos32; + return Flush(p, nowPos32); +} + + + +#define kBigHashDicLimit ((UInt32)1 << 24) + +static SRes LzmaEnc_Alloc(CLzmaEnc *p, UInt32 keepWindowSize, ISzAllocPtr alloc, ISzAllocPtr allocBig) +{ + UInt32 beforeSize = kNumOpts; + UInt32 dictSize; + + if (!RangeEnc_Alloc(&p->rc, alloc)) + return SZ_ERROR_MEM; + + #ifndef Z7_ST + p->mtMode = (p->multiThread && !p->fastMode && (MFB.btMode != 0)); + #endif + + { + unsigned lclp = p->lc + p->lp; + if (!p->litProbs || !p->saveState.litProbs || p->lclp != lclp) + { + LzmaEnc_FreeLits(p, alloc); + p->litProbs = (CLzmaProb *)ISzAlloc_Alloc(alloc, ((UInt32)0x300 << lclp) * sizeof(CLzmaProb)); + p->saveState.litProbs = (CLzmaProb *)ISzAlloc_Alloc(alloc, ((UInt32)0x300 << lclp) * sizeof(CLzmaProb)); + if (!p->litProbs || !p->saveState.litProbs) + { + LzmaEnc_FreeLits(p, alloc); + return SZ_ERROR_MEM; + } + p->lclp = lclp; + } + } + + MFB.bigHash = (Byte)(p->dictSize > kBigHashDicLimit ? 1 : 0); + + + dictSize = p->dictSize; + if (dictSize == ((UInt32)2 << 30) || + dictSize == ((UInt32)3 << 30)) + { + /* 21.03 : here we reduce the dictionary for 2 reasons: + 1) we don't want 32-bit back_distance matches in decoder for 2 GB dictionary. + 2) we want to elimate useless last MatchFinder_Normalize3() for corner cases, + where data size is aligned for 1 GB: 5/6/8 GB. + That reducing must be >= 1 for such corner cases. */ + dictSize -= 1; + } + + if (beforeSize + dictSize < keepWindowSize) + beforeSize = keepWindowSize - dictSize; + + /* in worst case we can look ahead for + max(LZMA_MATCH_LEN_MAX, numFastBytes + 1 + numFastBytes) bytes. + we send larger value for (keepAfter) to MantchFinder_Create(): + (numFastBytes + LZMA_MATCH_LEN_MAX + 1) + */ + + #ifndef Z7_ST + if (p->mtMode) + { + RINOK(MatchFinderMt_Create(&p->matchFinderMt, dictSize, beforeSize, + p->numFastBytes, LZMA_MATCH_LEN_MAX + 1 /* 18.04 */ + , allocBig)) + p->matchFinderObj = &p->matchFinderMt; + MFB.bigHash = (Byte)(MFB.hashMask >= 0xFFFFFF ? 1 : 0); + MatchFinderMt_CreateVTable(&p->matchFinderMt, &p->matchFinder); + } + else + #endif + { + if (!MatchFinder_Create(&MFB, dictSize, beforeSize, + p->numFastBytes, LZMA_MATCH_LEN_MAX + 1 /* 21.03 */ + , allocBig)) + return SZ_ERROR_MEM; + p->matchFinderObj = &MFB; + MatchFinder_CreateVTable(&MFB, &p->matchFinder); + } + + return SZ_OK; +} + +static void LzmaEnc_Init(CLzmaEnc *p) +{ + unsigned i; + p->state = 0; + p->reps[0] = + p->reps[1] = + p->reps[2] = + p->reps[3] = 1; + + RangeEnc_Init(&p->rc); + + for (i = 0; i < (1 << kNumAlignBits); i++) + p->posAlignEncoder[i] = kProbInitValue; + + for (i = 0; i < kNumStates; i++) + { + unsigned j; + for (j = 0; j < LZMA_NUM_PB_STATES_MAX; j++) + { + p->isMatch[i][j] = kProbInitValue; + p->isRep0Long[i][j] = kProbInitValue; + } + p->isRep[i] = kProbInitValue; + p->isRepG0[i] = kProbInitValue; + p->isRepG1[i] = kProbInitValue; + p->isRepG2[i] = kProbInitValue; + } + + { + for (i = 0; i < kNumLenToPosStates; i++) + { + CLzmaProb *probs = p->posSlotEncoder[i]; + unsigned j; + for (j = 0; j < (1 << kNumPosSlotBits); j++) + probs[j] = kProbInitValue; + } + } + { + for (i = 0; i < kNumFullDistances; i++) + p->posEncoders[i] = kProbInitValue; + } + + { + UInt32 num = (UInt32)0x300 << (p->lp + p->lc); + UInt32 k; + CLzmaProb *probs = p->litProbs; + for (k = 0; k < num; k++) + probs[k] = kProbInitValue; + } + + + LenEnc_Init(&p->lenProbs); + LenEnc_Init(&p->repLenProbs); + + p->optEnd = 0; + p->optCur = 0; + + { + for (i = 0; i < kNumOpts; i++) + p->opt[i].price = kInfinityPrice; + } + + p->additionalOffset = 0; + + p->pbMask = ((unsigned)1 << p->pb) - 1; + p->lpMask = ((UInt32)0x100 << p->lp) - ((unsigned)0x100 >> p->lc); + + // p->mf_Failure = False; +} + + +static void LzmaEnc_InitPrices(CLzmaEnc *p) +{ + if (!p->fastMode) + { + FillDistancesPrices(p); + FillAlignPrices(p); + } + + p->lenEnc.tableSize = + p->repLenEnc.tableSize = + p->numFastBytes + 1 - LZMA_MATCH_LEN_MIN; + + p->repLenEncCounter = REP_LEN_COUNT; + + LenPriceEnc_UpdateTables(&p->lenEnc, (unsigned)1 << p->pb, &p->lenProbs, p->ProbPrices); + LenPriceEnc_UpdateTables(&p->repLenEnc, (unsigned)1 << p->pb, &p->repLenProbs, p->ProbPrices); +} + +static SRes LzmaEnc_AllocAndInit(CLzmaEnc *p, UInt32 keepWindowSize, ISzAllocPtr alloc, ISzAllocPtr allocBig) +{ + unsigned i; + for (i = kEndPosModelIndex / 2; i < kDicLogSizeMax; i++) + if (p->dictSize <= ((UInt32)1 << i)) + break; + p->distTableSize = i * 2; + + p->finished = False; + p->result = SZ_OK; + p->nowPos64 = 0; + p->needInit = 1; + RINOK(LzmaEnc_Alloc(p, keepWindowSize, alloc, allocBig)) + LzmaEnc_Init(p); + LzmaEnc_InitPrices(p); + return SZ_OK; +} + +static SRes LzmaEnc_Prepare(CLzmaEncHandle p, + ISeqOutStreamPtr outStream, + ISeqInStreamPtr inStream, + ISzAllocPtr alloc, ISzAllocPtr allocBig) +{ + // GET_CLzmaEnc_p + MatchFinder_SET_STREAM(&MFB, inStream) + p->rc.outStream = outStream; + return LzmaEnc_AllocAndInit(p, 0, alloc, allocBig); +} + +SRes LzmaEnc_PrepareForLzma2(CLzmaEncHandle p, + ISeqInStreamPtr inStream, UInt32 keepWindowSize, + ISzAllocPtr alloc, ISzAllocPtr allocBig) +{ + // GET_CLzmaEnc_p + MatchFinder_SET_STREAM(&MFB, inStream) + return LzmaEnc_AllocAndInit(p, keepWindowSize, alloc, allocBig); +} + +SRes LzmaEnc_MemPrepare(CLzmaEncHandle p, + const Byte *src, SizeT srcLen, + UInt32 keepWindowSize, + ISzAllocPtr alloc, ISzAllocPtr allocBig) +{ + // GET_CLzmaEnc_p + MatchFinder_SET_DIRECT_INPUT_BUF(&MFB, src, srcLen) + LzmaEnc_SetDataSize(p, srcLen); + return LzmaEnc_AllocAndInit(p, keepWindowSize, alloc, allocBig); +} + +void LzmaEnc_Finish(CLzmaEncHandle p) +{ + #ifndef Z7_ST + // GET_CLzmaEnc_p + if (p->mtMode) + MatchFinderMt_ReleaseStream(&p->matchFinderMt); + #else + UNUSED_VAR(p) + #endif +} + + +typedef struct +{ + ISeqOutStream vt; + Byte *data; + size_t rem; + BoolInt overflow; +} CLzmaEnc_SeqOutStreamBuf; + +static size_t SeqOutStreamBuf_Write(ISeqOutStreamPtr pp, const void *data, size_t size) +{ + Z7_CONTAINER_FROM_VTBL_TO_DECL_VAR_pp_vt_p(CLzmaEnc_SeqOutStreamBuf) + if (p->rem < size) + { + size = p->rem; + p->overflow = True; + } + if (size != 0) + { + memcpy(p->data, data, size); + p->rem -= size; + p->data += size; + } + return size; +} + + +/* +UInt32 LzmaEnc_GetNumAvailableBytes(CLzmaEncHandle p) +{ + GET_const_CLzmaEnc_p + return p->matchFinder.GetNumAvailableBytes(p->matchFinderObj); +} +*/ + +const Byte *LzmaEnc_GetCurBuf(CLzmaEncHandle p) +{ + // GET_const_CLzmaEnc_p + return p->matchFinder.GetPointerToCurrentPos(p->matchFinderObj) - p->additionalOffset; +} + + +// (desiredPackSize == 0) is not allowed +SRes LzmaEnc_CodeOneMemBlock(CLzmaEncHandle p, BoolInt reInit, + Byte *dest, size_t *destLen, UInt32 desiredPackSize, UInt32 *unpackSize) +{ + // GET_CLzmaEnc_p + UInt64 nowPos64; + SRes res; + CLzmaEnc_SeqOutStreamBuf outStream; + + outStream.vt.Write = SeqOutStreamBuf_Write; + outStream.data = dest; + outStream.rem = *destLen; + outStream.overflow = False; + + p->writeEndMark = False; + p->finished = False; + p->result = SZ_OK; + + if (reInit) + LzmaEnc_Init(p); + LzmaEnc_InitPrices(p); + RangeEnc_Init(&p->rc); + p->rc.outStream = &outStream.vt; + nowPos64 = p->nowPos64; + + res = LzmaEnc_CodeOneBlock(p, desiredPackSize, *unpackSize); + + *unpackSize = (UInt32)(p->nowPos64 - nowPos64); + *destLen -= outStream.rem; + if (outStream.overflow) + return SZ_ERROR_OUTPUT_EOF; + + return res; +} + + +Z7_NO_INLINE +static SRes LzmaEnc_Encode2(CLzmaEnc *p, ICompressProgressPtr progress) +{ + SRes res = SZ_OK; + + #ifndef Z7_ST + Byte allocaDummy[0x300]; + allocaDummy[0] = 0; + allocaDummy[1] = allocaDummy[0]; + #endif + + for (;;) + { + res = LzmaEnc_CodeOneBlock(p, 0, 0); + if (res != SZ_OK || p->finished) + break; + if (progress) + { + res = ICompressProgress_Progress(progress, p->nowPos64, RangeEnc_GetProcessed(&p->rc)); + if (res != SZ_OK) + { + res = SZ_ERROR_PROGRESS; + break; + } + } + } + + LzmaEnc_Finish((CLzmaEncHandle)(void *)p); + + /* + if (res == SZ_OK && !Inline_MatchFinder_IsFinishedOK(&MFB)) + res = SZ_ERROR_FAIL; + } + */ + + return res; +} + + +SRes LzmaEnc_Encode(CLzmaEncHandle p, ISeqOutStreamPtr outStream, ISeqInStreamPtr inStream, ICompressProgressPtr progress, + ISzAllocPtr alloc, ISzAllocPtr allocBig) +{ + // GET_CLzmaEnc_p + RINOK(LzmaEnc_Prepare(p, outStream, inStream, alloc, allocBig)) + return LzmaEnc_Encode2(p, progress); +} + + +SRes LzmaEnc_WriteProperties(CLzmaEncHandle p, Byte *props, SizeT *size) +{ + if (*size < LZMA_PROPS_SIZE) + return SZ_ERROR_PARAM; + *size = LZMA_PROPS_SIZE; + { + // GET_CLzmaEnc_p + const UInt32 dictSize = p->dictSize; + UInt32 v; + props[0] = (Byte)((p->pb * 5 + p->lp) * 9 + p->lc); + + // we write aligned dictionary value to properties for lzma decoder + if (dictSize >= ((UInt32)1 << 21)) + { + const UInt32 kDictMask = ((UInt32)1 << 20) - 1; + v = (dictSize + kDictMask) & ~kDictMask; + if (v < dictSize) + v = dictSize; + } + else + { + unsigned i = 11 * 2; + do + { + v = (UInt32)(2 + (i & 1)) << (i >> 1); + i++; + } + while (v < dictSize); + } + + SetUi32(props + 1, v) + return SZ_OK; + } +} + + +unsigned LzmaEnc_IsWriteEndMark(CLzmaEncHandle p) +{ + // GET_CLzmaEnc_p + return (unsigned)p->writeEndMark; +} + + +SRes LzmaEnc_MemEncode(CLzmaEncHandle p, Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen, + int writeEndMark, ICompressProgressPtr progress, ISzAllocPtr alloc, ISzAllocPtr allocBig) +{ + SRes res; + // GET_CLzmaEnc_p + + CLzmaEnc_SeqOutStreamBuf outStream; + + outStream.vt.Write = SeqOutStreamBuf_Write; + outStream.data = dest; + outStream.rem = *destLen; + outStream.overflow = False; + + p->writeEndMark = writeEndMark; + p->rc.outStream = &outStream.vt; + + res = LzmaEnc_MemPrepare(p, src, srcLen, 0, alloc, allocBig); + + if (res == SZ_OK) + { + res = LzmaEnc_Encode2(p, progress); + if (res == SZ_OK && p->nowPos64 != srcLen) + res = SZ_ERROR_FAIL; + } + + *destLen -= (SizeT)outStream.rem; + if (outStream.overflow) + return SZ_ERROR_OUTPUT_EOF; + return res; +} + + +SRes LzmaEncode(Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen, + const CLzmaEncProps *props, Byte *propsEncoded, SizeT *propsSize, int writeEndMark, + ICompressProgressPtr progress, ISzAllocPtr alloc, ISzAllocPtr allocBig) +{ + CLzmaEncHandle p = LzmaEnc_Create(alloc); + SRes res; + if (!p) + return SZ_ERROR_MEM; + + res = LzmaEnc_SetProps(p, props); + if (res == SZ_OK) + { + res = LzmaEnc_WriteProperties(p, propsEncoded, propsSize); + if (res == SZ_OK) + res = LzmaEnc_MemEncode(p, dest, destLen, src, srcLen, + writeEndMark, progress, alloc, allocBig); + } + + LzmaEnc_Destroy(p, alloc, allocBig); + return res; +} + + +/* +#ifndef Z7_ST +void LzmaEnc_GetLzThreads(CLzmaEncHandle p, HANDLE lz_threads[2]) +{ + GET_const_CLzmaEnc_p + lz_threads[0] = p->matchFinderMt.hashSync.thread; + lz_threads[1] = p->matchFinderMt.btSync.thread; +} +#endif +*/ diff --git a/project/src/liblzma/LzmaEnc.h b/project/src/liblzma/LzmaEnc.h new file mode 100644 index 0000000..08711cb --- /dev/null +++ b/project/src/liblzma/LzmaEnc.h @@ -0,0 +1,83 @@ +/* LzmaEnc.h -- LZMA Encoder +2023-04-13 : Igor Pavlov : Public domain */ + +#ifndef ZIP7_INC_LZMA_ENC_H +#define ZIP7_INC_LZMA_ENC_H + +#include "7zTypes.h" + +EXTERN_C_BEGIN + +#define LZMA_PROPS_SIZE 5 + +typedef struct +{ + int level; /* 0 <= level <= 9 */ + UInt32 dictSize; /* (1 << 12) <= dictSize <= (1 << 27) for 32-bit version + (1 << 12) <= dictSize <= (3 << 29) for 64-bit version + default = (1 << 24) */ + int lc; /* 0 <= lc <= 8, default = 3 */ + int lp; /* 0 <= lp <= 4, default = 0 */ + int pb; /* 0 <= pb <= 4, default = 2 */ + int algo; /* 0 - fast, 1 - normal, default = 1 */ + int fb; /* 5 <= fb <= 273, default = 32 */ + int btMode; /* 0 - hashChain Mode, 1 - binTree mode - normal, default = 1 */ + int numHashBytes; /* 2, 3 or 4, default = 4 */ + unsigned numHashOutBits; /* default = ? */ + UInt32 mc; /* 1 <= mc <= (1 << 30), default = 32 */ + unsigned writeEndMark; /* 0 - do not write EOPM, 1 - write EOPM, default = 0 */ + int numThreads; /* 1 or 2, default = 2 */ + + // int _pad; + + UInt64 reduceSize; /* estimated size of data that will be compressed. default = (UInt64)(Int64)-1. + Encoder uses this value to reduce dictionary size */ + + UInt64 affinity; +} CLzmaEncProps; + +void LzmaEncProps_Init(CLzmaEncProps *p); +void LzmaEncProps_Normalize(CLzmaEncProps *p); +UInt32 LzmaEncProps_GetDictSize(const CLzmaEncProps *props2); + + +/* ---------- CLzmaEncHandle Interface ---------- */ + +/* LzmaEnc* functions can return the following exit codes: +SRes: + SZ_OK - OK + SZ_ERROR_MEM - Memory allocation error + SZ_ERROR_PARAM - Incorrect paramater in props + SZ_ERROR_WRITE - ISeqOutStream write callback error + SZ_ERROR_OUTPUT_EOF - output buffer overflow - version with (Byte *) output + SZ_ERROR_PROGRESS - some break from progress callback + SZ_ERROR_THREAD - error in multithreading functions (only for Mt version) +*/ + +typedef struct CLzmaEnc CLzmaEnc; +typedef CLzmaEnc * CLzmaEncHandle; +// Z7_DECLARE_HANDLE(CLzmaEncHandle) + +CLzmaEncHandle LzmaEnc_Create(ISzAllocPtr alloc); +void LzmaEnc_Destroy(CLzmaEncHandle p, ISzAllocPtr alloc, ISzAllocPtr allocBig); + +SRes LzmaEnc_SetProps(CLzmaEncHandle p, const CLzmaEncProps *props); +void LzmaEnc_SetDataSize(CLzmaEncHandle p, UInt64 expectedDataSiize); +SRes LzmaEnc_WriteProperties(CLzmaEncHandle p, Byte *properties, SizeT *size); +unsigned LzmaEnc_IsWriteEndMark(CLzmaEncHandle p); + +SRes LzmaEnc_Encode(CLzmaEncHandle p, ISeqOutStreamPtr outStream, ISeqInStreamPtr inStream, + ICompressProgressPtr progress, ISzAllocPtr alloc, ISzAllocPtr allocBig); +SRes LzmaEnc_MemEncode(CLzmaEncHandle p, Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen, + int writeEndMark, ICompressProgressPtr progress, ISzAllocPtr alloc, ISzAllocPtr allocBig); + + +/* ---------- One Call Interface ---------- */ + +SRes LzmaEncode(Byte *dest, SizeT *destLen, const Byte *src, SizeT srcLen, + const CLzmaEncProps *props, Byte *propsEncoded, SizeT *propsSize, int writeEndMark, + ICompressProgressPtr progress, ISzAllocPtr alloc, ISzAllocPtr allocBig); + +EXTERN_C_END + +#endif diff --git a/project/src/liblzma/Precomp.h b/project/src/liblzma/Precomp.h new file mode 100644 index 0000000..dc476aa --- /dev/null +++ b/project/src/liblzma/Precomp.h @@ -0,0 +1,10 @@ +/* Precomp.h -- StdAfx +2023-04-02 : Igor Pavlov : Public domain */ + +#ifndef ZIP7_INC_PRECOMP_H +#define ZIP7_INC_PRECOMP_H + +#include "Compiler.h" +/* #include "7zTypes.h" */ + +#endif diff --git a/project/src/liblzma/Threads.h b/project/src/liblzma/Threads.h new file mode 100644 index 0000000..c6aa6a8 --- /dev/null +++ b/project/src/liblzma/Threads.h @@ -0,0 +1,240 @@ +/* Threads.h -- multithreading library +2023-04-02 : Igor Pavlov : Public domain */ + +#ifndef ZIP7_INC_THREADS_H +#define ZIP7_INC_THREADS_H + +#ifdef _WIN32 +#include "7zWindows.h" + +#else + +#if defined(__linux__) +#if !defined(__APPLE__) && !defined(_AIX) && !defined(__ANDROID__) +#ifndef Z7_AFFINITY_DISABLE +#define Z7_AFFINITY_SUPPORTED +// #pragma message(" ==== Z7_AFFINITY_SUPPORTED") +// #define _GNU_SOURCE +#endif +#endif +#endif + +#include + +#endif + +#include "7zTypes.h" + +EXTERN_C_BEGIN + +#ifdef _WIN32 + +WRes HandlePtr_Close(HANDLE *h); +WRes Handle_WaitObject(HANDLE h); + +typedef HANDLE CThread; + +#define Thread_CONSTRUCT(p) { *(p) = NULL; } +#define Thread_WasCreated(p) (*(p) != NULL) +#define Thread_Close(p) HandlePtr_Close(p) +// #define Thread_Wait(p) Handle_WaitObject(*(p)) + +#ifdef UNDER_CE + // if (USE_THREADS_CreateThread is defined), we use _beginthreadex() + // if (USE_THREADS_CreateThread is not definned), we use CreateThread() + #define USE_THREADS_CreateThread +#endif + +typedef + #ifdef USE_THREADS_CreateThread + DWORD + #else + unsigned + #endif + THREAD_FUNC_RET_TYPE; + +#define THREAD_FUNC_RET_ZERO 0 + +typedef DWORD_PTR CAffinityMask; +typedef DWORD_PTR CCpuSet; + +#define CpuSet_Zero(p) *(p) = (0) +#define CpuSet_Set(p, cpu) *(p) |= ((DWORD_PTR)1 << (cpu)) + +#else // _WIN32 + +typedef struct +{ + pthread_t _tid; + int _created; +} CThread; + +#define Thread_CONSTRUCT(p) { (p)->_tid = 0; (p)->_created = 0; } +#define Thread_WasCreated(p) ((p)->_created != 0) +WRes Thread_Close(CThread *p); +// #define Thread_Wait Thread_Wait_Close + +typedef void * THREAD_FUNC_RET_TYPE; +#define THREAD_FUNC_RET_ZERO NULL + + +typedef UInt64 CAffinityMask; + +#ifdef Z7_AFFINITY_SUPPORTED + +typedef cpu_set_t CCpuSet; +#define CpuSet_Zero(p) CPU_ZERO(p) +#define CpuSet_Set(p, cpu) CPU_SET(cpu, p) +#define CpuSet_IsSet(p, cpu) CPU_ISSET(cpu, p) + +#else + +typedef UInt64 CCpuSet; +#define CpuSet_Zero(p) *(p) = (0) +#define CpuSet_Set(p, cpu) *(p) |= ((UInt64)1 << (cpu)) +#define CpuSet_IsSet(p, cpu) ((*(p) & ((UInt64)1 << (cpu))) != 0) + +#endif + + +#endif // _WIN32 + + +#define THREAD_FUNC_CALL_TYPE Z7_STDCALL + +#if defined(_WIN32) && defined(__GNUC__) +/* GCC compiler for x86 32-bit uses the rule: + the stack is 16-byte aligned before CALL instruction for function calling. + But only root function main() contains instructions that + set 16-byte alignment for stack pointer. And another functions + just keep alignment, if it was set in some parent function. + + The problem: + if we create new thread in MinGW (GCC) 32-bit x86 via _beginthreadex() or CreateThread(), + the root function of thread doesn't set 16-byte alignment. + And stack frames in all child functions also will be unaligned in that case. + + Here we set (force_align_arg_pointer) attribute for root function of new thread. + Do we need (force_align_arg_pointer) also for another systems? */ + + #define THREAD_FUNC_ATTRIB_ALIGN_ARG __attribute__((force_align_arg_pointer)) + // #define THREAD_FUNC_ATTRIB_ALIGN_ARG // for debug : bad alignment in SSE functions +#else + #define THREAD_FUNC_ATTRIB_ALIGN_ARG +#endif + +#define THREAD_FUNC_DECL THREAD_FUNC_ATTRIB_ALIGN_ARG THREAD_FUNC_RET_TYPE THREAD_FUNC_CALL_TYPE + +typedef THREAD_FUNC_RET_TYPE (THREAD_FUNC_CALL_TYPE * THREAD_FUNC_TYPE)(void *); +WRes Thread_Create(CThread *p, THREAD_FUNC_TYPE func, LPVOID param); +WRes Thread_Create_With_Affinity(CThread *p, THREAD_FUNC_TYPE func, LPVOID param, CAffinityMask affinity); +WRes Thread_Wait_Close(CThread *p); + +#ifdef _WIN32 +#define Thread_Create_With_CpuSet(p, func, param, cs) \ + Thread_Create_With_Affinity(p, func, param, *cs) +#else +WRes Thread_Create_With_CpuSet(CThread *p, THREAD_FUNC_TYPE func, LPVOID param, const CCpuSet *cpuSet); +#endif + + +#ifdef _WIN32 + +typedef HANDLE CEvent; +typedef CEvent CAutoResetEvent; +typedef CEvent CManualResetEvent; +#define Event_Construct(p) *(p) = NULL +#define Event_IsCreated(p) (*(p) != NULL) +#define Event_Close(p) HandlePtr_Close(p) +#define Event_Wait(p) Handle_WaitObject(*(p)) +WRes Event_Set(CEvent *p); +WRes Event_Reset(CEvent *p); +WRes ManualResetEvent_Create(CManualResetEvent *p, int signaled); +WRes ManualResetEvent_CreateNotSignaled(CManualResetEvent *p); +WRes AutoResetEvent_Create(CAutoResetEvent *p, int signaled); +WRes AutoResetEvent_CreateNotSignaled(CAutoResetEvent *p); + +typedef HANDLE CSemaphore; +#define Semaphore_Construct(p) *(p) = NULL +#define Semaphore_IsCreated(p) (*(p) != NULL) +#define Semaphore_Close(p) HandlePtr_Close(p) +#define Semaphore_Wait(p) Handle_WaitObject(*(p)) +WRes Semaphore_Create(CSemaphore *p, UInt32 initCount, UInt32 maxCount); +WRes Semaphore_OptCreateInit(CSemaphore *p, UInt32 initCount, UInt32 maxCount); +WRes Semaphore_ReleaseN(CSemaphore *p, UInt32 num); +WRes Semaphore_Release1(CSemaphore *p); + +typedef CRITICAL_SECTION CCriticalSection; +WRes CriticalSection_Init(CCriticalSection *p); +#define CriticalSection_Delete(p) DeleteCriticalSection(p) +#define CriticalSection_Enter(p) EnterCriticalSection(p) +#define CriticalSection_Leave(p) LeaveCriticalSection(p) + + +#else // _WIN32 + +typedef struct _CEvent +{ + int _created; + int _manual_reset; + int _state; + pthread_mutex_t _mutex; + pthread_cond_t _cond; +} CEvent; + +typedef CEvent CAutoResetEvent; +typedef CEvent CManualResetEvent; + +#define Event_Construct(p) (p)->_created = 0 +#define Event_IsCreated(p) ((p)->_created) + +WRes ManualResetEvent_Create(CManualResetEvent *p, int signaled); +WRes ManualResetEvent_CreateNotSignaled(CManualResetEvent *p); +WRes AutoResetEvent_Create(CAutoResetEvent *p, int signaled); +WRes AutoResetEvent_CreateNotSignaled(CAutoResetEvent *p); + +WRes Event_Set(CEvent *p); +WRes Event_Reset(CEvent *p); +WRes Event_Wait(CEvent *p); +WRes Event_Close(CEvent *p); + + +typedef struct _CSemaphore +{ + int _created; + UInt32 _count; + UInt32 _maxCount; + pthread_mutex_t _mutex; + pthread_cond_t _cond; +} CSemaphore; + +#define Semaphore_Construct(p) (p)->_created = 0 +#define Semaphore_IsCreated(p) ((p)->_created) + +WRes Semaphore_Create(CSemaphore *p, UInt32 initCount, UInt32 maxCount); +WRes Semaphore_OptCreateInit(CSemaphore *p, UInt32 initCount, UInt32 maxCount); +WRes Semaphore_ReleaseN(CSemaphore *p, UInt32 num); +#define Semaphore_Release1(p) Semaphore_ReleaseN(p, 1) +WRes Semaphore_Wait(CSemaphore *p); +WRes Semaphore_Close(CSemaphore *p); + + +typedef struct _CCriticalSection +{ + pthread_mutex_t _mutex; +} CCriticalSection; + +WRes CriticalSection_Init(CCriticalSection *p); +void CriticalSection_Delete(CCriticalSection *cs); +void CriticalSection_Enter(CCriticalSection *cs); +void CriticalSection_Leave(CCriticalSection *cs); + +LONG InterlockedIncrement(LONG volatile *addend); + +#endif // _WIN32 + +WRes AutoResetEvent_OptCreate_And_Reset(CAutoResetEvent *p); + +EXTERN_C_END + +#endif -- cgit v1.3.1