Path: blob/main/sys/contrib/openzfs/module/zstd/lib/common/fse.h
48774 views
// SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0-only1/* ******************************************************************2* FSE : Finite State Entropy codec3* Public Prototypes declaration4* Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.5*6* You can contact the author at :7* - Source repository : https://github.com/Cyan4973/FiniteStateEntropy8*9* This source code is licensed under both the BSD-style license (found in the10* LICENSE file in the root directory of this source tree) and the GPLv2 (found11* in the COPYING file in the root directory of this source tree).12* You may select, at your option, one of the above-listed licenses.13****************************************************************** */1415#if defined (__cplusplus)16extern "C" {17#endif1819#ifndef FSE_H20#define FSE_H212223/*-*****************************************24* Dependencies25******************************************/26#include <stddef.h> /* size_t, ptrdiff_t */272829/*-*****************************************30* FSE_PUBLIC_API : control library symbols visibility31******************************************/32#if defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) && defined(__GNUC__) && (__GNUC__ >= 4)33# define FSE_PUBLIC_API __attribute__ ((visibility ("default")))34#elif defined(FSE_DLL_EXPORT) && (FSE_DLL_EXPORT==1) /* Visual expected */35# define FSE_PUBLIC_API __declspec(dllexport)36#elif defined(FSE_DLL_IMPORT) && (FSE_DLL_IMPORT==1)37# define FSE_PUBLIC_API __declspec(dllimport) /* It isn't required but allows to generate better code, saving a function pointer load from the IAT and an indirect jump.*/38#else39# define FSE_PUBLIC_API40#endif4142/*------ Version ------*/43#define FSE_VERSION_MAJOR 044#define FSE_VERSION_MINOR 945#define FSE_VERSION_RELEASE 04647#define FSE_LIB_VERSION FSE_VERSION_MAJOR.FSE_VERSION_MINOR.FSE_VERSION_RELEASE48#define FSE_QUOTE(str) #str49#define FSE_EXPAND_AND_QUOTE(str) FSE_QUOTE(str)50#define FSE_VERSION_STRING FSE_EXPAND_AND_QUOTE(FSE_LIB_VERSION)5152#define FSE_VERSION_NUMBER (FSE_VERSION_MAJOR *100*100 + FSE_VERSION_MINOR *100 + FSE_VERSION_RELEASE)53FSE_PUBLIC_API unsigned FSE_versionNumber(void); /**< library version number; to be used when checking dll version */545556/*-****************************************57* FSE simple functions58******************************************/59/*! FSE_compress() :60Compress content of buffer 'src', of size 'srcSize', into destination buffer 'dst'.61'dst' buffer must be already allocated. Compression runs faster is dstCapacity >= FSE_compressBound(srcSize).62@return : size of compressed data (<= dstCapacity).63Special values : if return == 0, srcData is not compressible => Nothing is stored within dst !!!64if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression instead.65if FSE_isError(return), compression failed (more details using FSE_getErrorName())66*/67FSE_PUBLIC_API size_t FSE_compress(void* dst, size_t dstCapacity,68const void* src, size_t srcSize);6970/*! FSE_decompress():71Decompress FSE data from buffer 'cSrc', of size 'cSrcSize',72into already allocated destination buffer 'dst', of size 'dstCapacity'.73@return : size of regenerated data (<= maxDstSize),74or an error code, which can be tested using FSE_isError() .7576** Important ** : FSE_decompress() does not decompress non-compressible nor RLE data !!!77Why ? : making this distinction requires a header.78Header management is intentionally delegated to the user layer, which can better manage special cases.79*/80FSE_PUBLIC_API size_t FSE_decompress(void* dst, size_t dstCapacity,81const void* cSrc, size_t cSrcSize);828384/*-*****************************************85* Tool functions86******************************************/87FSE_PUBLIC_API size_t FSE_compressBound(size_t size); /* maximum compressed size */8889/* Error Management */90FSE_PUBLIC_API unsigned FSE_isError(size_t code); /* tells if a return value is an error code */91FSE_PUBLIC_API const char* FSE_getErrorName(size_t code); /* provides error code string (useful for debugging) */929394/*-*****************************************95* FSE advanced functions96******************************************/97/*! FSE_compress2() :98Same as FSE_compress(), but allows the selection of 'maxSymbolValue' and 'tableLog'99Both parameters can be defined as '0' to mean : use default value100@return : size of compressed data101Special values : if return == 0, srcData is not compressible => Nothing is stored within cSrc !!!102if return == 1, srcData is a single byte symbol * srcSize times. Use RLE compression.103if FSE_isError(return), it's an error code.104*/105FSE_PUBLIC_API size_t FSE_compress2 (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog);106107108/*-*****************************************109* FSE detailed API110******************************************/111/*!112FSE_compress() does the following:1131. count symbol occurrence from source[] into table count[] (see hist.h)1142. normalize counters so that sum(count[]) == Power_of_2 (2^tableLog)1153. save normalized counters to memory buffer using writeNCount()1164. build encoding table 'CTable' from normalized counters1175. encode the data stream using encoding table 'CTable'118119FSE_decompress() does the following:1201. read normalized counters with readNCount()1212. build decoding table 'DTable' from normalized counters1223. decode the data stream using decoding table 'DTable'123124The following API allows targeting specific sub-functions for advanced tasks.125For example, it's possible to compress several blocks using the same 'CTable',126or to save and provide normalized distribution using external method.127*/128129/* *** COMPRESSION *** */130131/*! FSE_optimalTableLog():132dynamically downsize 'tableLog' when conditions are met.133It saves CPU time, by using smaller tables, while preserving or even improving compression ratio.134@return : recommended tableLog (necessarily <= 'maxTableLog') */135FSE_PUBLIC_API unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue);136137/*! FSE_normalizeCount():138normalize counts so that sum(count[]) == Power_of_2 (2^tableLog)139'normalizedCounter' is a table of short, of minimum size (maxSymbolValue+1).140@return : tableLog,141or an errorCode, which can be tested using FSE_isError() */142FSE_PUBLIC_API size_t FSE_normalizeCount(short* normalizedCounter, unsigned tableLog,143const unsigned* count, size_t srcSize, unsigned maxSymbolValue);144145/*! FSE_NCountWriteBound():146Provides the maximum possible size of an FSE normalized table, given 'maxSymbolValue' and 'tableLog'.147Typically useful for allocation purpose. */148FSE_PUBLIC_API size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog);149150/*! FSE_writeNCount():151Compactly save 'normalizedCounter' into 'buffer'.152@return : size of the compressed table,153or an errorCode, which can be tested using FSE_isError(). */154FSE_PUBLIC_API size_t FSE_writeNCount (void* buffer, size_t bufferSize,155const short* normalizedCounter,156unsigned maxSymbolValue, unsigned tableLog);157158/*! Constructor and Destructor of FSE_CTable.159Note that FSE_CTable size depends on 'tableLog' and 'maxSymbolValue' */160typedef unsigned FSE_CTable; /* don't allocate that. It's only meant to be more restrictive than void* */161FSE_PUBLIC_API FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog);162FSE_PUBLIC_API void FSE_freeCTable (FSE_CTable* ct);163164/*! FSE_buildCTable():165Builds `ct`, which must be already allocated, using FSE_createCTable().166@return : 0, or an errorCode, which can be tested using FSE_isError() */167FSE_PUBLIC_API size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);168169/*! FSE_compress_usingCTable():170Compress `src` using `ct` into `dst` which must be already allocated.171@return : size of compressed data (<= `dstCapacity`),172or 0 if compressed data could not fit into `dst`,173or an errorCode, which can be tested using FSE_isError() */174FSE_PUBLIC_API size_t FSE_compress_usingCTable (void* dst, size_t dstCapacity, const void* src, size_t srcSize, const FSE_CTable* ct);175176/*!177Tutorial :178----------179The first step is to count all symbols. FSE_count() does this job very fast.180Result will be saved into 'count', a table of unsigned int, which must be already allocated, and have 'maxSymbolValuePtr[0]+1' cells.181'src' is a table of bytes of size 'srcSize'. All values within 'src' MUST be <= maxSymbolValuePtr[0]182maxSymbolValuePtr[0] will be updated, with its real value (necessarily <= original value)183FSE_count() will return the number of occurrence of the most frequent symbol.184This can be used to know if there is a single symbol within 'src', and to quickly evaluate its compressibility.185If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError()).186187The next step is to normalize the frequencies.188FSE_normalizeCount() will ensure that sum of frequencies is == 2 ^'tableLog'.189It also guarantees a minimum of 1 to any Symbol with frequency >= 1.190You can use 'tableLog'==0 to mean "use default tableLog value".191If you are unsure of which tableLog value to use, you can ask FSE_optimalTableLog(),192which will provide the optimal valid tableLog given sourceSize, maxSymbolValue, and a user-defined maximum (0 means "default").193194The result of FSE_normalizeCount() will be saved into a table,195called 'normalizedCounter', which is a table of signed short.196'normalizedCounter' must be already allocated, and have at least 'maxSymbolValue+1' cells.197The return value is tableLog if everything proceeded as expected.198It is 0 if there is a single symbol within distribution.199If there is an error (ex: invalid tableLog value), the function will return an ErrorCode (which can be tested using FSE_isError()).200201'normalizedCounter' can be saved in a compact manner to a memory area using FSE_writeNCount().202'buffer' must be already allocated.203For guaranteed success, buffer size must be at least FSE_headerBound().204The result of the function is the number of bytes written into 'buffer'.205If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError(); ex : buffer size too small).206207'normalizedCounter' can then be used to create the compression table 'CTable'.208The space required by 'CTable' must be already allocated, using FSE_createCTable().209You can then use FSE_buildCTable() to fill 'CTable'.210If there is an error, both functions will return an ErrorCode (which can be tested using FSE_isError()).211212'CTable' can then be used to compress 'src', with FSE_compress_usingCTable().213Similar to FSE_count(), the convention is that 'src' is assumed to be a table of char of size 'srcSize'214The function returns the size of compressed data (without header), necessarily <= `dstCapacity`.215If it returns '0', compressed data could not fit into 'dst'.216If there is an error, the function will return an ErrorCode (which can be tested using FSE_isError()).217*/218219220/* *** DECOMPRESSION *** */221222/*! FSE_readNCount():223Read compactly saved 'normalizedCounter' from 'rBuffer'.224@return : size read from 'rBuffer',225or an errorCode, which can be tested using FSE_isError().226maxSymbolValuePtr[0] and tableLogPtr[0] will also be updated with their respective values */227FSE_PUBLIC_API size_t FSE_readNCount (short* normalizedCounter,228unsigned* maxSymbolValuePtr, unsigned* tableLogPtr,229const void* rBuffer, size_t rBuffSize);230231/*! Constructor and Destructor of FSE_DTable.232Note that its size depends on 'tableLog' */233typedef unsigned FSE_DTable; /* don't allocate that. It's just a way to be more restrictive than void* */234FSE_PUBLIC_API FSE_DTable* FSE_createDTable(unsigned tableLog);235FSE_PUBLIC_API void FSE_freeDTable(FSE_DTable* dt);236237/*! FSE_buildDTable():238Builds 'dt', which must be already allocated, using FSE_createDTable().239return : 0, or an errorCode, which can be tested using FSE_isError() */240FSE_PUBLIC_API size_t FSE_buildDTable (FSE_DTable* dt, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog);241242/*! FSE_decompress_usingDTable():243Decompress compressed source `cSrc` of size `cSrcSize` using `dt`244into `dst` which must be already allocated.245@return : size of regenerated data (necessarily <= `dstCapacity`),246or an errorCode, which can be tested using FSE_isError() */247FSE_PUBLIC_API size_t FSE_decompress_usingDTable(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, const FSE_DTable* dt);248249/*!250Tutorial :251----------252(Note : these functions only decompress FSE-compressed blocks.253If block is uncompressed, use memcpy() instead254If block is a single repeated byte, use memset() instead )255256The first step is to obtain the normalized frequencies of symbols.257This can be performed by FSE_readNCount() if it was saved using FSE_writeNCount().258'normalizedCounter' must be already allocated, and have at least 'maxSymbolValuePtr[0]+1' cells of signed short.259In practice, that means it's necessary to know 'maxSymbolValue' beforehand,260or size the table to handle worst case situations (typically 256).261FSE_readNCount() will provide 'tableLog' and 'maxSymbolValue'.262The result of FSE_readNCount() is the number of bytes read from 'rBuffer'.263Note that 'rBufferSize' must be at least 4 bytes, even if useful information is less than that.264If there is an error, the function will return an error code, which can be tested using FSE_isError().265266The next step is to build the decompression tables 'FSE_DTable' from 'normalizedCounter'.267This is performed by the function FSE_buildDTable().268The space required by 'FSE_DTable' must be already allocated using FSE_createDTable().269If there is an error, the function will return an error code, which can be tested using FSE_isError().270271`FSE_DTable` can then be used to decompress `cSrc`, with FSE_decompress_usingDTable().272`cSrcSize` must be strictly correct, otherwise decompression will fail.273FSE_decompress_usingDTable() result will tell how many bytes were regenerated (<=`dstCapacity`).274If there is an error, the function will return an error code, which can be tested using FSE_isError(). (ex: dst buffer too small)275*/276277#endif /* FSE_H */278279#if defined(FSE_STATIC_LINKING_ONLY) && !defined(FSE_H_FSE_STATIC_LINKING_ONLY)280#define FSE_H_FSE_STATIC_LINKING_ONLY281282/* *** Dependency *** */283#include "bitstream.h"284285286/* *****************************************287* Static allocation288*******************************************/289/* FSE buffer bounds */290#define FSE_NCOUNTBOUND 512291#define FSE_BLOCKBOUND(size) (size + (size>>7) + 4 /* fse states */ + sizeof(size_t) /* bitContainer */)292#define FSE_COMPRESSBOUND(size) (FSE_NCOUNTBOUND + FSE_BLOCKBOUND(size)) /* Macro version, useful for static allocation */293294/* It is possible to statically allocate FSE CTable/DTable as a table of FSE_CTable/FSE_DTable using below macros */295#define FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) (1 + (1<<(maxTableLog-1)) + ((maxSymbolValue+1)*2))296#define FSE_DTABLE_SIZE_U32(maxTableLog) (1 + (1<<maxTableLog))297298/* or use the size to malloc() space directly. Pay attention to alignment restrictions though */299#define FSE_CTABLE_SIZE(maxTableLog, maxSymbolValue) (FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) * sizeof(FSE_CTable))300#define FSE_DTABLE_SIZE(maxTableLog) (FSE_DTABLE_SIZE_U32(maxTableLog) * sizeof(FSE_DTable))301302303/* *****************************************304* FSE advanced API305***************************************** */306307unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus);308/**< same as FSE_optimalTableLog(), which used `minus==2` */309310/* FSE_compress_wksp() :311* Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).312* FSE_WKSP_SIZE_U32() provides the minimum size required for `workSpace` as a table of FSE_CTable.313*/314#define FSE_WKSP_SIZE_U32(maxTableLog, maxSymbolValue) ( FSE_CTABLE_SIZE_U32(maxTableLog, maxSymbolValue) + ((maxTableLog > 12) ? (1 << (maxTableLog - 2)) : 1024) )315size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);316317size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits);318/**< build a fake FSE_CTable, designed for a flat distribution, where each symbol uses nbBits */319320size_t FSE_buildCTable_rle (FSE_CTable* ct, unsigned char symbolValue);321/**< build a fake FSE_CTable, designed to compress always the same symbolValue */322323/* FSE_buildCTable_wksp() :324* Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).325* `wkspSize` must be >= `(1<<tableLog)`.326*/327size_t FSE_buildCTable_wksp(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize);328329size_t FSE_buildDTable_raw (FSE_DTable* dt, unsigned nbBits);330/**< build a fake FSE_DTable, designed to read a flat distribution where each symbol uses nbBits */331332size_t FSE_buildDTable_rle (FSE_DTable* dt, unsigned char symbolValue);333/**< build a fake FSE_DTable, designed to always generate the same symbolValue */334335size_t FSE_decompress_wksp(void* dst, size_t dstCapacity, const void* cSrc, size_t cSrcSize, FSE_DTable* workSpace, unsigned maxLog);336/**< same as FSE_decompress(), using an externally allocated `workSpace` produced with `FSE_DTABLE_SIZE_U32(maxLog)` */337338typedef enum {339FSE_repeat_none, /**< Cannot use the previous table */340FSE_repeat_check, /**< Can use the previous table but it must be checked */341FSE_repeat_valid /**< Can use the previous table and it is assumed to be valid */342} FSE_repeat;343344/* *****************************************345* FSE symbol compression API346*******************************************/347/*!348This API consists of small unitary functions, which highly benefit from being inlined.349Hence their body are included in next section.350*/351typedef struct {352ptrdiff_t value;353const void* stateTable;354const void* symbolTT;355unsigned stateLog;356} FSE_CState_t;357358static void FSE_initCState(FSE_CState_t* CStatePtr, const FSE_CTable* ct);359360static void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* CStatePtr, unsigned symbol);361362static void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* CStatePtr);363364/**<365These functions are inner components of FSE_compress_usingCTable().366They allow the creation of custom streams, mixing multiple tables and bit sources.367368A key property to keep in mind is that encoding and decoding are done **in reverse direction**.369So the first symbol you will encode is the last you will decode, like a LIFO stack.370371You will need a few variables to track your CStream. They are :372373FSE_CTable ct; // Provided by FSE_buildCTable()374BIT_CStream_t bitStream; // bitStream tracking structure375FSE_CState_t state; // State tracking structure (can have several)376377378The first thing to do is to init bitStream and state.379size_t errorCode = BIT_initCStream(&bitStream, dstBuffer, maxDstSize);380FSE_initCState(&state, ct);381382Note that BIT_initCStream() can produce an error code, so its result should be tested, using FSE_isError();383You can then encode your input data, byte after byte.384FSE_encodeSymbol() outputs a maximum of 'tableLog' bits at a time.385Remember decoding will be done in reverse direction.386FSE_encodeByte(&bitStream, &state, symbol);387388At any time, you can also add any bit sequence.389Note : maximum allowed nbBits is 25, for compatibility with 32-bits decoders390BIT_addBits(&bitStream, bitField, nbBits);391392The above methods don't commit data to memory, they just store it into local register, for speed.393Local register size is 64-bits on 64-bits systems, 32-bits on 32-bits systems (size_t).394Writing data to memory is a manual operation, performed by the flushBits function.395BIT_flushBits(&bitStream);396397Your last FSE encoding operation shall be to flush your last state value(s).398FSE_flushState(&bitStream, &state);399400Finally, you must close the bitStream.401The function returns the size of CStream in bytes.402If data couldn't fit into dstBuffer, it will return a 0 ( == not compressible)403If there is an error, it returns an errorCode (which can be tested using FSE_isError()).404size_t size = BIT_closeCStream(&bitStream);405*/406407408/* *****************************************409* FSE symbol decompression API410*******************************************/411typedef struct {412size_t state;413const void* table; /* precise table may vary, depending on U16 */414} FSE_DState_t;415416417static void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt);418419static unsigned char FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);420421static unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr);422423/**<424Let's now decompose FSE_decompress_usingDTable() into its unitary components.425You will decode FSE-encoded symbols from the bitStream,426and also any other bitFields you put in, **in reverse order**.427428You will need a few variables to track your bitStream. They are :429430BIT_DStream_t DStream; // Stream context431FSE_DState_t DState; // State context. Multiple ones are possible432FSE_DTable* DTablePtr; // Decoding table, provided by FSE_buildDTable()433434The first thing to do is to init the bitStream.435errorCode = BIT_initDStream(&DStream, srcBuffer, srcSize);436437You should then retrieve your initial state(s)438(in reverse flushing order if you have several ones) :439errorCode = FSE_initDState(&DState, &DStream, DTablePtr);440441You can then decode your data, symbol after symbol.442For information the maximum number of bits read by FSE_decodeSymbol() is 'tableLog'.443Keep in mind that symbols are decoded in reverse order, like a LIFO stack (last in, first out).444unsigned char symbol = FSE_decodeSymbol(&DState, &DStream);445446You can retrieve any bitfield you eventually stored into the bitStream (in reverse order)447Note : maximum allowed nbBits is 25, for 32-bits compatibility448size_t bitField = BIT_readBits(&DStream, nbBits);449450All above operations only read from local register (which size depends on size_t).451Refueling the register from memory is manually performed by the reload method.452endSignal = FSE_reloadDStream(&DStream);453454BIT_reloadDStream() result tells if there is still some more data to read from DStream.455BIT_DStream_unfinished : there is still some data left into the DStream.456BIT_DStream_endOfBuffer : Dstream reached end of buffer. Its container may no longer be completely filled.457BIT_DStream_completed : Dstream reached its exact end, corresponding in general to decompression completed.458BIT_DStream_tooFar : Dstream went too far. Decompression result is corrupted.459460When reaching end of buffer (BIT_DStream_endOfBuffer), progress slowly, notably if you decode multiple symbols per loop,461to properly detect the exact end of stream.462After each decoded symbol, check if DStream is fully consumed using this simple test :463BIT_reloadDStream(&DStream) >= BIT_DStream_completed464465When it's done, verify decompression is fully completed, by checking both DStream and the relevant states.466Checking if DStream has reached its end is performed by :467BIT_endOfDStream(&DStream);468Check also the states. There might be some symbols left there, if some high probability ones (>50%) are possible.469FSE_endOfDState(&DState);470*/471472473/* *****************************************474* FSE unsafe API475*******************************************/476static unsigned char FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD);477/* faster, but works only if nbBits is always >= 1 (otherwise, result will be corrupted) */478479480/* *****************************************481* Implementation of inlined functions482*******************************************/483typedef struct {484int deltaFindState;485U32 deltaNbBits;486} FSE_symbolCompressionTransform; /* total 8 bytes */487488MEM_STATIC void FSE_initCState(FSE_CState_t* statePtr, const FSE_CTable* ct)489{490const void* ptr = ct;491const U16* u16ptr = (const U16*) ptr;492const U32 tableLog = MEM_read16(ptr);493statePtr->value = (ptrdiff_t)1<<tableLog;494statePtr->stateTable = u16ptr+2;495statePtr->symbolTT = ct + 1 + (tableLog ? (1<<(tableLog-1)) : 1);496statePtr->stateLog = tableLog;497}498499500/*! FSE_initCState2() :501* Same as FSE_initCState(), but the first symbol to include (which will be the last to be read)502* uses the smallest state value possible, saving the cost of this symbol */503MEM_STATIC void FSE_initCState2(FSE_CState_t* statePtr, const FSE_CTable* ct, U32 symbol)504{505FSE_initCState(statePtr, ct);506{ const FSE_symbolCompressionTransform symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];507const U16* stateTable = (const U16*)(statePtr->stateTable);508U32 nbBitsOut = (U32)((symbolTT.deltaNbBits + (1<<15)) >> 16);509statePtr->value = (nbBitsOut << 16) - symbolTT.deltaNbBits;510statePtr->value = stateTable[(statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];511}512}513514MEM_STATIC void FSE_encodeSymbol(BIT_CStream_t* bitC, FSE_CState_t* statePtr, unsigned symbol)515{516FSE_symbolCompressionTransform const symbolTT = ((const FSE_symbolCompressionTransform*)(statePtr->symbolTT))[symbol];517const U16* const stateTable = (const U16*)(statePtr->stateTable);518U32 const nbBitsOut = (U32)((statePtr->value + symbolTT.deltaNbBits) >> 16);519BIT_addBits(bitC, statePtr->value, nbBitsOut);520statePtr->value = stateTable[ (statePtr->value >> nbBitsOut) + symbolTT.deltaFindState];521}522523MEM_STATIC void FSE_flushCState(BIT_CStream_t* bitC, const FSE_CState_t* statePtr)524{525BIT_addBits(bitC, statePtr->value, statePtr->stateLog);526BIT_flushBits(bitC);527}528529530/* FSE_getMaxNbBits() :531* Approximate maximum cost of a symbol, in bits.532* Fractional get rounded up (i.e : a symbol with a normalized frequency of 3 gives the same result as a frequency of 2)533* note 1 : assume symbolValue is valid (<= maxSymbolValue)534* note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */535MEM_STATIC U32 FSE_getMaxNbBits(const void* symbolTTPtr, U32 symbolValue)536{537const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;538return (symbolTT[symbolValue].deltaNbBits + ((1<<16)-1)) >> 16;539}540541/* FSE_bitCost() :542* Approximate symbol cost, as fractional value, using fixed-point format (accuracyLog fractional bits)543* note 1 : assume symbolValue is valid (<= maxSymbolValue)544* note 2 : if freq[symbolValue]==0, @return a fake cost of tableLog+1 bits */545MEM_STATIC U32 FSE_bitCost(const void* symbolTTPtr, U32 tableLog, U32 symbolValue, U32 accuracyLog)546{547const FSE_symbolCompressionTransform* symbolTT = (const FSE_symbolCompressionTransform*) symbolTTPtr;548U32 const minNbBits = symbolTT[symbolValue].deltaNbBits >> 16;549U32 const threshold = (minNbBits+1) << 16;550assert(tableLog < 16);551assert(accuracyLog < 31-tableLog); /* ensure enough room for renormalization double shift */552{ U32 const tableSize = 1 << tableLog;553U32 const deltaFromThreshold = threshold - (symbolTT[symbolValue].deltaNbBits + tableSize);554U32 const normalizedDeltaFromThreshold = (deltaFromThreshold << accuracyLog) >> tableLog; /* linear interpolation (very approximate) */555U32 const bitMultiplier = 1 << accuracyLog;556assert(symbolTT[symbolValue].deltaNbBits + tableSize <= threshold);557assert(normalizedDeltaFromThreshold <= bitMultiplier);558return (minNbBits+1)*bitMultiplier - normalizedDeltaFromThreshold;559}560}561562563/* ====== Decompression ====== */564565typedef struct {566U16 tableLog;567U16 fastMode;568} FSE_DTableHeader; /* sizeof U32 */569570typedef struct571{572unsigned short newState;573unsigned char symbol;574unsigned char nbBits;575} FSE_decode_t; /* size == U32 */576577MEM_STATIC void FSE_initDState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD, const FSE_DTable* dt)578{579const void* ptr = dt;580const FSE_DTableHeader* const DTableH = (const FSE_DTableHeader*)ptr;581DStatePtr->state = BIT_readBits(bitD, DTableH->tableLog);582BIT_reloadDStream(bitD);583DStatePtr->table = dt + 1;584}585586MEM_STATIC BYTE FSE_peekSymbol(const FSE_DState_t* DStatePtr)587{588FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];589return DInfo.symbol;590}591592MEM_STATIC void FSE_updateState(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)593{594FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];595U32 const nbBits = DInfo.nbBits;596size_t const lowBits = BIT_readBits(bitD, nbBits);597DStatePtr->state = DInfo.newState + lowBits;598}599600MEM_STATIC BYTE FSE_decodeSymbol(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)601{602FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];603U32 const nbBits = DInfo.nbBits;604BYTE const symbol = DInfo.symbol;605size_t const lowBits = BIT_readBits(bitD, nbBits);606607DStatePtr->state = DInfo.newState + lowBits;608return symbol;609}610611/*! FSE_decodeSymbolFast() :612unsafe, only works if no symbol has a probability > 50% */613MEM_STATIC BYTE FSE_decodeSymbolFast(FSE_DState_t* DStatePtr, BIT_DStream_t* bitD)614{615FSE_decode_t const DInfo = ((const FSE_decode_t*)(DStatePtr->table))[DStatePtr->state];616U32 const nbBits = DInfo.nbBits;617BYTE const symbol = DInfo.symbol;618size_t const lowBits = BIT_readBitsFast(bitD, nbBits);619620DStatePtr->state = DInfo.newState + lowBits;621return symbol;622}623624MEM_STATIC unsigned FSE_endOfDState(const FSE_DState_t* DStatePtr)625{626return DStatePtr->state == 0;627}628629630631#ifndef FSE_COMMONDEFS_ONLY632633/* **************************************************************634* Tuning parameters635****************************************************************/636/*!MEMORY_USAGE :637* Memory usage formula : N->2^N Bytes (examples : 10 -> 1KB; 12 -> 4KB ; 16 -> 64KB; 20 -> 1MB; etc.)638* Increasing memory usage improves compression ratio639* Reduced memory usage can improve speed, due to cache effect640* Recommended max value is 14, for 16KB, which nicely fits into Intel x86 L1 cache */641#ifndef FSE_MAX_MEMORY_USAGE642# define FSE_MAX_MEMORY_USAGE 14643#endif644#ifndef FSE_DEFAULT_MEMORY_USAGE645# define FSE_DEFAULT_MEMORY_USAGE 13646#endif647648/*!FSE_MAX_SYMBOL_VALUE :649* Maximum symbol value authorized.650* Required for proper stack allocation */651#ifndef FSE_MAX_SYMBOL_VALUE652# define FSE_MAX_SYMBOL_VALUE 255653#endif654655/* **************************************************************656* template functions type & suffix657****************************************************************/658#define FSE_FUNCTION_TYPE BYTE659#define FSE_FUNCTION_EXTENSION660#define FSE_DECODE_TYPE FSE_decode_t661662663#endif /* !FSE_COMMONDEFS_ONLY */664665666/* ***************************************************************667* Constants668*****************************************************************/669#define FSE_MAX_TABLELOG (FSE_MAX_MEMORY_USAGE-2)670#define FSE_MAX_TABLESIZE (1U<<FSE_MAX_TABLELOG)671#define FSE_MAXTABLESIZE_MASK (FSE_MAX_TABLESIZE-1)672#define FSE_DEFAULT_TABLELOG (FSE_DEFAULT_MEMORY_USAGE-2)673#define FSE_MIN_TABLELOG 5674675#define FSE_TABLELOG_ABSOLUTE_MAX 15676#if FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX677# error "FSE_MAX_TABLELOG > FSE_TABLELOG_ABSOLUTE_MAX is not supported"678#endif679680#define FSE_TABLESTEP(tableSize) ((tableSize>>1) + (tableSize>>3) + 3)681682683#endif /* FSE_STATIC_LINKING_ONLY */684685686#if defined (__cplusplus)687}688#endif689690691