xref: /freebsd/sys/contrib/openzfs/module/zstd/lib/compress/fse_compress.c (revision 61145dc2b94f12f6a47344fb9aac702321880e43)
1*61145dc2SMartin Matuska // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0-only
2c03c5b1cSMartin Matuska /* ******************************************************************
3c03c5b1cSMartin Matuska  * FSE : Finite State Entropy encoder
4c03c5b1cSMartin Matuska  * Copyright (c) 2013-2020, Yann Collet, Facebook, Inc.
5c03c5b1cSMartin Matuska  *
6c03c5b1cSMartin Matuska  *  You can contact the author at :
7c03c5b1cSMartin Matuska  *  - FSE source repository : https://github.com/Cyan4973/FiniteStateEntropy
8c03c5b1cSMartin Matuska  *  - Public forum : https://groups.google.com/forum/#!forum/lz4c
9c03c5b1cSMartin Matuska  *
10c03c5b1cSMartin Matuska  * This source code is licensed under both the BSD-style license (found in the
11c03c5b1cSMartin Matuska  * LICENSE file in the root directory of this source tree) and the GPLv2 (found
12c03c5b1cSMartin Matuska  * in the COPYING file in the root directory of this source tree).
13c03c5b1cSMartin Matuska  * You may select, at your option, one of the above-listed licenses.
14c03c5b1cSMartin Matuska ****************************************************************** */
15c03c5b1cSMartin Matuska 
16c03c5b1cSMartin Matuska /* **************************************************************
17c03c5b1cSMartin Matuska *  Includes
18c03c5b1cSMartin Matuska ****************************************************************/
19c03c5b1cSMartin Matuska #include <stdlib.h>     /* malloc, free, qsort */
20c03c5b1cSMartin Matuska #include <string.h>     /* memcpy, memset */
21c03c5b1cSMartin Matuska #include "../common/compiler.h"
22c03c5b1cSMartin Matuska #include "../common/mem.h"        /* U32, U16, etc. */
23c03c5b1cSMartin Matuska #include "../common/debug.h"      /* assert, DEBUGLOG */
24c03c5b1cSMartin Matuska #include "hist.h"       /* HIST_count_wksp */
25c03c5b1cSMartin Matuska #include "../common/bitstream.h"
26c03c5b1cSMartin Matuska #define FSE_STATIC_LINKING_ONLY
27c03c5b1cSMartin Matuska #include "../common/fse.h"
28c03c5b1cSMartin Matuska #include "../common/error_private.h"
29c03c5b1cSMartin Matuska 
30c03c5b1cSMartin Matuska 
31c03c5b1cSMartin Matuska /* **************************************************************
32c03c5b1cSMartin Matuska *  Error Management
33c03c5b1cSMartin Matuska ****************************************************************/
34c03c5b1cSMartin Matuska #define FSE_isError ERR_isError
35c03c5b1cSMartin Matuska 
36c03c5b1cSMartin Matuska 
37c03c5b1cSMartin Matuska /* **************************************************************
38c03c5b1cSMartin Matuska *  Templates
39c03c5b1cSMartin Matuska ****************************************************************/
40c03c5b1cSMartin Matuska /*
41c03c5b1cSMartin Matuska   designed to be included
42c03c5b1cSMartin Matuska   for type-specific functions (template emulation in C)
43c03c5b1cSMartin Matuska   Objective is to write these functions only once, for improved maintenance
44c03c5b1cSMartin Matuska */
45c03c5b1cSMartin Matuska 
46c03c5b1cSMartin Matuska /* safety checks */
47c03c5b1cSMartin Matuska #ifndef FSE_FUNCTION_EXTENSION
48c03c5b1cSMartin Matuska #  error "FSE_FUNCTION_EXTENSION must be defined"
49c03c5b1cSMartin Matuska #endif
50c03c5b1cSMartin Matuska #ifndef FSE_FUNCTION_TYPE
51c03c5b1cSMartin Matuska #  error "FSE_FUNCTION_TYPE must be defined"
52c03c5b1cSMartin Matuska #endif
53c03c5b1cSMartin Matuska 
54c03c5b1cSMartin Matuska /* Function names */
55c03c5b1cSMartin Matuska #define FSE_CAT(X,Y) X##Y
56c03c5b1cSMartin Matuska #define FSE_FUNCTION_NAME(X,Y) FSE_CAT(X,Y)
57c03c5b1cSMartin Matuska #define FSE_TYPE_NAME(X,Y) FSE_CAT(X,Y)
58c03c5b1cSMartin Matuska 
59c03c5b1cSMartin Matuska 
60c03c5b1cSMartin Matuska /* Function templates */
61c03c5b1cSMartin Matuska 
62c03c5b1cSMartin Matuska /* FSE_buildCTable_wksp() :
63c03c5b1cSMartin Matuska  * Same as FSE_buildCTable(), but using an externally allocated scratch buffer (`workSpace`).
64c03c5b1cSMartin Matuska  * wkspSize should be sized to handle worst case situation, which is `1<<max_tableLog * sizeof(FSE_FUNCTION_TYPE)`
65c03c5b1cSMartin Matuska  * workSpace must also be properly aligned with FSE_FUNCTION_TYPE requirements
66c03c5b1cSMartin Matuska  */
FSE_buildCTable_wksp(FSE_CTable * ct,const short * normalizedCounter,unsigned maxSymbolValue,unsigned tableLog,void * workSpace,size_t wkspSize)67c03c5b1cSMartin Matuska size_t FSE_buildCTable_wksp(FSE_CTable* ct,
68c03c5b1cSMartin Matuska                       const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
69c03c5b1cSMartin Matuska                             void* workSpace, size_t wkspSize)
70c03c5b1cSMartin Matuska {
71c03c5b1cSMartin Matuska     U32 const tableSize = 1 << tableLog;
72c03c5b1cSMartin Matuska     U32 const tableMask = tableSize - 1;
73c03c5b1cSMartin Matuska     void* const ptr = ct;
74c03c5b1cSMartin Matuska     U16* const tableU16 = ( (U16*) ptr) + 2;
75c03c5b1cSMartin Matuska     void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableLog ? tableSize>>1 : 1) ;
76c03c5b1cSMartin Matuska     FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
77c03c5b1cSMartin Matuska     U32 const step = FSE_TABLESTEP(tableSize);
78c03c5b1cSMartin Matuska     U32 cumul[FSE_MAX_SYMBOL_VALUE+2];
79c03c5b1cSMartin Matuska 
80c03c5b1cSMartin Matuska     FSE_FUNCTION_TYPE* const tableSymbol = (FSE_FUNCTION_TYPE*)workSpace;
81c03c5b1cSMartin Matuska     U32 highThreshold = tableSize-1;
82c03c5b1cSMartin Matuska 
83c03c5b1cSMartin Matuska     /* CTable header */
84c03c5b1cSMartin Matuska     if (((size_t)1 << tableLog) * sizeof(FSE_FUNCTION_TYPE) > wkspSize) return ERROR(tableLog_tooLarge);
85c03c5b1cSMartin Matuska     tableU16[-2] = (U16) tableLog;
86c03c5b1cSMartin Matuska     tableU16[-1] = (U16) maxSymbolValue;
87c03c5b1cSMartin Matuska     assert(tableLog < 16);   /* required for threshold strategy to work */
88c03c5b1cSMartin Matuska 
89c03c5b1cSMartin Matuska     /* For explanations on how to distribute symbol values over the table :
90c03c5b1cSMartin Matuska      * http://fastcompression.blogspot.fr/2014/02/fse-distributing-symbol-values.html */
91c03c5b1cSMartin Matuska 
92c03c5b1cSMartin Matuska      #ifdef __clang_analyzer__
93c03c5b1cSMartin Matuska      memset(tableSymbol, 0, sizeof(*tableSymbol) * tableSize);   /* useless initialization, just to keep scan-build happy */
94c03c5b1cSMartin Matuska      #endif
95c03c5b1cSMartin Matuska 
96c03c5b1cSMartin Matuska     /* symbol start positions */
97c03c5b1cSMartin Matuska     {   U32 u;
98c03c5b1cSMartin Matuska         cumul[0] = 0;
99c03c5b1cSMartin Matuska         for (u=1; u <= maxSymbolValue+1; u++) {
100c03c5b1cSMartin Matuska             if (normalizedCounter[u-1]==-1) {  /* Low proba symbol */
101c03c5b1cSMartin Matuska                 cumul[u] = cumul[u-1] + 1;
102c03c5b1cSMartin Matuska                 tableSymbol[highThreshold--] = (FSE_FUNCTION_TYPE)(u-1);
103c03c5b1cSMartin Matuska             } else {
104c03c5b1cSMartin Matuska                 cumul[u] = cumul[u-1] + normalizedCounter[u-1];
105c03c5b1cSMartin Matuska         }   }
106c03c5b1cSMartin Matuska         cumul[maxSymbolValue+1] = tableSize+1;
107c03c5b1cSMartin Matuska     }
108c03c5b1cSMartin Matuska 
109c03c5b1cSMartin Matuska     /* Spread symbols */
110c03c5b1cSMartin Matuska     {   U32 position = 0;
111c03c5b1cSMartin Matuska         U32 symbol;
112c03c5b1cSMartin Matuska         for (symbol=0; symbol<=maxSymbolValue; symbol++) {
113c03c5b1cSMartin Matuska             int nbOccurrences;
114c03c5b1cSMartin Matuska             int const freq = normalizedCounter[symbol];
115c03c5b1cSMartin Matuska             for (nbOccurrences=0; nbOccurrences<freq; nbOccurrences++) {
116c03c5b1cSMartin Matuska                 tableSymbol[position] = (FSE_FUNCTION_TYPE)symbol;
117c03c5b1cSMartin Matuska                 position = (position + step) & tableMask;
118c03c5b1cSMartin Matuska                 while (position > highThreshold)
119c03c5b1cSMartin Matuska                     position = (position + step) & tableMask;   /* Low proba area */
120c03c5b1cSMartin Matuska         }   }
121c03c5b1cSMartin Matuska 
122c03c5b1cSMartin Matuska         assert(position==0);  /* Must have initialized all positions */
123c03c5b1cSMartin Matuska     }
124c03c5b1cSMartin Matuska 
125c03c5b1cSMartin Matuska     /* Build table */
126c03c5b1cSMartin Matuska     {   U32 u; for (u=0; u<tableSize; u++) {
127c03c5b1cSMartin Matuska         FSE_FUNCTION_TYPE s = tableSymbol[u];   /* note : static analyzer may not understand tableSymbol is properly initialized */
128c03c5b1cSMartin Matuska         tableU16[cumul[s]++] = (U16) (tableSize+u);   /* TableU16 : sorted by symbol order; gives next state value */
129c03c5b1cSMartin Matuska     }   }
130c03c5b1cSMartin Matuska 
131c03c5b1cSMartin Matuska     /* Build Symbol Transformation Table */
132c03c5b1cSMartin Matuska     {   unsigned total = 0;
133c03c5b1cSMartin Matuska         unsigned s;
134c03c5b1cSMartin Matuska         for (s=0; s<=maxSymbolValue; s++) {
135c03c5b1cSMartin Matuska             switch (normalizedCounter[s])
136c03c5b1cSMartin Matuska             {
137c03c5b1cSMartin Matuska             case  0:
138c03c5b1cSMartin Matuska                 /* filling nonetheless, for compatibility with FSE_getMaxNbBits() */
139c03c5b1cSMartin Matuska                 symbolTT[s].deltaNbBits = ((tableLog+1) << 16) - (1<<tableLog);
140c03c5b1cSMartin Matuska                 break;
141c03c5b1cSMartin Matuska 
142c03c5b1cSMartin Matuska             case -1:
143c03c5b1cSMartin Matuska             case  1:
144c03c5b1cSMartin Matuska                 symbolTT[s].deltaNbBits = (tableLog << 16) - (1<<tableLog);
145c03c5b1cSMartin Matuska                 symbolTT[s].deltaFindState = total - 1;
146c03c5b1cSMartin Matuska                 total ++;
147c03c5b1cSMartin Matuska                 break;
148c03c5b1cSMartin Matuska             default :
149c03c5b1cSMartin Matuska                 {
150c03c5b1cSMartin Matuska                     U32 const maxBitsOut = tableLog - BIT_highbit32 (normalizedCounter[s]-1);
151c03c5b1cSMartin Matuska                     U32 const minStatePlus = normalizedCounter[s] << maxBitsOut;
152c03c5b1cSMartin Matuska                     symbolTT[s].deltaNbBits = (maxBitsOut << 16) - minStatePlus;
153c03c5b1cSMartin Matuska                     symbolTT[s].deltaFindState = total - normalizedCounter[s];
154c03c5b1cSMartin Matuska                     total +=  normalizedCounter[s];
155c03c5b1cSMartin Matuska     }   }   }   }
156c03c5b1cSMartin Matuska 
157c03c5b1cSMartin Matuska #if 0  /* debug : symbol costs */
158c03c5b1cSMartin Matuska     DEBUGLOG(5, "\n --- table statistics : ");
159c03c5b1cSMartin Matuska     {   U32 symbol;
160c03c5b1cSMartin Matuska         for (symbol=0; symbol<=maxSymbolValue; symbol++) {
161c03c5b1cSMartin Matuska             DEBUGLOG(5, "%3u: w=%3i,   maxBits=%u, fracBits=%.2f",
162c03c5b1cSMartin Matuska                 symbol, normalizedCounter[symbol],
163c03c5b1cSMartin Matuska                 FSE_getMaxNbBits(symbolTT, symbol),
164c03c5b1cSMartin Matuska                 (double)FSE_bitCost(symbolTT, tableLog, symbol, 8) / 256);
165c03c5b1cSMartin Matuska         }
166c03c5b1cSMartin Matuska     }
167c03c5b1cSMartin Matuska #endif
168c03c5b1cSMartin Matuska 
169c03c5b1cSMartin Matuska     return 0;
170c03c5b1cSMartin Matuska }
171c03c5b1cSMartin Matuska 
172c03c5b1cSMartin Matuska 
FSE_buildCTable(FSE_CTable * ct,const short * normalizedCounter,unsigned maxSymbolValue,unsigned tableLog)173c03c5b1cSMartin Matuska size_t FSE_buildCTable(FSE_CTable* ct, const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
174c03c5b1cSMartin Matuska {
175c03c5b1cSMartin Matuska     FSE_FUNCTION_TYPE tableSymbol[FSE_MAX_TABLESIZE];   /* memset() is not necessary, even if static analyzer complain about it */
176c03c5b1cSMartin Matuska     return FSE_buildCTable_wksp(ct, normalizedCounter, maxSymbolValue, tableLog, tableSymbol, sizeof(tableSymbol));
177c03c5b1cSMartin Matuska }
178c03c5b1cSMartin Matuska 
179c03c5b1cSMartin Matuska 
180c03c5b1cSMartin Matuska 
181c03c5b1cSMartin Matuska #ifndef FSE_COMMONDEFS_ONLY
182c03c5b1cSMartin Matuska 
183c03c5b1cSMartin Matuska 
184c03c5b1cSMartin Matuska /*-**************************************************************
185c03c5b1cSMartin Matuska *  FSE NCount encoding
186c03c5b1cSMartin Matuska ****************************************************************/
FSE_NCountWriteBound(unsigned maxSymbolValue,unsigned tableLog)187c03c5b1cSMartin Matuska size_t FSE_NCountWriteBound(unsigned maxSymbolValue, unsigned tableLog)
188c03c5b1cSMartin Matuska {
189c03c5b1cSMartin Matuska     size_t const maxHeaderSize = (((maxSymbolValue+1) * tableLog) >> 3) + 3;
190c03c5b1cSMartin Matuska     return maxSymbolValue ? maxHeaderSize : FSE_NCOUNTBOUND;  /* maxSymbolValue==0 ? use default */
191c03c5b1cSMartin Matuska }
192c03c5b1cSMartin Matuska 
193c03c5b1cSMartin Matuska static size_t
FSE_writeNCount_generic(void * header,size_t headerBufferSize,const short * normalizedCounter,unsigned maxSymbolValue,unsigned tableLog,unsigned writeIsSafe)194c03c5b1cSMartin Matuska FSE_writeNCount_generic (void* header, size_t headerBufferSize,
195c03c5b1cSMartin Matuska                    const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog,
196c03c5b1cSMartin Matuska                          unsigned writeIsSafe)
197c03c5b1cSMartin Matuska {
198c03c5b1cSMartin Matuska     BYTE* const ostart = (BYTE*) header;
199c03c5b1cSMartin Matuska     BYTE* out = ostart;
200c03c5b1cSMartin Matuska     BYTE* const oend = ostart + headerBufferSize;
201c03c5b1cSMartin Matuska     int nbBits;
202c03c5b1cSMartin Matuska     const int tableSize = 1 << tableLog;
203c03c5b1cSMartin Matuska     int remaining;
204c03c5b1cSMartin Matuska     int threshold;
205c03c5b1cSMartin Matuska     U32 bitStream = 0;
206c03c5b1cSMartin Matuska     int bitCount = 0;
207c03c5b1cSMartin Matuska     unsigned symbol = 0;
208c03c5b1cSMartin Matuska     unsigned const alphabetSize = maxSymbolValue + 1;
209c03c5b1cSMartin Matuska     int previousIs0 = 0;
210c03c5b1cSMartin Matuska 
211c03c5b1cSMartin Matuska     /* Table Size */
212c03c5b1cSMartin Matuska     bitStream += (tableLog-FSE_MIN_TABLELOG) << bitCount;
213c03c5b1cSMartin Matuska     bitCount  += 4;
214c03c5b1cSMartin Matuska 
215c03c5b1cSMartin Matuska     /* Init */
216c03c5b1cSMartin Matuska     remaining = tableSize+1;   /* +1 for extra accuracy */
217c03c5b1cSMartin Matuska     threshold = tableSize;
218c03c5b1cSMartin Matuska     nbBits = tableLog+1;
219c03c5b1cSMartin Matuska 
220c03c5b1cSMartin Matuska     while ((symbol < alphabetSize) && (remaining>1)) {  /* stops at 1 */
221c03c5b1cSMartin Matuska         if (previousIs0) {
222c03c5b1cSMartin Matuska             unsigned start = symbol;
223c03c5b1cSMartin Matuska             while ((symbol < alphabetSize) && !normalizedCounter[symbol]) symbol++;
224c03c5b1cSMartin Matuska             if (symbol == alphabetSize) break;   /* incorrect distribution */
225c03c5b1cSMartin Matuska             while (symbol >= start+24) {
226c03c5b1cSMartin Matuska                 start+=24;
227c03c5b1cSMartin Matuska                 bitStream += 0xFFFFU << bitCount;
228c03c5b1cSMartin Matuska                 if ((!writeIsSafe) && (out > oend-2))
229c03c5b1cSMartin Matuska                     return ERROR(dstSize_tooSmall);   /* Buffer overflow */
230c03c5b1cSMartin Matuska                 out[0] = (BYTE) bitStream;
231c03c5b1cSMartin Matuska                 out[1] = (BYTE)(bitStream>>8);
232c03c5b1cSMartin Matuska                 out+=2;
233c03c5b1cSMartin Matuska                 bitStream>>=16;
234c03c5b1cSMartin Matuska             }
235c03c5b1cSMartin Matuska             while (symbol >= start+3) {
236c03c5b1cSMartin Matuska                 start+=3;
237c03c5b1cSMartin Matuska                 bitStream += 3 << bitCount;
238c03c5b1cSMartin Matuska                 bitCount += 2;
239c03c5b1cSMartin Matuska             }
240c03c5b1cSMartin Matuska             bitStream += (symbol-start) << bitCount;
241c03c5b1cSMartin Matuska             bitCount += 2;
242c03c5b1cSMartin Matuska             if (bitCount>16) {
243c03c5b1cSMartin Matuska                 if ((!writeIsSafe) && (out > oend - 2))
244c03c5b1cSMartin Matuska                     return ERROR(dstSize_tooSmall);   /* Buffer overflow */
245c03c5b1cSMartin Matuska                 out[0] = (BYTE)bitStream;
246c03c5b1cSMartin Matuska                 out[1] = (BYTE)(bitStream>>8);
247c03c5b1cSMartin Matuska                 out += 2;
248c03c5b1cSMartin Matuska                 bitStream >>= 16;
249c03c5b1cSMartin Matuska                 bitCount -= 16;
250c03c5b1cSMartin Matuska         }   }
251c03c5b1cSMartin Matuska         {   int count = normalizedCounter[symbol++];
252c03c5b1cSMartin Matuska             int const max = (2*threshold-1) - remaining;
253c03c5b1cSMartin Matuska             remaining -= count < 0 ? -count : count;
254c03c5b1cSMartin Matuska             count++;   /* +1 for extra accuracy */
255c03c5b1cSMartin Matuska             if (count>=threshold)
256c03c5b1cSMartin Matuska                 count += max;   /* [0..max[ [max..threshold[ (...) [threshold+max 2*threshold[ */
257c03c5b1cSMartin Matuska             bitStream += count << bitCount;
258c03c5b1cSMartin Matuska             bitCount  += nbBits;
259c03c5b1cSMartin Matuska             bitCount  -= (count<max);
260c03c5b1cSMartin Matuska             previousIs0  = (count==1);
261c03c5b1cSMartin Matuska             if (remaining<1) return ERROR(GENERIC);
262c03c5b1cSMartin Matuska             while (remaining<threshold) { nbBits--; threshold>>=1; }
263c03c5b1cSMartin Matuska         }
264c03c5b1cSMartin Matuska         if (bitCount>16) {
265c03c5b1cSMartin Matuska             if ((!writeIsSafe) && (out > oend - 2))
266c03c5b1cSMartin Matuska                 return ERROR(dstSize_tooSmall);   /* Buffer overflow */
267c03c5b1cSMartin Matuska             out[0] = (BYTE)bitStream;
268c03c5b1cSMartin Matuska             out[1] = (BYTE)(bitStream>>8);
269c03c5b1cSMartin Matuska             out += 2;
270c03c5b1cSMartin Matuska             bitStream >>= 16;
271c03c5b1cSMartin Matuska             bitCount -= 16;
272c03c5b1cSMartin Matuska     }   }
273c03c5b1cSMartin Matuska 
274c03c5b1cSMartin Matuska     if (remaining != 1)
275c03c5b1cSMartin Matuska         return ERROR(GENERIC);  /* incorrect normalized distribution */
276c03c5b1cSMartin Matuska     assert(symbol <= alphabetSize);
277c03c5b1cSMartin Matuska 
278c03c5b1cSMartin Matuska     /* flush remaining bitStream */
279c03c5b1cSMartin Matuska     if ((!writeIsSafe) && (out > oend - 2))
280c03c5b1cSMartin Matuska         return ERROR(dstSize_tooSmall);   /* Buffer overflow */
281c03c5b1cSMartin Matuska     out[0] = (BYTE)bitStream;
282c03c5b1cSMartin Matuska     out[1] = (BYTE)(bitStream>>8);
283c03c5b1cSMartin Matuska     out+= (bitCount+7) /8;
284c03c5b1cSMartin Matuska 
285c03c5b1cSMartin Matuska     return (out-ostart);
286c03c5b1cSMartin Matuska }
287c03c5b1cSMartin Matuska 
288c03c5b1cSMartin Matuska 
FSE_writeNCount(void * buffer,size_t bufferSize,const short * normalizedCounter,unsigned maxSymbolValue,unsigned tableLog)289c03c5b1cSMartin Matuska size_t FSE_writeNCount (void* buffer, size_t bufferSize,
290c03c5b1cSMartin Matuska                   const short* normalizedCounter, unsigned maxSymbolValue, unsigned tableLog)
291c03c5b1cSMartin Matuska {
292c03c5b1cSMartin Matuska     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported */
293c03c5b1cSMartin Matuska     if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported */
294c03c5b1cSMartin Matuska 
295c03c5b1cSMartin Matuska     if (bufferSize < FSE_NCountWriteBound(maxSymbolValue, tableLog))
296c03c5b1cSMartin Matuska         return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 0);
297c03c5b1cSMartin Matuska 
298c03c5b1cSMartin Matuska     return FSE_writeNCount_generic(buffer, bufferSize, normalizedCounter, maxSymbolValue, tableLog, 1 /* write in buffer is safe */);
299c03c5b1cSMartin Matuska }
300c03c5b1cSMartin Matuska 
301c03c5b1cSMartin Matuska 
302c03c5b1cSMartin Matuska /*-**************************************************************
303c03c5b1cSMartin Matuska *  FSE Compression Code
304c03c5b1cSMartin Matuska ****************************************************************/
305c03c5b1cSMartin Matuska 
FSE_createCTable(unsigned maxSymbolValue,unsigned tableLog)306c03c5b1cSMartin Matuska FSE_CTable* FSE_createCTable (unsigned maxSymbolValue, unsigned tableLog)
307c03c5b1cSMartin Matuska {
308716fd348SMartin Matuska     size_t size __attribute__ ((unused));
309c03c5b1cSMartin Matuska     if (tableLog > FSE_TABLELOG_ABSOLUTE_MAX) tableLog = FSE_TABLELOG_ABSOLUTE_MAX;
310c03c5b1cSMartin Matuska     size = FSE_CTABLE_SIZE_U32 (tableLog, maxSymbolValue) * sizeof(U32);
311c03c5b1cSMartin Matuska     return (FSE_CTable*)malloc(size);
312c03c5b1cSMartin Matuska }
313c03c5b1cSMartin Matuska 
FSE_freeCTable(FSE_CTable * ct)314c03c5b1cSMartin Matuska void FSE_freeCTable (FSE_CTable* ct) { free(ct); }
315c03c5b1cSMartin Matuska 
316c03c5b1cSMartin Matuska /* provides the minimum logSize to safely represent a distribution */
FSE_minTableLog(size_t srcSize,unsigned maxSymbolValue)317c03c5b1cSMartin Matuska static unsigned FSE_minTableLog(size_t srcSize, unsigned maxSymbolValue)
318c03c5b1cSMartin Matuska {
319c03c5b1cSMartin Matuska     U32 minBitsSrc = BIT_highbit32((U32)(srcSize)) + 1;
320c03c5b1cSMartin Matuska     U32 minBitsSymbols = BIT_highbit32(maxSymbolValue) + 2;
321c03c5b1cSMartin Matuska     U32 minBits = minBitsSrc < minBitsSymbols ? minBitsSrc : minBitsSymbols;
322c03c5b1cSMartin Matuska     assert(srcSize > 1); /* Not supported, RLE should be used instead */
323c03c5b1cSMartin Matuska     return minBits;
324c03c5b1cSMartin Matuska }
325c03c5b1cSMartin Matuska 
FSE_optimalTableLog_internal(unsigned maxTableLog,size_t srcSize,unsigned maxSymbolValue,unsigned minus)326c03c5b1cSMartin Matuska unsigned FSE_optimalTableLog_internal(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, unsigned minus)
327c03c5b1cSMartin Matuska {
328c03c5b1cSMartin Matuska     U32 maxBitsSrc = BIT_highbit32((U32)(srcSize - 1)) - minus;
329c03c5b1cSMartin Matuska     U32 tableLog = maxTableLog;
330c03c5b1cSMartin Matuska     U32 minBits = FSE_minTableLog(srcSize, maxSymbolValue);
331c03c5b1cSMartin Matuska     assert(srcSize > 1); /* Not supported, RLE should be used instead */
332c03c5b1cSMartin Matuska     if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
333c03c5b1cSMartin Matuska     if (maxBitsSrc < tableLog) tableLog = maxBitsSrc;   /* Accuracy can be reduced */
334c03c5b1cSMartin Matuska     if (minBits > tableLog) tableLog = minBits;   /* Need a minimum to safely represent all symbol values */
335c03c5b1cSMartin Matuska     if (tableLog < FSE_MIN_TABLELOG) tableLog = FSE_MIN_TABLELOG;
336c03c5b1cSMartin Matuska     if (tableLog > FSE_MAX_TABLELOG) tableLog = FSE_MAX_TABLELOG;
337c03c5b1cSMartin Matuska     return tableLog;
338c03c5b1cSMartin Matuska }
339c03c5b1cSMartin Matuska 
FSE_optimalTableLog(unsigned maxTableLog,size_t srcSize,unsigned maxSymbolValue)340c03c5b1cSMartin Matuska unsigned FSE_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue)
341c03c5b1cSMartin Matuska {
342c03c5b1cSMartin Matuska     return FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 2);
343c03c5b1cSMartin Matuska }
344c03c5b1cSMartin Matuska 
345c03c5b1cSMartin Matuska 
346c03c5b1cSMartin Matuska /* Secondary normalization method.
347c03c5b1cSMartin Matuska    To be used when primary method fails. */
348c03c5b1cSMartin Matuska 
FSE_normalizeM2(short * norm,U32 tableLog,const unsigned * count,size_t total,U32 maxSymbolValue)349c03c5b1cSMartin Matuska static size_t FSE_normalizeM2(short* norm, U32 tableLog, const unsigned* count, size_t total, U32 maxSymbolValue)
350c03c5b1cSMartin Matuska {
351c03c5b1cSMartin Matuska     short const NOT_YET_ASSIGNED = -2;
352c03c5b1cSMartin Matuska     U32 s;
353c03c5b1cSMartin Matuska     U32 distributed = 0;
354c03c5b1cSMartin Matuska     U32 ToDistribute;
355c03c5b1cSMartin Matuska 
356c03c5b1cSMartin Matuska     /* Init */
357c03c5b1cSMartin Matuska     U32 const lowThreshold = (U32)(total >> tableLog);
358c03c5b1cSMartin Matuska     U32 lowOne = (U32)((total * 3) >> (tableLog + 1));
359c03c5b1cSMartin Matuska 
360c03c5b1cSMartin Matuska     for (s=0; s<=maxSymbolValue; s++) {
361c03c5b1cSMartin Matuska         if (count[s] == 0) {
362c03c5b1cSMartin Matuska             norm[s]=0;
363c03c5b1cSMartin Matuska             continue;
364c03c5b1cSMartin Matuska         }
365c03c5b1cSMartin Matuska         if (count[s] <= lowThreshold) {
366c03c5b1cSMartin Matuska             norm[s] = -1;
367c03c5b1cSMartin Matuska             distributed++;
368c03c5b1cSMartin Matuska             total -= count[s];
369c03c5b1cSMartin Matuska             continue;
370c03c5b1cSMartin Matuska         }
371c03c5b1cSMartin Matuska         if (count[s] <= lowOne) {
372c03c5b1cSMartin Matuska             norm[s] = 1;
373c03c5b1cSMartin Matuska             distributed++;
374c03c5b1cSMartin Matuska             total -= count[s];
375c03c5b1cSMartin Matuska             continue;
376c03c5b1cSMartin Matuska         }
377c03c5b1cSMartin Matuska 
378c03c5b1cSMartin Matuska         norm[s]=NOT_YET_ASSIGNED;
379c03c5b1cSMartin Matuska     }
380c03c5b1cSMartin Matuska     ToDistribute = (1 << tableLog) - distributed;
381c03c5b1cSMartin Matuska 
382c03c5b1cSMartin Matuska     if (ToDistribute == 0)
383c03c5b1cSMartin Matuska         return 0;
384c03c5b1cSMartin Matuska 
385c03c5b1cSMartin Matuska     if ((total / ToDistribute) > lowOne) {
386c03c5b1cSMartin Matuska         /* risk of rounding to zero */
387c03c5b1cSMartin Matuska         lowOne = (U32)((total * 3) / (ToDistribute * 2));
388c03c5b1cSMartin Matuska         for (s=0; s<=maxSymbolValue; s++) {
389c03c5b1cSMartin Matuska             if ((norm[s] == NOT_YET_ASSIGNED) && (count[s] <= lowOne)) {
390c03c5b1cSMartin Matuska                 norm[s] = 1;
391c03c5b1cSMartin Matuska                 distributed++;
392c03c5b1cSMartin Matuska                 total -= count[s];
393c03c5b1cSMartin Matuska                 continue;
394c03c5b1cSMartin Matuska         }   }
395c03c5b1cSMartin Matuska         ToDistribute = (1 << tableLog) - distributed;
396c03c5b1cSMartin Matuska     }
397c03c5b1cSMartin Matuska 
398c03c5b1cSMartin Matuska     if (distributed == maxSymbolValue+1) {
399c03c5b1cSMartin Matuska         /* all values are pretty poor;
400c03c5b1cSMartin Matuska            probably incompressible data (should have already been detected);
401c03c5b1cSMartin Matuska            find max, then give all remaining points to max */
402c03c5b1cSMartin Matuska         U32 maxV = 0, maxC = 0;
403c03c5b1cSMartin Matuska         for (s=0; s<=maxSymbolValue; s++)
404c03c5b1cSMartin Matuska             if (count[s] > maxC) { maxV=s; maxC=count[s]; }
405c03c5b1cSMartin Matuska         norm[maxV] += (short)ToDistribute;
406c03c5b1cSMartin Matuska         return 0;
407c03c5b1cSMartin Matuska     }
408c03c5b1cSMartin Matuska 
409c03c5b1cSMartin Matuska     if (total == 0) {
410c03c5b1cSMartin Matuska         /* all of the symbols were low enough for the lowOne or lowThreshold */
411c03c5b1cSMartin Matuska         for (s=0; ToDistribute > 0; s = (s+1)%(maxSymbolValue+1))
412c03c5b1cSMartin Matuska             if (norm[s] > 0) { ToDistribute--; norm[s]++; }
413c03c5b1cSMartin Matuska         return 0;
414c03c5b1cSMartin Matuska     }
415c03c5b1cSMartin Matuska 
416c03c5b1cSMartin Matuska     {   U64 const vStepLog = 62 - tableLog;
417c03c5b1cSMartin Matuska         U64 const mid = (1ULL << (vStepLog-1)) - 1;
418c03c5b1cSMartin Matuska         U64 const rStep = ((((U64)1<<vStepLog) * ToDistribute) + mid) / total;   /* scale on remaining */
419c03c5b1cSMartin Matuska         U64 tmpTotal = mid;
420c03c5b1cSMartin Matuska         for (s=0; s<=maxSymbolValue; s++) {
421c03c5b1cSMartin Matuska             if (norm[s]==NOT_YET_ASSIGNED) {
422c03c5b1cSMartin Matuska                 U64 const end = tmpTotal + (count[s] * rStep);
423c03c5b1cSMartin Matuska                 U32 const sStart = (U32)(tmpTotal >> vStepLog);
424c03c5b1cSMartin Matuska                 U32 const sEnd = (U32)(end >> vStepLog);
425c03c5b1cSMartin Matuska                 U32 const weight = sEnd - sStart;
426c03c5b1cSMartin Matuska                 if (weight < 1)
427c03c5b1cSMartin Matuska                     return ERROR(GENERIC);
428c03c5b1cSMartin Matuska                 norm[s] = (short)weight;
429c03c5b1cSMartin Matuska                 tmpTotal = end;
430c03c5b1cSMartin Matuska     }   }   }
431c03c5b1cSMartin Matuska 
432c03c5b1cSMartin Matuska     return 0;
433c03c5b1cSMartin Matuska }
434c03c5b1cSMartin Matuska 
435c03c5b1cSMartin Matuska 
FSE_normalizeCount(short * normalizedCounter,unsigned tableLog,const unsigned * count,size_t total,unsigned maxSymbolValue)436c03c5b1cSMartin Matuska size_t FSE_normalizeCount (short* normalizedCounter, unsigned tableLog,
437c03c5b1cSMartin Matuska                            const unsigned* count, size_t total,
438c03c5b1cSMartin Matuska                            unsigned maxSymbolValue)
439c03c5b1cSMartin Matuska {
440c03c5b1cSMartin Matuska     /* Sanity checks */
441c03c5b1cSMartin Matuska     if (tableLog==0) tableLog = FSE_DEFAULT_TABLELOG;
442c03c5b1cSMartin Matuska     if (tableLog < FSE_MIN_TABLELOG) return ERROR(GENERIC);   /* Unsupported size */
443c03c5b1cSMartin Matuska     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);   /* Unsupported size */
444c03c5b1cSMartin Matuska     if (tableLog < FSE_minTableLog(total, maxSymbolValue)) return ERROR(GENERIC);   /* Too small tableLog, compression potentially impossible */
445c03c5b1cSMartin Matuska 
446c03c5b1cSMartin Matuska     {   static U32 const rtbTable[] = {     0, 473195, 504333, 520860, 550000, 700000, 750000, 830000 };
447c03c5b1cSMartin Matuska         U64 const scale = 62 - tableLog;
448c03c5b1cSMartin Matuska         U64 const step = ((U64)1<<62) / total;   /* <== here, one division ! */
449c03c5b1cSMartin Matuska         U64 const vStep = 1ULL<<(scale-20);
450c03c5b1cSMartin Matuska         int stillToDistribute = 1<<tableLog;
451c03c5b1cSMartin Matuska         unsigned s;
452c03c5b1cSMartin Matuska         unsigned largest=0;
453c03c5b1cSMartin Matuska         short largestP=0;
454c03c5b1cSMartin Matuska         U32 lowThreshold = (U32)(total >> tableLog);
455c03c5b1cSMartin Matuska 
456c03c5b1cSMartin Matuska         for (s=0; s<=maxSymbolValue; s++) {
457c03c5b1cSMartin Matuska             if (count[s] == total) return 0;   /* rle special case */
458c03c5b1cSMartin Matuska             if (count[s] == 0) { normalizedCounter[s]=0; continue; }
459c03c5b1cSMartin Matuska             if (count[s] <= lowThreshold) {
460c03c5b1cSMartin Matuska                 normalizedCounter[s] = -1;
461c03c5b1cSMartin Matuska                 stillToDistribute--;
462c03c5b1cSMartin Matuska             } else {
463c03c5b1cSMartin Matuska                 short proba = (short)((count[s]*step) >> scale);
464c03c5b1cSMartin Matuska                 if (proba<8) {
465c03c5b1cSMartin Matuska                     U64 restToBeat = vStep * rtbTable[proba];
466c03c5b1cSMartin Matuska                     proba += (count[s]*step) - ((U64)proba<<scale) > restToBeat;
467c03c5b1cSMartin Matuska                 }
468c03c5b1cSMartin Matuska                 if (proba > largestP) { largestP=proba; largest=s; }
469c03c5b1cSMartin Matuska                 normalizedCounter[s] = proba;
470c03c5b1cSMartin Matuska                 stillToDistribute -= proba;
471c03c5b1cSMartin Matuska         }   }
472c03c5b1cSMartin Matuska         if (-stillToDistribute >= (normalizedCounter[largest] >> 1)) {
473c03c5b1cSMartin Matuska             /* corner case, need another normalization method */
474c03c5b1cSMartin Matuska             size_t const errorCode = FSE_normalizeM2(normalizedCounter, tableLog, count, total, maxSymbolValue);
475c03c5b1cSMartin Matuska             if (FSE_isError(errorCode)) return errorCode;
476c03c5b1cSMartin Matuska         }
477c03c5b1cSMartin Matuska         else normalizedCounter[largest] += (short)stillToDistribute;
478c03c5b1cSMartin Matuska     }
479c03c5b1cSMartin Matuska 
480c03c5b1cSMartin Matuska #if 0
481c03c5b1cSMartin Matuska     {   /* Print Table (debug) */
482c03c5b1cSMartin Matuska         U32 s;
483c03c5b1cSMartin Matuska         U32 nTotal = 0;
484c03c5b1cSMartin Matuska         for (s=0; s<=maxSymbolValue; s++)
485c03c5b1cSMartin Matuska             RAWLOG(2, "%3i: %4i \n", s, normalizedCounter[s]);
486c03c5b1cSMartin Matuska         for (s=0; s<=maxSymbolValue; s++)
487c03c5b1cSMartin Matuska             nTotal += abs(normalizedCounter[s]);
488c03c5b1cSMartin Matuska         if (nTotal != (1U<<tableLog))
489c03c5b1cSMartin Matuska             RAWLOG(2, "Warning !!! Total == %u != %u !!!", nTotal, 1U<<tableLog);
490c03c5b1cSMartin Matuska         getchar();
491c03c5b1cSMartin Matuska     }
492c03c5b1cSMartin Matuska #endif
493c03c5b1cSMartin Matuska 
494c03c5b1cSMartin Matuska     return tableLog;
495c03c5b1cSMartin Matuska }
496c03c5b1cSMartin Matuska 
497c03c5b1cSMartin Matuska 
498c03c5b1cSMartin Matuska /* fake FSE_CTable, for raw (uncompressed) input */
FSE_buildCTable_raw(FSE_CTable * ct,unsigned nbBits)499c03c5b1cSMartin Matuska size_t FSE_buildCTable_raw (FSE_CTable* ct, unsigned nbBits)
500c03c5b1cSMartin Matuska {
501c03c5b1cSMartin Matuska     const unsigned tableSize = 1 << nbBits;
502c03c5b1cSMartin Matuska     const unsigned tableMask = tableSize - 1;
503c03c5b1cSMartin Matuska     const unsigned maxSymbolValue = tableMask;
504c03c5b1cSMartin Matuska     void* const ptr = ct;
505c03c5b1cSMartin Matuska     U16* const tableU16 = ( (U16*) ptr) + 2;
506c03c5b1cSMartin Matuska     void* const FSCT = ((U32*)ptr) + 1 /* header */ + (tableSize>>1);   /* assumption : tableLog >= 1 */
507c03c5b1cSMartin Matuska     FSE_symbolCompressionTransform* const symbolTT = (FSE_symbolCompressionTransform*) (FSCT);
508c03c5b1cSMartin Matuska     unsigned s;
509c03c5b1cSMartin Matuska 
510c03c5b1cSMartin Matuska     /* Sanity checks */
511c03c5b1cSMartin Matuska     if (nbBits < 1) return ERROR(GENERIC);             /* min size */
512c03c5b1cSMartin Matuska 
513c03c5b1cSMartin Matuska     /* header */
514c03c5b1cSMartin Matuska     tableU16[-2] = (U16) nbBits;
515c03c5b1cSMartin Matuska     tableU16[-1] = (U16) maxSymbolValue;
516c03c5b1cSMartin Matuska 
517c03c5b1cSMartin Matuska     /* Build table */
518c03c5b1cSMartin Matuska     for (s=0; s<tableSize; s++)
519c03c5b1cSMartin Matuska         tableU16[s] = (U16)(tableSize + s);
520c03c5b1cSMartin Matuska 
521c03c5b1cSMartin Matuska     /* Build Symbol Transformation Table */
522c03c5b1cSMartin Matuska     {   const U32 deltaNbBits = (nbBits << 16) - (1 << nbBits);
523c03c5b1cSMartin Matuska         for (s=0; s<=maxSymbolValue; s++) {
524c03c5b1cSMartin Matuska             symbolTT[s].deltaNbBits = deltaNbBits;
525c03c5b1cSMartin Matuska             symbolTT[s].deltaFindState = s-1;
526c03c5b1cSMartin Matuska     }   }
527c03c5b1cSMartin Matuska 
528c03c5b1cSMartin Matuska     return 0;
529c03c5b1cSMartin Matuska }
530c03c5b1cSMartin Matuska 
531c03c5b1cSMartin Matuska /* fake FSE_CTable, for rle input (always same symbol) */
FSE_buildCTable_rle(FSE_CTable * ct,BYTE symbolValue)532c03c5b1cSMartin Matuska size_t FSE_buildCTable_rle (FSE_CTable* ct, BYTE symbolValue)
533c03c5b1cSMartin Matuska {
534c03c5b1cSMartin Matuska     void* ptr = ct;
535c03c5b1cSMartin Matuska     U16* tableU16 = ( (U16*) ptr) + 2;
536c03c5b1cSMartin Matuska     void* FSCTptr = (U32*)ptr + 2;
537c03c5b1cSMartin Matuska     FSE_symbolCompressionTransform* symbolTT = (FSE_symbolCompressionTransform*) FSCTptr;
538c03c5b1cSMartin Matuska 
539c03c5b1cSMartin Matuska     /* header */
540c03c5b1cSMartin Matuska     tableU16[-2] = (U16) 0;
541c03c5b1cSMartin Matuska     tableU16[-1] = (U16) symbolValue;
542c03c5b1cSMartin Matuska 
543c03c5b1cSMartin Matuska     /* Build table */
544c03c5b1cSMartin Matuska     tableU16[0] = 0;
545c03c5b1cSMartin Matuska     tableU16[1] = 0;   /* just in case */
546c03c5b1cSMartin Matuska 
547c03c5b1cSMartin Matuska     /* Build Symbol Transformation Table */
548c03c5b1cSMartin Matuska     symbolTT[symbolValue].deltaNbBits = 0;
549c03c5b1cSMartin Matuska     symbolTT[symbolValue].deltaFindState = 0;
550c03c5b1cSMartin Matuska 
551c03c5b1cSMartin Matuska     return 0;
552c03c5b1cSMartin Matuska }
553c03c5b1cSMartin Matuska 
554c03c5b1cSMartin Matuska 
FSE_compress_usingCTable_generic(void * dst,size_t dstSize,const void * src,size_t srcSize,const FSE_CTable * ct,const unsigned fast)555c03c5b1cSMartin Matuska static size_t FSE_compress_usingCTable_generic (void* dst, size_t dstSize,
556c03c5b1cSMartin Matuska                            const void* src, size_t srcSize,
557c03c5b1cSMartin Matuska                            const FSE_CTable* ct, const unsigned fast)
558c03c5b1cSMartin Matuska {
559c03c5b1cSMartin Matuska     const BYTE* const istart = (const BYTE*) src;
560c03c5b1cSMartin Matuska     const BYTE* const iend = istart + srcSize;
561c03c5b1cSMartin Matuska     const BYTE* ip=iend;
562c03c5b1cSMartin Matuska 
563c03c5b1cSMartin Matuska     BIT_CStream_t bitC;
564c03c5b1cSMartin Matuska     FSE_CState_t CState1, CState2;
565c03c5b1cSMartin Matuska 
566c03c5b1cSMartin Matuska     /* init */
567c03c5b1cSMartin Matuska     if (srcSize <= 2) return 0;
568c03c5b1cSMartin Matuska     { size_t const initError = BIT_initCStream(&bitC, dst, dstSize);
569c03c5b1cSMartin Matuska       if (FSE_isError(initError)) return 0; /* not enough space available to write a bitstream */ }
570c03c5b1cSMartin Matuska 
571c03c5b1cSMartin Matuska #define FSE_FLUSHBITS(s)  (fast ? BIT_flushBitsFast(s) : BIT_flushBits(s))
572c03c5b1cSMartin Matuska 
573c03c5b1cSMartin Matuska     if (srcSize & 1) {
574c03c5b1cSMartin Matuska         FSE_initCState2(&CState1, ct, *--ip);
575c03c5b1cSMartin Matuska         FSE_initCState2(&CState2, ct, *--ip);
576c03c5b1cSMartin Matuska         FSE_encodeSymbol(&bitC, &CState1, *--ip);
577c03c5b1cSMartin Matuska         FSE_FLUSHBITS(&bitC);
578c03c5b1cSMartin Matuska     } else {
579c03c5b1cSMartin Matuska         FSE_initCState2(&CState2, ct, *--ip);
580c03c5b1cSMartin Matuska         FSE_initCState2(&CState1, ct, *--ip);
581c03c5b1cSMartin Matuska     }
582c03c5b1cSMartin Matuska 
583c03c5b1cSMartin Matuska     /* join to mod 4 */
584c03c5b1cSMartin Matuska     srcSize -= 2;
585c03c5b1cSMartin Matuska     if ((sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) && (srcSize & 2)) {  /* test bit 2 */
586c03c5b1cSMartin Matuska         FSE_encodeSymbol(&bitC, &CState2, *--ip);
587c03c5b1cSMartin Matuska         FSE_encodeSymbol(&bitC, &CState1, *--ip);
588c03c5b1cSMartin Matuska         FSE_FLUSHBITS(&bitC);
589c03c5b1cSMartin Matuska     }
590c03c5b1cSMartin Matuska 
591c03c5b1cSMartin Matuska     /* 2 or 4 encoding per loop */
592c03c5b1cSMartin Matuska     while ( ip>istart ) {
593c03c5b1cSMartin Matuska 
594c03c5b1cSMartin Matuska         FSE_encodeSymbol(&bitC, &CState2, *--ip);
595c03c5b1cSMartin Matuska 
596c03c5b1cSMartin Matuska         if (sizeof(bitC.bitContainer)*8 < FSE_MAX_TABLELOG*2+7 )   /* this test must be static */
597c03c5b1cSMartin Matuska             FSE_FLUSHBITS(&bitC);
598c03c5b1cSMartin Matuska 
599c03c5b1cSMartin Matuska         FSE_encodeSymbol(&bitC, &CState1, *--ip);
600c03c5b1cSMartin Matuska 
601c03c5b1cSMartin Matuska         if (sizeof(bitC.bitContainer)*8 > FSE_MAX_TABLELOG*4+7 ) {  /* this test must be static */
602c03c5b1cSMartin Matuska             FSE_encodeSymbol(&bitC, &CState2, *--ip);
603c03c5b1cSMartin Matuska             FSE_encodeSymbol(&bitC, &CState1, *--ip);
604c03c5b1cSMartin Matuska         }
605c03c5b1cSMartin Matuska 
606c03c5b1cSMartin Matuska         FSE_FLUSHBITS(&bitC);
607c03c5b1cSMartin Matuska     }
608c03c5b1cSMartin Matuska 
609c03c5b1cSMartin Matuska     FSE_flushCState(&bitC, &CState2);
610c03c5b1cSMartin Matuska     FSE_flushCState(&bitC, &CState1);
611c03c5b1cSMartin Matuska     return BIT_closeCStream(&bitC);
612c03c5b1cSMartin Matuska }
613c03c5b1cSMartin Matuska 
FSE_compress_usingCTable(void * dst,size_t dstSize,const void * src,size_t srcSize,const FSE_CTable * ct)614c03c5b1cSMartin Matuska size_t FSE_compress_usingCTable (void* dst, size_t dstSize,
615c03c5b1cSMartin Matuska                            const void* src, size_t srcSize,
616c03c5b1cSMartin Matuska                            const FSE_CTable* ct)
617c03c5b1cSMartin Matuska {
618c03c5b1cSMartin Matuska     unsigned const fast = (dstSize >= FSE_BLOCKBOUND(srcSize));
619c03c5b1cSMartin Matuska 
620c03c5b1cSMartin Matuska     if (fast)
621c03c5b1cSMartin Matuska         return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 1);
622c03c5b1cSMartin Matuska     else
623c03c5b1cSMartin Matuska         return FSE_compress_usingCTable_generic(dst, dstSize, src, srcSize, ct, 0);
624c03c5b1cSMartin Matuska }
625c03c5b1cSMartin Matuska 
626c03c5b1cSMartin Matuska 
FSE_compressBound(size_t size)627c03c5b1cSMartin Matuska size_t FSE_compressBound(size_t size) { return FSE_COMPRESSBOUND(size); }
628c03c5b1cSMartin Matuska 
629c03c5b1cSMartin Matuska /* FSE_compress_wksp() :
630c03c5b1cSMartin Matuska  * Same as FSE_compress2(), but using an externally allocated scratch buffer (`workSpace`).
631c03c5b1cSMartin Matuska  * `wkspSize` size must be `(1<<tableLog)`.
632c03c5b1cSMartin Matuska  */
FSE_compress_wksp(void * dst,size_t dstSize,const void * src,size_t srcSize,unsigned maxSymbolValue,unsigned tableLog,void * workSpace,size_t wkspSize)633c03c5b1cSMartin Matuska size_t FSE_compress_wksp (void* dst, size_t dstSize, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog, void* workSpace, size_t wkspSize)
634c03c5b1cSMartin Matuska {
635c03c5b1cSMartin Matuska     BYTE* const ostart = (BYTE*) dst;
636c03c5b1cSMartin Matuska     BYTE* op = ostart;
637c03c5b1cSMartin Matuska     BYTE* const oend = ostart + dstSize;
638c03c5b1cSMartin Matuska 
639c03c5b1cSMartin Matuska     unsigned count[FSE_MAX_SYMBOL_VALUE+1];
640c03c5b1cSMartin Matuska     S16   norm[FSE_MAX_SYMBOL_VALUE+1];
641c03c5b1cSMartin Matuska     FSE_CTable* CTable = (FSE_CTable*)workSpace;
642c03c5b1cSMartin Matuska     size_t const CTableSize = FSE_CTABLE_SIZE_U32(tableLog, maxSymbolValue);
643c03c5b1cSMartin Matuska     void* scratchBuffer = (void*)(CTable + CTableSize);
644c03c5b1cSMartin Matuska     size_t const scratchBufferSize = wkspSize - (CTableSize * sizeof(FSE_CTable));
645c03c5b1cSMartin Matuska 
646c03c5b1cSMartin Matuska     /* init conditions */
647c03c5b1cSMartin Matuska     if (wkspSize < FSE_WKSP_SIZE_U32(tableLog, maxSymbolValue)) return ERROR(tableLog_tooLarge);
648c03c5b1cSMartin Matuska     if (srcSize <= 1) return 0;  /* Not compressible */
649c03c5b1cSMartin Matuska     if (!maxSymbolValue) maxSymbolValue = FSE_MAX_SYMBOL_VALUE;
650c03c5b1cSMartin Matuska     if (!tableLog) tableLog = FSE_DEFAULT_TABLELOG;
651c03c5b1cSMartin Matuska 
652c03c5b1cSMartin Matuska     /* Scan input and build symbol stats */
653c03c5b1cSMartin Matuska     {   CHECK_V_F(maxCount, HIST_count_wksp(count, &maxSymbolValue, src, srcSize, scratchBuffer, scratchBufferSize) );
654c03c5b1cSMartin Matuska         if (maxCount == srcSize) return 1;   /* only a single symbol in src : rle */
655c03c5b1cSMartin Matuska         if (maxCount == 1) return 0;         /* each symbol present maximum once => not compressible */
656c03c5b1cSMartin Matuska         if (maxCount < (srcSize >> 7)) return 0;   /* Heuristic : not compressible enough */
657c03c5b1cSMartin Matuska     }
658c03c5b1cSMartin Matuska 
659c03c5b1cSMartin Matuska     tableLog = FSE_optimalTableLog(tableLog, srcSize, maxSymbolValue);
660c03c5b1cSMartin Matuska     CHECK_F( FSE_normalizeCount(norm, tableLog, count, srcSize, maxSymbolValue) );
661c03c5b1cSMartin Matuska 
662c03c5b1cSMartin Matuska     /* Write table description header */
663c03c5b1cSMartin Matuska     {   CHECK_V_F(nc_err, FSE_writeNCount(op, oend-op, norm, maxSymbolValue, tableLog) );
664c03c5b1cSMartin Matuska         op += nc_err;
665c03c5b1cSMartin Matuska     }
666c03c5b1cSMartin Matuska 
667c03c5b1cSMartin Matuska     /* Compress */
668c03c5b1cSMartin Matuska     CHECK_F( FSE_buildCTable_wksp(CTable, norm, maxSymbolValue, tableLog, scratchBuffer, scratchBufferSize) );
669c03c5b1cSMartin Matuska     {   CHECK_V_F(cSize, FSE_compress_usingCTable(op, oend - op, src, srcSize, CTable) );
670c03c5b1cSMartin Matuska         if (cSize == 0) return 0;   /* not enough space for compressed data */
671c03c5b1cSMartin Matuska         op += cSize;
672c03c5b1cSMartin Matuska     }
673c03c5b1cSMartin Matuska 
674c03c5b1cSMartin Matuska     /* check compressibility */
675c03c5b1cSMartin Matuska     if ( (size_t)(op-ostart) >= srcSize-1 ) return 0;
676c03c5b1cSMartin Matuska 
677c03c5b1cSMartin Matuska     return op-ostart;
678c03c5b1cSMartin Matuska }
679c03c5b1cSMartin Matuska 
680c03c5b1cSMartin Matuska typedef struct {
681c03c5b1cSMartin Matuska     FSE_CTable CTable_max[FSE_CTABLE_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE)];
682c03c5b1cSMartin Matuska     BYTE scratchBuffer[1 << FSE_MAX_TABLELOG];
683c03c5b1cSMartin Matuska } fseWkspMax_t;
684c03c5b1cSMartin Matuska 
FSE_compress2(void * dst,size_t dstCapacity,const void * src,size_t srcSize,unsigned maxSymbolValue,unsigned tableLog)685c03c5b1cSMartin Matuska size_t FSE_compress2 (void* dst, size_t dstCapacity, const void* src, size_t srcSize, unsigned maxSymbolValue, unsigned tableLog)
686c03c5b1cSMartin Matuska {
687c03c5b1cSMartin Matuska     fseWkspMax_t scratchBuffer;
688c03c5b1cSMartin Matuska     DEBUG_STATIC_ASSERT(sizeof(scratchBuffer) >= FSE_WKSP_SIZE_U32(FSE_MAX_TABLELOG, FSE_MAX_SYMBOL_VALUE));   /* compilation failures here means scratchBuffer is not large enough */
689c03c5b1cSMartin Matuska     if (tableLog > FSE_MAX_TABLELOG) return ERROR(tableLog_tooLarge);
690c03c5b1cSMartin Matuska     return FSE_compress_wksp(dst, dstCapacity, src, srcSize, maxSymbolValue, tableLog, &scratchBuffer, sizeof(scratchBuffer));
691c03c5b1cSMartin Matuska }
692c03c5b1cSMartin Matuska 
FSE_compress(void * dst,size_t dstCapacity,const void * src,size_t srcSize)693c03c5b1cSMartin Matuska size_t FSE_compress (void* dst, size_t dstCapacity, const void* src, size_t srcSize)
694c03c5b1cSMartin Matuska {
695c03c5b1cSMartin Matuska     return FSE_compress2(dst, dstCapacity, src, srcSize, FSE_MAX_SYMBOL_VALUE, FSE_DEFAULT_TABLELOG);
696c03c5b1cSMartin Matuska }
697c03c5b1cSMartin Matuska 
698c03c5b1cSMartin Matuska 
699c03c5b1cSMartin Matuska #endif   /* FSE_COMMONDEFS_ONLY */
700