2017-08-18 23:52:05 +00:00
|
|
|
/*
|
2016-08-30 17:04:33 +00:00
|
|
|
* Copyright (c) 2016-present, Yann Collet, Facebook, Inc.
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
2017-08-18 23:52:05 +00:00
|
|
|
* This source code is licensed under both the BSD-style license (found in the
|
|
|
|
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
|
|
|
|
* in the COPYING file in the root directory of this source tree).
|
2017-09-08 07:09:23 +00:00
|
|
|
* You may select, at your option, one of the above-listed licenses.
|
2016-08-30 17:04:33 +00:00
|
|
|
*/
|
2016-02-12 01:31:57 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
2016-12-21 12:23:34 +00:00
|
|
|
/* **************************************
|
|
|
|
* Compiler Warnings
|
|
|
|
****************************************/
|
|
|
|
#ifdef _MSC_VER
|
2017-09-14 22:12:57 +00:00
|
|
|
# pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
|
2016-12-21 12:23:34 +00:00
|
|
|
#endif
|
|
|
|
|
|
|
|
|
2016-02-12 01:31:57 +00:00
|
|
|
/*-*************************************
|
|
|
|
* Includes
|
|
|
|
***************************************/
|
2016-12-21 14:08:44 +00:00
|
|
|
#include "platform.h" /* Large Files support */
|
2016-12-21 12:23:34 +00:00
|
|
|
#include "util.h" /* UTIL_getFileSize, UTIL_getTotalFileSize */
|
2016-02-12 01:31:57 +00:00
|
|
|
#include <stdlib.h> /* malloc, free */
|
|
|
|
#include <string.h> /* memset */
|
|
|
|
#include <stdio.h> /* fprintf, fopen, ftello64 */
|
2016-07-06 14:27:17 +00:00
|
|
|
#include <errno.h> /* errno */
|
2018-08-15 21:35:38 +00:00
|
|
|
#include <assert.h>
|
2016-02-12 01:31:57 +00:00
|
|
|
|
2016-12-21 12:23:34 +00:00
|
|
|
#include "mem.h" /* read */
|
2016-02-12 01:31:57 +00:00
|
|
|
#include "error_private.h"
|
2016-04-22 10:43:18 +00:00
|
|
|
#include "dibio.h"
|
2016-02-12 01:31:57 +00:00
|
|
|
|
|
|
|
|
|
|
|
/*-*************************************
|
|
|
|
* Constants
|
|
|
|
***************************************/
|
|
|
|
#define KB *(1 <<10)
|
|
|
|
#define MB *(1 <<20)
|
|
|
|
#define GB *(1U<<30)
|
|
|
|
|
2016-12-18 10:58:23 +00:00
|
|
|
#define SAMPLESIZE_MAX (128 KB)
|
|
|
|
#define MEMMULT 11 /* rough estimation : memory cost to analyze 1 byte of sample */
|
2017-01-01 05:08:24 +00:00
|
|
|
#define COVER_MEMMULT 9 /* rough estimation : memory cost to analyze 1 byte of sample */
|
2018-08-23 19:06:20 +00:00
|
|
|
#define FASTCOVER_MEMMULT 1 /* rough estimation : memory cost to analyze 1 byte of sample */
|
2017-09-14 22:12:57 +00:00
|
|
|
static const size_t g_maxMemory = (sizeof(size_t) == 4) ? (2 GB - 64 MB) : ((size_t)(512 MB) << sizeof(size_t));
|
2016-02-12 01:31:57 +00:00
|
|
|
|
|
|
|
#define NOISELENGTH 32
|
|
|
|
|
|
|
|
|
|
|
|
/*-*************************************
|
|
|
|
* Console display
|
|
|
|
***************************************/
|
|
|
|
#define DISPLAY(...) fprintf(stderr, __VA_ARGS__)
|
2017-09-14 23:45:10 +00:00
|
|
|
#define DISPLAYLEVEL(l, ...) if (displayLevel>=l) { DISPLAY(__VA_ARGS__); }
|
2016-02-12 01:31:57 +00:00
|
|
|
|
2017-11-30 03:11:12 +00:00
|
|
|
static const U64 g_refreshRate = SEC_TO_MICRO / 6;
|
|
|
|
static UTIL_time_t g_displayClock = UTIL_TIME_INITIALIZER;
|
2016-05-09 02:44:45 +00:00
|
|
|
|
2017-11-30 03:11:12 +00:00
|
|
|
#define DISPLAYUPDATE(l, ...) { if (displayLevel>=l) { \
|
|
|
|
if ((UTIL_clockSpanMicro(g_displayClock) > g_refreshRate) || (displayLevel>=4)) \
|
|
|
|
{ g_displayClock = UTIL_getTime(); DISPLAY(__VA_ARGS__); \
|
|
|
|
if (displayLevel>=4) fflush(stderr); } } }
|
2016-02-12 01:31:57 +00:00
|
|
|
|
|
|
|
/*-*************************************
|
|
|
|
* Exceptions
|
|
|
|
***************************************/
|
|
|
|
#ifndef DEBUG
|
|
|
|
# define DEBUG 0
|
|
|
|
#endif
|
|
|
|
#define DEBUGOUTPUT(...) if (DEBUG) DISPLAY(__VA_ARGS__);
|
|
|
|
#define EXM_THROW(error, ...) \
|
|
|
|
{ \
|
|
|
|
DEBUGOUTPUT("Error defined at %s, line %i : \n", __FILE__, __LINE__); \
|
2017-09-14 23:45:10 +00:00
|
|
|
DISPLAY("Error %i : ", error); \
|
|
|
|
DISPLAY(__VA_ARGS__); \
|
|
|
|
DISPLAY("\n"); \
|
2016-02-12 01:31:57 +00:00
|
|
|
exit(error); \
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* ********************************************************
|
|
|
|
* Helper functions
|
|
|
|
**********************************************************/
|
2017-04-13 22:35:05 +00:00
|
|
|
#undef MIN
|
|
|
|
#define MIN(a,b) ((a) < (b) ? (a) : (b))
|
2016-07-06 13:41:03 +00:00
|
|
|
|
2016-02-12 01:31:57 +00:00
|
|
|
|
|
|
|
/* ********************************************************
|
|
|
|
* File related operations
|
|
|
|
**********************************************************/
|
2016-05-30 19:18:52 +00:00
|
|
|
/** DiB_loadFiles() :
|
2017-09-15 22:31:31 +00:00
|
|
|
* load samples from files listed in fileNamesTable into buffer.
|
|
|
|
* works even if buffer is too small to load all samples.
|
|
|
|
* Also provides the size of each sample into sampleSizes table
|
|
|
|
* which must be sized correctly, using DiB_fileStats().
|
|
|
|
* @return : nb of samples effectively loaded into `buffer`
|
|
|
|
* *bufferSizePtr is modified, it provides the amount data loaded within buffer.
|
|
|
|
* sampleSizes is filled with the size of each sample.
|
|
|
|
*/
|
2016-07-06 13:41:03 +00:00
|
|
|
static unsigned DiB_loadFiles(void* buffer, size_t* bufferSizePtr,
|
2017-09-15 22:31:31 +00:00
|
|
|
size_t* sampleSizes, unsigned sstSize,
|
2017-09-14 23:45:10 +00:00
|
|
|
const char** fileNamesTable, unsigned nbFiles, size_t targetChunkSize,
|
|
|
|
unsigned displayLevel)
|
2016-02-12 01:31:57 +00:00
|
|
|
{
|
2016-05-30 19:18:52 +00:00
|
|
|
char* const buff = (char*)buffer;
|
2016-02-12 01:31:57 +00:00
|
|
|
size_t pos = 0;
|
2017-09-14 23:45:10 +00:00
|
|
|
unsigned nbLoadedChunks = 0, fileIndex;
|
2016-02-12 01:31:57 +00:00
|
|
|
|
2017-09-14 23:45:10 +00:00
|
|
|
for (fileIndex=0; fileIndex<nbFiles; fileIndex++) {
|
|
|
|
const char* const fileName = fileNamesTable[fileIndex];
|
2016-07-06 13:41:03 +00:00
|
|
|
unsigned long long const fs64 = UTIL_getFileSize(fileName);
|
2017-10-17 23:14:25 +00:00
|
|
|
unsigned long long remainingToLoad = (fs64 == UTIL_FILESIZE_UNKNOWN) ? 0 : fs64;
|
2017-09-14 23:45:10 +00:00
|
|
|
U32 const nbChunks = targetChunkSize ? (U32)((fs64 + (targetChunkSize-1)) / targetChunkSize) : 1;
|
|
|
|
U64 const chunkSize = targetChunkSize ? MIN(targetChunkSize, fs64) : fs64;
|
2017-09-15 18:55:13 +00:00
|
|
|
size_t const maxChunkSize = (size_t)MIN(chunkSize, SAMPLESIZE_MAX);
|
2017-09-14 23:45:10 +00:00
|
|
|
U32 cnb;
|
|
|
|
FILE* const f = fopen(fileName, "rb");
|
|
|
|
if (f==NULL) EXM_THROW(10, "zstd: dictBuilder: %s %s ", fileName, strerror(errno));
|
|
|
|
DISPLAYUPDATE(2, "Loading %s... \r", fileName);
|
|
|
|
for (cnb=0; cnb<nbChunks; cnb++) {
|
2017-09-15 18:55:13 +00:00
|
|
|
size_t const toLoad = (size_t)MIN(maxChunkSize, remainingToLoad);
|
2017-09-14 23:45:10 +00:00
|
|
|
if (toLoad > *bufferSizePtr-pos) break;
|
|
|
|
{ size_t const readSize = fread(buff+pos, 1, toLoad, f);
|
|
|
|
if (readSize != toLoad) EXM_THROW(11, "Pb reading %s", fileName);
|
|
|
|
pos += readSize;
|
2017-09-15 22:31:31 +00:00
|
|
|
sampleSizes[nbLoadedChunks++] = toLoad;
|
2017-09-14 23:45:10 +00:00
|
|
|
remainingToLoad -= targetChunkSize;
|
2017-09-15 22:31:31 +00:00
|
|
|
if (nbLoadedChunks == sstSize) { /* no more space left in sampleSizes table */
|
|
|
|
fileIndex = nbFiles; /* stop there */
|
|
|
|
break;
|
|
|
|
}
|
2017-09-14 23:45:10 +00:00
|
|
|
if (toLoad < targetChunkSize) {
|
2017-09-15 17:16:26 +00:00
|
|
|
fseek(f, (long)(targetChunkSize - toLoad), SEEK_CUR);
|
2017-09-14 23:45:10 +00:00
|
|
|
} } }
|
|
|
|
fclose(f);
|
|
|
|
}
|
2017-01-01 05:08:24 +00:00
|
|
|
DISPLAYLEVEL(2, "\r%79s\r", "");
|
2016-07-06 13:41:03 +00:00
|
|
|
*bufferSizePtr = pos;
|
fix confusion between unsigned <-> U32
as suggested in #1441.
generally U32 and unsigned are the same thing,
except when they are not ...
case : 32-bit compilation for MIPS (uint32_t == unsigned long)
A vast majority of transformation consists in transforming U32 into unsigned.
In rare cases, it's the other way around (typically for internal code, such as seeds).
Among a few issues this patches solves :
- some parameters were declared with type `unsigned` in *.h,
but with type `U32` in their implementation *.c .
- some parameters have type unsigned*,
but the caller user a pointer to U32 instead.
These fixes are useful.
However, the bulk of changes is about %u formating,
which requires unsigned type,
but generally receives U32 values instead,
often just for brevity (U32 is shorter than unsigned).
These changes are generally minor, or even annoying.
As a consequence, the amount of code changed is larger than I would expect for such a patch.
Testing is also a pain :
it requires manually modifying `mem.h`,
in order to lie about `U32`
and force it to be an `unsigned long` typically.
On a 64-bit system, this will break the equivalence unsigned == U32.
Unfortunately, it will also break a few static_assert(), controlling structure sizes.
So it also requires modifying `debug.h` to make `static_assert()` a noop.
And then reverting these changes.
So it's inconvenient, and as a consequence,
this property is currently not checked during CI tests.
Therefore, these problems can emerge again in the future.
I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests.
It's another restriction for coding, adding more frustration during merge tests,
since most platforms don't need this distinction (hence contributor will not see it),
and while this can matter in theory, the number of platforms impacted seems minimal.
Thoughts ?
2018-12-22 00:19:44 +00:00
|
|
|
DISPLAYLEVEL(4, "loaded : %u KB \n", (unsigned)(pos >> 10))
|
2017-09-14 23:45:10 +00:00
|
|
|
return nbLoadedChunks;
|
2016-02-12 01:31:57 +00:00
|
|
|
}
|
|
|
|
|
2017-01-01 05:08:24 +00:00
|
|
|
#define DiB_rotl32(x,r) ((x << r) | (x >> (32 - r)))
|
|
|
|
static U32 DiB_rand(U32* src)
|
|
|
|
{
|
|
|
|
static const U32 prime1 = 2654435761U;
|
|
|
|
static const U32 prime2 = 2246822519U;
|
|
|
|
U32 rand32 = *src;
|
|
|
|
rand32 *= prime1;
|
|
|
|
rand32 ^= prime2;
|
|
|
|
rand32 = DiB_rotl32(rand32, 13);
|
|
|
|
*src = rand32;
|
|
|
|
return rand32 >> 5;
|
|
|
|
}
|
|
|
|
|
2017-09-14 22:12:57 +00:00
|
|
|
/* DiB_shuffle() :
|
|
|
|
* shuffle a table of file names in a semi-random way
|
|
|
|
* It improves dictionary quality by reducing "locality" impact, so if sample set is very large,
|
|
|
|
* it will load random elements from it, instead of just the first ones. */
|
2017-01-01 05:08:24 +00:00
|
|
|
static void DiB_shuffle(const char** fileNamesTable, unsigned nbFiles) {
|
2017-09-14 22:12:57 +00:00
|
|
|
U32 seed = 0xFD2FB528;
|
|
|
|
unsigned i;
|
2018-08-15 21:35:38 +00:00
|
|
|
assert(nbFiles >= 1);
|
2017-09-14 22:12:57 +00:00
|
|
|
for (i = nbFiles - 1; i > 0; --i) {
|
|
|
|
unsigned const j = DiB_rand(&seed) % (i + 1);
|
|
|
|
const char* const tmp = fileNamesTable[j];
|
|
|
|
fileNamesTable[j] = fileNamesTable[i];
|
|
|
|
fileNamesTable[i] = tmp;
|
|
|
|
}
|
2017-01-01 05:08:24 +00:00
|
|
|
}
|
|
|
|
|
2016-02-12 01:31:57 +00:00
|
|
|
|
|
|
|
/*-********************************************************
|
|
|
|
* Dictionary training functions
|
|
|
|
**********************************************************/
|
|
|
|
static size_t DiB_findMaxMem(unsigned long long requiredMem)
|
|
|
|
{
|
2016-05-30 19:18:52 +00:00
|
|
|
size_t const step = 8 MB;
|
2016-02-12 01:31:57 +00:00
|
|
|
void* testmem = NULL;
|
|
|
|
|
|
|
|
requiredMem = (((requiredMem >> 23) + 1) << 23);
|
2016-07-06 13:41:03 +00:00
|
|
|
requiredMem += step;
|
2017-09-14 22:12:57 +00:00
|
|
|
if (requiredMem > g_maxMemory) requiredMem = g_maxMemory;
|
2016-02-12 01:31:57 +00:00
|
|
|
|
|
|
|
while (!testmem) {
|
|
|
|
testmem = malloc((size_t)requiredMem);
|
2016-07-06 13:41:03 +00:00
|
|
|
requiredMem -= step;
|
2016-02-12 01:31:57 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
free(testmem);
|
2016-07-06 13:41:03 +00:00
|
|
|
return (size_t)requiredMem;
|
2016-02-12 01:31:57 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void DiB_fillNoise(void* buffer, size_t length)
|
|
|
|
{
|
2016-07-06 13:41:03 +00:00
|
|
|
unsigned const prime1 = 2654435761U;
|
|
|
|
unsigned const prime2 = 2246822519U;
|
|
|
|
unsigned acc = prime1;
|
2016-02-12 01:31:57 +00:00
|
|
|
size_t p=0;;
|
|
|
|
|
|
|
|
for (p=0; p<length; p++) {
|
2016-07-06 13:41:03 +00:00
|
|
|
acc *= prime2;
|
2016-02-12 01:31:57 +00:00
|
|
|
((unsigned char*)buffer)[p] = (unsigned char)(acc >> 21);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void DiB_saveDict(const char* dictFileName,
|
|
|
|
const void* buff, size_t buffSize)
|
|
|
|
{
|
2016-05-30 19:18:52 +00:00
|
|
|
FILE* const f = fopen(dictFileName, "wb");
|
2016-02-12 01:31:57 +00:00
|
|
|
if (f==NULL) EXM_THROW(3, "cannot open %s ", dictFileName);
|
|
|
|
|
2016-05-09 02:44:45 +00:00
|
|
|
{ size_t const n = fwrite(buff, 1, buffSize, f);
|
|
|
|
if (n!=buffSize) EXM_THROW(4, "%s : write error", dictFileName) }
|
2016-02-12 01:31:57 +00:00
|
|
|
|
2016-05-09 02:44:45 +00:00
|
|
|
{ size_t const n = (size_t)fclose(f);
|
|
|
|
if (n!=0) EXM_THROW(5, "%s : flush error", dictFileName) }
|
2016-02-12 01:31:57 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2017-09-14 23:45:10 +00:00
|
|
|
typedef struct {
|
|
|
|
U64 totalSizeToLoad;
|
|
|
|
unsigned oneSampleTooLarge;
|
2017-09-15 22:31:31 +00:00
|
|
|
unsigned nbSamples;
|
2017-09-14 23:45:10 +00:00
|
|
|
} fileStats;
|
|
|
|
|
2017-09-15 22:31:31 +00:00
|
|
|
/*! DiB_fileStats() :
|
|
|
|
* Given a list of files, and a chunkSize (0 == no chunk, whole files)
|
|
|
|
* provides the amount of data to be loaded and the resulting nb of samples.
|
|
|
|
* This is useful primarily for allocation purpose => sample buffer, and sample sizes table.
|
|
|
|
*/
|
2017-09-14 23:45:10 +00:00
|
|
|
static fileStats DiB_fileStats(const char** fileNamesTable, unsigned nbFiles, size_t chunkSize, unsigned displayLevel)
|
2016-12-18 10:58:23 +00:00
|
|
|
{
|
2017-09-14 23:45:10 +00:00
|
|
|
fileStats fs;
|
2016-12-18 10:58:23 +00:00
|
|
|
unsigned n;
|
2017-09-14 23:45:10 +00:00
|
|
|
memset(&fs, 0, sizeof(fs));
|
2016-12-18 10:58:23 +00:00
|
|
|
for (n=0; n<nbFiles; n++) {
|
|
|
|
U64 const fileSize = UTIL_getFileSize(fileNamesTable[n]);
|
2017-10-17 23:14:25 +00:00
|
|
|
U64 const srcSize = (fileSize == UTIL_FILESIZE_UNKNOWN) ? 0 : fileSize;
|
|
|
|
U32 const nbSamples = (U32)(chunkSize ? (srcSize + (chunkSize-1)) / chunkSize : 1);
|
|
|
|
U64 const chunkToLoad = chunkSize ? MIN(chunkSize, srcSize) : srcSize;
|
2017-09-15 18:55:13 +00:00
|
|
|
size_t const cappedChunkSize = (size_t)MIN(chunkToLoad, SAMPLESIZE_MAX);
|
2017-09-15 22:31:31 +00:00
|
|
|
fs.totalSizeToLoad += cappedChunkSize * nbSamples;
|
2017-09-14 23:45:10 +00:00
|
|
|
fs.oneSampleTooLarge |= (chunkSize > 2*SAMPLESIZE_MAX);
|
2017-09-15 22:31:31 +00:00
|
|
|
fs.nbSamples += nbSamples;
|
2016-12-18 10:58:23 +00:00
|
|
|
}
|
fix confusion between unsigned <-> U32
as suggested in #1441.
generally U32 and unsigned are the same thing,
except when they are not ...
case : 32-bit compilation for MIPS (uint32_t == unsigned long)
A vast majority of transformation consists in transforming U32 into unsigned.
In rare cases, it's the other way around (typically for internal code, such as seeds).
Among a few issues this patches solves :
- some parameters were declared with type `unsigned` in *.h,
but with type `U32` in their implementation *.c .
- some parameters have type unsigned*,
but the caller user a pointer to U32 instead.
These fixes are useful.
However, the bulk of changes is about %u formating,
which requires unsigned type,
but generally receives U32 values instead,
often just for brevity (U32 is shorter than unsigned).
These changes are generally minor, or even annoying.
As a consequence, the amount of code changed is larger than I would expect for such a patch.
Testing is also a pain :
it requires manually modifying `mem.h`,
in order to lie about `U32`
and force it to be an `unsigned long` typically.
On a 64-bit system, this will break the equivalence unsigned == U32.
Unfortunately, it will also break a few static_assert(), controlling structure sizes.
So it also requires modifying `debug.h` to make `static_assert()` a noop.
And then reverting these changes.
So it's inconvenient, and as a consequence,
this property is currently not checked during CI tests.
Therefore, these problems can emerge again in the future.
I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests.
It's another restriction for coding, adding more frustration during merge tests,
since most platforms don't need this distinction (hence contributor will not see it),
and while this can matter in theory, the number of platforms impacted seems minimal.
Thoughts ?
2018-12-22 00:19:44 +00:00
|
|
|
DISPLAYLEVEL(4, "Preparing to load : %u KB \n", (unsigned)(fs.totalSizeToLoad >> 10));
|
2017-09-14 23:45:10 +00:00
|
|
|
return fs;
|
2016-12-18 10:58:23 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2017-06-27 04:07:14 +00:00
|
|
|
/*! ZDICT_trainFromBuffer_unsafe_legacy() :
|
2016-02-12 19:19:48 +00:00
|
|
|
Strictly Internal use only !!
|
2017-06-27 04:07:14 +00:00
|
|
|
Same as ZDICT_trainFromBuffer_legacy(), but does not control `samplesBuffer`.
|
2016-02-12 19:19:48 +00:00
|
|
|
`samplesBuffer` must be followed by noisy guard band to avoid out-of-buffer reads.
|
|
|
|
@return : size of dictionary stored into `dictBuffer` (<= `dictBufferCapacity`)
|
|
|
|
or an error code.
|
|
|
|
*/
|
2017-06-27 04:07:14 +00:00
|
|
|
size_t ZDICT_trainFromBuffer_unsafe_legacy(void* dictBuffer, size_t dictBufferCapacity,
|
|
|
|
const void* samplesBuffer, const size_t* samplesSizes, unsigned nbSamples,
|
|
|
|
ZDICT_legacy_params_t parameters);
|
2016-02-12 19:19:48 +00:00
|
|
|
|
|
|
|
|
2016-02-12 01:31:57 +00:00
|
|
|
int DiB_trainFromFiles(const char* dictFileName, unsigned maxDictSize,
|
2017-09-14 23:45:10 +00:00
|
|
|
const char** fileNamesTable, unsigned nbFiles, size_t chunkSize,
|
2018-08-23 19:06:20 +00:00
|
|
|
ZDICT_legacy_params_t* params, ZDICT_cover_params_t* coverParams,
|
|
|
|
ZDICT_fastCover_params_t* fastCoverParams, int optimize)
|
2016-02-12 01:31:57 +00:00
|
|
|
{
|
2017-09-15 22:31:31 +00:00
|
|
|
unsigned const displayLevel = params ? params->zParams.notificationLevel :
|
|
|
|
coverParams ? coverParams->zParams.notificationLevel :
|
2018-08-23 19:06:20 +00:00
|
|
|
fastCoverParams ? fastCoverParams->zParams.notificationLevel :
|
2017-09-15 22:31:31 +00:00
|
|
|
0; /* should never happen */
|
2016-05-30 19:18:52 +00:00
|
|
|
void* const dictBuffer = malloc(maxDictSize);
|
2017-09-14 23:45:10 +00:00
|
|
|
fileStats const fs = DiB_fileStats(fileNamesTable, nbFiles, chunkSize, displayLevel);
|
2017-09-15 22:31:31 +00:00
|
|
|
size_t* const sampleSizes = (size_t*)malloc(fs.nbSamples * sizeof(size_t));
|
2018-08-23 19:06:20 +00:00
|
|
|
size_t const memMult = params ? MEMMULT :
|
|
|
|
coverParams ? COVER_MEMMULT:
|
|
|
|
FASTCOVER_MEMMULT;
|
2017-09-14 23:45:10 +00:00
|
|
|
size_t const maxMem = DiB_findMaxMem(fs.totalSizeToLoad * memMult) / memMult;
|
|
|
|
size_t loadedSize = (size_t) MIN ((unsigned long long)maxMem, fs.totalSizeToLoad);
|
|
|
|
void* const srcBuffer = malloc(loadedSize+NOISELENGTH);
|
2016-02-12 01:31:57 +00:00
|
|
|
int result = 0;
|
|
|
|
|
2016-05-30 19:18:52 +00:00
|
|
|
/* Checks */
|
2017-09-15 22:31:31 +00:00
|
|
|
if ((!sampleSizes) || (!srcBuffer) || (!dictBuffer))
|
2017-09-14 22:12:57 +00:00
|
|
|
EXM_THROW(12, "not enough memory for DiB_trainFiles"); /* should not happen */
|
2017-09-14 23:45:10 +00:00
|
|
|
if (fs.oneSampleTooLarge) {
|
|
|
|
DISPLAYLEVEL(2, "! Warning : some sample(s) are very large \n");
|
|
|
|
DISPLAYLEVEL(2, "! Note that dictionary is only useful for small samples. \n");
|
|
|
|
DISPLAYLEVEL(2, "! As a consequence, only the first %u bytes of each sample are loaded \n", SAMPLESIZE_MAX);
|
2016-12-18 10:58:23 +00:00
|
|
|
}
|
2017-09-15 22:31:31 +00:00
|
|
|
if (fs.nbSamples < 5) {
|
2016-08-18 13:02:11 +00:00
|
|
|
DISPLAYLEVEL(2, "! Warning : nb of samples too low for proper processing ! \n");
|
|
|
|
DISPLAYLEVEL(2, "! Please provide _one file per sample_. \n");
|
2017-09-15 23:23:50 +00:00
|
|
|
DISPLAYLEVEL(2, "! Alternatively, split files into fixed-size blocks representative of samples, with -B# \n");
|
2017-09-14 23:45:10 +00:00
|
|
|
EXM_THROW(14, "nb of samples too low"); /* we now clearly forbid this case */
|
|
|
|
}
|
|
|
|
if (fs.totalSizeToLoad < (unsigned long long)(8 * maxDictSize)) {
|
|
|
|
DISPLAYLEVEL(2, "! Warning : data size of samples too small for target dictionary size \n");
|
|
|
|
DISPLAYLEVEL(2, "! Samples should be about 100x larger than target dictionary size \n");
|
2016-07-27 10:35:29 +00:00
|
|
|
}
|
2016-05-30 19:18:52 +00:00
|
|
|
|
2016-02-12 01:31:57 +00:00
|
|
|
/* init */
|
2017-09-14 23:45:10 +00:00
|
|
|
if (loadedSize < fs.totalSizeToLoad)
|
|
|
|
DISPLAYLEVEL(1, "Not enough memory; training on %u MB only...\n", (unsigned)(loadedSize >> 20));
|
2016-02-12 01:31:57 +00:00
|
|
|
|
|
|
|
/* Load input buffer */
|
2017-01-01 05:08:24 +00:00
|
|
|
DISPLAYLEVEL(3, "Shuffling input files\n");
|
|
|
|
DiB_shuffle(fileNamesTable, nbFiles);
|
2018-08-23 19:06:20 +00:00
|
|
|
|
2018-08-15 21:35:38 +00:00
|
|
|
DiB_loadFiles(srcBuffer, &loadedSize, sampleSizes, fs.nbSamples, fileNamesTable, nbFiles, chunkSize, displayLevel);
|
2016-02-12 01:31:57 +00:00
|
|
|
|
2017-09-14 22:12:57 +00:00
|
|
|
{ size_t dictSize;
|
2017-01-01 05:08:24 +00:00
|
|
|
if (params) {
|
2017-09-14 23:45:10 +00:00
|
|
|
DiB_fillNoise((char*)srcBuffer + loadedSize, NOISELENGTH); /* guard band, for end of buffer condition */
|
2017-06-27 04:07:14 +00:00
|
|
|
dictSize = ZDICT_trainFromBuffer_unsafe_legacy(dictBuffer, maxDictSize,
|
2017-09-15 22:31:31 +00:00
|
|
|
srcBuffer, sampleSizes, fs.nbSamples,
|
2017-06-27 04:07:14 +00:00
|
|
|
*params);
|
2018-08-23 19:06:20 +00:00
|
|
|
} else if (coverParams) {
|
|
|
|
if (optimize) {
|
|
|
|
dictSize = ZDICT_optimizeTrainFromBuffer_cover(dictBuffer, maxDictSize,
|
|
|
|
srcBuffer, sampleSizes, fs.nbSamples,
|
|
|
|
coverParams);
|
|
|
|
if (!ZDICT_isError(dictSize)) {
|
|
|
|
unsigned splitPercentage = (unsigned)(coverParams->splitPoint * 100);
|
|
|
|
DISPLAYLEVEL(2, "k=%u\nd=%u\nsteps=%u\nsplit=%u\n", coverParams->k, coverParams->d,
|
|
|
|
coverParams->steps, splitPercentage);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
dictSize = ZDICT_trainFromBuffer_cover(dictBuffer, maxDictSize, srcBuffer,
|
|
|
|
sampleSizes, fs.nbSamples, *coverParams);
|
2017-01-01 05:08:24 +00:00
|
|
|
}
|
|
|
|
} else {
|
2018-08-23 19:06:20 +00:00
|
|
|
assert(fastCoverParams != NULL);
|
|
|
|
if (optimize) {
|
|
|
|
dictSize = ZDICT_optimizeTrainFromBuffer_fastCover(dictBuffer, maxDictSize,
|
|
|
|
srcBuffer, sampleSizes, fs.nbSamples,
|
|
|
|
fastCoverParams);
|
|
|
|
if (!ZDICT_isError(dictSize)) {
|
|
|
|
unsigned splitPercentage = (unsigned)(fastCoverParams->splitPoint * 100);
|
|
|
|
DISPLAYLEVEL(2, "k=%u\nd=%u\nf=%u\nsteps=%u\nsplit=%u\naccel=%u\n", fastCoverParams->k,
|
|
|
|
fastCoverParams->d, fastCoverParams->f, fastCoverParams->steps, splitPercentage,
|
|
|
|
fastCoverParams->accel);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
dictSize = ZDICT_trainFromBuffer_fastCover(dictBuffer, maxDictSize, srcBuffer,
|
|
|
|
sampleSizes, fs.nbSamples, *fastCoverParams);
|
|
|
|
}
|
2017-01-01 05:08:24 +00:00
|
|
|
}
|
2016-05-30 19:18:52 +00:00
|
|
|
if (ZDICT_isError(dictSize)) {
|
|
|
|
DISPLAYLEVEL(1, "dictionary training failed : %s \n", ZDICT_getErrorName(dictSize)); /* should not happen */
|
|
|
|
result = 1;
|
|
|
|
goto _cleanup;
|
|
|
|
}
|
|
|
|
/* save dict */
|
fix confusion between unsigned <-> U32
as suggested in #1441.
generally U32 and unsigned are the same thing,
except when they are not ...
case : 32-bit compilation for MIPS (uint32_t == unsigned long)
A vast majority of transformation consists in transforming U32 into unsigned.
In rare cases, it's the other way around (typically for internal code, such as seeds).
Among a few issues this patches solves :
- some parameters were declared with type `unsigned` in *.h,
but with type `U32` in their implementation *.c .
- some parameters have type unsigned*,
but the caller user a pointer to U32 instead.
These fixes are useful.
However, the bulk of changes is about %u formating,
which requires unsigned type,
but generally receives U32 values instead,
often just for brevity (U32 is shorter than unsigned).
These changes are generally minor, or even annoying.
As a consequence, the amount of code changed is larger than I would expect for such a patch.
Testing is also a pain :
it requires manually modifying `mem.h`,
in order to lie about `U32`
and force it to be an `unsigned long` typically.
On a 64-bit system, this will break the equivalence unsigned == U32.
Unfortunately, it will also break a few static_assert(), controlling structure sizes.
So it also requires modifying `debug.h` to make `static_assert()` a noop.
And then reverting these changes.
So it's inconvenient, and as a consequence,
this property is currently not checked during CI tests.
Therefore, these problems can emerge again in the future.
I wonder if it is worth ensuring proper distinction of U32 != unsigned in CI tests.
It's another restriction for coding, adding more frustration during merge tests,
since most platforms don't need this distinction (hence contributor will not see it),
and while this can matter in theory, the number of platforms impacted seems minimal.
Thoughts ?
2018-12-22 00:19:44 +00:00
|
|
|
DISPLAYLEVEL(2, "Save dictionary of size %u into file %s \n", (unsigned)dictSize, dictFileName);
|
2016-05-30 19:18:52 +00:00
|
|
|
DiB_saveDict(dictFileName, dictBuffer, dictSize);
|
2016-02-12 01:31:57 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/* clean up */
|
|
|
|
_cleanup:
|
|
|
|
free(srcBuffer);
|
2017-09-15 22:31:31 +00:00
|
|
|
free(sampleSizes);
|
2016-02-12 01:31:57 +00:00
|
|
|
free(dictBuffer);
|
|
|
|
return result;
|
|
|
|
}
|