extzstd 0.2 → 0.3
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/HISTORY.ja.md +13 -0
- data/README.md +17 -14
- data/contrib/zstd/{NEWS → CHANGELOG} +115 -2
- data/contrib/zstd/CODE_OF_CONDUCT.md +5 -0
- data/contrib/zstd/Makefile +99 -53
- data/contrib/zstd/README.md +59 -39
- data/contrib/zstd/TESTING.md +1 -1
- data/contrib/zstd/appveyor.yml +17 -6
- data/contrib/zstd/lib/BUCK +29 -2
- data/contrib/zstd/lib/Makefile +118 -21
- data/contrib/zstd/lib/README.md +84 -44
- data/contrib/zstd/lib/common/bitstream.h +17 -33
- data/contrib/zstd/lib/common/compiler.h +62 -8
- data/contrib/zstd/lib/common/cpu.h +215 -0
- data/contrib/zstd/lib/common/debug.c +44 -0
- data/contrib/zstd/lib/common/debug.h +134 -0
- data/contrib/zstd/lib/common/entropy_common.c +16 -1
- data/contrib/zstd/lib/common/error_private.c +7 -0
- data/contrib/zstd/lib/common/fse.h +48 -44
- data/contrib/zstd/lib/common/fse_decompress.c +3 -3
- data/contrib/zstd/lib/common/huf.h +169 -113
- data/contrib/zstd/lib/common/mem.h +20 -2
- data/contrib/zstd/lib/common/pool.c +135 -49
- data/contrib/zstd/lib/common/pool.h +40 -21
- data/contrib/zstd/lib/common/threading.c +2 -2
- data/contrib/zstd/lib/common/threading.h +12 -12
- data/contrib/zstd/lib/common/xxhash.c +3 -2
- data/contrib/zstd/lib/common/zstd_common.c +3 -6
- data/contrib/zstd/lib/common/zstd_errors.h +17 -7
- data/contrib/zstd/lib/common/zstd_internal.h +76 -48
- data/contrib/zstd/lib/compress/fse_compress.c +89 -209
- data/contrib/zstd/lib/compress/hist.c +203 -0
- data/contrib/zstd/lib/compress/hist.h +95 -0
- data/contrib/zstd/lib/compress/huf_compress.c +188 -80
- data/contrib/zstd/lib/compress/zstd_compress.c +2500 -1203
- data/contrib/zstd/lib/compress/zstd_compress_internal.h +463 -62
- data/contrib/zstd/lib/compress/zstd_double_fast.c +321 -131
- data/contrib/zstd/lib/compress/zstd_double_fast.h +13 -4
- data/contrib/zstd/lib/compress/zstd_fast.c +335 -108
- data/contrib/zstd/lib/compress/zstd_fast.h +12 -6
- data/contrib/zstd/lib/compress/zstd_lazy.c +654 -313
- data/contrib/zstd/lib/compress/zstd_lazy.h +44 -16
- data/contrib/zstd/lib/compress/zstd_ldm.c +310 -420
- data/contrib/zstd/lib/compress/zstd_ldm.h +63 -26
- data/contrib/zstd/lib/compress/zstd_opt.c +773 -325
- data/contrib/zstd/lib/compress/zstd_opt.h +31 -5
- data/contrib/zstd/lib/compress/zstdmt_compress.c +1468 -518
- data/contrib/zstd/lib/compress/zstdmt_compress.h +96 -45
- data/contrib/zstd/lib/decompress/huf_decompress.c +518 -282
- data/contrib/zstd/lib/decompress/zstd_ddict.c +240 -0
- data/contrib/zstd/lib/decompress/zstd_ddict.h +44 -0
- data/contrib/zstd/lib/decompress/zstd_decompress.c +613 -1513
- data/contrib/zstd/lib/decompress/zstd_decompress_block.c +1311 -0
- data/contrib/zstd/lib/decompress/zstd_decompress_block.h +59 -0
- data/contrib/zstd/lib/decompress/zstd_decompress_internal.h +175 -0
- data/contrib/zstd/lib/dictBuilder/cover.c +194 -113
- data/contrib/zstd/lib/dictBuilder/cover.h +112 -0
- data/contrib/zstd/lib/dictBuilder/divsufsort.c +3 -3
- data/contrib/zstd/lib/dictBuilder/fastcover.c +740 -0
- data/contrib/zstd/lib/dictBuilder/zdict.c +142 -106
- data/contrib/zstd/lib/dictBuilder/zdict.h +115 -49
- data/contrib/zstd/lib/legacy/zstd_legacy.h +44 -12
- data/contrib/zstd/lib/legacy/zstd_v01.c +41 -10
- data/contrib/zstd/lib/legacy/zstd_v01.h +12 -7
- data/contrib/zstd/lib/legacy/zstd_v02.c +37 -12
- data/contrib/zstd/lib/legacy/zstd_v02.h +12 -7
- data/contrib/zstd/lib/legacy/zstd_v03.c +38 -12
- data/contrib/zstd/lib/legacy/zstd_v03.h +12 -7
- data/contrib/zstd/lib/legacy/zstd_v04.c +55 -174
- data/contrib/zstd/lib/legacy/zstd_v04.h +12 -7
- data/contrib/zstd/lib/legacy/zstd_v05.c +59 -31
- data/contrib/zstd/lib/legacy/zstd_v05.h +12 -7
- data/contrib/zstd/lib/legacy/zstd_v06.c +48 -20
- data/contrib/zstd/lib/legacy/zstd_v06.h +10 -5
- data/contrib/zstd/lib/legacy/zstd_v07.c +62 -29
- data/contrib/zstd/lib/legacy/zstd_v07.h +10 -5
- data/contrib/zstd/lib/zstd.h +1346 -832
- data/ext/extzstd.c +27 -19
- data/ext/extzstd_stream.c +20 -4
- data/ext/zstd_compress.c +1 -0
- data/ext/zstd_decompress.c +4 -0
- data/ext/zstd_dictbuilder.c +4 -0
- data/ext/zstd_dictbuilder_fastcover.c +5 -0
- data/lib/extzstd.rb +52 -220
- data/lib/extzstd/version.rb +1 -1
- metadata +21 -7
- data/contrib/zstd/circle.yml +0 -63
@@ -0,0 +1,112 @@
|
|
1
|
+
#include <stdio.h> /* fprintf */
|
2
|
+
#include <stdlib.h> /* malloc, free, qsort */
|
3
|
+
#include <string.h> /* memset */
|
4
|
+
#include <time.h> /* clock */
|
5
|
+
#include "mem.h" /* read */
|
6
|
+
#include "pool.h"
|
7
|
+
#include "threading.h"
|
8
|
+
#include "zstd_internal.h" /* includes zstd.h */
|
9
|
+
#ifndef ZDICT_STATIC_LINKING_ONLY
|
10
|
+
#define ZDICT_STATIC_LINKING_ONLY
|
11
|
+
#endif
|
12
|
+
#include "zdict.h"
|
13
|
+
|
14
|
+
/**
|
15
|
+
* COVER_best_t is used for two purposes:
|
16
|
+
* 1. Synchronizing threads.
|
17
|
+
* 2. Saving the best parameters and dictionary.
|
18
|
+
*
|
19
|
+
* All of the methods except COVER_best_init() are thread safe if zstd is
|
20
|
+
* compiled with multithreaded support.
|
21
|
+
*/
|
22
|
+
typedef struct COVER_best_s {
|
23
|
+
ZSTD_pthread_mutex_t mutex;
|
24
|
+
ZSTD_pthread_cond_t cond;
|
25
|
+
size_t liveJobs;
|
26
|
+
void *dict;
|
27
|
+
size_t dictSize;
|
28
|
+
ZDICT_cover_params_t parameters;
|
29
|
+
size_t compressedSize;
|
30
|
+
} COVER_best_t;
|
31
|
+
|
32
|
+
/**
|
33
|
+
* A segment is a range in the source as well as the score of the segment.
|
34
|
+
*/
|
35
|
+
typedef struct {
|
36
|
+
U32 begin;
|
37
|
+
U32 end;
|
38
|
+
U32 score;
|
39
|
+
} COVER_segment_t;
|
40
|
+
|
41
|
+
/**
|
42
|
+
*Number of epochs and size of each epoch.
|
43
|
+
*/
|
44
|
+
typedef struct {
|
45
|
+
U32 num;
|
46
|
+
U32 size;
|
47
|
+
} COVER_epoch_info_t;
|
48
|
+
|
49
|
+
/**
|
50
|
+
* Computes the number of epochs and the size of each epoch.
|
51
|
+
* We will make sure that each epoch gets at least 10 * k bytes.
|
52
|
+
*
|
53
|
+
* The COVER algorithms divide the data up into epochs of equal size and
|
54
|
+
* select one segment from each epoch.
|
55
|
+
*
|
56
|
+
* @param maxDictSize The maximum allowed dictionary size.
|
57
|
+
* @param nbDmers The number of dmers we are training on.
|
58
|
+
* @param k The parameter k (segment size).
|
59
|
+
* @param passes The target number of passes over the dmer corpus.
|
60
|
+
* More passes means a better dictionary.
|
61
|
+
*/
|
62
|
+
COVER_epoch_info_t COVER_computeEpochs(U32 maxDictSize, U32 nbDmers,
|
63
|
+
U32 k, U32 passes);
|
64
|
+
|
65
|
+
/**
|
66
|
+
* Warns the user when their corpus is too small.
|
67
|
+
*/
|
68
|
+
void COVER_warnOnSmallCorpus(size_t maxDictSize, size_t nbDmers, int displayLevel);
|
69
|
+
|
70
|
+
/**
|
71
|
+
* Checks total compressed size of a dictionary
|
72
|
+
*/
|
73
|
+
size_t COVER_checkTotalCompressedSize(const ZDICT_cover_params_t parameters,
|
74
|
+
const size_t *samplesSizes, const BYTE *samples,
|
75
|
+
size_t *offsets,
|
76
|
+
size_t nbTrainSamples, size_t nbSamples,
|
77
|
+
BYTE *const dict, size_t dictBufferCapacity);
|
78
|
+
|
79
|
+
/**
|
80
|
+
* Returns the sum of the sample sizes.
|
81
|
+
*/
|
82
|
+
size_t COVER_sum(const size_t *samplesSizes, unsigned nbSamples) ;
|
83
|
+
|
84
|
+
/**
|
85
|
+
* Initialize the `COVER_best_t`.
|
86
|
+
*/
|
87
|
+
void COVER_best_init(COVER_best_t *best);
|
88
|
+
|
89
|
+
/**
|
90
|
+
* Wait until liveJobs == 0.
|
91
|
+
*/
|
92
|
+
void COVER_best_wait(COVER_best_t *best);
|
93
|
+
|
94
|
+
/**
|
95
|
+
* Call COVER_best_wait() and then destroy the COVER_best_t.
|
96
|
+
*/
|
97
|
+
void COVER_best_destroy(COVER_best_t *best);
|
98
|
+
|
99
|
+
/**
|
100
|
+
* Called when a thread is about to be launched.
|
101
|
+
* Increments liveJobs.
|
102
|
+
*/
|
103
|
+
void COVER_best_start(COVER_best_t *best);
|
104
|
+
|
105
|
+
/**
|
106
|
+
* Called when a thread finishes executing, both on error or success.
|
107
|
+
* Decrements liveJobs and signals any waiting threads if liveJobs == 0.
|
108
|
+
* If this dictionary is the best so far save it and its parameters.
|
109
|
+
*/
|
110
|
+
void COVER_best_finish(COVER_best_t *best, size_t compressedSize,
|
111
|
+
ZDICT_cover_params_t parameters, void *dict,
|
112
|
+
size_t dictSize);
|
@@ -1637,7 +1637,7 @@ construct_SA(const unsigned char *T, int *SA,
|
|
1637
1637
|
if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
|
1638
1638
|
k = SA + BUCKET_B(c2 = c0, c1);
|
1639
1639
|
}
|
1640
|
-
assert(k < j);
|
1640
|
+
assert(k < j); assert(k != NULL);
|
1641
1641
|
*k-- = s;
|
1642
1642
|
} else {
|
1643
1643
|
assert(((s == 0) && (T[s] == c1)) || (s < 0));
|
@@ -1701,7 +1701,7 @@ construct_BWT(const unsigned char *T, int *SA,
|
|
1701
1701
|
if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
|
1702
1702
|
k = SA + BUCKET_B(c2 = c0, c1);
|
1703
1703
|
}
|
1704
|
-
assert(k < j);
|
1704
|
+
assert(k < j); assert(k != NULL);
|
1705
1705
|
*k-- = s;
|
1706
1706
|
} else if(s != 0) {
|
1707
1707
|
*j = ~s;
|
@@ -1785,7 +1785,7 @@ construct_BWT_indexes(const unsigned char *T, int *SA,
|
|
1785
1785
|
if(0 <= c2) { BUCKET_B(c2, c1) = k - SA; }
|
1786
1786
|
k = SA + BUCKET_B(c2 = c0, c1);
|
1787
1787
|
}
|
1788
|
-
assert(k < j);
|
1788
|
+
assert(k < j); assert(k != NULL);
|
1789
1789
|
*k-- = s;
|
1790
1790
|
} else if(s != 0) {
|
1791
1791
|
*j = ~s;
|
@@ -0,0 +1,740 @@
|
|
1
|
+
/*-*************************************
|
2
|
+
* Dependencies
|
3
|
+
***************************************/
|
4
|
+
#include <stdio.h> /* fprintf */
|
5
|
+
#include <stdlib.h> /* malloc, free, qsort */
|
6
|
+
#include <string.h> /* memset */
|
7
|
+
#include <time.h> /* clock */
|
8
|
+
|
9
|
+
#include "mem.h" /* read */
|
10
|
+
#include "pool.h"
|
11
|
+
#include "threading.h"
|
12
|
+
#include "cover.h"
|
13
|
+
#include "zstd_internal.h" /* includes zstd.h */
|
14
|
+
#ifndef ZDICT_STATIC_LINKING_ONLY
|
15
|
+
#define ZDICT_STATIC_LINKING_ONLY
|
16
|
+
#endif
|
17
|
+
#include "zdict.h"
|
18
|
+
|
19
|
+
|
20
|
+
/*-*************************************
|
21
|
+
* Constants
|
22
|
+
***************************************/
|
23
|
+
#define FASTCOVER_MAX_SAMPLES_SIZE (sizeof(size_t) == 8 ? ((unsigned)-1) : ((unsigned)1 GB))
|
24
|
+
#define FASTCOVER_MAX_F 31
|
25
|
+
#define FASTCOVER_MAX_ACCEL 10
|
26
|
+
#define DEFAULT_SPLITPOINT 0.75
|
27
|
+
#define DEFAULT_F 20
|
28
|
+
#define DEFAULT_ACCEL 1
|
29
|
+
|
30
|
+
|
31
|
+
/*-*************************************
|
32
|
+
* Console display
|
33
|
+
***************************************/
|
34
|
+
static int g_displayLevel = 2;
|
35
|
+
#define DISPLAY(...) \
|
36
|
+
{ \
|
37
|
+
fprintf(stderr, __VA_ARGS__); \
|
38
|
+
fflush(stderr); \
|
39
|
+
}
|
40
|
+
#define LOCALDISPLAYLEVEL(displayLevel, l, ...) \
|
41
|
+
if (displayLevel >= l) { \
|
42
|
+
DISPLAY(__VA_ARGS__); \
|
43
|
+
} /* 0 : no display; 1: errors; 2: default; 3: details; 4: debug */
|
44
|
+
#define DISPLAYLEVEL(l, ...) LOCALDISPLAYLEVEL(g_displayLevel, l, __VA_ARGS__)
|
45
|
+
|
46
|
+
#define LOCALDISPLAYUPDATE(displayLevel, l, ...) \
|
47
|
+
if (displayLevel >= l) { \
|
48
|
+
if ((clock() - g_time > refreshRate) || (displayLevel >= 4)) { \
|
49
|
+
g_time = clock(); \
|
50
|
+
DISPLAY(__VA_ARGS__); \
|
51
|
+
} \
|
52
|
+
}
|
53
|
+
#define DISPLAYUPDATE(l, ...) LOCALDISPLAYUPDATE(g_displayLevel, l, __VA_ARGS__)
|
54
|
+
static const clock_t refreshRate = CLOCKS_PER_SEC * 15 / 100;
|
55
|
+
static clock_t g_time = 0;
|
56
|
+
|
57
|
+
|
58
|
+
/*-*************************************
|
59
|
+
* Hash Functions
|
60
|
+
***************************************/
|
61
|
+
static const U64 prime6bytes = 227718039650203ULL;
|
62
|
+
static size_t ZSTD_hash6(U64 u, U32 h) { return (size_t)(((u << (64-48)) * prime6bytes) >> (64-h)) ; }
|
63
|
+
static size_t ZSTD_hash6Ptr(const void* p, U32 h) { return ZSTD_hash6(MEM_readLE64(p), h); }
|
64
|
+
|
65
|
+
static const U64 prime8bytes = 0xCF1BBCDCB7A56463ULL;
|
66
|
+
static size_t ZSTD_hash8(U64 u, U32 h) { return (size_t)(((u) * prime8bytes) >> (64-h)) ; }
|
67
|
+
static size_t ZSTD_hash8Ptr(const void* p, U32 h) { return ZSTD_hash8(MEM_readLE64(p), h); }
|
68
|
+
|
69
|
+
|
70
|
+
/**
|
71
|
+
* Hash the d-byte value pointed to by p and mod 2^f
|
72
|
+
*/
|
73
|
+
static size_t FASTCOVER_hashPtrToIndex(const void* p, U32 h, unsigned d) {
|
74
|
+
if (d == 6) {
|
75
|
+
return ZSTD_hash6Ptr(p, h) & ((1 << h) - 1);
|
76
|
+
}
|
77
|
+
return ZSTD_hash8Ptr(p, h) & ((1 << h) - 1);
|
78
|
+
}
|
79
|
+
|
80
|
+
|
81
|
+
/*-*************************************
|
82
|
+
* Acceleration
|
83
|
+
***************************************/
|
84
|
+
typedef struct {
|
85
|
+
unsigned finalize; /* Percentage of training samples used for ZDICT_finalizeDictionary */
|
86
|
+
unsigned skip; /* Number of dmer skipped between each dmer counted in computeFrequency */
|
87
|
+
} FASTCOVER_accel_t;
|
88
|
+
|
89
|
+
|
90
|
+
static const FASTCOVER_accel_t FASTCOVER_defaultAccelParameters[FASTCOVER_MAX_ACCEL+1] = {
|
91
|
+
{ 100, 0 }, /* accel = 0, should not happen because accel = 0 defaults to accel = 1 */
|
92
|
+
{ 100, 0 }, /* accel = 1 */
|
93
|
+
{ 50, 1 }, /* accel = 2 */
|
94
|
+
{ 34, 2 }, /* accel = 3 */
|
95
|
+
{ 25, 3 }, /* accel = 4 */
|
96
|
+
{ 20, 4 }, /* accel = 5 */
|
97
|
+
{ 17, 5 }, /* accel = 6 */
|
98
|
+
{ 14, 6 }, /* accel = 7 */
|
99
|
+
{ 13, 7 }, /* accel = 8 */
|
100
|
+
{ 11, 8 }, /* accel = 9 */
|
101
|
+
{ 10, 9 }, /* accel = 10 */
|
102
|
+
};
|
103
|
+
|
104
|
+
|
105
|
+
/*-*************************************
|
106
|
+
* Context
|
107
|
+
***************************************/
|
108
|
+
typedef struct {
|
109
|
+
const BYTE *samples;
|
110
|
+
size_t *offsets;
|
111
|
+
const size_t *samplesSizes;
|
112
|
+
size_t nbSamples;
|
113
|
+
size_t nbTrainSamples;
|
114
|
+
size_t nbTestSamples;
|
115
|
+
size_t nbDmers;
|
116
|
+
U32 *freqs;
|
117
|
+
unsigned d;
|
118
|
+
unsigned f;
|
119
|
+
FASTCOVER_accel_t accelParams;
|
120
|
+
} FASTCOVER_ctx_t;
|
121
|
+
|
122
|
+
|
123
|
+
/*-*************************************
|
124
|
+
* Helper functions
|
125
|
+
***************************************/
|
126
|
+
/**
|
127
|
+
* Selects the best segment in an epoch.
|
128
|
+
* Segments of are scored according to the function:
|
129
|
+
*
|
130
|
+
* Let F(d) be the frequency of all dmers with hash value d.
|
131
|
+
* Let S_i be hash value of the dmer at position i of segment S which has length k.
|
132
|
+
*
|
133
|
+
* Score(S) = F(S_1) + F(S_2) + ... + F(S_{k-d+1})
|
134
|
+
*
|
135
|
+
* Once the dmer with hash value d is in the dictionary we set F(d) = 0.
|
136
|
+
*/
|
137
|
+
static COVER_segment_t FASTCOVER_selectSegment(const FASTCOVER_ctx_t *ctx,
|
138
|
+
U32 *freqs, U32 begin, U32 end,
|
139
|
+
ZDICT_cover_params_t parameters,
|
140
|
+
U16* segmentFreqs) {
|
141
|
+
/* Constants */
|
142
|
+
const U32 k = parameters.k;
|
143
|
+
const U32 d = parameters.d;
|
144
|
+
const U32 f = ctx->f;
|
145
|
+
const U32 dmersInK = k - d + 1;
|
146
|
+
|
147
|
+
/* Try each segment (activeSegment) and save the best (bestSegment) */
|
148
|
+
COVER_segment_t bestSegment = {0, 0, 0};
|
149
|
+
COVER_segment_t activeSegment;
|
150
|
+
|
151
|
+
/* Reset the activeDmers in the segment */
|
152
|
+
/* The activeSegment starts at the beginning of the epoch. */
|
153
|
+
activeSegment.begin = begin;
|
154
|
+
activeSegment.end = begin;
|
155
|
+
activeSegment.score = 0;
|
156
|
+
|
157
|
+
/* Slide the activeSegment through the whole epoch.
|
158
|
+
* Save the best segment in bestSegment.
|
159
|
+
*/
|
160
|
+
while (activeSegment.end < end) {
|
161
|
+
/* Get hash value of current dmer */
|
162
|
+
const size_t idx = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.end, f, d);
|
163
|
+
|
164
|
+
/* Add frequency of this index to score if this is the first occurrence of index in active segment */
|
165
|
+
if (segmentFreqs[idx] == 0) {
|
166
|
+
activeSegment.score += freqs[idx];
|
167
|
+
}
|
168
|
+
/* Increment end of segment and segmentFreqs*/
|
169
|
+
activeSegment.end += 1;
|
170
|
+
segmentFreqs[idx] += 1;
|
171
|
+
/* If the window is now too large, drop the first position */
|
172
|
+
if (activeSegment.end - activeSegment.begin == dmersInK + 1) {
|
173
|
+
/* Get hash value of the dmer to be eliminated from active segment */
|
174
|
+
const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
|
175
|
+
segmentFreqs[delIndex] -= 1;
|
176
|
+
/* Subtract frequency of this index from score if this is the last occurrence of this index in active segment */
|
177
|
+
if (segmentFreqs[delIndex] == 0) {
|
178
|
+
activeSegment.score -= freqs[delIndex];
|
179
|
+
}
|
180
|
+
/* Increment start of segment */
|
181
|
+
activeSegment.begin += 1;
|
182
|
+
}
|
183
|
+
|
184
|
+
/* If this segment is the best so far save it */
|
185
|
+
if (activeSegment.score > bestSegment.score) {
|
186
|
+
bestSegment = activeSegment;
|
187
|
+
}
|
188
|
+
}
|
189
|
+
|
190
|
+
/* Zero out rest of segmentFreqs array */
|
191
|
+
while (activeSegment.begin < end) {
|
192
|
+
const size_t delIndex = FASTCOVER_hashPtrToIndex(ctx->samples + activeSegment.begin, f, d);
|
193
|
+
segmentFreqs[delIndex] -= 1;
|
194
|
+
activeSegment.begin += 1;
|
195
|
+
}
|
196
|
+
|
197
|
+
{
|
198
|
+
/* Zero the frequency of hash value of each dmer covered by the chosen segment. */
|
199
|
+
U32 pos;
|
200
|
+
for (pos = bestSegment.begin; pos != bestSegment.end; ++pos) {
|
201
|
+
const size_t i = FASTCOVER_hashPtrToIndex(ctx->samples + pos, f, d);
|
202
|
+
freqs[i] = 0;
|
203
|
+
}
|
204
|
+
}
|
205
|
+
|
206
|
+
return bestSegment;
|
207
|
+
}
|
208
|
+
|
209
|
+
|
210
|
+
static int FASTCOVER_checkParameters(ZDICT_cover_params_t parameters,
|
211
|
+
size_t maxDictSize, unsigned f,
|
212
|
+
unsigned accel) {
|
213
|
+
/* k, d, and f are required parameters */
|
214
|
+
if (parameters.d == 0 || parameters.k == 0) {
|
215
|
+
return 0;
|
216
|
+
}
|
217
|
+
/* d has to be 6 or 8 */
|
218
|
+
if (parameters.d != 6 && parameters.d != 8) {
|
219
|
+
return 0;
|
220
|
+
}
|
221
|
+
/* k <= maxDictSize */
|
222
|
+
if (parameters.k > maxDictSize) {
|
223
|
+
return 0;
|
224
|
+
}
|
225
|
+
/* d <= k */
|
226
|
+
if (parameters.d > parameters.k) {
|
227
|
+
return 0;
|
228
|
+
}
|
229
|
+
/* 0 < f <= FASTCOVER_MAX_F*/
|
230
|
+
if (f > FASTCOVER_MAX_F || f == 0) {
|
231
|
+
return 0;
|
232
|
+
}
|
233
|
+
/* 0 < splitPoint <= 1 */
|
234
|
+
if (parameters.splitPoint <= 0 || parameters.splitPoint > 1) {
|
235
|
+
return 0;
|
236
|
+
}
|
237
|
+
/* 0 < accel <= 10 */
|
238
|
+
if (accel > 10 || accel == 0) {
|
239
|
+
return 0;
|
240
|
+
}
|
241
|
+
return 1;
|
242
|
+
}
|
243
|
+
|
244
|
+
|
245
|
+
/**
|
246
|
+
* Clean up a context initialized with `FASTCOVER_ctx_init()`.
|
247
|
+
*/
|
248
|
+
static void
|
249
|
+
FASTCOVER_ctx_destroy(FASTCOVER_ctx_t* ctx)
|
250
|
+
{
|
251
|
+
if (!ctx) return;
|
252
|
+
|
253
|
+
free(ctx->freqs);
|
254
|
+
ctx->freqs = NULL;
|
255
|
+
|
256
|
+
free(ctx->offsets);
|
257
|
+
ctx->offsets = NULL;
|
258
|
+
}
|
259
|
+
|
260
|
+
|
261
|
+
/**
|
262
|
+
* Calculate for frequency of hash value of each dmer in ctx->samples
|
263
|
+
*/
|
264
|
+
static void
|
265
|
+
FASTCOVER_computeFrequency(U32* freqs, const FASTCOVER_ctx_t* ctx)
|
266
|
+
{
|
267
|
+
const unsigned f = ctx->f;
|
268
|
+
const unsigned d = ctx->d;
|
269
|
+
const unsigned skip = ctx->accelParams.skip;
|
270
|
+
const unsigned readLength = MAX(d, 8);
|
271
|
+
size_t i;
|
272
|
+
assert(ctx->nbTrainSamples >= 5);
|
273
|
+
assert(ctx->nbTrainSamples <= ctx->nbSamples);
|
274
|
+
for (i = 0; i < ctx->nbTrainSamples; i++) {
|
275
|
+
size_t start = ctx->offsets[i]; /* start of current dmer */
|
276
|
+
size_t const currSampleEnd = ctx->offsets[i+1];
|
277
|
+
while (start + readLength <= currSampleEnd) {
|
278
|
+
const size_t dmerIndex = FASTCOVER_hashPtrToIndex(ctx->samples + start, f, d);
|
279
|
+
freqs[dmerIndex]++;
|
280
|
+
start = start + skip + 1;
|
281
|
+
}
|
282
|
+
}
|
283
|
+
}
|
284
|
+
|
285
|
+
|
286
|
+
/**
|
287
|
+
* Prepare a context for dictionary building.
|
288
|
+
* The context is only dependent on the parameter `d` and can used multiple
|
289
|
+
* times.
|
290
|
+
* Returns 1 on success or zero on error.
|
291
|
+
* The context must be destroyed with `FASTCOVER_ctx_destroy()`.
|
292
|
+
*/
|
293
|
+
static int
|
294
|
+
FASTCOVER_ctx_init(FASTCOVER_ctx_t* ctx,
|
295
|
+
const void* samplesBuffer,
|
296
|
+
const size_t* samplesSizes, unsigned nbSamples,
|
297
|
+
unsigned d, double splitPoint, unsigned f,
|
298
|
+
FASTCOVER_accel_t accelParams)
|
299
|
+
{
|
300
|
+
const BYTE* const samples = (const BYTE*)samplesBuffer;
|
301
|
+
const size_t totalSamplesSize = COVER_sum(samplesSizes, nbSamples);
|
302
|
+
/* Split samples into testing and training sets */
|
303
|
+
const unsigned nbTrainSamples = splitPoint < 1.0 ? (unsigned)((double)nbSamples * splitPoint) : nbSamples;
|
304
|
+
const unsigned nbTestSamples = splitPoint < 1.0 ? nbSamples - nbTrainSamples : nbSamples;
|
305
|
+
const size_t trainingSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes, nbTrainSamples) : totalSamplesSize;
|
306
|
+
const size_t testSamplesSize = splitPoint < 1.0 ? COVER_sum(samplesSizes + nbTrainSamples, nbTestSamples) : totalSamplesSize;
|
307
|
+
|
308
|
+
/* Checks */
|
309
|
+
if (totalSamplesSize < MAX(d, sizeof(U64)) ||
|
310
|
+
totalSamplesSize >= (size_t)FASTCOVER_MAX_SAMPLES_SIZE) {
|
311
|
+
DISPLAYLEVEL(1, "Total samples size is too large (%u MB), maximum size is %u MB\n",
|
312
|
+
(unsigned)(totalSamplesSize >> 20), (FASTCOVER_MAX_SAMPLES_SIZE >> 20));
|
313
|
+
return 0;
|
314
|
+
}
|
315
|
+
|
316
|
+
/* Check if there are at least 5 training samples */
|
317
|
+
if (nbTrainSamples < 5) {
|
318
|
+
DISPLAYLEVEL(1, "Total number of training samples is %u and is invalid\n", nbTrainSamples);
|
319
|
+
return 0;
|
320
|
+
}
|
321
|
+
|
322
|
+
/* Check if there's testing sample */
|
323
|
+
if (nbTestSamples < 1) {
|
324
|
+
DISPLAYLEVEL(1, "Total number of testing samples is %u and is invalid.\n", nbTestSamples);
|
325
|
+
return 0;
|
326
|
+
}
|
327
|
+
|
328
|
+
/* Zero the context */
|
329
|
+
memset(ctx, 0, sizeof(*ctx));
|
330
|
+
DISPLAYLEVEL(2, "Training on %u samples of total size %u\n", nbTrainSamples,
|
331
|
+
(unsigned)trainingSamplesSize);
|
332
|
+
DISPLAYLEVEL(2, "Testing on %u samples of total size %u\n", nbTestSamples,
|
333
|
+
(unsigned)testSamplesSize);
|
334
|
+
|
335
|
+
ctx->samples = samples;
|
336
|
+
ctx->samplesSizes = samplesSizes;
|
337
|
+
ctx->nbSamples = nbSamples;
|
338
|
+
ctx->nbTrainSamples = nbTrainSamples;
|
339
|
+
ctx->nbTestSamples = nbTestSamples;
|
340
|
+
ctx->nbDmers = trainingSamplesSize - MAX(d, sizeof(U64)) + 1;
|
341
|
+
ctx->d = d;
|
342
|
+
ctx->f = f;
|
343
|
+
ctx->accelParams = accelParams;
|
344
|
+
|
345
|
+
/* The offsets of each file */
|
346
|
+
ctx->offsets = (size_t*)calloc((nbSamples + 1), sizeof(size_t));
|
347
|
+
if (ctx->offsets == NULL) {
|
348
|
+
DISPLAYLEVEL(1, "Failed to allocate scratch buffers \n");
|
349
|
+
FASTCOVER_ctx_destroy(ctx);
|
350
|
+
return 0;
|
351
|
+
}
|
352
|
+
|
353
|
+
/* Fill offsets from the samplesSizes */
|
354
|
+
{ U32 i;
|
355
|
+
ctx->offsets[0] = 0;
|
356
|
+
assert(nbSamples >= 5);
|
357
|
+
for (i = 1; i <= nbSamples; ++i) {
|
358
|
+
ctx->offsets[i] = ctx->offsets[i - 1] + samplesSizes[i - 1];
|
359
|
+
}
|
360
|
+
}
|
361
|
+
|
362
|
+
/* Initialize frequency array of size 2^f */
|
363
|
+
ctx->freqs = (U32*)calloc(((U64)1 << f), sizeof(U32));
|
364
|
+
if (ctx->freqs == NULL) {
|
365
|
+
DISPLAYLEVEL(1, "Failed to allocate frequency table \n");
|
366
|
+
FASTCOVER_ctx_destroy(ctx);
|
367
|
+
return 0;
|
368
|
+
}
|
369
|
+
|
370
|
+
DISPLAYLEVEL(2, "Computing frequencies\n");
|
371
|
+
FASTCOVER_computeFrequency(ctx->freqs, ctx);
|
372
|
+
|
373
|
+
return 1;
|
374
|
+
}
|
375
|
+
|
376
|
+
|
377
|
+
/**
|
378
|
+
* Given the prepared context build the dictionary.
|
379
|
+
*/
|
380
|
+
static size_t
|
381
|
+
FASTCOVER_buildDictionary(const FASTCOVER_ctx_t* ctx,
|
382
|
+
U32* freqs,
|
383
|
+
void* dictBuffer, size_t dictBufferCapacity,
|
384
|
+
ZDICT_cover_params_t parameters,
|
385
|
+
U16* segmentFreqs)
|
386
|
+
{
|
387
|
+
BYTE *const dict = (BYTE *)dictBuffer;
|
388
|
+
size_t tail = dictBufferCapacity;
|
389
|
+
/* Divide the data into epochs. We will select one segment from each epoch. */
|
390
|
+
const COVER_epoch_info_t epochs = COVER_computeEpochs(
|
391
|
+
(U32)dictBufferCapacity, (U32)ctx->nbDmers, parameters.k, 1);
|
392
|
+
const size_t maxZeroScoreRun = 10;
|
393
|
+
size_t zeroScoreRun = 0;
|
394
|
+
size_t epoch;
|
395
|
+
DISPLAYLEVEL(2, "Breaking content into %u epochs of size %u\n",
|
396
|
+
(U32)epochs.num, (U32)epochs.size);
|
397
|
+
/* Loop through the epochs until there are no more segments or the dictionary
|
398
|
+
* is full.
|
399
|
+
*/
|
400
|
+
for (epoch = 0; tail > 0; epoch = (epoch + 1) % epochs.num) {
|
401
|
+
const U32 epochBegin = (U32)(epoch * epochs.size);
|
402
|
+
const U32 epochEnd = epochBegin + epochs.size;
|
403
|
+
size_t segmentSize;
|
404
|
+
/* Select a segment */
|
405
|
+
COVER_segment_t segment = FASTCOVER_selectSegment(
|
406
|
+
ctx, freqs, epochBegin, epochEnd, parameters, segmentFreqs);
|
407
|
+
|
408
|
+
/* If the segment covers no dmers, then we are out of content.
|
409
|
+
* There may be new content in other epochs, for continue for some time.
|
410
|
+
*/
|
411
|
+
if (segment.score == 0) {
|
412
|
+
if (++zeroScoreRun >= maxZeroScoreRun) {
|
413
|
+
break;
|
414
|
+
}
|
415
|
+
continue;
|
416
|
+
}
|
417
|
+
zeroScoreRun = 0;
|
418
|
+
|
419
|
+
/* Trim the segment if necessary and if it is too small then we are done */
|
420
|
+
segmentSize = MIN(segment.end - segment.begin + parameters.d - 1, tail);
|
421
|
+
if (segmentSize < parameters.d) {
|
422
|
+
break;
|
423
|
+
}
|
424
|
+
|
425
|
+
/* We fill the dictionary from the back to allow the best segments to be
|
426
|
+
* referenced with the smallest offsets.
|
427
|
+
*/
|
428
|
+
tail -= segmentSize;
|
429
|
+
memcpy(dict + tail, ctx->samples + segment.begin, segmentSize);
|
430
|
+
DISPLAYUPDATE(
|
431
|
+
2, "\r%u%% ",
|
432
|
+
(unsigned)(((dictBufferCapacity - tail) * 100) / dictBufferCapacity));
|
433
|
+
}
|
434
|
+
DISPLAYLEVEL(2, "\r%79s\r", "");
|
435
|
+
return tail;
|
436
|
+
}
|
437
|
+
|
438
|
+
|
439
|
+
/**
|
440
|
+
* Parameters for FASTCOVER_tryParameters().
|
441
|
+
*/
|
442
|
+
typedef struct FASTCOVER_tryParameters_data_s {
|
443
|
+
const FASTCOVER_ctx_t* ctx;
|
444
|
+
COVER_best_t* best;
|
445
|
+
size_t dictBufferCapacity;
|
446
|
+
ZDICT_cover_params_t parameters;
|
447
|
+
} FASTCOVER_tryParameters_data_t;
|
448
|
+
|
449
|
+
|
450
|
+
/**
|
451
|
+
* Tries a set of parameters and updates the COVER_best_t with the results.
|
452
|
+
* This function is thread safe if zstd is compiled with multithreaded support.
|
453
|
+
* It takes its parameters as an *OWNING* opaque pointer to support threading.
|
454
|
+
*/
|
455
|
+
static void FASTCOVER_tryParameters(void *opaque)
|
456
|
+
{
|
457
|
+
/* Save parameters as local variables */
|
458
|
+
FASTCOVER_tryParameters_data_t *const data = (FASTCOVER_tryParameters_data_t *)opaque;
|
459
|
+
const FASTCOVER_ctx_t *const ctx = data->ctx;
|
460
|
+
const ZDICT_cover_params_t parameters = data->parameters;
|
461
|
+
size_t dictBufferCapacity = data->dictBufferCapacity;
|
462
|
+
size_t totalCompressedSize = ERROR(GENERIC);
|
463
|
+
/* Initialize array to keep track of frequency of dmer within activeSegment */
|
464
|
+
U16* segmentFreqs = (U16 *)calloc(((U64)1 << ctx->f), sizeof(U16));
|
465
|
+
/* Allocate space for hash table, dict, and freqs */
|
466
|
+
BYTE *const dict = (BYTE * const)malloc(dictBufferCapacity);
|
467
|
+
U32 *freqs = (U32*) malloc(((U64)1 << ctx->f) * sizeof(U32));
|
468
|
+
if (!segmentFreqs || !dict || !freqs) {
|
469
|
+
DISPLAYLEVEL(1, "Failed to allocate buffers: out of memory\n");
|
470
|
+
goto _cleanup;
|
471
|
+
}
|
472
|
+
/* Copy the frequencies because we need to modify them */
|
473
|
+
memcpy(freqs, ctx->freqs, ((U64)1 << ctx->f) * sizeof(U32));
|
474
|
+
/* Build the dictionary */
|
475
|
+
{ const size_t tail = FASTCOVER_buildDictionary(ctx, freqs, dict, dictBufferCapacity,
|
476
|
+
parameters, segmentFreqs);
|
477
|
+
const unsigned nbFinalizeSamples = (unsigned)(ctx->nbTrainSamples * ctx->accelParams.finalize / 100);
|
478
|
+
dictBufferCapacity = ZDICT_finalizeDictionary(
|
479
|
+
dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
|
480
|
+
ctx->samples, ctx->samplesSizes, nbFinalizeSamples, parameters.zParams);
|
481
|
+
if (ZDICT_isError(dictBufferCapacity)) {
|
482
|
+
DISPLAYLEVEL(1, "Failed to finalize dictionary\n");
|
483
|
+
goto _cleanup;
|
484
|
+
}
|
485
|
+
}
|
486
|
+
/* Check total compressed size */
|
487
|
+
totalCompressedSize = COVER_checkTotalCompressedSize(parameters, ctx->samplesSizes,
|
488
|
+
ctx->samples, ctx->offsets,
|
489
|
+
ctx->nbTrainSamples, ctx->nbSamples,
|
490
|
+
dict, dictBufferCapacity);
|
491
|
+
_cleanup:
|
492
|
+
COVER_best_finish(data->best, totalCompressedSize, parameters, dict,
|
493
|
+
dictBufferCapacity);
|
494
|
+
free(data);
|
495
|
+
free(segmentFreqs);
|
496
|
+
free(dict);
|
497
|
+
free(freqs);
|
498
|
+
}
|
499
|
+
|
500
|
+
|
501
|
+
static void
|
502
|
+
FASTCOVER_convertToCoverParams(ZDICT_fastCover_params_t fastCoverParams,
|
503
|
+
ZDICT_cover_params_t* coverParams)
|
504
|
+
{
|
505
|
+
coverParams->k = fastCoverParams.k;
|
506
|
+
coverParams->d = fastCoverParams.d;
|
507
|
+
coverParams->steps = fastCoverParams.steps;
|
508
|
+
coverParams->nbThreads = fastCoverParams.nbThreads;
|
509
|
+
coverParams->splitPoint = fastCoverParams.splitPoint;
|
510
|
+
coverParams->zParams = fastCoverParams.zParams;
|
511
|
+
}
|
512
|
+
|
513
|
+
|
514
|
+
static void
|
515
|
+
FASTCOVER_convertToFastCoverParams(ZDICT_cover_params_t coverParams,
|
516
|
+
ZDICT_fastCover_params_t* fastCoverParams,
|
517
|
+
unsigned f, unsigned accel)
|
518
|
+
{
|
519
|
+
fastCoverParams->k = coverParams.k;
|
520
|
+
fastCoverParams->d = coverParams.d;
|
521
|
+
fastCoverParams->steps = coverParams.steps;
|
522
|
+
fastCoverParams->nbThreads = coverParams.nbThreads;
|
523
|
+
fastCoverParams->splitPoint = coverParams.splitPoint;
|
524
|
+
fastCoverParams->f = f;
|
525
|
+
fastCoverParams->accel = accel;
|
526
|
+
fastCoverParams->zParams = coverParams.zParams;
|
527
|
+
}
|
528
|
+
|
529
|
+
|
530
|
+
ZDICTLIB_API size_t
|
531
|
+
ZDICT_trainFromBuffer_fastCover(void* dictBuffer, size_t dictBufferCapacity,
|
532
|
+
const void* samplesBuffer,
|
533
|
+
const size_t* samplesSizes, unsigned nbSamples,
|
534
|
+
ZDICT_fastCover_params_t parameters)
|
535
|
+
{
|
536
|
+
BYTE* const dict = (BYTE*)dictBuffer;
|
537
|
+
FASTCOVER_ctx_t ctx;
|
538
|
+
ZDICT_cover_params_t coverParams;
|
539
|
+
FASTCOVER_accel_t accelParams;
|
540
|
+
/* Initialize global data */
|
541
|
+
g_displayLevel = parameters.zParams.notificationLevel;
|
542
|
+
/* Assign splitPoint and f if not provided */
|
543
|
+
parameters.splitPoint = 1.0;
|
544
|
+
parameters.f = parameters.f == 0 ? DEFAULT_F : parameters.f;
|
545
|
+
parameters.accel = parameters.accel == 0 ? DEFAULT_ACCEL : parameters.accel;
|
546
|
+
/* Convert to cover parameter */
|
547
|
+
memset(&coverParams, 0 , sizeof(coverParams));
|
548
|
+
FASTCOVER_convertToCoverParams(parameters, &coverParams);
|
549
|
+
/* Checks */
|
550
|
+
if (!FASTCOVER_checkParameters(coverParams, dictBufferCapacity, parameters.f,
|
551
|
+
parameters.accel)) {
|
552
|
+
DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
|
553
|
+
return ERROR(GENERIC);
|
554
|
+
}
|
555
|
+
if (nbSamples == 0) {
|
556
|
+
DISPLAYLEVEL(1, "FASTCOVER must have at least one input file\n");
|
557
|
+
return ERROR(GENERIC);
|
558
|
+
}
|
559
|
+
if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
|
560
|
+
DISPLAYLEVEL(1, "dictBufferCapacity must be at least %u\n",
|
561
|
+
ZDICT_DICTSIZE_MIN);
|
562
|
+
return ERROR(dstSize_tooSmall);
|
563
|
+
}
|
564
|
+
/* Assign corresponding FASTCOVER_accel_t to accelParams*/
|
565
|
+
accelParams = FASTCOVER_defaultAccelParameters[parameters.accel];
|
566
|
+
/* Initialize context */
|
567
|
+
if (!FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples,
|
568
|
+
coverParams.d, parameters.splitPoint, parameters.f,
|
569
|
+
accelParams)) {
|
570
|
+
DISPLAYLEVEL(1, "Failed to initialize context\n");
|
571
|
+
return ERROR(GENERIC);
|
572
|
+
}
|
573
|
+
COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, g_displayLevel);
|
574
|
+
/* Build the dictionary */
|
575
|
+
DISPLAYLEVEL(2, "Building dictionary\n");
|
576
|
+
{
|
577
|
+
/* Initialize array to keep track of frequency of dmer within activeSegment */
|
578
|
+
U16* segmentFreqs = (U16 *)calloc(((U64)1 << parameters.f), sizeof(U16));
|
579
|
+
const size_t tail = FASTCOVER_buildDictionary(&ctx, ctx.freqs, dictBuffer,
|
580
|
+
dictBufferCapacity, coverParams, segmentFreqs);
|
581
|
+
const unsigned nbFinalizeSamples = (unsigned)(ctx.nbTrainSamples * ctx.accelParams.finalize / 100);
|
582
|
+
const size_t dictionarySize = ZDICT_finalizeDictionary(
|
583
|
+
dict, dictBufferCapacity, dict + tail, dictBufferCapacity - tail,
|
584
|
+
samplesBuffer, samplesSizes, nbFinalizeSamples, coverParams.zParams);
|
585
|
+
if (!ZSTD_isError(dictionarySize)) {
|
586
|
+
DISPLAYLEVEL(2, "Constructed dictionary of size %u\n",
|
587
|
+
(unsigned)dictionarySize);
|
588
|
+
}
|
589
|
+
FASTCOVER_ctx_destroy(&ctx);
|
590
|
+
free(segmentFreqs);
|
591
|
+
return dictionarySize;
|
592
|
+
}
|
593
|
+
}
|
594
|
+
|
595
|
+
|
596
|
+
ZDICTLIB_API size_t
|
597
|
+
ZDICT_optimizeTrainFromBuffer_fastCover(
|
598
|
+
void* dictBuffer, size_t dictBufferCapacity,
|
599
|
+
const void* samplesBuffer,
|
600
|
+
const size_t* samplesSizes, unsigned nbSamples,
|
601
|
+
ZDICT_fastCover_params_t* parameters)
|
602
|
+
{
|
603
|
+
ZDICT_cover_params_t coverParams;
|
604
|
+
FASTCOVER_accel_t accelParams;
|
605
|
+
/* constants */
|
606
|
+
const unsigned nbThreads = parameters->nbThreads;
|
607
|
+
const double splitPoint =
|
608
|
+
parameters->splitPoint <= 0.0 ? DEFAULT_SPLITPOINT : parameters->splitPoint;
|
609
|
+
const unsigned kMinD = parameters->d == 0 ? 6 : parameters->d;
|
610
|
+
const unsigned kMaxD = parameters->d == 0 ? 8 : parameters->d;
|
611
|
+
const unsigned kMinK = parameters->k == 0 ? 50 : parameters->k;
|
612
|
+
const unsigned kMaxK = parameters->k == 0 ? 2000 : parameters->k;
|
613
|
+
const unsigned kSteps = parameters->steps == 0 ? 40 : parameters->steps;
|
614
|
+
const unsigned kStepSize = MAX((kMaxK - kMinK) / kSteps, 1);
|
615
|
+
const unsigned kIterations =
|
616
|
+
(1 + (kMaxD - kMinD) / 2) * (1 + (kMaxK - kMinK) / kStepSize);
|
617
|
+
const unsigned f = parameters->f == 0 ? DEFAULT_F : parameters->f;
|
618
|
+
const unsigned accel = parameters->accel == 0 ? DEFAULT_ACCEL : parameters->accel;
|
619
|
+
/* Local variables */
|
620
|
+
const int displayLevel = parameters->zParams.notificationLevel;
|
621
|
+
unsigned iteration = 1;
|
622
|
+
unsigned d;
|
623
|
+
unsigned k;
|
624
|
+
COVER_best_t best;
|
625
|
+
POOL_ctx *pool = NULL;
|
626
|
+
int warned = 0;
|
627
|
+
/* Checks */
|
628
|
+
if (splitPoint <= 0 || splitPoint > 1) {
|
629
|
+
LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect splitPoint\n");
|
630
|
+
return ERROR(GENERIC);
|
631
|
+
}
|
632
|
+
if (accel == 0 || accel > FASTCOVER_MAX_ACCEL) {
|
633
|
+
LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect accel\n");
|
634
|
+
return ERROR(GENERIC);
|
635
|
+
}
|
636
|
+
if (kMinK < kMaxD || kMaxK < kMinK) {
|
637
|
+
LOCALDISPLAYLEVEL(displayLevel, 1, "Incorrect k\n");
|
638
|
+
return ERROR(GENERIC);
|
639
|
+
}
|
640
|
+
if (nbSamples == 0) {
|
641
|
+
LOCALDISPLAYLEVEL(displayLevel, 1, "FASTCOVER must have at least one input file\n");
|
642
|
+
return ERROR(GENERIC);
|
643
|
+
}
|
644
|
+
if (dictBufferCapacity < ZDICT_DICTSIZE_MIN) {
|
645
|
+
LOCALDISPLAYLEVEL(displayLevel, 1, "dictBufferCapacity must be at least %u\n",
|
646
|
+
ZDICT_DICTSIZE_MIN);
|
647
|
+
return ERROR(dstSize_tooSmall);
|
648
|
+
}
|
649
|
+
if (nbThreads > 1) {
|
650
|
+
pool = POOL_create(nbThreads, 1);
|
651
|
+
if (!pool) {
|
652
|
+
return ERROR(memory_allocation);
|
653
|
+
}
|
654
|
+
}
|
655
|
+
/* Initialization */
|
656
|
+
COVER_best_init(&best);
|
657
|
+
memset(&coverParams, 0 , sizeof(coverParams));
|
658
|
+
FASTCOVER_convertToCoverParams(*parameters, &coverParams);
|
659
|
+
accelParams = FASTCOVER_defaultAccelParameters[accel];
|
660
|
+
/* Turn down global display level to clean up display at level 2 and below */
|
661
|
+
g_displayLevel = displayLevel == 0 ? 0 : displayLevel - 1;
|
662
|
+
/* Loop through d first because each new value needs a new context */
|
663
|
+
LOCALDISPLAYLEVEL(displayLevel, 2, "Trying %u different sets of parameters\n",
|
664
|
+
kIterations);
|
665
|
+
for (d = kMinD; d <= kMaxD; d += 2) {
|
666
|
+
/* Initialize the context for this value of d */
|
667
|
+
FASTCOVER_ctx_t ctx;
|
668
|
+
LOCALDISPLAYLEVEL(displayLevel, 3, "d=%u\n", d);
|
669
|
+
if (!FASTCOVER_ctx_init(&ctx, samplesBuffer, samplesSizes, nbSamples, d, splitPoint, f, accelParams)) {
|
670
|
+
LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to initialize context\n");
|
671
|
+
COVER_best_destroy(&best);
|
672
|
+
POOL_free(pool);
|
673
|
+
return ERROR(GENERIC);
|
674
|
+
}
|
675
|
+
if (!warned) {
|
676
|
+
COVER_warnOnSmallCorpus(dictBufferCapacity, ctx.nbDmers, displayLevel);
|
677
|
+
warned = 1;
|
678
|
+
}
|
679
|
+
/* Loop through k reusing the same context */
|
680
|
+
for (k = kMinK; k <= kMaxK; k += kStepSize) {
|
681
|
+
/* Prepare the arguments */
|
682
|
+
FASTCOVER_tryParameters_data_t *data = (FASTCOVER_tryParameters_data_t *)malloc(
|
683
|
+
sizeof(FASTCOVER_tryParameters_data_t));
|
684
|
+
LOCALDISPLAYLEVEL(displayLevel, 3, "k=%u\n", k);
|
685
|
+
if (!data) {
|
686
|
+
LOCALDISPLAYLEVEL(displayLevel, 1, "Failed to allocate parameters\n");
|
687
|
+
COVER_best_destroy(&best);
|
688
|
+
FASTCOVER_ctx_destroy(&ctx);
|
689
|
+
POOL_free(pool);
|
690
|
+
return ERROR(GENERIC);
|
691
|
+
}
|
692
|
+
data->ctx = &ctx;
|
693
|
+
data->best = &best;
|
694
|
+
data->dictBufferCapacity = dictBufferCapacity;
|
695
|
+
data->parameters = coverParams;
|
696
|
+
data->parameters.k = k;
|
697
|
+
data->parameters.d = d;
|
698
|
+
data->parameters.splitPoint = splitPoint;
|
699
|
+
data->parameters.steps = kSteps;
|
700
|
+
data->parameters.zParams.notificationLevel = g_displayLevel;
|
701
|
+
/* Check the parameters */
|
702
|
+
if (!FASTCOVER_checkParameters(data->parameters, dictBufferCapacity,
|
703
|
+
data->ctx->f, accel)) {
|
704
|
+
DISPLAYLEVEL(1, "FASTCOVER parameters incorrect\n");
|
705
|
+
free(data);
|
706
|
+
continue;
|
707
|
+
}
|
708
|
+
/* Call the function and pass ownership of data to it */
|
709
|
+
COVER_best_start(&best);
|
710
|
+
if (pool) {
|
711
|
+
POOL_add(pool, &FASTCOVER_tryParameters, data);
|
712
|
+
} else {
|
713
|
+
FASTCOVER_tryParameters(data);
|
714
|
+
}
|
715
|
+
/* Print status */
|
716
|
+
LOCALDISPLAYUPDATE(displayLevel, 2, "\r%u%% ",
|
717
|
+
(unsigned)((iteration * 100) / kIterations));
|
718
|
+
++iteration;
|
719
|
+
}
|
720
|
+
COVER_best_wait(&best);
|
721
|
+
FASTCOVER_ctx_destroy(&ctx);
|
722
|
+
}
|
723
|
+
LOCALDISPLAYLEVEL(displayLevel, 2, "\r%79s\r", "");
|
724
|
+
/* Fill the output buffer and parameters with output of the best parameters */
|
725
|
+
{
|
726
|
+
const size_t dictSize = best.dictSize;
|
727
|
+
if (ZSTD_isError(best.compressedSize)) {
|
728
|
+
const size_t compressedSize = best.compressedSize;
|
729
|
+
COVER_best_destroy(&best);
|
730
|
+
POOL_free(pool);
|
731
|
+
return compressedSize;
|
732
|
+
}
|
733
|
+
FASTCOVER_convertToFastCoverParams(best.parameters, parameters, f, accel);
|
734
|
+
memcpy(dictBuffer, best.dict, dictSize);
|
735
|
+
COVER_best_destroy(&best);
|
736
|
+
POOL_free(pool);
|
737
|
+
return dictSize;
|
738
|
+
}
|
739
|
+
|
740
|
+
}
|