4
* This is an implementation of LZ compression for PostgreSQL.
5
* It uses a simple history table and generates 2-3 byte tags
6
* capable of backward copy information for 3-273 bytes with
7
* a max offset of 4095.
12
* pglz_compress(const char *source, int32 slen, PGLZ_Header *dest,
13
* const PGLZ_Strategy *strategy);
15
* source is the input data to be compressed.
17
* slen is the length of the input data.
19
* dest is the output area for the compressed result.
20
* It must be at least as big as PGLZ_MAX_OUTPUT(slen).
22
* strategy is a pointer to some information controlling
23
* the compression algorithm. If NULL, the compiled
24
* in default strategy is used.
26
* The return value is TRUE if compression succeeded,
27
* FALSE if not; in the latter case the contents of dest
31
* pglz_decompress(const PGLZ_Header *source, char *dest)
33
* source is the compressed input.
35
* dest is the area where the uncompressed data will be
36
* written to. It is the callers responsibility to
37
* provide enough space. The required amount can be
38
* obtained with the macro PGLZ_RAW_SIZE(source).
40
* The data is written to buff exactly as it was handed
41
* to pglz_compress(). No terminating zero byte is added.
43
* The decompression algorithm and internal data format:
45
* PGLZ_Header is defined as
47
* typedef struct PGLZ_Header {
52
* The header is followed by the compressed data itself.
54
* The data representation is easiest explained by describing
55
* the process of decompression.
57
* If VARSIZE(x) == rawsize + sizeof(PGLZ_Header), then the data
58
* is stored uncompressed as plain bytes. Thus, the decompressor
59
* simply copies rawsize bytes from the location after the
60
* header to the destination.
62
* Otherwise the first byte after the header tells what to do
63
* the next 8 times. We call this the control byte.
65
* An unset bit in the control byte means, that one uncompressed
66
* byte follows, which is copied from input to output.
68
* A set bit in the control byte means, that a tag of 2-3 bytes
69
* follows. A tag contains information to copy some bytes, that
70
* are already in the output buffer, to the current location in
71
* the output. Let's call the three tag bytes T1, T2 and T3. The
72
* position of the data to copy is coded as an offset from the
73
* actual output position.
75
* The offset is in the upper nibble of T1 and in T2.
76
* The length is in the lower nibble of T1.
78
* So the 16 bits of a 2 byte tag are coded as
83
* This limits the offset to 1-4095 (12 bits) and the length
84
* to 3-18 (4 bits) because 3 is always added to it. To emit
85
* a tag of 2 bytes with a length of 2 only saves one control
86
* bit. But we lose one byte in the possible length of a tag.
88
* In the actual implementation, the 2 byte tag's length is
89
* limited to 3-17, because the value 0xF in the length nibble
90
* has special meaning. It means, that the next following
91
* byte (T3) has to be added to the length value of 18. That
92
* makes total limits of 1-4095 for offset and 3-273 for length.
94
* Now that we have successfully decoded a tag. We simply copy
95
* the output that occurred <offset> bytes back to the current
96
* output location in the specified <length>. Thus, a
97
* sequence of 200 spaces (think about bpchar fields) could be
98
* coded in 4 bytes. One literal space and a three byte tag to
99
* copy 199 bytes with a -1 offset. Whow - that's a compression
100
* rate of 98%! Well, the implementation needs to save the
101
* original data size too, so we need another 4 bytes for it
102
* and end up with a total compression rate of 96%, what's still
105
* The compression algorithm
107
* The following uses numbers used in the default strategy.
109
* The compressor works best for attributes of a size between
110
* 1K and 1M. For smaller items there's not that much chance of
111
* redundancy in the character sequence (except for large areas
112
* of identical bytes like trailing spaces) and for bigger ones
113
* our 4K maximum look-back distance is too small.
115
* The compressor creates a table for 8192 lists of positions.
116
* For each input position (except the last 3), a hash key is
117
* built from the 4 next input bytes and the position remembered
118
* in the appropriate list. Thus, the table points to linked
119
* lists of likely to be at least in the first 4 characters
120
* matching strings. This is done on the fly while the input
121
* is compressed into the output area. Table entries are only
122
* kept for the last 4096 input positions, since we cannot use
123
* back-pointers larger than that anyway.
125
* For each byte in the input, it's hash key (built from this
126
* byte and the next 3) is used to find the appropriate list
127
* in the table. The lists remember the positions of all bytes
128
* that had the same hash key in the past in increasing backward
129
* offset order. Now for all entries in the used lists, the
130
* match length is computed by comparing the characters from the
131
* entries position with the characters from the actual input
134
* The compressor starts with a so called "good_match" of 128.
135
* It is a "prefer speed against compression ratio" optimizer.
136
* So if the first entry looked at already has 128 or more
137
* matching characters, the lookup stops and that position is
138
* used for the next tag in the output.
140
* For each subsequent entry in the history list, the "good_match"
141
* is lowered by 10%. So the compressor will be more happy with
142
* short matches the farer it has to go back in the history.
143
* Another "speed against ratio" preference characteristic of
146
* Thus there are 3 stop conditions for the lookup of matches:
148
* - a match >= good_match is found
149
* - there are no more history entries to look at
150
* - the next history entry is already too far back
151
* to be coded into a tag.
153
* Finally the match algorithm checks that at least a match
154
* of 3 or more bytes has been found, because thats the smallest
155
* amount of copy information to code into a tag. If so, a tag
156
* is omitted and all the input bytes covered by that are just
157
* scanned for the history add's, otherwise a literal character
158
* is omitted and only his history entry added.
162
* Many thanks to Adisak Pochanayon, who's article about SLZ
163
* inspired me to write the PostgreSQL compression this way.
167
* Copyright (c) 1999-2009, PostgreSQL Global Development Group
172
#include "postgres.h"
176
#include "utils/pg_lzcompress.h"
183
#define PGLZ_HISTORY_LISTS 8192 /* must be power of 2 */
184
#define PGLZ_HISTORY_MASK (PGLZ_HISTORY_LISTS - 1)
185
#define PGLZ_HISTORY_SIZE 4096
186
#define PGLZ_MAX_MATCH 273
192
* Linked list for the backward history lookup
194
* All the entries sharing a hash key are linked in a doubly linked list.
195
* This makes it easy to remove an entry when it's time to recycle it
196
* (because it's more than 4K positions old).
199
typedef struct PGLZ_HistEntry
201
struct PGLZ_HistEntry *next; /* links for my hash key's list */
202
struct PGLZ_HistEntry *prev;
203
int hindex; /* my current hash key */
204
const char *pos; /* my input position */
209
* The provided standard strategies
212
static const PGLZ_Strategy strategy_default_data = {
213
32, /* Data chunks less than 32 bytes are not compressed */
214
INT_MAX, /* No upper limit on what we'll try to compress */
215
25, /* Require 25% compression rate, or not worth it */
216
1024, /* Give up if no compression in the first 1KB */
217
128, /* Stop history lookup if a match of 128 bytes is found */
218
10 /* Lower good match size by 10% at every loop iteration */
220
const PGLZ_Strategy *const PGLZ_strategy_default = &strategy_default_data;
223
static const PGLZ_Strategy strategy_always_data = {
224
0, /* Chunks of any size are compressed */
226
0, /* It's enough to save one single byte */
227
INT_MAX, /* Never give up early */
228
128, /* Stop history lookup if a match of 128 bytes is found */
229
6 /* Look harder for a good match */
231
const PGLZ_Strategy *const PGLZ_strategy_always = &strategy_always_data;
235
* Statically allocated work arrays for history
238
static PGLZ_HistEntry *hist_start[PGLZ_HISTORY_LISTS];
239
static PGLZ_HistEntry hist_entries[PGLZ_HISTORY_SIZE];
245
* Computes the history table slot for the lookup by the next 4
246
* characters in the input.
248
* NB: because we use the next 4 characters, we are not guaranteed to
249
* find 3-character matches; they very possibly will be in the wrong
250
* hash list. This seems an acceptable tradeoff for spreading out the
254
#define pglz_hist_idx(_s,_e) ( \
255
((((_e) - (_s)) < 4) ? (int) (_s)[0] : \
256
(((_s)[0] << 9) ^ ((_s)[1] << 6) ^ \
257
((_s)[2] << 3) ^ (_s)[3])) & (PGLZ_HISTORY_MASK) \
264
* Adds a new entry to the history table.
266
* If _recycle is true, then we are recycling a previously used entry,
267
* and must first delink it from its old hashcode's linked list.
269
* NOTE: beware of multiple evaluations of macro's arguments, and note that
270
* _hn and _recycle are modified in the macro.
273
#define pglz_hist_add(_hs,_he,_hn,_recycle,_s,_e) \
275
int __hindex = pglz_hist_idx((_s),(_e)); \
276
PGLZ_HistEntry **__myhsp = &(_hs)[__hindex]; \
277
PGLZ_HistEntry *__myhe = &(_he)[_hn]; \
279
if (__myhe->prev == NULL) \
280
(_hs)[__myhe->hindex] = __myhe->next; \
282
__myhe->prev->next = __myhe->next; \
283
if (__myhe->next != NULL) \
284
__myhe->next->prev = __myhe->prev; \
286
__myhe->next = *__myhsp; \
287
__myhe->prev = NULL; \
288
__myhe->hindex = __hindex; \
289
__myhe->pos = (_s); \
290
if (*__myhsp != NULL) \
291
(*__myhsp)->prev = __myhe; \
293
if (++(_hn) >= PGLZ_HISTORY_SIZE) { \
303
* Outputs the last and allocates a new control byte if needed.
306
#define pglz_out_ctrl(__ctrlp,__ctrlb,__ctrl,__buf) \
308
if ((__ctrl & 0xff) == 0) \
310
*(__ctrlp) = __ctrlb; \
311
__ctrlp = (__buf)++; \
321
* Outputs a literal byte to the destination buffer including the
322
* appropriate control bit.
325
#define pglz_out_literal(_ctrlp,_ctrlb,_ctrl,_buf,_byte) \
327
pglz_out_ctrl(_ctrlp,_ctrlb,_ctrl,_buf); \
328
*(_buf)++ = (unsigned char)(_byte); \
336
* Outputs a backward reference tag of 2-4 bytes (depending on
337
* offset and length) to the destination buffer including the
338
* appropriate control bit.
341
#define pglz_out_tag(_ctrlp,_ctrlb,_ctrl,_buf,_len,_off) \
343
pglz_out_ctrl(_ctrlp,_ctrlb,_ctrl,_buf); \
348
(_buf)[0] = (unsigned char)((((_off) & 0xf00) >> 4) | 0x0f); \
349
(_buf)[1] = (unsigned char)(((_off) & 0xff)); \
350
(_buf)[2] = (unsigned char)((_len) - 18); \
353
(_buf)[0] = (unsigned char)((((_off) & 0xf00) >> 4) | ((_len) - 3)); \
354
(_buf)[1] = (unsigned char)((_off) & 0xff); \
363
* Lookup the history table if the actual input stream matches
364
* another sequence of characters, starting somewhere earlier
365
* in the input buffer.
369
pglz_find_match(PGLZ_HistEntry **hstart, const char *input, const char *end,
370
int *lenp, int *offp, int good_match, int good_drop)
372
PGLZ_HistEntry *hent;
377
* Traverse the linked history list until a good enough match is found.
379
hent = hstart[pglz_hist_idx(input, end)];
382
const char *ip = input;
383
const char *hp = hent->pos;
388
* Stop if the offset does not fit into our tag anymore.
391
if (thisoff >= 0x0fff)
395
* Determine length of match. A better match must be larger than the
396
* best so far. And if we already have a match of 16 or more bytes,
397
* it's worth the call overhead to use memcmp() to check if this match
398
* is equal for the same size. After that we must fallback to
399
* character by character comparison to know the exact position where
405
if (memcmp(ip, hp, len) == 0)
410
while (ip < end && *ip == *hp && thislen < PGLZ_MAX_MATCH)
420
while (ip < end && *ip == *hp && thislen < PGLZ_MAX_MATCH)
429
* Remember this match as the best (if it is)
438
* Advance to the next history entry
443
* Be happy with lesser good matches the more entries we visited. But
444
* no point in doing calculation if we're at end of list.
448
if (len >= good_match)
450
good_match -= (good_match * good_drop) / 100;
455
* Return match information only if it results at least in one byte
472
* Compresses source into dest using strategy.
476
pglz_compress(const char *source, int32 slen, PGLZ_Header *dest,
477
const PGLZ_Strategy *strategy)
479
unsigned char *bp = ((unsigned char *) dest) + sizeof(PGLZ_Header);
480
unsigned char *bstart = bp;
482
bool hist_recycle = false;
483
const char *dp = source;
484
const char *dend = source + slen;
485
unsigned char ctrl_dummy = 0;
486
unsigned char *ctrlp = &ctrl_dummy;
487
unsigned char ctrlb = 0;
488
unsigned char ctrl = 0;
489
bool found_match = false;
499
* Our fallback strategy is the default.
501
if (strategy == NULL)
502
strategy = PGLZ_strategy_default;
505
* If the strategy forbids compression (at all or if source chunk size
506
* out of range), fail.
508
if (strategy->match_size_good <= 0 ||
509
slen < strategy->min_input_size ||
510
slen > strategy->max_input_size)
514
* Save the original source size in the header.
516
dest->rawsize = slen;
519
* Limit the match parameters to the supported range.
521
good_match = strategy->match_size_good;
522
if (good_match > PGLZ_MAX_MATCH)
523
good_match = PGLZ_MAX_MATCH;
524
else if (good_match < 17)
527
good_drop = strategy->match_size_drop;
530
else if (good_drop > 100)
533
need_rate = strategy->min_comp_rate;
536
else if (need_rate > 99)
540
* Compute the maximum result size allowed by the strategy, namely
541
* the input size minus the minimum wanted compression rate. This had
542
* better be <= slen, else we might overrun the provided output buffer.
544
if (slen > (INT_MAX/100))
546
/* Approximate to avoid overflow */
547
result_max = (slen / 100) * (100 - need_rate);
550
result_max = (slen * (100 - need_rate)) / 100;
553
* Initialize the history lists to empty. We do not need to zero the
554
* hist_entries[] array; its entries are initialized as they are used.
556
memset(hist_start, 0, sizeof(hist_start));
559
* Compress the source directly into the output buffer.
564
* If we already exceeded the maximum result size, fail.
566
* We check once per loop; since the loop body could emit as many as 4
567
* bytes (a control byte and 3-byte tag), PGLZ_MAX_OUTPUT() had better
568
* allow 4 slop bytes.
570
if (bp - bstart >= result_max)
574
* If we've emitted more than first_success_by bytes without finding
575
* anything compressible at all, fail. This lets us fall out
576
* reasonably quickly when looking at incompressible input (such as
577
* pre-compressed data).
579
if (!found_match && bp - bstart >= strategy->first_success_by)
583
* Try to find a match in the history
585
if (pglz_find_match(hist_start, dp, dend, &match_len,
586
&match_off, good_match, good_drop))
589
* Create the tag and add history entries for all matched
592
pglz_out_tag(ctrlp, ctrlb, ctrl, bp, match_len, match_off);
595
pglz_hist_add(hist_start, hist_entries,
596
hist_next, hist_recycle,
598
dp++; /* Do not do this ++ in the line above! */
599
/* The macro would do it four times - Jan. */
606
* No match found. Copy one literal byte.
608
pglz_out_literal(ctrlp, ctrlb, ctrl, bp, *dp);
609
pglz_hist_add(hist_start, hist_entries,
610
hist_next, hist_recycle,
612
dp++; /* Do not do this ++ in the line above! */
613
/* The macro would do it four times - Jan. */
618
* Write out the last control byte and check that we haven't overrun the
619
* output size allowed by the strategy.
622
result_size = bp - bstart;
623
if (result_size >= result_max)
627
* Success - need only fill in the actual length of the compressed datum.
629
SET_VARSIZE_COMPRESSED(dest, result_size + sizeof(PGLZ_Header));
638
* Decompresses source into dest.
642
pglz_decompress(const PGLZ_Header *source, char *dest)
644
const unsigned char *sp;
645
const unsigned char *srcend;
647
unsigned char *destend;
649
sp = ((const unsigned char *) source) + sizeof(PGLZ_Header);
650
srcend = ((const unsigned char *) source) + VARSIZE(source);
651
dp = (unsigned char *) dest;
652
destend = dp + source->rawsize;
654
while (sp < srcend && dp < destend)
657
* Read one control byte and process the next 8 items (or as many
658
* as remain in the compressed input).
660
unsigned char ctrl = *sp++;
663
for (ctrlc = 0; ctrlc < 8 && sp < srcend; ctrlc++)
668
* Otherwise it contains the match length minus 3 and the
669
* upper 4 bits of the offset. The next following byte
670
* contains the lower 8 bits of the offset. If the length is
671
* coded as 18, another extension tag byte tells how much
672
* longer the match really was (0-255).
677
len = (sp[0] & 0x0f) + 3;
678
off = ((sp[0] & 0xf0) << 4) | sp[1];
684
* Check for output buffer overrun, to ensure we don't
685
* clobber memory in case of corrupt input. Note: we must
686
* advance dp here to ensure the error is detected below
687
* the loop. We don't simply put the elog inside the loop
688
* since that will probably interfere with optimization.
690
if (dp + len > destend)
697
* Now we copy the bytes specified by the tag from OUTPUT to
698
* OUTPUT. It is dangerous and platform dependent to use
699
* memcpy() here, because the copied areas could overlap
711
* An unset control bit means LITERAL BYTE. So we just copy
712
* one from INPUT to OUTPUT.
714
if (dp >= destend) /* check for buffer overrun */
715
break; /* do not clobber memory */
721
* Advance the control bit
728
* Check we decompressed the right amount.
730
if (dp != destend || sp != srcend)
731
elog(ERROR, "compressed data is corrupt");