zstd/contrib/long_distance_matching/ldm.c

453 lines
11 KiB
C
Raw Normal View History

2017-07-05 20:57:07 +00:00
#include <stdlib.h>
#include <string.h>
#include <stdint.h>
#include <stdio.h>
2017-07-05 20:57:07 +00:00
#include "ldm.h"
2017-07-08 00:09:28 +00:00
#define HASH_EVERY 7
#define LDM_MEMORY_USAGE 14
#define LDM_HASHLOG (LDM_MEMORY_USAGE-2)
#define LDM_HASHTABLESIZE (1 << (LDM_MEMORY_USAGE))
#define LDM_HASHTABLESIZE_U32 ((LDM_HASHTABLESIZE) >> 2)
#define LDM_HASH_SIZE_U32 (1 << (LDM_HASHLOG))
2017-07-07 21:52:40 +00:00
#define WINDOW_SIZE (1 << 20)
#define MAX_WINDOW_SIZE 31
2017-07-08 00:09:28 +00:00
#define HASH_SIZE 8
#define MINMATCH 8
#define ML_BITS 4
#define ML_MASK ((1U<<ML_BITS)-1)
#define RUN_BITS (8-ML_BITS)
#define RUN_MASK ((1U<<RUN_BITS)-1)
2017-07-06 23:47:08 +00:00
//#define LDM_DEBUG
2017-07-05 20:57:07 +00:00
typedef uint8_t BYTE;
typedef uint16_t U16;
typedef uint32_t U32;
typedef int32_t S32;
typedef uint64_t U64;
typedef uint64_t tag;
2017-07-10 13:50:49 +00:00
static unsigned LDM_isLittleEndian(void) {
const union { U32 u; BYTE c[4]; } one = { 1 };
return one.c[0];
}
static U16 LDM_read16(const void *memPtr) {
U16 val;
memcpy(&val, memPtr, sizeof(val));
return val;
}
static U16 LDM_readLE16(const void *memPtr) {
if (LDM_isLittleEndian()) {
return LDM_read16(memPtr);
} else {
const BYTE *p = (const BYTE *)memPtr;
return (U16)((U16)p[0] + (p[1] << 8));
}
}
2017-07-10 13:50:49 +00:00
static void LDM_write16(void *memPtr, U16 value){
memcpy(memPtr, &value, sizeof(value));
}
2017-07-10 13:50:49 +00:00
static void LDM_write32(void *memPtr, U32 value) {
memcpy(memPtr, &value, sizeof(value));
}
static void LDM_writeLE16(void *memPtr, U16 value) {
if (LDM_isLittleEndian()) {
LDM_write16(memPtr, value);
} else {
BYTE* p = (BYTE*)memPtr;
p[0] = (BYTE) value;
p[1] = (BYTE)(value>>8);
}
}
static U32 LDM_read32(const void *ptr) {
return *(const U32 *)ptr;
}
2017-07-07 21:52:40 +00:00
static U64 LDM_read64(const void *ptr) {
return *(const U64 *)ptr;
}
static void LDM_copy8(void *dst, const void *src) {
memcpy(dst, src, 8);
}
2017-07-08 00:09:28 +00:00
typedef struct compress_stats {
U32 num_matches;
U32 total_match_length;
U32 total_literal_length;
U64 total_offset;
} compress_stats;
2017-07-10 13:50:49 +00:00
static void LDM_printCompressStats(const compress_stats *stats) {
2017-07-08 00:09:28 +00:00
printf("=====================\n");
printf("Compression statistics\n");
printf("Total number of matches: %u\n", stats->num_matches);
printf("Average match length: %.1f\n", ((double)stats->total_match_length) /
(double)stats->num_matches);
printf("Average literal length: %.1f\n",
((double)stats->total_literal_length) / (double)stats->num_matches);
printf("Average offset length: %.1f\n",
((double)stats->total_offset) / (double)stats->num_matches);
printf("=====================\n");
}
2017-07-10 13:50:49 +00:00
// TODO: unused.
2017-07-05 20:57:07 +00:00
struct hash_entry {
U64 offset;
tag t;
};
static U32 LDM_hash(U32 sequence) {
return ((sequence * 2654435761U) >> ((32)-LDM_HASHLOG));
}
2017-07-05 20:57:07 +00:00
2017-07-07 21:52:40 +00:00
static U32 LDM_hash5(U64 sequence) {
static const U64 prime5bytes = 889523592379ULL;
static const U64 prime8bytes = 11400714785074694791ULL;
const U32 hashLog = LDM_HASHLOG;
if (LDM_isLittleEndian())
return (U32)(((sequence << 24) * prime5bytes) >> (64 - hashLog));
else
return (U32)(((sequence >> 24) * prime8bytes) >> (64 - hashLog));
}
static U32 LDM_hash_position(const void * const p) {
return LDM_hash(LDM_read32(p));
}
static void LDM_put_position_on_hash(const BYTE *p, U32 h, void *tableBase,
const BYTE *srcBase) {
2017-07-08 00:09:28 +00:00
if (((p - srcBase) & HASH_EVERY) != HASH_EVERY) {
return;
}
U32 *hashTable = (U32 *) tableBase;
hashTable[h] = (U32)(p - srcBase);
}
static void LDM_put_position(const BYTE *p, void *tableBase,
const BYTE *srcBase) {
2017-07-08 00:09:28 +00:00
if (((p - srcBase) & HASH_EVERY) != HASH_EVERY) {
return;
}
U32 const h = LDM_hash_position(p);
LDM_put_position_on_hash(p, h, tableBase, srcBase);
}
2017-07-05 20:57:07 +00:00
static const BYTE *LDM_get_position_on_hash(
U32 h, void *tableBase, const BYTE *srcBase) {
const U32 * const hashTable = (U32*)tableBase;
return hashTable[h] + srcBase;
}
2017-07-05 20:57:07 +00:00
static BYTE LDM_read_byte(const void *memPtr) {
BYTE val;
memcpy(&val, memPtr, 1);
return val;
2017-07-05 20:57:07 +00:00
}
static unsigned LDM_count(const BYTE *pIn, const BYTE *pMatch,
const BYTE *pInLimit) {
const BYTE * const pStart = pIn;
while (pIn < pInLimit - 1) {
BYTE const diff = LDM_read_byte(pMatch) ^ LDM_read_byte(pIn);
if (!diff) {
pIn++;
pMatch++;
continue;
}
return (unsigned)(pIn - pStart);
}
return (unsigned)(pIn - pStart);
}
2017-07-10 13:50:49 +00:00
void LDM_read_header(const void *src, size_t *compressSize,
size_t *decompressSize) {
2017-07-10 13:32:29 +00:00
const U32 *ip = (const U32 *)src;
2017-07-10 13:50:49 +00:00
*compressSize = *ip++;
*decompressSize = *ip;
}
2017-07-10 13:32:29 +00:00
// TODO: maxDstSize is unused
size_t LDM_compress(const void *src, size_t srcSize,
void *dst, size_t maxDstSize) {
const BYTE * const istart = (const BYTE*)src;
const BYTE *ip = istart;
2017-07-10 13:32:29 +00:00
const BYTE * const iend = istart + srcSize;
const BYTE *ilimit = iend - HASH_SIZE;
const BYTE * const matchlimit = iend - HASH_SIZE;
2017-07-07 21:14:01 +00:00
const BYTE * const mflimit = iend - MINMATCH;
2017-07-10 13:32:29 +00:00
BYTE *op = (BYTE*) dst;
2017-07-08 00:09:28 +00:00
compress_stats compressStats = { 0 };
U32 hashTable[LDM_HASHTABLESIZE_U32];
memset(hashTable, 0, sizeof(hashTable));
2017-07-10 13:32:29 +00:00
const BYTE *anchor = (const BYTE *)src;
// struct LDM_cctx cctx;
size_t output_size = 0;
U32 forwardH;
/* Hash first byte: put into hash table */
LDM_put_position(ip, hashTable, istart);
2017-07-07 21:52:40 +00:00
const BYTE *lastHash = ip;
ip++;
forwardH = LDM_hash_position(ip);
2017-07-07 21:14:01 +00:00
//TODO Loop terminates before ip>=ilimit.
while (ip < ilimit) {
const BYTE *match;
BYTE *token;
2017-07-06 23:47:08 +00:00
/* Find a match */
{
const BYTE *forwardIp = ip;
unsigned step = 1;
do {
U32 const h = forwardH;
ip = forwardIp;
forwardIp += step;
2017-07-07 21:14:01 +00:00
if (forwardIp > mflimit) {
goto _last_literals;
}
match = LDM_get_position_on_hash(h, hashTable, istart);
forwardH = LDM_hash_position(forwardIp);
LDM_put_position_on_hash(ip, h, hashTable, istart);
2017-07-07 21:52:40 +00:00
lastHash = ip;
} while (ip - match > WINDOW_SIZE ||
2017-07-07 21:52:40 +00:00
LDM_read64(match) != LDM_read64(ip));
}
2017-07-08 00:09:28 +00:00
compressStats.num_matches++;
2017-07-08 00:09:28 +00:00
/* Catchup: look back to extend match from found match */
2017-07-07 21:14:01 +00:00
while (ip > anchor && match > istart && ip[-1] == match[-1]) {
ip--;
match--;
}
/* Encode literals */
{
unsigned const litLength = (unsigned)(ip - anchor);
token = op++;
2017-07-08 00:09:28 +00:00
compressStats.total_literal_length += litLength;
2017-07-06 23:47:08 +00:00
#ifdef LDM_DEBUG
printf("Cur position: %zu\n", anchor - istart);
printf("LitLength %zu. (Match offset). %zu\n", litLength, ip - match);
2017-07-06 23:47:08 +00:00
#endif
if (litLength >= RUN_MASK) {
int len = (int)litLength - RUN_MASK;
*token = (RUN_MASK << ML_BITS);
for (; len >= 255; len -= 255) {
2017-07-06 23:47:08 +00:00
*op++ = 255;
}
2017-07-06 23:47:08 +00:00
*op++ = (BYTE)len;
} else {
*token = (BYTE)(litLength << ML_BITS);
}
2017-07-06 23:47:08 +00:00
#ifdef LDM_DEBUG
printf("Literals ");
fwrite(anchor, litLength, 1, stdout);
printf("\n");
2017-07-06 23:47:08 +00:00
#endif
2017-07-07 21:14:01 +00:00
memcpy(op, anchor, litLength);
op += litLength;
}
_next_match:
/* Encode offset */
{
2017-07-08 00:09:28 +00:00
/*
LDM_writeLE16(op, ip-match);
op += 2;
*/
2017-07-07 21:52:40 +00:00
LDM_write32(op, ip - match);
op += 4;
2017-07-08 00:09:28 +00:00
compressStats.total_offset += (ip - match);
}
/* Encode Match Length */
{
unsigned matchCode;
matchCode = LDM_count(ip + MINMATCH, match + MINMATCH,
matchlimit);
2017-07-06 23:47:08 +00:00
#ifdef LDM_DEBUG
printf("Match length %zu\n", matchCode + MINMATCH);
fwrite(ip, MINMATCH + matchCode, 1, stdout);
printf("\n");
2017-07-06 23:47:08 +00:00
#endif
2017-07-08 00:09:28 +00:00
compressStats.total_match_length += matchCode + MINMATCH;
2017-07-07 21:52:40 +00:00
unsigned ctr = 1;
ip++;
for (; ctr < MINMATCH + matchCode; ip++, ctr++) {
LDM_put_position(ip, hashTable, istart);
}
// ip += MINMATCH + matchCode;
if (matchCode >= ML_MASK) {
*token += ML_MASK;
matchCode -= ML_MASK;
LDM_write32(op, 0xFFFFFFFF);
while (matchCode >= 4*0xFF) {
op += 4;
LDM_write32(op, 0xffffffff);
matchCode -= 4*0xFF;
}
op += matchCode / 255;
*op++ = (BYTE)(matchCode % 255);
} else {
*token += (BYTE)(matchCode);
}
2017-07-06 23:47:08 +00:00
#ifdef LDM_DEBUG
printf("\n");
2017-07-08 00:09:28 +00:00
2017-07-06 23:47:08 +00:00
#endif
}
anchor = ip;
LDM_put_position(ip, hashTable, istart);
forwardH = LDM_hash_position(++ip);
2017-07-07 21:52:40 +00:00
lastHash = ip;
}
2017-07-07 21:14:01 +00:00
_last_literals:
/* Encode last literals */
{
size_t const lastRun = (size_t)(iend - anchor);
if (lastRun >= RUN_MASK) {
size_t accumulator = lastRun - RUN_MASK;
*op++ = RUN_MASK << ML_BITS;
for(; accumulator >= 255; accumulator -= 255) {
*op++ = 255;
}
2017-07-07 21:14:01 +00:00
*op++ = (BYTE)accumulator;
} else {
*op++ = (BYTE)(lastRun << ML_BITS);
}
memcpy(op, anchor, lastRun);
op += lastRun;
}
2017-07-10 13:50:49 +00:00
LDM_printCompressStats(&compressStats);
2017-07-10 13:32:29 +00:00
return (op - (BYTE *)dst);
}
2017-07-10 13:50:49 +00:00
typedef struct LDM_DCtx {
2017-07-10 14:38:09 +00:00
const BYTE *ibase; /* Pointer to base of input */
const BYTE *ip; /* Pointer to current input position */
const BYTE *iend; /* End of source */
2017-07-10 13:50:49 +00:00
2017-07-10 14:38:09 +00:00
const BYTE *obase; /* Pointer to base of output */
BYTE *op; /* Pointer to output */
const BYTE *oend; /* Pointer to end of output */
size_t compressSize;
size_t maxDecompressSize;
2017-07-10 13:50:49 +00:00
} LDM_DCtx;
2017-07-10 14:38:09 +00:00
static void LDM_initializeDCtx(LDM_DCtx *dctx,
const void *src, size_t compressSize,
void *dst, size_t maxDecompressSize) {
dctx->ibase = src;
dctx->ip = (const BYTE *)src;
dctx->iend = dctx->ip + compressSize;
dctx->op = dst;
dctx->oend = dctx->op + maxDecompressSize;
dctx->compressSize = compressSize;
dctx->maxDecompressSize = maxDecompressSize;
}
size_t LDM_decompress(const void *src, size_t compressSize,
void *dst, size_t maxDecompressSize) {
LDM_DCtx dctx;
LDM_initializeDCtx(&dctx, src, compressSize, dst, maxDecompressSize);
BYTE *cpy;
2017-07-10 14:38:09 +00:00
while (dctx.ip < dctx.iend) {
size_t length;
const BYTE *match;
size_t offset;
/* get literal length */
2017-07-10 14:38:09 +00:00
unsigned const token = *(dctx.ip)++;
if ((length=(token >> ML_BITS)) == RUN_MASK) {
unsigned s;
do {
2017-07-10 14:38:09 +00:00
s = *(dctx.ip)++;
length += s;
} while (s == 255);
}
2017-07-06 23:47:08 +00:00
#ifdef LDM_DEBUG
printf("Literal length: %zu\n", length);
2017-07-06 23:47:08 +00:00
#endif
/* copy literals */
2017-07-10 14:38:09 +00:00
cpy = dctx.op + length;
2017-07-06 23:47:08 +00:00
#ifdef LDM_DEBUG
printf("Literals ");
2017-07-10 14:38:09 +00:00
fwrite(dctx.ip, length, 1, stdout);
2017-07-06 23:47:08 +00:00
printf("\n");
#endif
2017-07-10 14:38:09 +00:00
memcpy(dctx.op, dctx.ip, length);
dctx.ip += length;
dctx.op = cpy;
/* get offset */
2017-07-08 00:09:28 +00:00
/*
2017-07-10 14:38:09 +00:00
offset = LDM_readLE16(dctx.ip);
dctx.ip += 2;
2017-07-08 00:09:28 +00:00
*/
2017-07-10 14:38:09 +00:00
offset = LDM_read32(dctx.ip);
dctx.ip += 4;
2017-07-06 23:47:08 +00:00
#ifdef LDM_DEBUG
printf("Offset: %zu\n", offset);
2017-07-06 23:47:08 +00:00
#endif
2017-07-10 14:38:09 +00:00
match = dctx.op - offset;
// LDM_write32(op, (U32)offset);
/* get matchlength */
length = token & ML_MASK;
if (length == ML_MASK) {
unsigned s;
do {
2017-07-10 14:38:09 +00:00
s = *(dctx.ip)++;
length += s;
} while (s == 255);
}
length += MINMATCH;
2017-07-06 23:47:08 +00:00
#ifdef LDM_DEBUG
printf("Match length: %zu\n", length);
#endif
/* copy match */
2017-07-10 14:38:09 +00:00
cpy = dctx.op + length;
2017-07-06 23:47:08 +00:00
// Inefficient for now
2017-07-10 14:38:09 +00:00
while (match < cpy - offset && dctx.op < dctx.oend) {
*(dctx.op)++ = *match++;
2017-07-06 23:47:08 +00:00
}
}
2017-07-10 14:38:09 +00:00
return dctx.op - (BYTE *)dst;
2017-07-05 20:57:07 +00:00
}