mirror of
https://github.com/emmanuel-marty/lzsa.git
synced 2024-11-22 05:33:58 +00:00
Speed up portable decompressor a bit
This commit is contained in:
emmanuel-marty
parent
3e5639afaf
commit
fb79d319cb
183
src/expand.c
183
src/expand.c
@ -32,37 +32,50 @@
|
||||
#define FORCE_INLINE __attribute__((always_inline))
|
||||
#endif /* _MSC_VER */
|
||||
|
||||
static inline FORCE_INLINE int lzsa_expand_literals_slow(const unsigned char **ppInBlock, const unsigned char *pInBlockEnd, int nLiterals, unsigned char **ppCurOutData, const unsigned char *pOutDataEnd) {
|
||||
static inline FORCE_INLINE int lzsa_expand_literals_slow(const unsigned char **ppInBlock, const unsigned char *pInBlockEnd, unsigned int nLiterals, unsigned char **ppCurOutData, const unsigned char *pOutDataEnd) {
|
||||
const unsigned char *pInBlock = *ppInBlock;
|
||||
unsigned char *pCurOutData = *ppCurOutData;
|
||||
|
||||
if (nLiterals == LITERALS_RUN_LEN) {
|
||||
unsigned char nByte;
|
||||
|
||||
if (pInBlock >= pInBlockEnd) return -1;
|
||||
nByte = *pInBlock++;
|
||||
nLiterals += (int)((unsigned int)nByte);
|
||||
if (pInBlock < pInBlockEnd) {
|
||||
nByte = *pInBlock++;
|
||||
nLiterals += ((unsigned int)nByte);
|
||||
|
||||
if (nByte == 254) {
|
||||
if (pInBlock >= pInBlockEnd) return -1;
|
||||
nLiterals += (int)((unsigned int)*pInBlock++);
|
||||
if (nByte == 254) {
|
||||
if (pInBlock < pInBlockEnd) {
|
||||
nLiterals += ((unsigned int)*pInBlock++);
|
||||
}
|
||||
else {
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
else if (nByte == 255) {
|
||||
if ((pInBlock + 1) < pInBlockEnd) {
|
||||
nLiterals = ((unsigned int)*pInBlock++);
|
||||
nLiterals |= (((unsigned int)*pInBlock++) << 8);
|
||||
}
|
||||
else {
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
}
|
||||
else if (nByte == 255) {
|
||||
if ((pInBlock + 1) >= pInBlockEnd) return -1;
|
||||
nLiterals = ((unsigned int)*pInBlock++);
|
||||
nLiterals |= (((unsigned int)*pInBlock++) << 8);
|
||||
else {
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
|
||||
if (nLiterals != 0) {
|
||||
if ((pInBlock + nLiterals) > pInBlockEnd ||
|
||||
(pCurOutData + nLiterals) > pOutDataEnd) {
|
||||
if ((pInBlock + nLiterals) <= pInBlockEnd &&
|
||||
(pCurOutData + nLiterals) <= pOutDataEnd) {
|
||||
memcpy(pCurOutData, pInBlock, nLiterals);
|
||||
pInBlock += nLiterals;
|
||||
pCurOutData += nLiterals;
|
||||
}
|
||||
else {
|
||||
return -1;
|
||||
}
|
||||
|
||||
memcpy(pCurOutData, pInBlock, nLiterals);
|
||||
pInBlock += nLiterals;
|
||||
pCurOutData += nLiterals;
|
||||
}
|
||||
|
||||
*ppInBlock = pInBlock;
|
||||
@ -70,76 +83,76 @@ static inline FORCE_INLINE int lzsa_expand_literals_slow(const unsigned char **p
|
||||
return 0;
|
||||
}
|
||||
|
||||
static inline FORCE_INLINE int lzsa_expand_match_slow(const unsigned char **ppInBlock, const unsigned char *pInBlockEnd, const unsigned char *pSrc, int nMatchLen, unsigned char **ppCurOutData, const unsigned char *pOutDataEnd, const unsigned char *pOutDataFastEnd) {
|
||||
static inline FORCE_INLINE int lzsa_expand_match_slow(const unsigned char **ppInBlock, const unsigned char *pInBlockEnd, const unsigned char *pSrc, unsigned int nMatchLen, unsigned char **ppCurOutData, const unsigned char *pOutDataEnd, const unsigned char *pOutDataFastEnd) {
|
||||
const unsigned char *pInBlock = *ppInBlock;
|
||||
unsigned char *pCurOutData = *ppCurOutData;
|
||||
|
||||
if (nMatchLen == MATCH_RUN_LEN) {
|
||||
unsigned char nByte;
|
||||
|
||||
if (pInBlock >= pInBlockEnd) return -1;
|
||||
nByte = *pInBlock++;
|
||||
nMatchLen += (int)((unsigned int)nByte);
|
||||
if (pInBlock < pInBlockEnd) {
|
||||
nByte = *pInBlock++;
|
||||
nMatchLen += ((unsigned int)nByte);
|
||||
|
||||
if (nByte == 254) {
|
||||
if (pInBlock >= pInBlockEnd) return -1;
|
||||
nMatchLen += (int)((unsigned int)*pInBlock++);
|
||||
if (nByte == 254) {
|
||||
if (pInBlock < pInBlockEnd) {
|
||||
nMatchLen += ((unsigned int)*pInBlock++);
|
||||
}
|
||||
else {
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
else if (nByte == 255) {
|
||||
if ((pInBlock + 1) < pInBlockEnd) {
|
||||
nMatchLen = ((unsigned int)*pInBlock++);
|
||||
nMatchLen |= (((unsigned int)*pInBlock++) << 8);
|
||||
}
|
||||
else {
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
}
|
||||
else if (nByte == 255) {
|
||||
if ((pInBlock + 1) >= pInBlockEnd) return -1;
|
||||
nMatchLen = ((unsigned int)*pInBlock++);
|
||||
nMatchLen |= (((unsigned int)*pInBlock++) << 8);
|
||||
else {
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
|
||||
nMatchLen += MIN_MATCH_SIZE;
|
||||
|
||||
if ((pCurOutData + nMatchLen) > pOutDataEnd) {
|
||||
return -1;
|
||||
}
|
||||
|
||||
if ((pSrc + 1) == pCurOutData && nMatchLen >= 16) {
|
||||
/* One-byte RLE */
|
||||
memset(pCurOutData, *pSrc, nMatchLen);
|
||||
pCurOutData += nMatchLen;
|
||||
}
|
||||
else {
|
||||
if ((pCurOutData + nMatchLen) <= pOutDataEnd) {
|
||||
/* Do a deterministic, left to right byte copy instead of memcpy() so as to handle overlaps */
|
||||
|
||||
int nMaxFast = nMatchLen;
|
||||
if (nMaxFast > (pCurOutData - pSrc))
|
||||
nMaxFast = (int)(pCurOutData - pSrc);
|
||||
if ((pCurOutData + nMaxFast) > (pOutDataFastEnd - 15))
|
||||
nMaxFast = (int)(pOutDataFastEnd - 15 - pCurOutData);
|
||||
|
||||
if (nMaxFast > 0) {
|
||||
if ((pCurOutData - pSrc) >= 8 && (pCurOutData + nMatchLen) < (pOutDataFastEnd - 15)) {
|
||||
const unsigned char *pCopySrc = pSrc;
|
||||
unsigned char *pCopyDst = pCurOutData;
|
||||
const unsigned char *pCopyEndDst = pCurOutData + nMaxFast;
|
||||
const unsigned char *pCopyEndDst = pCurOutData + nMatchLen;
|
||||
|
||||
do {
|
||||
memcpy(pCopyDst, pCopySrc, 16);
|
||||
memcpy(pCopyDst, pCopySrc, 8);
|
||||
memcpy(pCopyDst + 8, pCopySrc + 8, 8);
|
||||
pCopySrc += 16;
|
||||
pCopyDst += 16;
|
||||
} while (pCopyDst < pCopyEndDst);
|
||||
|
||||
pCurOutData += nMaxFast;
|
||||
pSrc += nMaxFast;
|
||||
nMatchLen -= nMaxFast;
|
||||
pCurOutData += nMatchLen;
|
||||
}
|
||||
|
||||
while (nMatchLen >= 4) {
|
||||
*pCurOutData++ = *pSrc++;
|
||||
*pCurOutData++ = *pSrc++;
|
||||
*pCurOutData++ = *pSrc++;
|
||||
*pCurOutData++ = *pSrc++;
|
||||
nMatchLen -= 4;
|
||||
}
|
||||
while (nMatchLen > 0) {
|
||||
*pCurOutData++ = *pSrc++;
|
||||
nMatchLen--;
|
||||
else {
|
||||
while (nMatchLen >= 4) {
|
||||
*pCurOutData++ = *pSrc++;
|
||||
*pCurOutData++ = *pSrc++;
|
||||
*pCurOutData++ = *pSrc++;
|
||||
*pCurOutData++ = *pSrc++;
|
||||
nMatchLen -= 4;
|
||||
}
|
||||
while (nMatchLen) {
|
||||
*pCurOutData++ = *pSrc++;
|
||||
nMatchLen--;
|
||||
}
|
||||
}
|
||||
}
|
||||
else {
|
||||
return -1;
|
||||
}
|
||||
|
||||
*ppInBlock = pInBlock;
|
||||
*ppCurOutData = pCurOutData;
|
||||
@ -159,16 +172,16 @@ static inline FORCE_INLINE int lzsa_expand_match_slow(const unsigned char **ppIn
|
||||
*/
|
||||
int lzsa_expand_block(const unsigned char *pInBlock, int nBlockSize, unsigned char *pOutData, int nOutDataOffset, int nBlockMaxSize) {
|
||||
const unsigned char *pInBlockEnd = pInBlock + nBlockSize;
|
||||
const unsigned char *pInBlockFastEnd = pInBlock + nBlockSize - 16;
|
||||
const unsigned char *pInBlockFastEnd = pInBlock + nBlockSize - 8;
|
||||
unsigned char *pCurOutData = pOutData + nOutDataOffset;
|
||||
const unsigned char *pOutDataEnd = pCurOutData + nBlockMaxSize;
|
||||
const unsigned char *pOutDataFastEnd = pOutDataEnd - 16;
|
||||
const unsigned char *pOutDataFastEnd = pOutDataEnd - 20;
|
||||
|
||||
/* Fast loop */
|
||||
|
||||
while (pInBlock < pInBlockFastEnd && pCurOutData < pOutDataFastEnd) {
|
||||
const unsigned char token = *pInBlock++;
|
||||
int nLiterals = (int)((unsigned int)((token & 0x70) >> 4));
|
||||
unsigned int nLiterals = (unsigned int)((token & 0x70) >> 4);
|
||||
|
||||
if (nLiterals < LITERALS_RUN_LEN) {
|
||||
memcpy(pCurOutData, pInBlock, 8);
|
||||
@ -180,28 +193,31 @@ int lzsa_expand_block(const unsigned char *pInBlock, int nBlockSize, unsigned ch
|
||||
return -1;
|
||||
}
|
||||
|
||||
if (pInBlock < pInBlockEnd) { /* The last token in the block does not include match information */
|
||||
if ((pInBlock + 1) < pInBlockEnd) { /* The last token in the block does not include match information */
|
||||
int nMatchOffset;
|
||||
|
||||
nMatchOffset = ((unsigned int)*pInBlock++);
|
||||
if (token & 0x80) {
|
||||
if (pInBlock >= pInBlockEnd) return -1;
|
||||
nMatchOffset |= (((unsigned int)*pInBlock++) << 8);
|
||||
}
|
||||
nMatchOffset++;
|
||||
|
||||
const unsigned char *pSrc = pCurOutData - nMatchOffset;
|
||||
if (pSrc < pOutData)
|
||||
return -1;
|
||||
|
||||
int nMatchLen = (int)((unsigned int)(token & 0x0f));
|
||||
if (nMatchLen < (16 - MIN_MATCH_SIZE + 1) && (pSrc + MIN_MATCH_SIZE + nMatchLen) < pCurOutData && pCurOutData < pOutDataFastEnd) {
|
||||
memcpy(pCurOutData, pSrc, 16);
|
||||
pCurOutData += (MIN_MATCH_SIZE + nMatchLen);
|
||||
if (pSrc >= pOutData) {
|
||||
unsigned int nMatchLen = (unsigned int)(token & 0x0f);
|
||||
if (nMatchLen < MATCH_RUN_LEN && nMatchOffset >= 8 && pCurOutData < pOutDataFastEnd) {
|
||||
memcpy(pCurOutData, pSrc, 8);
|
||||
memcpy(pCurOutData + 8, pSrc + 8, 8);
|
||||
memcpy(pCurOutData + 16, pSrc + 16, 4);
|
||||
pCurOutData += (MIN_MATCH_SIZE + nMatchLen);
|
||||
}
|
||||
else {
|
||||
if (lzsa_expand_match_slow(&pInBlock, pInBlockEnd, pSrc, nMatchLen, &pCurOutData, pOutDataEnd, pOutDataFastEnd))
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
else {
|
||||
if (lzsa_expand_match_slow(&pInBlock, pInBlockEnd, pSrc, nMatchLen, &pCurOutData, pOutDataEnd, pOutDataFastEnd))
|
||||
return -1;
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
}
|
||||
@ -210,28 +226,29 @@ int lzsa_expand_block(const unsigned char *pInBlock, int nBlockSize, unsigned ch
|
||||
|
||||
while (pInBlock < pInBlockEnd) {
|
||||
const unsigned char token = *pInBlock++;
|
||||
int nLiterals = (int)((unsigned int)((token & 0x70) >> 4));
|
||||
unsigned int nLiterals = (unsigned int)((token & 0x70) >> 4);
|
||||
|
||||
if (lzsa_expand_literals_slow(&pInBlock, pInBlockEnd, nLiterals, &pCurOutData, pOutDataEnd))
|
||||
return -1;
|
||||
|
||||
if (pInBlock < pInBlockEnd) { /* The last token in the block does not include match information */
|
||||
if ((pInBlock + 1) < pInBlockEnd) { /* The last token in the block does not include match information */
|
||||
int nMatchOffset;
|
||||
|
||||
nMatchOffset = ((unsigned int)*pInBlock++);
|
||||
if (token & 0x80) {
|
||||
if (pInBlock >= pInBlockEnd) return -1;
|
||||
nMatchOffset |= (((unsigned int)*pInBlock++) << 8);
|
||||
}
|
||||
nMatchOffset++;
|
||||
|
||||
const unsigned char *pSrc = pCurOutData - nMatchOffset;
|
||||
if (pSrc < pOutData)
|
||||
return -1;
|
||||
|
||||
int nMatchLen = (int)((unsigned int)(token & 0x0f));
|
||||
if (lzsa_expand_match_slow(&pInBlock, pInBlockEnd, pSrc, nMatchLen, &pCurOutData, pOutDataEnd, pOutDataFastEnd))
|
||||
if (pSrc >= pOutData) {
|
||||
unsigned int nMatchLen = (unsigned int)(token & 0x0f);
|
||||
if (lzsa_expand_match_slow(&pInBlock, pInBlockEnd, pSrc, nMatchLen, &pCurOutData, pOutDataEnd, pOutDataFastEnd))
|
||||
return -1;
|
||||
}
|
||||
else {
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user