shaba/lvm2
1
0
Fork 0
mirror of git://sourceware.org/git/lvm2.git synced 2024-10-26 08:55:06 +03:00
Code Releases Activity

crc: add newer zlib code

Browse Source 
This code is faster when calculating crc32 checksum for larger
block areas. There is also SIMD variant present in the code,
however ATM the influence on performance of lvm2 is not that big..
This commit is contained in:
Zdenek Kabelac 2024-10-04 14:03:16 +02:00
parent f55c949410
commit b185578321
1 changed files with 86 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

86
lib/misc/crc.c
View File

@ -58,6 +58,90 @@ static const uint32_t _crctab[] = {
0xb3667a2e, 0xc4614ab8, 0x5d681b02, 0x2a6f2b94, 0xb40bbe37, 0xc30c8ea1, 0x5a05df1b, 0x2d02ef8d,
};
#ifdef __x86_64__
/*
* Note that the CRC-32 checksum is merely used for error detection in
* transmission and storage. It is not intended to guard against the malicious
* modification of files (i.e., it is not a cryptographic hash). !!!
*
* This code is based on zlib code from:
*
* https://github.com/vlastavesely/crc32sum
* https://github.com/chromium/chromium/blob/master/third_party/zlib/
*
* SPDX-License-Identifier: GPL-2.0
*/
/*
* ATM Use this code only for X86_64 arch where it was tested
* TODO: check if it speeds also non X86_64 arch
*/
static unsigned int _crc32_lookup[16][256] = { 0 };
static void _initialise_crc32(void)
{
unsigned int i, j;
if (_crc32_lookup[0][1])
return;
for (i = 0; i < 256; i++)
_crc32_lookup[0][i] = _crctab[i];
for (i = 0; i < 256; i++)
for (j = 1; j < 16; j++)
_crc32_lookup[j][i] = (_crc32_lookup[j - 1][i] >> 8) ^
_crc32_lookup[0][_crc32_lookup[j - 1][i] & 0xff];
}
#ifndef DEBUG_CRC32
uint32_t calc_crc(uint32_t initial, const uint8_t *buf, uint32_t size)
#else
static uint32_t _calc_crc_new(uint32_t initial, const uint8_t *buf, uint32_t size)
#endif
{
const uint32_t *ptr = (const uint32_t *) buf;
uint32_t a, b, c, d;
uint32_t crc = initial;
_initialise_crc32();
for (;size >= 16; size -= 16) {
a = xlate32(*ptr++) ^ crc;
b = xlate32(*ptr++);
c = xlate32(*ptr++);
d = xlate32(*ptr++);
crc = _crc32_lookup[ 0][(d >> 24) & 0xff] ^
_crc32_lookup[ 1][(d >> 16) & 0xff] ^
_crc32_lookup[ 2][(d >> 8) & 0xff] ^
_crc32_lookup[ 3][ d & 0xff] ^
_crc32_lookup[ 4][(c >> 24) & 0xff] ^
_crc32_lookup[ 5][(c >> 16) & 0xff] ^
_crc32_lookup[ 6][(c >> 8) & 0xff] ^
_crc32_lookup[ 7][ c & 0xff] ^
_crc32_lookup[ 8][(b >> 24) & 0xff] ^
_crc32_lookup[ 9][(b >> 16) & 0xff] ^
_crc32_lookup[10][(b >> 8) & 0xff] ^
_crc32_lookup[11][ b & 0xff] ^
_crc32_lookup[12][(a >> 24) & 0xff] ^
_crc32_lookup[13][(a >> 16) & 0xff] ^
_crc32_lookup[14][(a >> 8) & 0xff] ^
_crc32_lookup[15][ a & 0xff];
}
buf = (const uint8_t *) ptr;
while (size--)
crc = _crc32_lookup[0][((unsigned char) crc ^ *(buf++))] ^ (crc >> 8);
return crc;
}
#else // __x86_64__
/* Calculate an endian-independent CRC of supplied buffer */
#ifndef DEBUG_CRC32
uint32_t calc_crc(uint32_t initial, const uint8_t *buf, uint32_t size)
@ -89,6 +173,8 @@ static uint32_t _calc_crc_new(uint32_t initial, const uint8_t *buf, uint32_t siz
return crc;
}
#endif // __x86_64__
#ifdef DEBUG_CRC32
static uint32_t _calc_crc_old(uint32_t initial, const uint8_t *buf, uint32_t size)
{