AuroraMiddleware/glibc
1
0
Fork 0
mirror of https://sourceware.org/git/glibc.git synced 2024-12-04 02:40:06 +00:00
Code Issues Packages Projects Releases Wiki Activity
a7fcc2f8ed
glibc/sysdeps/tile/tilegx/string-endian.h

59 lines
2.1 KiB
C
Raw Normal View History

Update copyright dates with scripts/update-copyrights.
2015-01-02 16:28:19 +00:00
/* Copyright (C) 2011-2015 Free Software Foundation, Inc.
Add sysdeps/tile, sysdeps/unix/sysv/linux/tile. Changes are tracked in ChangeLog.tile. Also add data/c++-types-tile-linux-gnu.data.
2011-12-04 02:14:25 +00:00
This file is part of the GNU C Library.
Contributed by Chris Metcalf <cmetcalf@tilera.com>, 2011.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
Replace FSF snail mail address by URL.
2012-03-09 23:56:38 +00:00
License along with the GNU C Library. If not, see
<http://www.gnu.org/licenses/>. */
Add sysdeps/tile, sysdeps/unix/sysv/linux/tile. Changes are tracked in ChangeLog.tile. Also add data/c++-types-tile-linux-gnu.data.
2011-12-04 02:14:25 +00:00
tilegx: provide optimized strnlen, strstr, and strcasestr strnlen() is based on the existing tile strlen() with length checking added. It speeds up by up to 5x, but on average across the benchtest corpus by around 35%. No regressions are seen. strstr() does 8-byte aligned loads and compares using a 2-byte filter on the first two bytes of the needle and then testing the remaining bytes in needle using memcmp(). It speeds up about 5x in the best case (for "found" needles), about 2x looking at benchtest as a whole, with some slowdowns as much as 45%. on a few cases (including the "fail" case for 128KB search). strcasestr() is based on strstr() but uses a SIMD tolower routine to convert 8-bytes to lower case in 5 instructions. It also uses a 2-byte filter and then strncasecmp() for the remaining bytes. strncasecmp() is not optimized for SIMD, so there is futher room for improvement. However, it is still up to 16x faster for "found" needles, averaging 2x faster on the whole corpus of benchtests. It does slow down by up to 35% on a few cases, similarly to strstr().
2014-09-16 00:10:18 +00:00
#include <endian.h>
#include <stdint.h>
/* Provide a set of macros to help keep endianness #ifdefs out of
the string functions.
MASK: Provide a mask based on the pointer alignment that
Add sysdeps/tile, sysdeps/unix/sysv/linux/tile. Changes are tracked in ChangeLog.tile. Also add data/c++-types-tile-linux-gnu.data.
2011-12-04 02:14:25 +00:00
sets up non-zero bytes before the beginning of the string.
The MASK expression works because shift counts are taken mod 64.
tilegx: provide optimized strnlen, strstr, and strcasestr strnlen() is based on the existing tile strlen() with length checking added. It speeds up by up to 5x, but on average across the benchtest corpus by around 35%. No regressions are seen. strstr() does 8-byte aligned loads and compares using a 2-byte filter on the first two bytes of the needle and then testing the remaining bytes in needle using memcmp(). It speeds up about 5x in the best case (for "found" needles), about 2x looking at benchtest as a whole, with some slowdowns as much as 45%. on a few cases (including the "fail" case for 128KB search). strcasestr() is based on strstr() but uses a SIMD tolower routine to convert 8-bytes to lower case in 5 instructions. It also uses a 2-byte filter and then strncasecmp() for the remaining bytes. strncasecmp() is not optimized for SIMD, so there is futher room for improvement. However, it is still up to 16x faster for "found" needles, averaging 2x faster on the whole corpus of benchtests. It does slow down by up to 35% on a few cases, similarly to strstr().
2014-09-16 00:10:18 +00:00
NULMASK: Clear bytes beyond a given point in the string.
CFZ: Find the first zero bit in the 8 string bytes in a long.
REVCZ: Find the last zero bit in the 8 string bytes in a long.
STRSHIFT: Shift N bits towards the start of the string. */
Add #include <stdint.h> for uint[32|64]_t usage (except installed headers).
2013-05-01 15:46:34 +00:00
tilegx: provide optimized strnlen, strstr, and strcasestr strnlen() is based on the existing tile strlen() with length checking added. It speeds up by up to 5x, but on average across the benchtest corpus by around 35%. No regressions are seen. strstr() does 8-byte aligned loads and compares using a 2-byte filter on the first two bytes of the needle and then testing the remaining bytes in needle using memcmp(). It speeds up about 5x in the best case (for "found" needles), about 2x looking at benchtest as a whole, with some slowdowns as much as 45%. on a few cases (including the "fail" case for 128KB search). strcasestr() is based on strstr() but uses a SIMD tolower routine to convert 8-bytes to lower case in 5 instructions. It also uses a 2-byte filter and then strncasecmp() for the remaining bytes. strncasecmp() is not optimized for SIMD, so there is futher room for improvement. However, it is still up to 16x faster for "found" needles, averaging 2x faster on the whole corpus of benchtests. It does slow down by up to 35% on a few cases, similarly to strstr().
2014-09-16 00:10:18 +00:00
#if __BYTE_ORDER == __LITTLE_ENDIAN
Add sysdeps/tile, sysdeps/unix/sysv/linux/tile. Changes are tracked in ChangeLog.tile. Also add data/c++-types-tile-linux-gnu.data.
2011-12-04 02:14:25 +00:00
#define MASK(x) (__insn_shl(1ULL, (x << 3)) - 1)
#define NULMASK(x) ((2ULL << x) - 1)
#define CFZ(x) __insn_ctz(x)
#define REVCZ(x) __insn_clz(x)
tilegx: provide optimized strnlen, strstr, and strcasestr strnlen() is based on the existing tile strlen() with length checking added. It speeds up by up to 5x, but on average across the benchtest corpus by around 35%. No regressions are seen. strstr() does 8-byte aligned loads and compares using a 2-byte filter on the first two bytes of the needle and then testing the remaining bytes in needle using memcmp(). It speeds up about 5x in the best case (for "found" needles), about 2x looking at benchtest as a whole, with some slowdowns as much as 45%. on a few cases (including the "fail" case for 128KB search). strcasestr() is based on strstr() but uses a SIMD tolower routine to convert 8-bytes to lower case in 5 instructions. It also uses a 2-byte filter and then strncasecmp() for the remaining bytes. strncasecmp() is not optimized for SIMD, so there is futher room for improvement. However, it is still up to 16x faster for "found" needles, averaging 2x faster on the whole corpus of benchtests. It does slow down by up to 35% on a few cases, similarly to strstr().
2014-09-16 00:10:18 +00:00
#define STRSHIFT(x,n) ((x) >> n)
Add sysdeps/tile, sysdeps/unix/sysv/linux/tile. Changes are tracked in ChangeLog.tile. Also add data/c++-types-tile-linux-gnu.data.
2011-12-04 02:14:25 +00:00
#else
#define MASK(x) (__insn_shl(-2LL, ((-x << 3) - 1)))
#define NULMASK(x) (-2LL << (63 - x))
#define CFZ(x) __insn_clz(x)
#define REVCZ(x) __insn_ctz(x)
tilegx: provide optimized strnlen, strstr, and strcasestr strnlen() is based on the existing tile strlen() with length checking added. It speeds up by up to 5x, but on average across the benchtest corpus by around 35%. No regressions are seen. strstr() does 8-byte aligned loads and compares using a 2-byte filter on the first two bytes of the needle and then testing the remaining bytes in needle using memcmp(). It speeds up about 5x in the best case (for "found" needles), about 2x looking at benchtest as a whole, with some slowdowns as much as 45%. on a few cases (including the "fail" case for 128KB search). strcasestr() is based on strstr() but uses a SIMD tolower routine to convert 8-bytes to lower case in 5 instructions. It also uses a 2-byte filter and then strncasecmp() for the remaining bytes. strncasecmp() is not optimized for SIMD, so there is futher room for improvement. However, it is still up to 16x faster for "found" needles, averaging 2x faster on the whole corpus of benchtests. It does slow down by up to 35% on a few cases, similarly to strstr().
2014-09-16 00:10:18 +00:00
#define STRSHIFT(x,n) ((x) << n)
Add sysdeps/tile, sysdeps/unix/sysv/linux/tile. Changes are tracked in ChangeLog.tile. Also add data/c++-types-tile-linux-gnu.data.
2011-12-04 02:14:25 +00:00
#endif
tilegx: small performance fix for string routines We were multiplying a byte by 0x0101010101010101ULL to create a constant for SIMD ops, but the compiler isn't good at optimizing this case (the fact that one operand is a byte is lost by the time it would be possible to do the optimization). So instead we add a helper routine that explicitly uses SIMD ops to create the constant.
2012-05-11 22:03:58 +00:00
tilegx: optimize string copy_byte() internal function We can use one "shufflebytes" instruction instead of 3 "bfins" instructions to optimize the string functions.
2014-09-16 00:02:50 +00:00
/* Create eight copies of the byte in a uint64_t. Byte Shuffle uses
the bytes of srcB as the index into the dest vector to select a
byte. With all indices of zero, the first byte is copied into all
the other bytes. */
tilegx: small performance fix for string routines We were multiplying a byte by 0x0101010101010101ULL to create a constant for SIMD ops, but the compiler isn't good at optimizing this case (the fact that one operand is a byte is lost by the time it would be possible to do the optimization). So instead we add a helper routine that explicitly uses SIMD ops to create the constant.
2012-05-11 22:03:58 +00:00
static inline uint64_t copy_byte(uint8_t byte)
{
tilegx: optimize string copy_byte() internal function We can use one "shufflebytes" instruction instead of 3 "bfins" instructions to optimize the string functions.
2014-09-16 00:02:50 +00:00
return __insn_shufflebytes(byte, 0, 0);
tilegx: small performance fix for string routines We were multiplying a byte by 0x0101010101010101ULL to create a constant for SIMD ops, but the compiler isn't good at optimizing this case (the fact that one operand is a byte is lost by the time it would be possible to do the optimization). So instead we add a helper routine that explicitly uses SIMD ops to create the constant.
2012-05-11 22:03:58 +00:00
}
Reference in New Issue Copy Permalink