mirror/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-01-15 19:42:28 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions
mariadb/include/byte_order_generic_x86_64.h

126 lines
4 KiB
C
Raw Permalink Normal View History

Temporary commit of 10.0-merge
2013-03-25 23:03:13 +01:00
/* Copyright (c) 2001, 2012, Oracle and/or its affiliates. All rights reserved.
MDEV-21981 Replace arithmetic + with bitwise OR when possible Several macros such as sint2korr() and uint4korr() are using the arithmetic + operator while a bitwise or operator would suffice. GCC 5 and clang 5 and later can detect patterns consisting of bitwise or and shifts by multiples of 8 bits, such as those used in the InnoDB function mach_read_from_4(). They actually translate that verbose low-level code into high-level machine language (i486 bswap instruction or fused into the Haswell movbe instruction). We should do the same for MariaDB Server code that is outside InnoDB. Note: The Microsoft C compiler is lacking this optimization. There, we might consider using _byteswap_ushort(), _byteswap_ulong(), _byteswap_uint64(). But, those would lead to unaligned reads, which are bad for reasons stated in MDEV-20277. Besides, outside InnoDB, most data is already being stored in the native little-endian format of that compiler.
2020-03-19 06:52:35 +01:00
Copyright (c) 2020, MariaDB Corporation.
Temporary commit of 10.0-merge
2013-03-25 23:03:13 +01:00
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program 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 General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Update FSF address
2019-05-11 18:25:02 +02:00
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1335 USA */
Temporary commit of 10.0-merge
2013-03-25 23:03:13 +01:00
/*
Optimized function-like macros for the x86 architecture (_WIN32 included).
*/
Optimized intkorr() and intstore(functions) for intel 64 bits. (Benchmarked, about 30 % faster and 50 % smaller than original)
2016-01-29 23:11:27 +01:00
Temporary commit of 10.0-merge
2013-03-25 23:03:13 +01:00
#define sint2korr(A) (int16) (*((int16 *) (A)))
#define sint3korr(A) ((int32) ((((uchar) (A)[2]) & 128) ? \
(((uint32) 255L << 24) | \
(((uint32) (uchar) (A)[2]) << 16) |\
(((uint32) (uchar) (A)[1]) << 8) | \
((uint32) (uchar) (A)[0])) : \
(((uint32) (uchar) (A)[2]) << 16) |\
(((uint32) (uchar) (A)[1]) << 8) | \
((uint32) (uchar) (A)[0])))
#define sint4korr(A) (int32) (*((int32 *) (A)))
#define uint2korr(A) (uint16) (*((uint16 *) (A)))
MDEV-21981 Replace arithmetic + with bitwise OR when possible Several macros such as sint2korr() and uint4korr() are using the arithmetic + operator while a bitwise or operator would suffice. GCC 5 and clang 5 and later can detect patterns consisting of bitwise or and shifts by multiples of 8 bits, such as those used in the InnoDB function mach_read_from_4(). They actually translate that verbose low-level code into high-level machine language (i486 bswap instruction or fused into the Haswell movbe instruction). We should do the same for MariaDB Server code that is outside InnoDB. Note: The Microsoft C compiler is lacking this optimization. There, we might consider using _byteswap_ushort(), _byteswap_ulong(), _byteswap_uint64(). But, those would lead to unaligned reads, which are bad for reasons stated in MDEV-20277. Besides, outside InnoDB, most data is already being stored in the native little-endian format of that compiler.
2020-03-19 06:52:35 +01:00
#define uint3korr(A) (uint32) (((uint32) ((uchar) (A)[0])) |\
(((uint32) ((uchar) (A)[1])) << 8) |\
Temporary commit of 10.0-merge
2013-03-25 23:03:13 +01:00
(((uint32) ((uchar) (A)[2])) << 16))
#define uint4korr(A) (uint32) (*((uint32 *) (A)))
Optimized intkorr() and intstore(functions) for intel 64 bits. (Benchmarked, about 30 % faster and 50 % smaller than original)
2016-01-29 23:11:27 +01:00
static inline ulonglong uint5korr(const void *p)
{
ulonglong a= *(uint32 *) p;
ulonglong b= *(4 + (uchar *) p);
return a | (b << 32);
}
static inline ulonglong uint6korr(const void *p)
{
ulonglong a= *(uint32 *) p;
ulonglong b= *(uint16 *) (4 + (char *) p);
return a | (b << 32);
}
Temporary commit of 10.0-merge
2013-03-25 23:03:13 +01:00
#define uint8korr(A) (ulonglong) (*((ulonglong *) (A)))
#define sint8korr(A) (longlong) (*((longlong *) (A)))
#define int2store(T,A) do { uchar *pT= (uchar*)(T);\
*((uint16*)(pT))= (uint16) (A);\
} while (0)
MDEV-21981 Replace arithmetic + with bitwise OR when possible Several macros such as sint2korr() and uint4korr() are using the arithmetic + operator while a bitwise or operator would suffice. GCC 5 and clang 5 and later can detect patterns consisting of bitwise or and shifts by multiples of 8 bits, such as those used in the InnoDB function mach_read_from_4(). They actually translate that verbose low-level code into high-level machine language (i486 bswap instruction or fused into the Haswell movbe instruction). We should do the same for MariaDB Server code that is outside InnoDB. Note: The Microsoft C compiler is lacking this optimization. There, we might consider using _byteswap_ushort(), _byteswap_ulong(), _byteswap_uint64(). But, those would lead to unaligned reads, which are bad for reasons stated in MDEV-20277. Besides, outside InnoDB, most data is already being stored in the native little-endian format of that compiler.
2020-03-19 06:52:35 +01:00
Temporary commit of 10.0-merge
2013-03-25 23:03:13 +01:00
#define int3store(T,A) do { *(T)= (uchar) ((A));\
*(T+1)=(uchar) (((uint) (A) >> 8));\
*(T+2)=(uchar) (((A) >> 16));\
} while (0)
Optimized intkorr() and intstore(functions) for intel 64 bits. (Benchmarked, about 30 % faster and 50 % smaller than original)
2016-01-29 23:11:27 +01:00
Temporary commit of 10.0-merge
2013-03-25 23:03:13 +01:00
#define int4store(T,A) do { uchar *pT= (uchar*)(T);\
*((uint32 *) (pT))= (uint32) (A); \
} while (0)
Optimized intkorr() and intstore(functions) for intel 64 bits. (Benchmarked, about 30 % faster and 50 % smaller than original)
2016-01-29 23:11:27 +01:00
#define int5store(T,A) do { uchar *pT= (uchar*)(T);\
*((uint32 *) (pT))= (uint32) (A); \
*((pT)+4)=(uchar) (((A) >> 32));\
} while (0)
#define int6store(T,A) do { uchar *pT= (uchar*)(T);\
*((uint32 *) (pT))= (uint32) (A); \
*((uint16*)(pT+4))= (uint16) (A >> 32);\
} while (0)
Temporary commit of 10.0-merge
2013-03-25 23:03:13 +01:00
#define int8store(T,A) do { uchar *pT= (uchar*)(T);\
*((ulonglong *) (pT))= (ulonglong) (A);\
} while(0)
Optimized intkorr() and intstore(functions) for intel 64 bits. (Benchmarked, about 30 % faster and 50 % smaller than original)
2016-01-29 23:11:27 +01:00
#if defined(__GNUC__)
#define HAVE_mi_uint5korr
#define HAVE_mi_uint6korr
#define HAVE_mi_uint7korr
typo fixed. HAVE_mi_uint8korr
2023-11-02 04:33:48 +01:00
#define HAVE_mi_uint8korr
Optimized intkorr() and intstore(functions) for intel 64 bits. (Benchmarked, about 30 % faster and 50 % smaller than original)
2016-01-29 23:11:27 +01:00
/* Read numbers stored in high-bytes-first order */
static inline ulonglong mi_uint5korr(const void *p)
{
ulonglong a= *(uint32 *) p;
ulonglong b= *(4 + (uchar *) p);
ulonglong v= (a | (b << 32)) << 24;
asm ("bswapq %0" : "=r" (v) : "0" (v));
return v;
}
static inline ulonglong mi_uint6korr(const void *p)
{
ulonglong a= *(uint32 *) p;
ulonglong b= *(uint16 *) (4 + (char *) p);
ulonglong v= (a | (b << 32)) << 16;
asm ("bswapq %0" : "=r" (v) : "0" (v));
return v;
}
static inline ulonglong mi_uint7korr(const void *p)
{
ulonglong a= *(uint32 *) p;
ulonglong b= *(uint16 *) (4 + (char *) p);
ulonglong c= *(6 + (uchar *) p);
ulonglong v= (a | (b << 32) | (c << 48)) << 8;
asm ("bswapq %0" : "=r" (v) : "0" (v));
return v;
}
static inline ulonglong mi_uint8korr(const void *p)
{
ulonglong v= *(ulonglong *) p;
asm ("bswapq %0" : "=r" (v) : "0" (v));
return v;
}
#endif
Reference in a new issue Copy permalink