| 1 |
148 |
jeremybenn |
/* High precision, low overhead timing functions. i686 version.
|
| 2 |
|
|
Copyright (C) 1998 Free Software Foundation, Inc.
|
| 3 |
|
|
This file is part of the GNU C Library.
|
| 4 |
|
|
Contributed by Ulrich Drepper <drepper@cygnus.com>, 1998.
|
| 5 |
|
|
|
| 6 |
|
|
The GNU C Library is free software; you can redistribute it and/or
|
| 7 |
|
|
modify it under the terms of the GNU Lesser General Public
|
| 8 |
|
|
License as published by the Free Software Foundation; either
|
| 9 |
|
|
version 2.1 of the License, or (at your option) any later version.
|
| 10 |
|
|
|
| 11 |
|
|
The GNU C Library is distributed in the hope that it will be useful,
|
| 12 |
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 13 |
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
| 14 |
|
|
Lesser General Public License for more details.
|
| 15 |
|
|
|
| 16 |
|
|
You should have received a copy of the GNU Lesser General Public
|
| 17 |
|
|
License along with the GNU C Library; if not, write to the Free
|
| 18 |
|
|
Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
|
| 19 |
|
|
02111-1307 USA. */
|
| 20 |
|
|
|
| 21 |
|
|
/* Modified for newlib by Jeff Johnston - June 27, 2002 */
|
| 22 |
|
|
|
| 23 |
|
|
#ifndef _HP_TIMING_H
|
| 24 |
|
|
#define _HP_TIMING_H 1
|
| 25 |
|
|
|
| 26 |
|
|
#include <string.h>
|
| 27 |
|
|
#include <stdio.h>
|
| 28 |
|
|
#include <sys/param.h>
|
| 29 |
|
|
|
| 30 |
|
|
#ifdef __i686__
|
| 31 |
|
|
|
| 32 |
|
|
/* The macros defined here use the timestamp counter in i586 and up versions
|
| 33 |
|
|
of the x86 processors. They provide a very accurate way to measure the
|
| 34 |
|
|
time with very little overhead. The time values themself have no real
|
| 35 |
|
|
meaning, only differences are interesting.
|
| 36 |
|
|
|
| 37 |
|
|
This version is for the i686 processors. The difference to the i586
|
| 38 |
|
|
version is that the timerstamp register is unconditionally used. This is
|
| 39 |
|
|
not the case for the i586 version where we have to perform runtime test
|
| 40 |
|
|
whether the processor really has this capability. We have to make this
|
| 41 |
|
|
distinction since the sysdeps/i386/i586 code is supposed to work on all
|
| 42 |
|
|
platforms while the i686 already contains i686-specific code.
|
| 43 |
|
|
|
| 44 |
|
|
The list of macros we need includes the following:
|
| 45 |
|
|
|
| 46 |
|
|
- HP_TIMING_AVAIL: test for availability.
|
| 47 |
|
|
|
| 48 |
|
|
- HP_TIMING_INLINE: this macro is non-zero if the functionality is not
|
| 49 |
|
|
implemented using function calls but instead uses some inlined code
|
| 50 |
|
|
which might simply consist of a few assembler instructions. We have to
|
| 51 |
|
|
know this since we might want to use the macros here in places where we
|
| 52 |
|
|
cannot make function calls.
|
| 53 |
|
|
|
| 54 |
|
|
- hp_timing_t: This is the type for variables used to store the time
|
| 55 |
|
|
values.
|
| 56 |
|
|
|
| 57 |
|
|
- HP_TIMING_ZERO: clear `hp_timing_t' object.
|
| 58 |
|
|
|
| 59 |
|
|
- HP_TIMING_NOW: place timestamp for current time in variable given as
|
| 60 |
|
|
parameter.
|
| 61 |
|
|
|
| 62 |
|
|
- HP_TIMING_DIFF_INIT: do whatever is necessary to be able to use the
|
| 63 |
|
|
HP_TIMING_DIFF macro.
|
| 64 |
|
|
|
| 65 |
|
|
- HP_TIMING_DIFF: compute difference between two times and store it
|
| 66 |
|
|
in a third. Source and destination might overlap.
|
| 67 |
|
|
|
| 68 |
|
|
- HP_TIMING_ACCUM: add time difference to another variable. This might
|
| 69 |
|
|
be a bit more complicated to implement for some platforms as the
|
| 70 |
|
|
operation should be thread-safe and 64bit arithmetic on 32bit platforms
|
| 71 |
|
|
is not.
|
| 72 |
|
|
|
| 73 |
|
|
- HP_TIMING_ACCUM_NT: this is the variant for situations where we know
|
| 74 |
|
|
there are no threads involved.
|
| 75 |
|
|
|
| 76 |
|
|
- HP_TIMING_PRINT: write decimal representation of the timing value into
|
| 77 |
|
|
the given string. This operation need not be inline even though
|
| 78 |
|
|
HP_TIMING_INLINE is specified.
|
| 79 |
|
|
|
| 80 |
|
|
*/
|
| 81 |
|
|
|
| 82 |
|
|
/* We always assume having the timestamp register. */
|
| 83 |
|
|
#define HP_TIMING_AVAIL (1)
|
| 84 |
|
|
|
| 85 |
|
|
/* We indeed have inlined functions. */
|
| 86 |
|
|
#define HP_TIMING_INLINE (1)
|
| 87 |
|
|
|
| 88 |
|
|
/* We use 64bit values for the times. */
|
| 89 |
|
|
typedef unsigned long long int hp_timing_t;
|
| 90 |
|
|
|
| 91 |
|
|
/* Internal variable used to store the overhead of the measurement
|
| 92 |
|
|
opcodes. */
|
| 93 |
|
|
extern hp_timing_t __libc_hp_timing_overhead;
|
| 94 |
|
|
|
| 95 |
|
|
/* Set timestamp value to zero. */
|
| 96 |
|
|
#define HP_TIMING_ZERO(Var) (Var) = (0)
|
| 97 |
|
|
|
| 98 |
|
|
/* That's quite simple. Use the `rdtsc' instruction. Note that the value
|
| 99 |
|
|
might not be 100% accurate since there might be some more instructions
|
| 100 |
|
|
running in this moment. This could be changed by using a barrier like
|
| 101 |
|
|
'cpuid' right before the `rdtsc' instruciton. But we are not interested
|
| 102 |
|
|
in accurate clock cycles here so we don't do this. */
|
| 103 |
|
|
#define HP_TIMING_NOW(Var) __asm__ __volatile__ ("rdtsc" : "=A" (Var))
|
| 104 |
|
|
|
| 105 |
|
|
/* Use two 'rdtsc' instructions in a row to find out how long it takes. */
|
| 106 |
|
|
#define HP_TIMING_DIFF_INIT() \
|
| 107 |
|
|
do { \
|
| 108 |
|
|
int __cnt = 5; \
|
| 109 |
|
|
__libc_hp_timing_overhead = ~0ull; \
|
| 110 |
|
|
do \
|
| 111 |
|
|
{ \
|
| 112 |
|
|
hp_timing_t __t1, __t2; \
|
| 113 |
|
|
HP_TIMING_NOW (__t1); \
|
| 114 |
|
|
HP_TIMING_NOW (__t2); \
|
| 115 |
|
|
if (__t2 - __t1 < __libc_hp_timing_overhead) \
|
| 116 |
|
|
__libc_hp_timing_overhead = __t2 - __t1; \
|
| 117 |
|
|
} \
|
| 118 |
|
|
while (--__cnt > 0); \
|
| 119 |
|
|
} while (0)
|
| 120 |
|
|
|
| 121 |
|
|
/* It's simple arithmetic for us. */
|
| 122 |
|
|
#define HP_TIMING_DIFF(Diff, Start, End) (Diff) = ((End) - (Start))
|
| 123 |
|
|
|
| 124 |
|
|
/* We have to jump through hoops to get this correctly implemented. */
|
| 125 |
|
|
#define HP_TIMING_ACCUM(Sum, Diff) \
|
| 126 |
|
|
do { \
|
| 127 |
|
|
char __not_done; \
|
| 128 |
|
|
hp_timing_t __oldval = (Sum); \
|
| 129 |
|
|
hp_timing_t __diff = (Diff) - __libc_hp_timing_overhead; \
|
| 130 |
|
|
do \
|
| 131 |
|
|
{ \
|
| 132 |
|
|
hp_timing_t __newval = __oldval + __diff; \
|
| 133 |
|
|
int __temp0, __temp1; \
|
| 134 |
|
|
__asm__ __volatile__ ("xchgl %4, %%ebx\n\t" \
|
| 135 |
|
|
"lock; cmpxchg8b %1\n\t" \
|
| 136 |
|
|
"sete %0\n\t" \
|
| 137 |
|
|
"movl %4, %%ebx" \
|
| 138 |
|
|
: "=q" (__not_done), "=m" (Sum), \
|
| 139 |
|
|
"=A" (__oldval), "=c" (__temp0), \
|
| 140 |
|
|
"=SD" (__temp1) \
|
| 141 |
|
|
: "1" (Sum), "2" (__oldval), \
|
| 142 |
|
|
"3" (__newval >> 32), \
|
| 143 |
|
|
"4" (__newval & 0xffffffff) \
|
| 144 |
|
|
: "memory"); \
|
| 145 |
|
|
} \
|
| 146 |
|
|
while (__not_done); \
|
| 147 |
|
|
} while (0)
|
| 148 |
|
|
|
| 149 |
|
|
/* No threads, no extra work. */
|
| 150 |
|
|
#define HP_TIMING_ACCUM_NT(Sum, Diff) (Sum) += (Diff)
|
| 151 |
|
|
|
| 152 |
|
|
/* Print the time value. */
|
| 153 |
|
|
#define HP_TIMING_PRINT(Buf, Len, Val) \
|
| 154 |
|
|
do { \
|
| 155 |
|
|
char __buf[20]; \
|
| 156 |
|
|
char *__cp = __buf + sizeof (__buf); \
|
| 157 |
|
|
int __len = (Len); \
|
| 158 |
|
|
char *__dest = (Buf); \
|
| 159 |
|
|
do { \
|
| 160 |
|
|
*--__cp = Val % 10; \
|
| 161 |
|
|
Val /= 10; \
|
| 162 |
|
|
} while (Val > 0); \
|
| 163 |
|
|
while (__len-- > 0 && __cp < __buf + sizeof (__buf)) \
|
| 164 |
|
|
*__dest++ = *__cp++; \
|
| 165 |
|
|
memcpy (__dest, " clock cycles", MIN (__len, sizeof (" clock cycles"))); \
|
| 166 |
|
|
} while (0)
|
| 167 |
|
|
|
| 168 |
|
|
#else /* !__i686__ */
|
| 169 |
|
|
|
| 170 |
|
|
/* Provide dummy definitions. */
|
| 171 |
|
|
#define HP_TIMING_AVAIL (0)
|
| 172 |
|
|
#define HP_TIMING_INLINE (0)
|
| 173 |
|
|
typedef int hp_timing_t;
|
| 174 |
|
|
#define HP_TIMING_ZERO(Var)
|
| 175 |
|
|
#define HP_TIMING_NOW(var)
|
| 176 |
|
|
#define HP_TIMING_DIFF_INIT()
|
| 177 |
|
|
#define HP_TIMING_DIFF(Diff, Start, End)
|
| 178 |
|
|
#define HP_TIMING_ACCUM(Sum, Diff)
|
| 179 |
|
|
#define HP_TIMING_ACCUM_NT(Sum, Diff)
|
| 180 |
|
|
#define HP_TIMING_PRINT(Buf, Len, Val)
|
| 181 |
|
|
|
| 182 |
|
|
/* Since this implementation is not available we tell the user about it. */
|
| 183 |
|
|
#define HP_TIMING_NONAVAIL 1
|
| 184 |
|
|
|
| 185 |
|
|
#endif
|
| 186 |
|
|
|
| 187 |
|
|
#endif /* hp-timing.h */
|
