1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
|
/*
Copyright 2005, 2006, 2007, 2008 Free Software Foundation, Inc.
Contributed by Patrick Pelissier, INRIA.
This file is part of the MPFR Library.
The MPFR 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 MPFR 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 License
along with the MPFR Library; see the file COPYING.LIB. If not, write to
the Free Software Foundation, Inc., 51 Franklin St, Fifth Floor, Boston,
MA 02110-1301, USA. */
#ifndef __TIMP__H__
#define __TIMP__H__
/* Usage:
* Before doing the measure, call TIMP_OVERHEAD ();
* Then unsigned long long t = TIMP_MEASURE (f(x));
* to measure the # of cycles taken by the call to f(x).
*/
#define TIMP_VERSION 1*100+0*10+0
#ifndef __GNUC__
# error CC != GCC
#endif
/* High accuracy timing */
#if defined (__i386__) || defined(__amd64__)
#define timp_rdtsc_before(time) \
__asm__ __volatile__( \
".align 64\n\t" \
"xorl %%eax,%%eax\n\t" \
"cpuid\n\t" \
"rdtsc\n\t" \
"movl %%eax,(%0)\n\t" \
"movl %%edx,4(%0)\n\t" \
"xorl %%eax,%%eax\n\t" \
"cpuid\n\t" \
: /* no output */ \
: "S"(&time) \
: "eax", "ebx", "ecx", "edx", "memory")
#define timp_rdtsc_after(time) \
__asm__ __volatile__( \
"xorl %%eax,%%eax\n\t" \
"cpuid\n\t" \
"rdtsc\n\t" \
"movl %%eax,(%0)\n\t" \
"movl %%edx,4(%0)\n\t" \
"xorl %%eax,%%eax\n\t" \
"cpuid\n\t" \
: /* no output */ \
: "S"(&time) \
: "eax", "ebx", "ecx", "edx", "memory")
#elif defined (__ia64)
#define timp_rdtsc() \
({ unsigned long long int x; \
__asm__ __volatile__("mov %0=ar.itc" : "=r"(x) :: "memory"); \
x; })
#define timp_rdtsc_before(time) (time = timp_rdtsc())
#define timp_rdtsc_after(time) (time = timp_rdtsc())
#elif defined (__alpha)
#define timp_rdtsc() \
({ unsigned long long int x; \
__asm__ volatile ("rpcc %0\n\t" : "=r" (x)); \
x; })
#define timp_rdtsc_before(time) (time = tpp_rdtsc())
#define timp_rdtsc_after(time) (time = tpp_rdtsc())
#else
# error Unsupported CPU
#endif
/* We do several measures and keep the minimum to avoid counting
* hardware interruption cycles.
* The filling of the CPU cache is done because we do several loops,
* and get the minimum.
* Declaring num_cycle as "volatile" is to avoid optimisation when it is
* possible (To properly calcul overhead).
* overhead is calculated outside by a call to:
* overhead = MEASURE("overhead", ;)
* Use a lot the preprocessor.
* It is a macro to be very flexible.
*/
static unsigned long long int timp_overhead = 0;
#define TIMP_NUM_TRY 4327
#define TIMP_MAX_WAIT_FOR_MEASURE 10000000ULL
#define TIMP_MEASURE(CODE) \
({ \
volatile unsigned long long int num_cycle, num_cycle2; \
unsigned long long min_num_cycle, start_num_cycle; \
int _i; \
timp_rdtsc_before (start_num_cycle); \
min_num_cycle = 0xFFFFFFFFFFFFFFFFLL; \
for(_i = 0 ; _i < TIMP_NUM_TRY ; _i++) { \
timp_rdtsc_before(num_cycle); \
CODE; \
timp_rdtsc_after(num_cycle2); \
num_cycle = num_cycle2 - num_cycle; \
if (num_cycle < min_num_cycle) \
min_num_cycle = num_cycle; \
if (num_cycle2 - start_num_cycle > TIMP_MAX_WAIT_FOR_MEASURE) \
break; \
} \
min_num_cycle - timp_overhead; })
#define TIMP_OVERHEAD() \
(timp_overhead = 0, timp_overhead = TIMP_MEASURE((void) 0) )
#endif /* __TIMP__H__ */
|
