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
|
// Copyright 2013 Alex Reece.
//
// A cross platform monotonic timer.
// see https://github.com/awreece/monotonic_timer
#include <unistd.h>
#include "Timer.h"
#define NANOS_PER_SECF 1000000000.0
#define USECS_PER_SEC 1000000
#if _POSIX_TIMERS > 0 && defined(_POSIX_MONOTONIC_CLOCK)
// If we have it, use clock_gettime and CLOCK_MONOTONIC.
#include <time.h>
double uscxml::Timer::monotonic_seconds() {
struct timespec time;
// Note: Make sure to link with -lrt to define clock_gettime.
clock_gettime(CLOCK_MONOTONIC, &time);
return ((double) time.tv_sec) + ((double) time.tv_nsec / (NANOS_PER_SECF));
}
#elif defined(__APPLE__)
// If we don't have CLOCK_MONOTONIC, we might be on a Mac. There we instead
// use mach_absolute_time().
#include <mach/mach_time.h>
static mach_timebase_info_data_t info;
static void __attribute__((constructor)) init_info() {
mach_timebase_info(&info);
}
double uscxml::Timer::monotonic_seconds() {
uint64_t time = mach_absolute_time();
double dtime = (double) time;
dtime *= (double) info.numer;
dtime /= (double) info.denom;
return dtime / NANOS_PER_SECF;
}
#elif defined(_MSC_VER)
// On Windows, use QueryPerformanceCounter and QueryPerformanceFrequency.
#include <windows.h>
static double PCFreq = 0.0;
// According to http://stackoverflow.com/q/1113409/447288, this will
// make this function a constructor.
// TODO(awreece) Actually attempt to compile on windows.
static void __cdecl init_pcfreq();
__declspec(allocate(".CRT$XCU")) void (__cdecl*init_pcfreq_)() = init_pcfreq;
static void __cdecl init_pcfreq() {
// Accoring to http://stackoverflow.com/a/1739265/447288, this will
// properly initialize the QueryPerformanceCounter.
LARGE_INTEGER li;
int has_qpc = QueryPerformanceFrequency(&li);
assert(has_qpc);
PCFreq = ((double) li.QuadPart) / 1000.0;
}
double uscxml::Timer::monotonic_seconds() {
LARGE_INTEGER li;
QueryPerformanceCounter(&li);
return ((double) li.QuadPart) / PCFreq;
}
#else
// Fall back to rdtsc. The reason we don't use clock() is this scary message
// from the man page:
// "On several other implementations, the value returned by clock() also
// includes the times of any children whose status has been collected via
// wait(2) (or another wait-type call)."
//
// Also, clock() only has microsecond accuracy.
//
// This whitepaper offered excellent advice on how to use rdtscp for
// profiling: http://download.intel.com/embedded/software/IA/324264.pdf
//
// Unfortunately, we can't follow its advice exactly with our semantics,
// so we're just going to use rdtscp with cpuid.
//
// Note that rdtscp will only be available on new processors.
#include <stdint.h>
static inline uint64_t rdtsc() {
uint32_t hi, lo;
asm volatile("rdtscp\n"
"movl %%edx, %0\n"
"movl %%eax, %1\n"
"cpuid"
: "=r" (hi), "=r" (lo) : : "%rax", "%rbx", "%rcx", "%rdx");
return (((uint64_t)hi) << 32) | (uint64_t)lo;
}
static uint64_t rdtsc_per_sec = 0;
static void __attribute__((constructor)) init_rdtsc_per_sec() {
uint64_t before, after;
before = rdtsc();
usleep(USECS_PER_SEC);
after = rdtsc();
rdtsc_per_sec = after - before;
}
double uscxml::Timer::monotonic_seconds() {
return (double) rdtsc() / (double) rdtsc_per_sec;
}
#endif
|
