File: perf-inl.h

package info (click to toggle)
pytorch-cuda 2.6.0%2Bdfsg-7
  • links: PTS, VCS
  • area: contrib
  • in suites: forky, sid, trixie
  • size: 161,620 kB
  • sloc: python: 1,278,832; cpp: 900,322; ansic: 82,710; asm: 7,754; java: 3,363; sh: 2,811; javascript: 2,443; makefile: 597; ruby: 195; xml: 84; objc: 68
file content (68 lines) | stat: -rw-r--r-- 1,404 bytes parent folder | download | duplicates (3)
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
#pragma once

#if defined(__ANDROID__) || defined(__linux__)

#include <unistd.h>

#include <sys/ioctl.h>
#include <sys/syscall.h>

#include <linux/perf_event.h>

#endif /* __ANDROID__ || __linux__ */

#include <torch/csrc/profiler/perf.h>

#include <limits>

namespace torch::profiler::impl::linux_perf {

/*
 * PerfEvent
 * ---------
 */

inline void PerfEvent::Disable() const {
#if defined(__ANDROID__) || defined(__linux__)
  ioctl(fd_, PERF_EVENT_IOC_DISABLE, 0);
#endif /* __ANDROID__ || __linux__ */
}

inline void PerfEvent::Enable() const {
#if defined(__ANDROID__) || defined(__linux__)
  ioctl(fd_, PERF_EVENT_IOC_ENABLE, 0);
#endif /* __ANDROID__ || __linux__ */
}

inline void PerfEvent::Reset() const {
#if defined(__ANDROID__) || defined(__linux__)
  ioctl(fd_, PERF_EVENT_IOC_RESET, 0);
#endif /* __ANDROID__ || __linux__ */
}

/*
 * PerfProfiler
 * ------------
 */

inline uint64_t PerfProfiler::CalcDelta(uint64_t start, uint64_t end) const {
  if (end < start) { // overflow
    return end + (std::numeric_limits<uint64_t>::max() - start);
  }
  // not possible to wrap around start for a 64b cycle counter
  return end - start;
}

inline void PerfProfiler::StartCounting() const {
  for (auto& e : events_) {
    e.Enable();
  }
}

inline void PerfProfiler::StopCounting() const {
  for (auto& e : events_) {
    e.Disable();
  }
}

} // namespace torch::profiler::impl::linux_perf