Profiling

Learn performance measurement with profiling tools.

Profiling

Profiling measures where a program spends its time. Profile to act on data rather than guesses; often an unexpected function is the bottleneck.

In C++ time is measured with std::chrono::steady_clock: auto t0 = std::chrono::steady_clock::now(); for the start, auto t1 = std::chrono::steady_clock::now(); for the end. The elapsed time is the difference t1 - t0. On Linux use perf, with GCC use gprof, with Valgrind use callgrind.

Measuring with std::chrono

std::chrono::steady_clock is a monotonic clock; it never goes back. The now() call returns a time point; the difference between two points gives a duration.

To convert the duration to microseconds: auto us = std::chrono::duration_cast<std::chrono::microseconds>(t1 - t0).count();. Here .count() extracts the number. This idiom is placed inside multiplier (benchmark) loops to measure a function's average time.

Exercise Measure a time point

Add #include <chrono>. In main write auto t0 = std::chrono::steady_clock::now(); and auto t1 = std::chrono::steady_clock::now();. Then write std::cout << (t1 >= t0);.

Show example solution

#include <iostream>
#include <chrono>

int main() {
    auto t0 = std::chrono::steady_clock::now();
    auto t1 = std::chrono::steady_clock::now();
    std::cout << (t1 >= t0);
    return 0;
}

Test yourself

What does profiling measure?
```
auto t0 = std::chrono::steady_clock::now();
// ... work ...
auto t1 = std::chrono::steady_clock::now();
```
- Where time is spent
- Compile time
- Symbol count
- Memory size
Show answer

Where time is spent — Profiling shows the bottleneck with data.

Complete the now() call

auto t0 = std::chrono::steady_clock::____();

Show answer

auto t0 = std::chrono::steady_clock::now();

Match each tool to its role

perf

Linux profiler
gprof

GCC profiler
callgrind

Valgrind profiler
std::chrono

Duration measure

Next lesson: Cache

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies