Using the Profile class

The Profile class allows you to instrument your code for profiling, enabling you to measure various metrics like time, FLOPs, memory allocations, and more. It provides functions to mark the beginning and end of profiling blocks and to increment profiling counters. This tutorial will guide you through using the Profile class in your code. For more advanced usage and additional features, please refer to the Profile class API in profile.hpp.

Enabling and Disabling Profiling

You can enable or disable profiling using the Enable function. By default, profiling is disabled.

sctl::Profile::Enable(true)  // Enable profiling

Marking the Start and End of Profiling Blocks

Use Tic to mark the start of a profiling block and Toc to mark its end.

sctl::Profile::Tic("BlockName")  // Start of profiling block
// Your code to be profiled
sctl::Profile::Toc()  // End of profiling block

Incrementing Profiling Counters

You can manually increment profiling counters using IncrementCounter.

sctl::Profile::IncrementCounter(ProfileCounter::FLOP, numFlops)

Displaying Profiling Results

You can print the profiling results using the print function. It displays various profiling metrics.

sctl::Profile::print()

Example Usage

Let’s take a look at how you can use the Profile class in your code:

#include "sctl.hpp"

int main(int argc, char** argv) {
  sctl::Comm::MPI_Init(&argc, &argv);

  // Enable profiling
  sctl::Profile::Enable(true);

  // Start profiling block
  sctl::Profile::Tic("TestBlock");

  // Your code to be profiled
  // ...

  // End profiling block
  sctl::Profile::Toc();

  // Print profiling results
  sctl::Profile::print();

  sctl::Comm::MPI_Finalize();
  return 0;
}

Additional Notes

• You can create a scoped profiling block using the Scoped struct, which automatically marks the beginning and end of a block within its scope.

• There are various predefined profiling counters like time, FLOPs, heap allocations, etc., that you can use.