RooFit::Detail::CudaHelpers Namespace Reference

Functions
std::pair< std::chrono::microseconds, std::chrono::microseconds >	memcpyBenchmark (std::size_t nBytes)
	Measure the time taken to perform memory copy operations between host and device memory.

Function Documentation

memcpyBenchmark()

std::pair< std::chrono::microseconds, std::chrono::microseconds > RooFit::Detail::CudaHelpers::memcpyBenchmark

(

std::size_t

nBytes

)

Measure the time taken to perform memory copy operations between host and device memory.

Parameters

nBytes

The number of bytes to be copied between host and device memory.

Returns

A pair of durations representing the average time taken for host-to-device and device-to-host copies.

This function measures the time taken to copy data between host and device memory using the CUDA API. It performs a series of copy operations and calculates the average time for both directions. The input parameter nBytes specifies the size of the data to be copied in bytes. The function returns a pair of durations, where the first duration represents the average time taken for host-to-device copies and the second duration represents the average time taken for device-to-host copies.

Example usage:

auto copyTimes = RooFit::Detail::CudaHelpers::memcpyBenchmark(1024 * 1024); // Measure copy time for 1 MB of data
std::cout << "Average Host-to-Device Copy Time: " << copyTimes.first.count() << " microseconds" << std::endl;
std::cout << "Average Device-to-Host Copy Time: " << copyTimes.second.count() << " microseconds" << std::endl;

Definition at line 37 of file CudaHelpers.cxx.