fft_backend_benchmark.cpp File Reference

Benchmark CPU (FFTW) vs GPU (CUDA) FFT performance. More...

#include <chrono>
#include <iomanip>
#include <iostream>
#include <memory>
#include <vector>
#include "openpfc/core/databuffer.hpp"
#include "openpfc/core/decomposition.hpp"
#include "openpfc/core/world.hpp"
#include "openpfc/fft.hpp"

Include dependency graph for fft_backend_benchmark.cpp:

Functions
double	benchmark_fft (fft::Backend backend, const World &world, const decomposition::Decomposition &decomp, int rank_id)
	Benchmark FFT performance for a given backend.
int	main (int argc, char *argv[])

Variables
constexpr int	GRID_SIZE = 128
constexpr int	NUM_ITERATIONS = 10

Detailed Description

Benchmark CPU (FFTW) vs GPU (CUDA) FFT performance.

This example demonstrates:

• Runtime FFT backend selection
• Performance measurement using std::chrono
• Speedup comparison between CPU and GPU
• Proper usage of DataBuffer for GPU operations

Compile with: cmake -B build -DOpenPFC_ENABLE_CUDA=ON cmake –build build –target fft_backend_benchmark

Run: mpirun -np 1 ./examples/fft_backend_benchmark

Function Documentation

benchmark_fft()

double benchmark_fft	(	fft::Backend	backend,
		const World &	world,
		const decomposition::Decomposition &	decomp,
		int	rank_id
	)

Benchmark FFT performance for a given backend.

Parameters

backend	The FFT backend to test (FFTW or CUDA)
world	The computational domain
decomp	Domain decomposition
rank_id	MPI rank ID

Returns

Average time per forward+backward transform pair (in milliseconds)

Here is the call graph for this function: