Testing vDSP's FFT vs FFTW's

main.cpp

#include "ofMain.h"
#include "ofxFft.h"
#include <Accelerate/Accelerate.h>
#include <mach/mach_time.h>

double mach2ms(uint64_t time)
{
    mach_timebase_info_data_t timebase;
    mach_timebase_info(&timebase);
    return (double)time * (double)timebase.numer / (double)timebase.denom / 1e6;
}

int main() {
	UInt32 min_log2N = 10;
	UInt32 max_log2N = 20;
	UInt32 max_N = (1 << max_log2N);
	
	// Create re-useable vDSP bits
	COMPLEX_SPLIT FFTData;
	FFTData.realp         = (float *) malloc(sizeof(float) * max_N/2);
	FFTData.imagp         = (float *) malloc(sizeof(float) * max_N/2);
	float * hammingWindow = (float *) malloc(sizeof(float) * max_N);
	float * A             = (float *) malloc(sizeof(float) * max_N);
	float * B             = (float *) malloc(sizeof(float) * max_N);
	FFTSetup FFTSetup = vDSP_create_fftsetup(max_log2N, kFFTRadix2);
	
	// Create a fake signal
	for(UInt32 i = 0; i < max_N; i++){
		A[i] = ofRandom(-1, 1);
	}
	
	for(UInt32 log2N = min_log2N; log2N < max_log2N; log2N++)
	{
		UInt32 N = (1 << log2N);
		
		// Prep vDSP
		vDSP_hamm_window(hammingWindow, N, 0);
		
		// Prep ofxFft
		ofxFft * fft = ofxFft::create(N, OF_FFT_WINDOW_HAMMING, OF_FFT_FFTW);
		
		uint64_t accel_start = mach_absolute_time();
		vDSP_vmul(A, 1, hammingWindow, 1, B, 1, N); // A -> hamming -> B
		vDSP_ctoz((COMPLEX *) B, 2, &FFTData, 1, N/2);
		vDSP_fft_zrip(FFTSetup, &FFTData, 1, log2N, kFFTDirection_Forward);
		uint64_t accel_end = mach_absolute_time();
		
		uint64_t fftw_start = mach_absolute_time();
		fft->setSignal(A);
		fft->getImaginary();
		uint64_t fftw_end = mach_absolute_time();
		
		cout
		<< " N = "    << N
		<< " Accel: " << mach2ms(accel_end - accel_start)
		<< " FFTW: "  << mach2ms(fftw_end - fftw_start) << endl;
		
		delete fft;
	}
	
	free(FFTData.realp);
	free(FFTData.imagp);
	free(hammingWindow);
	free(A);
	free(B);
	vDSP_destroy_fftsetup(FFTSetup);
	return 0;
}

Output

 N = 1024 Accel: 0.05265 FFTW: 0.026213
 N = 2048 Accel: 0.042565 FFTW: 0.049334
 N = 4096 Accel: 0.049206 FFTW: 0.082782
 N = 8192 Accel: 0.118918 FFTW: 0.173506
 N = 16384 Accel: 0.200201 FFTW: 0.371488
 N = 32768 Accel: 0.356826 FFTW: 1.01589
 N = 65536 Accel: 0.844955 FFTW: 1.79316
 N = 131072 Accel: 1.95759 FFTW: 3.99565
 N = 262144 Accel: 4.34179 FFTW: 9.87287
 N = 524288 Accel: 9.31858 FFTW: 19.6675

Specs = Mid-2009 MBP, Core 2 Duo, 2.66GHz