testGPU_FFT.m

% This m-file compiles and test
% a MEX file making use of the GPUFFTW 
% source to do FFTs on the GPU in matlab

% It assumes a working GPUFFTW installation
% (http://gamma.cs.unc.edu/GPUFFTW/)
% in the same folder than this testGPU_FFT.m 
% The gpu_FFT.cpp and fftGPU.m file should also be in this folder.

% Mods required to built the C++ example3:

% LINUX
%   1)
%   in /usr/lib64,   freeglut (libglut.so.3)
%   Was replaced by glut 3.7, see www.opengl.org

%   2)
%   in the /lib/Linux folder of the GPUFFTW 
%   installation we had to make a symlink to 
%   our /usr/lib64/libglut.so.3

% WINDOWS
%   1)  install glut 3.7g, lib named glut32.lib
%       there's a glut 64 bit for win64

% Our first system is Fedora Core 5, 64 bits
% NVIDIA linux driver installed.
% You may have to change the /usr/lib64/ path 
% depending upon your system.

% Our second system is Windows XP pro, 64 bits
% NVIDIA driver installed.
% You may have to change the glut32.lib path 
% depending upon your system.

% Read fftGPU.m for some restrictions for the fft 

% Known bugs: on Linux, GPUFFTW is unable to 
% retreive the VRAM size and assumes 256 MB.
% You can change the assuming size in GPUFFTW.cpp line (132)

% Simon Potvin and Jerome Genest, July 2006
% Centre d'optique, photonique et laser (COPL)
% Universite Laval
% Quebec, Canada

				


close all
clear all
clc

%%% LINUX MEX COMPILATION
lib =   '/usr/lib64/';

mex('-IGPUFFTW/include', '-IGPUFFTW/src', ... % Headers
    'gpu_FFT.cpp', 'GPUFFTW/src/GPUFFTW.cpp', 'GPUFFTW/src/arbfprog.cpp', 'GPUFFTW/src/stopwatch.cpp',...  % sources
    [lib 'libglut.so.3'], [lib 'libGLU.so'], [lib 'libGL.so'], [lib 'libXi.so.6'], [lib 'libXmu.so'])      % Shared libs


%%% WINDOWS MEX COMPILATION
% mex ('-IGPUFFTW/include', '-IGPUFFTW/src', ... % Headers
%     'gpu_FFT.cpp', 'GPUFFTW/src/GPUFFTW.cpp', 'GPUFFTW/src/arbfprog.cpp', 'GPUFFTW/src/stopwatch.cpp',... % sources
%     'C:\Program Files (x86)\Microsoft Visual Studio 8\VC\lib\amd64\glut32.lib') % Shared libs



N=5:20;

t_gpu_c = [];
t_fft_c = [];
t_gpu_r = [];
t_fft_r = [];

Nombre=2^22./2.^N; % assuming VRAM = 256 MB 
%Nombre=[1];

for i=1:length(N)
    N(i)

%%% COMPLEX TEST
    a=rand(2^N(i),Nombre(i));
    a=a+sqrt(-1).*rand(2^N(i),Nombre(i));
    
    tic
    d1=fftGPU(a,1);
    t_gpu_c = [t_gpu_c toc/Nombre(i)];

%%% REAL TEST
    b=rand(2^(N(i)+1),Nombre(i));
    
    tic
    d2=fftGPU(b,1);
    t_gpu_r = [t_gpu_r toc/Nombre(i)];

%%% MATLAB
    tic
    c1=fft(a);
    t_fft_c = [t_fft_c toc/Nombre(i)];


    tic
    c2=fft(b);
    t_fft_r = [t_fft_r toc/Nombre(i)];

end
  
%%% ERROR
    error_real=c2-d2;
    error_complex=c1-d1;
    
%%% PLOT SPEED
    plot(2.^N,t_gpu_c,'-*b',2.^N,t_fft_c,'-+g',2.^(N+1),t_gpu_r,'-or',2.^(N+1),t_fft_r,'-xm')
    figure
    plot(N,t_gpu_c,'-*b' ,N,t_fft_c,'-+g',(N+1),t_gpu_r,'-or',(N+1),t_fft_r,'-xm')