fftGPU

Execute FFT faster on a NVIDIA GPU using the library of Naga K. Govindaraju.

Description

This function implements vectorized FFTs on a GPU. The answer is the same as the fft and ifft matlab functions.

You need to install the GPUFFTW2.0 library of Naga K. Govindaraju http://gamma.cs.unc.edu/GPUFFTW/ . On Naga's wed site, you can see the system requirements. It works on linux and windows.

This function support real and complex FFTs.
 
I don't recommend this program for small fft. It really worth it for fft over 2^20 points. It does the fft in SINGLE PRECISION. The fft length must be a power of two.

For obtaining reasonably high performance, Naga K. Govindaraju recommend a PC with AGP8X/PCI-Express NVIDIA GeForce 6800 GT or faster GPU. The Video RAM will determine the maximum array length of fft that can done on the GPU. For example, on a card with 256 MB VRAM, the maximum-length array is 256/32 = 8 Million real values or 4M complex values.

• MATLAB release: MATLAB 7.2 (R2006a)
• Other requirements: You need to install the GPUFFTW2.0 library of Naga K. Govindaraju http://gamma.cs.unc.edu/GPUFFTW/

[shanu vadivu]

 Sir

    i am doing Ph.d in D.S.P ...
 i want Radix -4 Dft program in matlab . please send that one to my mail i.d [e-mail]

[Siyi Deng]

tried this:
d = randn(64,3);
gpuf = fftGPU(d,1);
matf = fft(d);

figure, plot(abs(gpuf));
figure, plot(abs(matf));

seems inconsistent with matlab fft.