MKL 2D fft's vs. FFTW?

MKL 2D fft's vs. FFTW?

The MKL 2D fft's don't seem that fast for non-powers of 2. They aren'tfaster than fftw 3.0 . The benchmark results displayed on the MKL web page are for the itanium processor. Does anyone have experience with the 32 bit or EM64T processors?

Also, is there a way of determining what the preferred non-power of 2 lengths for the MKL ffts are?

Doing non-power of 2 lengths are crucial for ffts. If I have a fft that is 20% slower, but can do length 140, or 145, that is much better than having to go up to length 256.

2 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

I am not certain which version of MKL you are using. In MKL 8.0 we added radix 7 and 11 kernels. We have had radix 2, 3, and 5. One of your transforms - 140 - should profit from the radix 7 transform. The second transform - 145 - should be zero-filled to size 150.

Bruce

Leave a Comment

Please sign in to add a comment. Not a member? Join today