Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.

FFT Kernels

Lui_S_
Beginner
160 Views

Hi everybody,

quick question:

https://software.intel.com/en-us/articles/fft-length-and-layout-advisor

states: "The Intel® MKL FFT implementation employs radix 2, 3, 4, 5, 7, 8, 11, 13, 16 kernels, and several other larger size kernels."

https://software.intel.com/en-us/articles/mkl-ipp-choosing-an-fft

states: "2,3,5,7,11 and 13 kernels (**) ** - Both libraries support arbitrary radix in optimized manner, that is O(N log N), but these specific radixes are better optimized than others. 

So are only the radixes 2,3,5,7,11 and 13 optimized or 2, 3, 4, 5, 7, 8, 11, 13, 16 and several other? I do not find the answer to that anywhere. And before I start optimizing and get a suboptimal result (because I didn't know better) - I thought I'd ask.

Thanks!

Lui

0 Kudos
4 Replies
Gennady_F_Intel
Moderator
160 Views

FFT with radixes 2, 4, 8, 16 ..... (power of 2) are most well optimized for both of  MKL and IPP libs ..

Lui_S_
Beginner
160 Views

Thanks! This is really helpful!

Is IPP generally faster than MKL on the same "task"? So if I can use IPP - should I use IPP versus MKL? I am talking about FFTs. 

Have a nice weekend!

Gennady_F_Intel
Moderator
160 Views

we expected the same performance for IPP and MKL.

Lui_S_
Beginner
160 Views

Thanks a lot! This was really helpful!

Reply