Intel® Math Kernel Library

MKL performance w/regard to transform length



The description of MKL perforce states:

"Fast Fourier Transforms
Intel MKL FFTs include many optimizations and should provide significant performance gains over other libraries ****** for medium and large transform sizes. ******" (emphasis mine).

What is the expected value of "medium"? I am comparing performance of MKL FFT against FFTPACK 5.1 and am seeing a break-even value that *I* think is quite large; way bigger than I expected.

Custom DLL


i just downloaded the composer trial and now i am struggling to built

the mkl dlls, the provided visual studio project files don't seem to work,

is there a step by step tutorial how to built the dlls - the hints i found so far

on the intel pages like removing and reincluding files to the projects didn't work,

thank you

inconsistency between mkl and matmul

We have compared the results for matrix-vector recurrence relations of the type v_i = A*v_(i-1), as calculated by matmul and mkl routines.

After some iterations the calculations seem to diverge exponentially in most of the cases, but the outcome is machine dependent.

We have make comparisons for dgemv, dgemm, zgemm, and dzgemv. The problem occurs more frequently for the transpose multiplication case.

The tests were made on the following code.

MKL threads don't seem to exit properly.


We have been chasing this problem forever and hope to get some direction for this problem. We use Intel MKL for one of our internal FORTRAN programs and recently started to use MKL-Parallel option (/Qmkl:parallel) and have been experiencing problems:

Sometimes, the program seems to crash / hangs executing the math functions and sometimes it doesn't come back as if it is silently crashing. 

- How can we gracefully capture the fatal errors from the calls to MKL Libraries?

- How do we handle the cases where the threads do not exit? 

best way to compute products of the form A*S*A' where S is symmetric?

What's the best way to efficiently compute a product of the form A*S*A' where S is symmetric with MKL?

Does MKL provide anything similar to the sykrx function that's part of cublas ( so that you can use it along with symm to compute the product without having to recompute symmetric entries?

Problem of multithreading mkl_?csrcsc

Hey Guys

When I use mkl_?csrcsc to transpose my sparse matrix, I think the multithreading is supported internally: just like multithreading BLAS function cblas_dgemm(), call omp_set_num_threads() to set the number of threads before cblas_dgemm(). Unfortunately, no matter how many threads I set in omp_set_num_threads(), the performance of mkl_?csrcsc looks consistent.  I'm wondering how I can enable the multithreading for mkl_?csrcsc() functions? 

I use Intel compiler 13.0.1 on CentOS release 6.3, and my CPU is Intel(R) Xeon(R) CPU E5-2670.

Fortran FFT with MKL

I'm currently messing with some code that uses an old F77 looking FFT subroutine full of goto's and other nasty stuff. I have been looking around at using MKL's FFT routines, but I am having trouble finding some examples of how they are used. My current FFT routine uses only double precision data, so I don't need a FFT dealing with complex. Can anyone point me to any example code using a 2D MKL FFT routine? (possibly within /opt/intel/mkl/examples)

Thanks in advance!

(For any interested, here is the current FFT routine being used)

Iscriversi a Intel® Math Kernel Library