imkl-FFTW¶

Intel® oneAPI Math Kernel Library (oneMKL) offers FFTW2 and FFTW3 interfaces to Intel® oneAPI Math Kernel Library (oneMKL) Fast Fourier Transform and Trigonometric Transform functionality. The purpose of these interfaces is to enable applications using FFTW (www.fftw.org) to gain performance with Intel® oneAPI Math Kernel Library (oneMKL) without changing the program source code.

Both FFTW2 and FFTW3 interfaces are provided in open source as FFTW wrappers to Intel® oneAPI Math Kernel Library (oneMKL). For ease of use, FFTW3 interface is also integrated in Intel® oneAPI Math Kernel Library (oneMKL).

Usage¶

imkl-FFTW module¶

imkl-FFTW has to be loaded using Lmod prior to running it.

$ module load imkl-FFTW

You can also search the module with spider

$ module spider imkl-FFTW

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  imkl-FFTW:
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
    Description:
      FFTW interfaces using Intel oneAPI Math Kernel Library

     Versions:
        imkl-FFTW/2021.4.0-iimpi-2021b
        imkl-FFTW/2022.1.0-iimpi-2022a
        imkl-FFTW/2022.2.1-iimpi-2022b
        imkl-FFTW/2023.1.0-iimpi-2023a

------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
  For detailed information about a specific "imkl-FFTW" package (including how to load the modules) use the module's full name.
  Note that names that have a trailing (E) are extensions provided by other modules.
  For example:

     $ module spider imkl-FFTW/2023.1.0-iimpi-2023a
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Software version¶

Here you can check the available versions for imkl-FFTW in the different clusters

Atlas FDRAtlas EDRHyperion

imkl-FFTW/2021.4.0-iimpi-2021b
imkl-FFTW/2022.1.0-iimpi-2022a

imkl-FFTW/2021.4.0-iimpi-2021b
imkl-FFTW/2022.1.0-iimpi-2022a

imkl-FFTW/2021.4.0-iimpi-2021b
imkl-FFTW/2022.1.0-iimpi-2022a
imkl-FFTW/2022.2.1-iimpi-2022b
imkl-FFTW/2023.1.0-iimpi-2023a

How to use imkl-FFTW¶

In order to use it, you will need to compile your code against the mkl and FFTL libraries.

Let's see how to use it with an example, using the following code:

Example code using imkl-FFTW

#include <stdio.h>
#include <stdlib.h>
#include <mkl.h>
#include "mkl_dfti.h"

int main() {
    MKL_LONG N = 8;
    MKL_LONG status;
    DFTI_DESCRIPTOR_HANDLE handle;
    MKL_Complex16 *inout;

    // Allocate memory for the input/output array
    inout = (MKL_Complex16*)mkl_malloc(N * sizeof(MKL_Complex16), 64);
    if (inout == NULL) {
        printf("Error allocating memory\n");
        return 1;
    }

    // Initialize the input array with some values
    for (int i = 0; i < N; i++) {
        inout[i].real = i;
        inout[i].imag = 0.0;
    }

    // Create a descriptor for a double-precision complex 1D forward FFT
    status = DftiCreateDescriptor(&handle, DFTI_DOUBLE, DFTI_COMPLEX, 1, N);
    if (status != DFTI_NO_ERROR) {
        printf("Error creating descriptor: %ld\n", status);
        return 1;
    }

    // Commit the descriptor
    status = DftiCommitDescriptor(handle);
    if (status != DFTI_NO_ERROR) {
        printf("Error committing descriptor: %ld\n", status);
        return 1;
    }

    // Perform the FFT
    status = DftiComputeForward(handle, inout);
    if (status != DFTI_NO_ERROR) {
        printf("Error computing forward FFT: %ld\n", status);
        return 1;
    }

    // Print the results
    printf("Results of FFT:\n");
    for (int i = 0; i < N; i++) {
        printf("out[%d] = (%f, %f)\n", i, inout[i].real, inout[i].imag);
    }

    // Free the descriptor
    status = DftiFreeDescriptor(&handle);
    if (status != DFTI_NO_ERROR) {
        printf("Error freeing descriptor: %ld\n", status);
        return 1;
    }

    // Free the allocated memory
    mkl_free(inout);

    return 0;
}

First, we load the module:

$ module load imkl-FFTW/2023.1.0-iimpi-2023a

To compile our code, we will use the following line:

$ icc -qmkl main.c -o example -I/scicomp/builds/Rocky/8.7/IceLake/software/imkl/2023.1.0/mkl/2023.1.0/include

Note

The example code requires to an include file from mkl. The path to the folder containing it must be specified by the -I flag.

Once compiled, we can execute it using the following submission script:

Batch script for a parallel execution using FFTW

#!/bin/bash
#SBATCH --qos=regular
#SBATCH --job-name=IMKL_FFTW_JOB
#SBATCH --mem=20gb
#SBATCH --cpus-per-task=1
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=5
#SBATCH --output=%x-%j.out
#SBATCH --error=%x-%j.err

module load imkl-FFTW/2023.1.0-iimpi-2023a
mpirun -np 5 ./example

Where:

-np: number of MPI tasks.