VASP Compilation with Intel oneAPI: Is Preparation of Intel Optimised BLAS95, LAPACK95 and FFTW Required?

[hszhao.cn@gmail.com]

Dear VASP Community,

Currently, I am undertaking the task of compiling VASP.6.4.2 with the Intel oneAPI toolkit. I've come across multiple instances in conventional instructions indicating the necessity to prepare and compile Intel Optimised BLAS95, LAPACK95, and FFTW libraries, inclusive of their respective Interface Wrapper Libraries, say, the tutorial here:

Step 3: Prepare the Intel Optimised BLAS95, LAPACK95 and FFTW.

Compiling Intel BLAS95 and LAPACK95 Interface Wrapper Library
Compiling Intel FFTW3 and FFTW2 Interface Wrapper Library

However, my understanding suggests that the Math Kernel Library (MKL) included in Intel's oneAPI encompasses optimized implementations of BLAS, LAPACK, and FFTW. The MKL offers Fortran 95 interfaces for BLAS and LAPACK and has FFTW interface wrappers for FFTW, enabling FFTW API calls to run MKL's FFT routines.

Considering the above, I would appreciate it if you can clarify whether it is still necessary to compile Intel Optimised BLAS95, LAPACK95, and FFTW libraries separately when utilizing Intel oneAPI toolkits inclusive of MKL. Can I proceed utilizing the MKL libraries directly? Would the necessity be contingent on whether VASP is adapted to Intel's oneAPI?

Your guidance and clarification on this matter would be highly beneficial as I aim to deploy the most efficient approach for this compilation process.

Best regards,
Zhao

[hszhao.cn@gmail.com]

See here for the related discussion.

[svijay]

Dear Zhao,

Looking through the documentation for the FFTW interface of the intel oneAPI toolkit, it does appear that the wrappers are integrated into the toolkit (I am looking at the appendix titled oneAPI Math Kernel Library (oneMKL) Developer Reference from the manual here: https://d1hdbi2t0py8f.cloudfront.net/mk ... index.html). From their documentation

The FFTW3 wrappers are integrated in Intel® oneAPI Math Kernel Library (oneMKL). The only change required to use Intel® oneAPI Math Kernel Library (oneMKL) through the FFTW3 wrappers is to link your application using FFTW3 against Intel® oneAPI Math Kernel Library (oneMKL).

which I believe it taken care of with the appropriate intel-specific makefile.include in the arch folder. In case you are unable to compile with this configuration (i.e. directly using oneAPI), please provide the makefile.include that you use in your compilation.

Sudarshan

[hszhao.cn@gmail.com]

Dear Sudarshan,

which I believe it taken care of with the appropriate intel-specific makefile.include in the arch folder. In case you are unable to compile with this configuration (i.e. directly using oneAPI), please provide the makefile.include that you use in your compilation.

I can successfully compile VASP based on the related makefile.include that comes with VASP, without needing to additionally compile these interface wrappers discussed here. My original intention for raising this question is: If we compile and use these interface wrapper libraries, does it mean that we can make VASP perform better?

Regards,
Zhao

[svijay]

Dear Zhao,

It appears that the necessary wrappers are already integrated into the oneAPI toolkit; there should now be no need for any other interface wrapper libraries if the link is correctly done in the VASP makefile.include.

Sudarshan

[hszhao.cn@gmail.com]

Thank you for your confirmation.

Regards,
Zhao