Developer Guide

Developer Guide for Intel® oneAPI Math Kernel Library Linux*

ID 766690
Date 7/13/2023
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

Using a Custom Message-Passing Interface

While different message-passing interface (MPI) libraries are compatible at the application programming interface (API) level, they are often incompatible at the application binary interface (ABI) level. Therefore, Intel® oneAPI Math Kernel Library (oneMKL) provides a set of prebuilt BLACS libraries that support certain MPI libraries, but this, however, does not enable use of Intel® oneAPI Math Kernel Library (oneMKL) with other MPI libraries. To fill this gap, Intel® oneAPI Math Kernel Library (oneMKL) also includes the MKL MPI wrapper, which provides an MPI-independent ABI to Intel® oneAPI Math Kernel Library (oneMKL). The adaptor is provided as source code. To use Intel® oneAPI Math Kernel Library (oneMKL) with an MPI library that is not supported by default, you can use the adapter to build custom static or dynamic BLACS libraries and use them similarly to the prebuilt libraries.

Building a Custom BLACS Library

The MKL MPI wrapper is located in the <mkl directory>/interfaces/mklmpi directory.

To build a custom BLACS library, from the above directory run the make command.

For example: the command

make libintel64

builds a static custom BLACS library libmkl_blacs_custom_lp64.a using the MPI compiler from the current shell environment. Look into the <mkl directory>/interfaces/mklmpi/makefile for targets and variables that define how to build the custom library. In particular, you can specify the compiler through the MPICC variable.

For more control over the building process, refer to the documentation available through the command

make help

Using a Custom BLACS Library

Use custom BLACS libraries exactly the same way as you use the prebuilt BLACS libraries, but pass the custom library to the linker. For example, instead of passing the libmkl_blacs_intelmpi_lp64.a library, pass libmkl_blacs_custom_lp64.a.