Using ScaLAPACK and Cluster FFT on Intel Xeon Phi Coprocessors

Intel MKL ScaLAPACK and Cluster FFT support only Intel MPI Library on Intel Xeon Phi coprocessors.

The Intel MPI library can treat each Intel Xeon Phi coprocessor as a regular node in a cluster of processors based on Intel 64 architecture and Intel Xeon Phi coprocessors and enables a straightforward way to run an MPI application on clusters that contain both processors and coprocessors as compute nodes.

The documentation for the Intel MPI library recommends the following steps to run an MPI application on the specific Intel Xeon Phi coprocessor and the host node if the nodes are properly specified on the cluster and the network protocols and environment are properly set up:

1. Build the application for the Intel 64 architecture.

2. Build the application for the Intel MIC Architecture.

3. Launch the application from the host computer.

For more details, check the Intel MPI Library documentation, available in the Intel Software Documentation Library.

To run a dynamically linked application natively, perform the Scripts to Set Environment Variables Setting Environment Variables step of the Getting Started process. In addition to other environment variables, it sets:

• LD_LIBRARY_PATH to contain <mkl directory>/lib/intel64_lin

• MIC_LD_LIBRARY_PATH to contain <mkl directory>/lib/intel64_lin_mic

When building your application that uses Intel MKL ScaLAPACK or Cluster FFT, follow the linking guidelines in the Linking with Intel MKL Cluster Software, but be aware of Intel MKL specifics on Intel Xeon Phi coprocessors:

• Only Intel compiler and Intel MPI are supported
• Only OpenMP threading layer for Intel compilers and Intel® Threading Building Blocks (Intel® TBB) threading layer are provided

You can find a full list of Intel MKL libraries for Intel MIC architecture in Detailed Directory Structure of the lib/intel64_win_intel64_lin_mic Directory. Be aware that coprocessors run a Unix* operating system.