Running Intel MKL on an Intel Xeon Phi Coprocessor in Native Mode

Some applications can benefit from running on Intel Xeon Phi coprocessors in native mode. In this mode, the application runs directly on a coprocessor and its Linux* operating system without being offloaded from a host system. To run on Intel MIC Architecture in the native mode, an application requires minimal changes to the source code.

Because in the native mode the code runs exclusively on a coprocessor, binaries built for native runs contain only the code to be run on a coprocessor. Intel compilers provide a specialized option to support building applications to be run in the native mode.

To build an application that calls Intel MKL and natively run it on a coprocessor, you need to perform these high-level steps:

1. On the host system, compile and build the application using the -mmic option.

2. Transfer the executable and all the dynamic libraries it requires to the coprocessor:

   • The Intel MKL libraries in the <mkl directory>/lib/intel64_lin_mic directory.
   • libiomp5.so in the <parent directory>/compiler/lib/intel64_lin_mic directory.

3. Use the Secure Shell (SSH) or Telnet protocol to execute on the coprocessor and add the paths to the dynamic libraries transferred to the coprocessor in step 2 and to the value of the LD_LIBRARY_PATH environment variable.

4. Set the number of threads and the thread affinity using your threading run-time library.

5. Execute just as you would on a standard Linux* system.

For more information, see Intel® Compiler User and Reference Guides, available in the Intel Software Documentation Library.