Using MKL on a heterogeneous cluster

Using MKL on a heterogeneous cluster


I intend using MKL70 on a heterogeneous mosix (linux) cluster (PIII, PIV and Xeons).

Currently I compile the fortan source code using a specific MKL library (eg mkl_p4) and use mosrun to run the compiled binaries on a specific node which has the appropriate type processor.

Now the questions:
1. I would like to know if there is any way of linking/compiling the fortran source code in such a way that the correct MKL routines will be automatically used, depending on which processor the binary is running. (without having to compile several binaries).

2. If the code is running on a specific node (say PIV) and MKL has chosen the appropriate routines for that processor, is it possible to migrate the running process to a different node (e.g PIII), without the process crashing? In other words, will MKL be able to change from one processor to another while it is ALREADY RUNNING?

Thank you


1 post / 0 new
For more complete information about compiler optimizations, see our Optimization Notice.