This version is outdated by a newer approved version.This version (2012/09/04 13:10) is a draft.
Approvals: 0/1
Approvals: 0/1
This is an old revision of the document!
BLAS libraries
There are several implementations of the Basic Linear Algebra Subprograms (BLAS) libraries available. These provide highly optimized routines for matrix and vector operations and are a key to high performance applications.
We recommend to use one of the fastest available libraries:
- GOTO BLAS by Kazushige Goto (http://www.tacc.utexas.edu/tacc-projects/gotoblas2) available in the servers and compute nodes by
-L/opt/goto/ifort -lgoto2_barcelonap-r1.13
You may want to set THREADS=1
.
- Intel Math Kernel Library (MKL): to use e.g. with
-L/opt/intel/composerxe/mkl/lib/intel64/ -lmkl_intel_lp64 -lmkl_sequential -lmkl_core
or – with the Intel compiler suite – by simply using
-mkl
- AMD Core Math Library (ACML): to use e.g. with
-L/opt/acml5.1.0/ifort64/lib/ -lacml
The GOTO and MKL libraries exist in single and multi threaded versions.
- To use GOTO multi threaded version use
-L/opt/goto/ifort -lgoto2_barcelonap-r1.13 -lpthread
- To use the MKL multi threaded version use
-lmkl_threads
instead of-lmkl_sequential
The reference BLAS is installed on some nodes (-lblas
) but significantly slower. We recommend not to use it, even if available.
If shared libraries are used, setting the variable LD_LIBRARY_PATH
is required, e.g.
export LD_LIBRARY_PATH=/opt/goto/ifort:$LD_LIBRARY_PATH