Difference between revisions of "BLAS and LAPACK"
Albensoeder (talk | contribs) |
Albensoeder (talk | contribs) |
||
| (One intermediate revision by the same user not shown) | |||
| Line 1: | Line 1: | ||
The libraries BLAS ''Basic Linear Algebra Subprograms'' and LAPACK — ''Linear Algebra Package'' are not installed on FLOW and HERO as single modules. The | The libraries BLAS ''Basic Linear Algebra Subprograms'' and LAPACK — ''Linear Algebra Package'' are not installed on FLOW and HERO as single modules. The reason for this is the worse performance of self compiled version. However, both libraries in a highly optimized version are available within the Intel MKL (Math Kernel Library) and in the ACML (AMD Core Math Library). These optimized libraries are significantly faster than self compiled LAPACK and BLAS libraries. So it is worth to use these libraries which are optimized for the used Intel processors. | ||
| Line 6: | Line 6: | ||
ics/2013_sp1.3.174/64 | ics/2013_sp1.3.174/64 | ||
Inside the MKL all BLAS and LAPACK routines are included. The MKL can be linked independently if you are using the PGI, GNU compilers or the Intel compilers. | Inside the MKL all BLAS and LAPACK routines are included. The MKL can be linked independently if you are using the PGI, GNU compilers or the Intel compilers. | ||
| Line 15: | Line 15: | ||
pgi/13.10 | pgi/13.10 | ||
Inside the ACML all BLAS and LAPACK routines are included. Again, the ACML can be linked independently if you are using the PGI, GNU compilers or the Intel compilers. | Inside the ACML all BLAS and LAPACK routines are included. Again, the ACML can be linked independently if you are using the PGI, GNU compilers or the Intel compilers. | ||
In case of trouble when compiling/linking please contact your HPC support. | In case of trouble when compiling/linking please contact your HPC support. | ||
Latest revision as of 16:55, 6 February 2015
The libraries BLAS Basic Linear Algebra Subprograms and LAPACK — Linear Algebra Package are not installed on FLOW and HERO as single modules. The reason for this is the worse performance of self compiled version. However, both libraries in a highly optimized version are available within the Intel MKL (Math Kernel Library) and in the ACML (AMD Core Math Library). These optimized libraries are significantly faster than self compiled LAPACK and BLAS libraries. So it is worth to use these libraries which are optimized for the used Intel processors.
Intel MKL
The Intel MKL library is includes in the Intel Cluster Studio module
ics/2013_sp1.3.174/64
Inside the MKL all BLAS and LAPACK routines are included. The MKL can be linked independently if you are using the PGI, GNU compilers or the Intel compilers.
Unfortunately the linking of the MKL against your code is not so easy and depends on the release of the MKL. Luckyly there exist a nice configurator from Intel to get the right compiler and linker flags. In case of trouble when compiling/linking please contact your HPC support.
ACML
The ACML library is includes in the PGI Compiler Suite module
pgi/13.10
Inside the ACML all BLAS and LAPACK routines are included. Again, the ACML can be linked independently if you are using the PGI, GNU compilers or the Intel compilers. In case of trouble when compiling/linking please contact your HPC support.
