| 1 |
ou.wang |
1.1 |
> Obtaining optimized BLAS routines for HPC platforms |
| 2 |
|
|
> (e.g. O3K, Altix, Opteron, ...) |
| 3 |
|
|
> heimbach@mit.edu, 9-Sep-2004 |
| 4 |
|
|
> |
| 5 |
|
|
> 1. |
| 6 |
|
|
> Go to |
| 7 |
|
|
> http://www.cs.utexas.edu/users/kgoto/ |
| 8 |
|
|
> which has a set of high performance BLAS libraries for |
| 9 |
|
|
> all kinds of platforms. |
| 10 |
|
|
> |
| 11 |
|
|
> 2. |
| 12 |
|
|
> Download (e.g. for Opteron) |
| 13 |
|
|
> http://www.cs.utexas.edu/users/kgoto/libraries/libgoto_opt64p-r0.94-2.so.gz |
| 14 |
|
|
> which seems to be the suitable one for your Opteron 64 bit multi-threaded |
| 15 |
|
|
> and put them in a directory, e.g. ~/mylib/ |
| 16 |
|
|
> |
| 17 |
|
|
> 3. |
| 18 |
|
|
> As per website instruction, |
| 19 |
|
|
> also need the file xerbla.f |
| 20 |
|
|
> http://www.cs.utexas.edu/users/kgoto/libraries/xerbla.f |
| 21 |
|
|
> Download it into the MITgcm/optim/ directory as |
| 22 |
|
|
> xerbla.F (note capital F) |
| 23 |
|
|
> (maybe cleaner to generate an object file and keep it in ~/mylib/ ) |
| 24 |
|
|
> |
| 25 |
|
|
> 4. |
| 26 |
|
|
> in .bashrc added line |
| 27 |
|
|
> export LD_LIBRARY_PATH=~/mylib |
| 28 |
|
|
> (consistent with the directory where I keep the library) |
| 29 |
|
|
> |
| 30 |
|
|
> 5. |
| 31 |
|
|
> in MITgcm/optim/ |
| 32 |
|
|
> edit the Makefile as follows |
| 33 |
|
|
> * add xerbla.F to $SRC |
| 34 |
|
|
> * add -L~/mylib/ $LIBDIRS |
| 35 |
|
|
> * add -lgoto_opt64-r0.94-2 $LIBS |
| 36 |
|
|
> and ignore |
| 37 |
|
|
> -lblas1 |
| 38 |
|
|
> |
| 39 |
|
|
> That works. |
| 40 |
|
|
|
| 41 |
|
|
Some BLAS libraries provided in lsopt/ are |
| 42 |
|
|
SGI O3K: libblas1.a_IRIX64.gz |
| 43 |
|
|
SUN: libblas1.a_SUN.gz |
| 44 |
|
|
SGI Altix: libgoto_it2-r0.95.so.gz |
| 45 |
|
|
PC Pentium4: libgoto_p4_512-r0.94.so.gz |