|
|
|||||||||
LAPACK & BLASLAPACK and BLAS routines were ftped from http://www.netlib.org, or http://www.att.com. We unpack them into a directory and modify the make.inc file to add an extension to the library names (ULT, SOL) and also look at the flags for compilations. We did not have to change any flags for the compilations. We need to read the release notes to verify any special considerations for the architecture and the os version we are trying to compile on. If the documentation says that the LAPACK version has been verified on the platform, you may want to comment out the testing lines in the Makefile. These tests take a long time and do not give us much information at the end.cd INSTALL rm *.o rm test* cd .. make all cp lapack.a /local/lib/liblapack.a cp blas.a /local/lib/libblas.a cp tmglib.a /local/lib/libtmglib.aAt this point, run testslamch and testdlamch. The output of these programs should be saved in a logfile, and gives values for various machine-dependent parameters. For example, on neptune, the slamch parameters should be: Tests completed Epsilon = 5.96046E-08 Safe minimum = 1.17549E-38 Base = 2.00000 Precision = 1.19209E-07 Number of digits in mantissa = 24.0000 Rounding mode = 1.00000 Minimum exponent = -125.000 Underflow threshold = 1.17549E-38 Largest exponent = 128.000 Overflow threshold = 3.40282E+38 Reciprocal of safe minimum = 8.50706E+37These values can be hard-coded into slamch.f so that they don't have to be recalculated later. The dlamch.f file can be altered in a similar fashion. For Digital UNIX, a machine optimized blas and lapack libraries are made avialable as Digital Extended Math Library and we have a license for this through the CSLG program. You can install this software using setld and the library is known as libdxml.a and the shared library is known as libdxml.so. |
|||||||||