We recently encountered tremendous performance troubles when
parallelizing our FEM/BEM-code on a V2200 under HP-UX11 using the
#pragma _CNX directives for C. In some cases it works fine in other not.
Not working fine meens that there is good parallelization but with a
tremendous overhead. When we compare the sequential version of our code
to the parallel version running with just one processor we see
differences ranging from 1.05 (= works fine) up to 10 (= works very bad,
not usefull). Did anybody encounter similar problems?
Best regards from Vienna
_/_/ _/ _/ _/ _/ _/ | Numerical Analysis and Design GmbH & Co KG
_/ _/_/ _/_/_/_/ _/ _/ | Landskrongasse 5, A-1010 Wien, Austria/Europe
_/ _/ _/ _/ _/_/_/ | Tel.: ++43.1.5335306-0 FAX: ++43.1.5335306-22