[Pw_forum] How to convince QE from local lapack/blas?

[Paolo Giannozzi]

On Jul 27, 2011, at 5:40 , Michael Sullivan wrote:

> If I'm not wrong, that just means that you're not using ESSL.

it is true that by linking lapack before essl, the lapack version
of lapack routines is linked, not the essl version. However
the fft and blas from essl are linked (unless there is a
copy of blas into lapack, which shouldn't be the case).
The resulting loss of performances should be minor.
If fft and blas from essl are not linked, the loss of
performances is much larger.

Final clarification:

-D__ESSL means that the fft routines of essl are used, and
that the calls of some lapack routines follow the essl syntax,
not the lapack syntax;

-D__LINUX_ESSL means that the fft routines of essl are used,
and that the calls of lapack routines follow the lapack syntax.

In all cases, it is convenient to link essl before lapack (essl does
not contain all lapack routines, just a subset, so not linking lapack
may result in missing symbols). If the lapack replacements in essl
do not follow the lapack syntax (IBM SP machines, typically),
-D__ESSL should be used. If they follow the lapack syntax, or if
they do not contain lapack replacements (IBM BlueGene and other
strange IBM machines), -D__LINUX_ESSL should be used.

Complicated? blame the geniuses who wrote lapack replacements
with the same names but called in a different way

P.
---
Paolo Giannozzi, Dept of Chemistry&Physics&Environment,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222