Newsgroups: comp.parallel.pvm
From: Dima Pasechnik <dima@duti515a.twi.tudelft.nl>
Subject: PVM on  32-bit *and* 64-bit machines - double/float question
Organization: Delft University of Technology
Date: 04 Aug 1998 19:21:05 +0200
Message-ID: <t3r9ywoe7y.fsf@duti515a.twi.tudelft.nl>

We are running PVM on a network that has 
Linux (Pentium II, 32-bit)            PVM 3.3.11 (public release)
HPPA (HP9000, 32-bit)                 PVM 3.3.11 (public release)
Unicos (CRAY J90, 64-bit)             PVM 3.3.10 (CRAY release)
hosts.

There appear to be a kind of incompatibility with the floating point
data within PVM.  On the 1st two types of machines, PVM_DOUBLE (and C
double) is 8 byte, while on CRAY it's 16 byte.  There is no need to
use the CRAY double, it would suffice to use CRAY float for our
problems.

However, PVM does not seem to be able to permit data that is
packed via pvm_pkdouble from the 32-bit machine to be (properly)
unpacked on 64-bit machine as (8-byte) float. We have to use CRAY
double instead, that is not really what we want 
(as well as much slower and memory-hungry).

Is there any way around this?
(surely I'm not the first to come across such a situation.)
All the pointers etc are most appreciated.

Dmitrii.

Dmitrii V. Pasechnik	e-mail: D.Pasechnik@twi.tudelft.nl
SSOR, TWI		office location:		
TU Delft			HB 05.150
Postbus 5031			Mekelweg 4
2600 GA Delft			2628 CD Delft
The Netherlands			phone: +31-(0)15-2782520
                                fax:   +31-(0)15-2787255
				http://ssor.twi.tudelft.nl/~dima


