Newsgroups: comp.parallel.pvm
From: orenl@mangal.cs.huji.ac.il (Oren Laden)
Subject: Re: using pvm_pking for float
Organization: The Hebrew U. of Jerusalem, Computer Science Dept.
Date: 21 Mar 1995 16:41:09 GMT

In article <3k604j$6n8@magus.cs.utah.edu>, hsong@asylum.cs.utah.edu (Hwal Song) writes:
|> would this work?
|>
|>   float a10, b;
|>
|>   ........
|>   pvm_pkint(a, 10*sizeof(float)/sizeof(int), 1)
|>   pvm_pkint(&b, sizeof(float)/sizeof(int), 1);
|>
|>
|> do you any difficulties in unpacking?
|> what would cause any confilicts with implementational definition?

Of course I do !

One of the main advantages of PVM is heterogeneousity support: it does all
the hard work of translating from one data representation of machine A to
another data representation of machine B (e.g. little endian to big endian).
This means that what you did is like sending a raw stream of bytes, (as in
   pvm_pkchar(a, 10*sizeof(float)/sizeof(char), 1) ...)

The bottom line is that you can use this method only when communicating
between two machines with the same data representation (alternatively, you
can use the PvmDataRaw and save some convertion time ...).

However, the main issue here is not whether or not one can use this style,
but completely different:  Why should the user do something like this :-) ?
As I said, there are better means for doing it.

Oren.

