Message-ID: <354F8705.CF8897F4@lyapunov.ucsd.edu>
Date: Tue, 05 May 1998 14:39:17 -0700
From: Matt Kennel <kennel AT lyapunov DOT ucsd DOT edu>
Organization: Institute for Nonlinear Science, UCSD
MIME-Version: 1.0
To: beastium-list AT Desk DOT nl
Subject: Linking with GCC 2.7.2 compiled libraries???
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Sender: Marc Lehmann <pcg AT goof DOT com>
Resent-From: Marc Lehmann <pcg AT goof DOT com>
Resent-Date: Wed, 6 May 1998 00:34:20 +0200
Resent-To: Marc Lehmann <root AT cerebro DOT laendle>
Resent-Message-Id: <E0yWqI6-0000Iu-00 DOT pgcc-forever-1998-05-06-00-34-26-pgcc-forever AT cerebro>
Status: RO
Content-Length: 2078
Lines: 58

Hello, I am having some problems with 'pgcc' compiled programs crashing,
I believe
when linking to existing gcc compiled libraries.

My situation is unusual, let me explain.  I am using the NAG Fortran 90
compiler
on Linux.  It emits C code which is then compiled with the system C
compiler
(in my case, I made it pgcc based on egcs 1.0.2) and then linked with
various
NAG supplied system libraries, presumably compiled with standard GNU CC.

With 'egcs 1.0.2' I am not having any major problems but with pgcc, I
am.
The program crashes at various places, but the syndrome seems to be
that the NAG run-time libraries are not getting the proper arguments.

In other words, if possible, I need flags to make sure that the calling
convention for pgcc compiled prorams is as identical as possible to
that compiled with standard gcc, so I can link to the NAG binary
libraries.

No I do not have source code for them and I will not be able to obtain
it.

By the way, with gcc, adding the flag '-malign-double' does improve
performance
with gcc 2.7, and does not result in any run-time errors.

If this is not the problem, I seem to have found some wierd bugs in
pgcc,
but I'm not sure the C source would be that informative.


I am not on the mailing list here, please reply directly to
   kennel AT lyapunov DOT ucsd DOT edu

I really do need maximum run-time performance for floating point: each
run
of our simulation takes eight to twelve hours with modern hardware.  We
have
investigated Linux/Alpha but found that it was NOT good for our problem,

specifically "sin/cos/exp" are VERY poor on the Alpha, resulting in an
overall
speed no faster than a Pentium II 300 on our programs.

Mucho thanks in advance,
matt

--
*        Matthew B. Kennel/Institute for Nonlinear Science, UCSD           -
* "People who send spam to Emperor Cartagia... vanish! _They say_ that
* there's a room where he has their heads, lined up in a row on a desk...
* _They say_ that late at night, he goes there, and talks to them... _they
*- say_ he asks them, 'Now tell me again, how _do_ you make money fast?'"