[gmx-developers] CVS code seg fault on x86_64 system

[David Bostick]

Dear all,

I have written a few previous emails about problems compiling on an x86_64 
system, and am writing a follow-up email. As I mentioned previously, I was 
able to finally compile. However,  mdrun compiled from the latest CVS code 
(checked out yesterday) breaks whether or not I run in parallel. When running 
a single-processor mdrun executable, the result is a segmentation fault.

As a test, I compiled version 3.2.1 on the same machine(s) using exactly 
the same lam-mpi and fftw. This version of gromacs works. Thus the problem 
must be in mdrun, itself.

David
  


-- 
===================================|
David Bostick			   |
The Scripps Research Institute     |
Dept. of Molecular Biology, TPC 6  |
10550 N. Torrey Pines Rd. 	   |
La Jolla, CA 92037		   |	
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=|
Phone: 858-784-9614		   |
Fax:   858-784-8688		   |
Email: dbostick at scripps.edu	   |
===================================|

---------- Forwarded message ----------
Date: Fri, 24 Jun 2005 14:14:53 -0700 (PDT)
From: David Bostick <dbostick at vangogh.scripps.edu>
To: Discussion list for GROMACS development <gmx-developers at gromacs.org>
Subject: CVS compile error


Hello all,

I again have a question pertaining to the previous emails on a CVS 
gmx code compile error where I get the following message:

nb_kernel410_x86_64_sse.s: Assembler messages:
nb_kernel410_x86_64_sse.s:440: Error: invalid operand for 'mulps' ('z' unexpected)
nb_kernel410_x86_64_sse.s:1579: Error: invalid operand for 'mulps' ('z' unexpected)

The machine I'm using is an EMT 64 machine (a 36 bit machine). Each node 
in the cluster has an Intel Xeon 3.4 GHz processor. The cluster is running 
SUSE Linux. The gcc version is 3.3.5, and the assembler info is as 
follows:
GNU assembler version 2.15.94.0.2.2 (x86_64-suse-linux) using BFD version 
2.15.94.0.2.2 20041220 (SuSE Linux)

My solution was to comment out the 'z' on lines 440 and 1579 in 
nb_kernel410_x86_64_sse.s with a ;#. This allowed the code to compile. 
However, mdrun_mpi exits without any helpful message.

Should "commenting" out the 'z' in this file nb_kernel410_x86_64_sse.s 
cause the code to break? If so, then I can pursue other routes to solving 
this problem.

As a sidenote, I have access to a nearly identical (slightly 
different) cluster of Intel Xeon 3.2 GHZ processors. This cluster runs 
gcc 3.3.4 and an assembler with the following info:
GNU assembler version 2.15.91.0.2 (x86_64-suse-linux) using BFD version 
2.15.91.0.2 20040727 (SuSE Linux)

On this machine everything compiles fine with no error messages.

David

-- 
===================================|
David Bostick			   |
The Scripps Research Institute     |
Dept. of Molecular Biology, TPC 6  |
10550 N. Torrey Pines Rd. 	   |
La Jolla, CA 92037		   |	
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=|
Phone: 858-784-9614		   |
Fax:   858-784-8688		   |
Email: dbostick at scripps.edu	   |
===================================|