Development Environment --HELP!

[gwayne]

I am currently tasked with parallelizing (where possible) parts of the ISIS codebase and apps we use at LPI. The target platform is an older generation Scyld beowulf. Right now I am developing on RH9, but have apparently run into problems with glibc-2.3.2. I have moved my development environment onto the beowulf but continue to experience similar problems and new ones resulting from older tools.

I intend to use both OpenMP and MPI and therefore need a compiler that supports OpenMP, like Intel's. I have read through the build scripts and setup my environment using them, but I must be missing something. The majority of problems seem to be missing ENV VARS. I'm not completely new to UNIX/Linux development, but I've hit a few stumpers. I've tested both with gcc and icc with the same results -- mostly undefined references. Even the build/Firstbuild scripts fail to build all the libraries.

Anyone else in a similar situation? Any tips on platform, environment or other tools? Compilers?

Thanks,
Glenn

[gwayne]

Interesting thing about icc, when you have a -I with a blank path, it produces undefined references errors when it would otherwise compile. I was able to clear up that problem, but now when trying to link apps against ISIS libs, I get undefined refs. When I link all the necessary libs, I get even more undefined refs from them.

Any clues would be helpful.

Glenn

[janderson]

Hi Glenn,

We are planning a new release of Isis 2.1.3 around June 15th. In this release we have modified source code and the make system so that Isis will build on the newer version of the Linux kernel and compilers. In particular:

Kernel.................2.4.21-99
gcc-g77-g++........3.3.1-24
glibc....................2.3.2-88
ncurses................5.3-110
perl.....................5.8.1-46

Hopefully these changes will make your project a bit easier. We did experience undefined reference problems when we tried to use g++ for compiling and ld for linking and building shared libraries. We discovered we needed to use g++ for all three tasks: compiling, linking, and building shared libraries.

Unfortunately I don't have any experience parallelizing code, but have plenty of exp paralyzing code , so I can't weigh in on this one. Good luck and hopefully somebody else has some valuable comments/experience.

Jeff Anderson.

[gwayne]

I have done quite a bit of rework on the Makefiles and build scripts too. In particular, I created an autoconf/automake script, but it's still in the works. I look forward to comparing your changes.

The results of the parallelization will benefit anyone who's interested, but right now the D.E. is giving me fits, so I have to get that fixed first.


Thanks,
Glenn

[karlmitchell]

Hi Glenn,

We have also been looking into parallelising parts of ISIS using MPI, but haven't actually had the time to make a start yet. We have access to both ix86-linux and sparc-solaris clusters, and so if there is anything we can do to help regarding testing the system, please let me know.

Cheers,

-k

Dr Karl Mitchell
Lancaster University, UK

[gwayne]

Once I get the build environment established with the forthcoming release, I'll begin my parallelization efforts. We have a small x86 beowulf and Sparc cluster available also.

Have you identified any candidates for parallelization? I am not an ISIS expert, nor do I have extensive experience with its code. I plan to do some profiling and code coverage after talking to the heavy users to see where improvements are needed most.

Glenn

[ansley]

I have done a bit of work on the Makefiles and build scripts. In particular, I created an autoconf/automake script, but it's still in the works. I look forward to comparing your changes.