Project

General

Profile

Bug #583

mdrun crash on old AMD Processors.

Added by Kyle Beauchamp about 9 years ago. Updated about 9 years ago.

Status:
Closed
Priority:
Normal
Assignee:
Erik Lindahl
Category:
mdrun
Target version:
Affected version - extra info:
Affected version:
Difficulty:
uncategorized
Close

Description

The crash occurs immediately when running mdrun.exe. No files are created.

If we have any ideas for further tests, I have a user who is able to test things on a windows XP laptop with an AthlonXP processor.

Thus, if we have things that you need tested for such a machine, let me know and I can arrange it.

History

#1 Updated by Roland Schulz about 9 years ago

Do you get any error message or any other output?
Is this with cygwin?
Are you able to produce a stack trace or core dump?

#2 Updated by Kyle Beauchamp about 9 years ago

My build is with visual studio 2008+ICC+MKL.

So I don't currently have one. This user is a contributer on folding@home who has offered to help me debug this issue. He already installed windbg, but I still haven't walked him through the process of loading symbols to get a stack trace.

I'm away at a conference for the next 2 days, so I won't get a chance to work on this until then.

I would assume this has something to do with SSE2 being missing, but I'm actually surprised that mdrun dies at startup instead of during the MD phase.

#3 Updated by Erik Lindahl about 9 years ago

Hi Kyle,

I'm starting to worry that this might not be as simple as an easy SSE2-ability-check. If it crashes instantly it could be the compiler doing some (stupid?) initialization things, and if it happens before the actual program starts executing we will never be able to catch it.

1. Is this only visible with Generalized Born projects, or normal runs too?
2. What happens if you try to quit/return directly in do_md(), without actually calling any routines? Do we still get a crash? (then we haven't called any of our potential SSE2-code).

#4 Updated by Kyle Beauchamp about 9 years ago

Sorry if I was unclear, as I'm interpreting someone else's description.

It really seems to be instant--right now, it appears to be starting the actual mdrun executable. Like, before any sort of file input or output occurs. And again, this is only one machine.

I recently found an old AthlonXP in our lab, but the machine doesn't post so I'm not sure if I'll be able to get things running on that.

Again, I've got a workaround for this right now, which is to avoid using machines that are pre-SSE2.

When I get back from this conference, I'll try to find a way to directly verify and investigate this.

Thanks again

#5 Updated by Szilárd Páll about 9 years ago

@Kyle: Have you managed to get more details about the issue?

#6 Updated by Kyle Beauchamp about 9 years ago

Hi,

I have no more details, as we chose to require SSE2 before letting people use 4.5.1.

We recently acquired a machine with an older AMD processor (actually, we acquired about 15 different machines with a variety of hardware configurations). Once that's up, I may be able to test things out.

PS you won't offend me if you call this one closed.

#7 Updated by Szilárd Páll about 9 years ago

(In reply to comment #6)

Hi,

I have no more details, as we chose to require SSE2 before letting people use
4.5.1.

We recently acquired a machine with an older AMD processor (actually, we
acquired about 15 different machines with a variety of hardware
configurations). Once that's up, I may be able to test things out.

PS you won't offend me if you call this one closed.

Hi Kyle,

Thanks for the update!

I'll change the status to "LATER". Please give us an update if/when you manage to confirm the bug.

Sz.

Also available in: Atom PDF