Project

General

Profile

Bug #1703

be less vocal about affinity where it is not relevant

Added by Mark Abraham over 4 years ago. Updated over 4 years ago.

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

Description

On (at least) BG/Q and (I think) PowerPC generally, mdrun does issue various warnings about affinity setting that are useless because threads cannot migrate. We should be more sensitive about that. Am I right about PowerPC? x86 and ARM can migrate.

Associated revisions

Revision 8a5f25a9 (diff)
Added by Erik Lindahl over 4 years ago

Avoid warning about affinity on Blue Gene, reduce warnings

- Blue Gene does not migrate threads, so don't warn about
affinity setting not being available.
- To avoid flooding users with warnings, just use warnings
when it is very likely the user did something incorrect
when setting pinning options, but notes when it might be
a property of the platform, or when it is not severe.

Fixes #1703.

Change-Id: Idaa781959e0997f6737593a8fd9243d6a979a493

History

#1 Updated by Szilárd Páll over 4 years ago

I don't know of any other architecture that we support/use where threads can't migrate. Does the pthreads API provide any means for checking this? Otherwise, it would need to be a (single) platform-specific hack which is not too nice.

#2 Updated by Erik Lindahl over 4 years ago

Another alternative is to just warn on platforms where we know it can happen.

I think it is starting to become a problem that a non-expert trying to compile Gromacs can be faced with lots of warnings both when configuring, preparing, and using Gromacs. While the intention of improving performance is admirable, a novice user will rather come away with the impression that Gromacs is difficult to use and that you need to read up on lots of things - and then they are still likely to ignore most warnings since they don't know how to fix them.

In particular, we should limit warnings to things where the scientific integrity of the output is at risk. Mere performance considerations should be a note (just as we do when telling the user we could use a better SIMD instruction set on this hist).

#3 Updated by Gerrit Code Review Bot over 4 years ago

Gerrit received a related patchset '1' for Issue #1703.
Uploader: Erik Lindahl ()
Change-Id: Idaa781959e0997f6737593a8fd9243d6a979a493
Gerrit URL: https://gerrit.gromacs.org/4711

#4 Updated by Erik Lindahl over 4 years ago

  • Status changed from New to Fix uploaded

#5 Updated by Erik Lindahl over 4 years ago

  • Status changed from Fix uploaded to Resolved

#6 Updated by Erik Lindahl over 4 years ago

  • Status changed from Resolved to Closed

Also available in: Atom PDF