Check/fix possible race conditions

[Martyn Gigg]

We have been getting random unit tests failures recently that would indicate we have some race conditions somewhere in the code.

Tools such as helgrind, drd can help with this.

[Martyn Gigg]

The attached file shows the output after running the convert units test through helgrind. It inidcates a number of problems most notably around the parameter map caching.

[Martyn Gigg]

Refs #4133. Fixed some race conditions in various places.

The valgrind tool helgrind flagged up some race conditions in the code. This is not a complete list of those that need fixing but some major ones that would affect large portions of the code

Changeset: b04a3eaa36407e9caabf0ea45ae1e76a3eb1b44a

[Russell Taylor]

Not surprisingly, this has had an impact on the performance tests. Looking at the RHEL6 tests run at ORNL, the following tests show noticeable slowdowns:

• CreateGroupingWorkspaceTestPerformance.test_TOPAZ_2010
• DiffractionFocussing2TestPerformance.test_SNAP_event_one_group_dontPreserveEvents
• DiffractionFocussing2TestPerformance.test_SNAP_event_six_groups
• DiffractionFocussing2TestPerformance.test_SNAP_event_six_groups_dontPreserveEvents
• Q1D2TestPerformance.test_slow_performance
• InstrumentRayTracerTestPerformance.test_TOPAZ

It's not all doom and gloom though. The following tests got faster at this point!

• LoadEventNexusTestPerformance.testDefaultLoad
• LoadEventPreNexusTestPerformance.testDefaultLoad
• MaskDetectorsInShapeTestPerformance.testMaskingLotsOfDetectors
• NearestNeighboursTestPerformance.testUsingNumberOfNeighbours

[Martyn Gigg]

Refs #4133. Swap assign for a fill to avoid race condition warning.

Changeset: a99e0c533e418b8c0d4c738a3eb0aa523c84dab7

[Martyn Gigg]

I'm guessing the ones that have slowed down all do a lot of getPos calls which means they now have to go through the critical block to get in to the cache. I suppose one way of avoiding that would be switching to storing the absolute positions rather relative positions. That way it wouldn't have to go into the cache. Although if you moved something it would still have to go into the parameter map so I don't really have a feel for how much gain we would get from that

[Martyn Gigg]

I ran the WISH script through helgrind last night and it didn't flag anything up either.

[Martyn Gigg]

Refs #4133. More race conditions found.

The HFIR system test randomly failed on RHEL last night. Helgrind flagged up a problem with the ParameterMap access again. This will probably affect the performance but correctness is of higher importance.

Changeset: 9775b281b348f086b4d5b47a01ba9bbc782a330a

[Peter Peterson]

Looks good to me.

[Stuart Campbell]

This ticket has been transferred to github issue ​4980