#2426: g++-10 possibly miscompiles parts of Carpet
| Reporter: | Roland Haas |
| Status: | new |
| Milestone: | |
| Version: | |
| Type: | bug |
| Priority: | major |
| Component: | Other |
Comment (by Roland Haas):
From what I understand QueenBee is currently “QueenBee 2” but LSU kept both the hostname and the name used on documentation the same. I am told “QueenBee 3” has just been deployed (see eg. http://www.hpc.lsu.edu/resources/hpc/system.php?system=QueenBee vs http://www.hpc.lsu.edu/docs/guides.php?system=QB2).
The Intel compiler we use is very old 2013.1.046 (with 14.0.2 being the machine default) however there are much newer ones available (up to 2018.1) which we could easily use.
My approach would be to first update the machine definition files for queenbee (and other clusters in simfactory) to use the newer compiler then update the flesh with the breaking change, then update Carpet. This would keep the time of breaking things in the ET to a minimum.
The other option would be to make the breaking change in the flesh, and count on QB users that care about their system remaining usable with the ET to update the simfactory files.
Currently qb.ini says:
# last-tested-on: 2016-06-04 # last-tested-by: Erik Schnetter <schnetter@gmail.com>
which is not quite correct anymore.
--
Ticket URL: https://bitbucket.org/einsteintoolkit/tickets/issues/2426/g-10-possibly-miscompiles-parts-of-carpet