Avoid use of MPI_2INT - Githubissues

MP-Gadget / bigfile

A reproducible massively parallel IO library for hierarchical data

BSD 2-Clause "Simplified" License

18 stars 6 forks source link

Avoid use of MPI_2INT #33

Closed sbird closed 5 years ago

sbird commented 5 years ago

The error broadcasting algorithm with MPI_IN_PLACE and MPI_2INT continues to give trouble on Frontera with intel MPI 18. Just avoid it. It is an obscure enough part of the MPI standard that it is likely often buggy.

It seems easier to just not do this than to keep arguing with Frontera and intel!

rainwoodman commented 5 years ago

The travis failure was due to conda. I added a temporary workaround on master -- it shall work if you rebase.

sbird commented 5 years ago

Should be fixed now! This does make Frontera more stable.

sbird commented 5 years ago

Can you merge this please? I don't have commit access.

rainwoodman commented 5 years ago

I see. You've fixed the loc bug. Merging!