search

ForestClaw / forestclaw

Quadtree/octree adaptive PDE solver based based on p4est.
http://www.forestclaw.org
BSD 2-Clause "Simplified" License
58 stars 21 forks source link

Parallel runs do not have bitwise agreement #137

Closed donnaaboise closed 6 years ago

donnaaboise commented 6 years ago

At one time, results from runs on different proc counts agreed to machine precision. This no longer seems to be the case. Look into this!

donnaaboise commented 6 years ago

There is a difference between the asynchronous ghost fill routine and the non-async ghost fill. With the async ghost fill (the one that is the most well tested), the results compare exactly. Clear that the logic in these two is different, but not clear where. This will be a separate issue.