-
Notifications
You must be signed in to change notification settings - Fork 171
Firedrake meeting 2020 12 02
Date and time 2020-12-02 16:00UTC (16:00GMT)
- Pick Chair and Minuter.
- ALL: (ongoing) triage the open issues and confirm if they are indeed still open (and perhaps provide labels)
- ??: Build master on centos to catch errors
Present: CW, RK, KK, KS, DH, JB, RNH, DS, SV, PK, CC, LM
Apologies:
LM: Excalibur performance workshop
Might be of interest to some (organised at Durham). Daniel Ruprecht (now at Hamburg) got in touch because he has a PhD student who would be keen to join at Firedrake team.
Do we want to send a virtual 'team'?
There are tutorials in the morning on profiling tools. In the afternoon you try to get it working on your code.
CW is interested and maybe others so a team will definitely be sent.
It might actually be a bug that has appeared in PETSc, but we don't yet have a concrete "always works" testcase.
Florian Wechsung (NYU) and Angus Gibson (ANU) have both been reporting issues recently.
Infiniband libraries do not have forking.
A nice solution to this would be to have a libclang JIT thing that you could call.
The less nice option is to fork before calling MPI_Init.
This would involve handling Firedrake imports to make sure they don't touch subprocess. In particular PyOP2.
Status of JIT C compilation is unknown but likely to be painful.
Shared by PK: https://clang.llvm.org/docs/LibTooling.html
DH: libgcc JIT looks promising. CW to look into it this week.
Can now drop volume potentials into UFL. The next thing is to use non-local operators.
The code is currently a bit messy and needs tidying before merge into Naseem's branch.
Several merged.
DH is talking at 3pm (UK) on Monday about dual spaces/interpolation. This has implications for simplifying UFL and the relevant algebra.
2020-12-09 16:00UTC (16:00GMT)