Firedrake meeting 2021 09 09

Date and time 2021-09-09 15:00UTC (16:00BST)

Action Items

1. Pick Chair and Minuter.
2. ALL: (ongoing) triage the open issues and confirm if they are indeed still open (and perhaps provide labels)
3. DH: Firedrake 2021 Ask the other three (out of 24) if they'd like to give a talk.

Minutes

Present: DH, DRS, SV, KS, JB, SK, NB, RK

Apologies: LM

Firedrake 21

Have a schedule, it's been distributed.

Action DH: Need to send links for gather.town & zoom. Also setup retrium. Allocate chairs

DRS: build + compile fail on M1

#2196 shows the hacks needed to make Firedrake compile and run on an M1 mac. The change to the build script is fairly straightforward. What to do about -march=native failing on M1? Remove this flag, even for other macs?

I tried to debug this problem by running a simple test script with an unmodified PyOP2 and with PYOP2_DEBUG=1, but this makes the problem go away.

JB: Can use -mcpu=<something> to optimise for M1

DH: How should we sniff the CPU in PyOP2?

JB: platform.uname().machine is used in the install script

Action DRS: Create PR for M1 fixes

Merge PRs:

PR #2208: Should be fine, merged

PR #2202: RNH required

PR #2201: Failing: Subtle bug, fixed in Firedrake, now failing in pyadjoint

PR #2200: Failing: Part of refactor, related to TSFC PR #261. Big diff but commits are small, have meeting to resolve.

PR #2190: We need to think carefully about how we want users to cite Firedrake from the repo, the 2016 paper isn't sufficient.

PR #2176: Waiting on upstream PETSc change, PETSc has been merged, just remove branch pointer (done live)

PR #2101: KS: This needs a fix, but don't know how

AOB

Performance regression

DH: Finat dual evaluation results in regression on the CI, something goes wrong with GCC. We will just wait for sparse tensor stuff to land.

JB: Can we just update GCC in the docker container?

Fun fact

DH: test_tlm.py::test_time_dependent is not, in fact, testing the tlm. It's testing the adjoint. It's also failing on Melina's Branch

Date of next meeting

NO MEETING NEXT WEEK DUE TO FIREDRAKE21

1500 UTC (1600 BST) 2021-09-23