[gentoo-soc] Weekly Report: MPI Overlay, Week 5

[gentoo-soc] Weekly Report: MPI Overlay, Week 5

[Michael Gilroy]

[-- Attachment #1: Type: text/plain, Size: 4456 bytes --]

Hi all,

*Summary:*

Last week's plans:
1) Loop over commands in the build process for multiple implementations.
2) Add more mpi implementation's versions to the overlay, testing with
mpi-providers
3) Adjust libraries/binaries installations globally. This will later be for
mpi-dependent things only

I have split this work into two different versions- one which emulates
multibuild and one which exports src_* in a more normal fashion. I did this
considering multibuild is more complex to fully understand, and that it may
be a better course of action to start with a more normal src_* procedure,
then add multibuild functionality as needed.

Next week's plan:
1) Successfully export src_* and test with mpich.
2) Continue testing against mpi-providers and adding more versions of
openmpi and mpich to the overlay.
3) Continue adding to wiki/README as functionality is added. Will talk to
mentors about other means of documentation as well.

*Day-by-day breakdown:*

*Monday's Goals:*

Complete loop to repeat commands per selected implementation.
Work in progress. Currently using a multibuild-like system, needs further
testing with mpi implementations.

Implement src_* per abi, per mpi implementation.

Further discussion needed. The main question here- how will users be
interacting with this eclass? I was thinking something along the lines of
${implementation}_foreach_version to find the correct range of versions
that need to be installed. Perhaps this could be its own function that acts
globally among all implementations, and pulls the existing Portage tree
versions to find ones that need to be installed. I'll bring this up in the
IRC.



*Tuesday's Goals:*

I will be taking the most of the day off due to an upcoming data science
exam on Wednesday. Work will continue on Wednesday as normal, with hours
shifted to reach over 35 hours for the week. However, I will likely be
asking questions in the IRC. I mainly want to clarify design tomorrow.


*Wednesday's Goals:*

Fix everything that is currently broken in the overlay. Repoman among other
things. Get master to a stable testing point again.
Complete. I am still investigating the "partial QA payment" warnings, and
still need to look into fortran uses in different gcc versions.
Improvements were made to the point where testing can continue.

Mail about mpi-select design. Make revisions and additions accordingly.
Awaiting response. I was intending on sending this much earlier than I did
because I ended up working much later than intended.

Add to wiki.

Work in progress. I updated it to match the README, which is minimal. Next
steps are to add instructions/guidance using proper wiki formatting.



*Thursday's Goals:*

Test new eclass with different versions of different implementations (i.e.
multilib vs. no multilib, fortran vs. no fortran, etc.)

Work in progress. With such varied testing there are still many errors to
be fixed. These will be apart of tomorrow's work.


Ask more questions regarding 2nd eclass as needed. Add to eclass based on
design responses.

Completed, again rather late. I hope to continue this dialogue more in our
weekly meeting. The current dialogue has been very helpful- it has given me
a lot more to research today.


Update wiki to use better formatting to better explain things (i.e. code
blocks)

Incomplete. As we are discussing new ways of doing things and clarifying
design choices, I thought it would be best to postpone the wiki overhaul
until more resolutions have been agreed upon. I'm planning on getting
enough info to overhaul the wiki tomorrow.


*Friday's Goals:*

Further clarify user-end execution in IRC chat. Modify eclass to properly
support make.conf MPI_TARGETS editing as needed.
Incomplete. I failed to set up a meeting. I will try to do this earlier, as
I know differing time zones make things less convenient. My Thursdays and
Fridays are preferable for meetings, as I do not have to leave for class
these days.

Test & edit mpi-select to the point where it is more stable for various
openmpi and mpich versions. Add any working versions to the overlay.
Complete. Worth noting- I have been testing against mpi-providers rather
than mpi-select. mdpi-select needs to have more functionality in order to
warrant bringing something else into the overlay from a successful test.

Finally overhaul the wiki based on current discussions :)
Complete. Will be adding to this as time moves on, of course.



Thanks,
Michael

[-- Attachment #2: Type: text/html, Size: 8495 bytes --]