[gentoo-soc] Weekly Progress: MPI Overlay, Week 8

[gentoo-soc] Weekly Progress: MPI Overlay, Week 8

[Michael Gilroy]

[-- Attachment #1: Type: text/plain, Size: 3271 bytes --]

Hi,

*Summary:*
1) Fix bugs from build.log output- general eclass housekeeping.
2) Expand upon implementation-specifc src_* phases while using multilib
functions to iterate through implementations.
3) Discuss further design & next steps (installing mpi-dependent libs to
certain location, etc)

A lot of this week entailed fixing broken bits of code to avoid any
errors/warnings. There is still some work to be done in this regard,
particularly with the exported functions.

Next week's plans:
1) Continue work on exported functions and respective environment variables.
2) Start to objdump (among other things) to try and link against only
MPI-dependent libs.
3) Continue ironing out bugs and errors from build.log.

*Day-by-day breakdown:*

*Monday Plans:*

Sort out issues I've been having with has_version in mpi_dependencies and
function returns in get_mpicc. Via email if not available by IRC.

Work in progress- though I am feeling a bit stuck. Even calling the most
basic of functions (such as get_mpicc) fails to return a simple string.
Perhaps there is something large that I am missing.


Revisit mpi_src_* to start tailoring builds to their respective
implementations

As per our emails, building hpl directly from source seems like a helpful
task to get a better understanding of what is needed at each phase. I will
be doing this first thing on Tuesday.


*Tuesday's Plans:*

Build hpl directly from source, document needed environment variable
changes.
I need to continue tinkering with this, particularly with the TUNING
options (using hpl-2.1.tar.gz from ebuild)

Continue trying to debug current issues with eclass function calls.
Work in progress. The feedback I've gotten has been helpful, though I may
come with more questions tomorrow.


*Wednesday's Plans:*

Continue work on building hpl from source & tuning to specifics of my
machine, reference eselect-mpi for env variables.

Work in progress. I believe I have a much better understanding of the build
process, and will work on proper environment variable setting tomorrow in
the eclass.


Re-add mpi_foreach_implementation from before and test until it properly
iterates over MPI_TARGETS.

Work in progress. This will need much more testing as it is a bit of a
transition going from build-in multilib functions to something a bit more
custom.


*Thursday's Plans:*

Set proper environment variables in functions like src_prepare and
mpi_wrapper.
Incomplete. I've been working on my final project for my computer science
class, and did not have time to work on this today. The project is due at
noon on Friday and is my last assignment for that class.

Continue development on mpi_foreach_implementation, add loop for all "${@}"
as well as for all MPI_TARGETS.
Needs more testing, but I believe the loop implemented will account for the
implementations as well as the multilib-ABI information.


*Friday's Plans:*

Test the mpi_src_* functions with new mpi_foreach loop.

This is still broken and needs revising. The new custom function
(mpi_foreach) will need some updating, particularly in the inner function
where commands are executed.


Set proper environment variables in functions like src_prepare and
mpi_wrapper.

Work in progress. This work will continue on Monday.


Thanks,
Michael

[-- Attachment #2: Type: text/html, Size: 8097 bytes --]