| 1 |
Hi all, |
| 2 |
|
| 3 |
*Summary:* |
| 4 |
Last week's plans: |
| 5 |
1) Test various make.conf configurations to test against hpl builds. |
| 6 |
2) Set proper LDFLAGS to detect libmpi.so in its new location. |
| 7 |
3) Verify implementations listed in make.conf, expand upon |
| 8 |
dependency-appending function. |
| 9 |
|
| 10 |
I believe a lot of good progress was made. The hpl builds are going much |
| 11 |
more smoothly, and mpi-select now offers much more functionality during the |
| 12 |
compilation phases. There are some more minor hurdles that I need to clear |
| 13 |
to polish up certain functions, like the dependencies as well as locating |
| 14 |
mpicc effectively. The LDFLAGS should be working now but require further |
| 15 |
testing. Make.conf configurations are also coming along but also require |
| 16 |
more testing. |
| 17 |
|
| 18 |
Next week's plans: |
| 19 |
1) Fix bugs from build.log output- general eclass housekeeping. |
| 20 |
2) Expand upon implementation-specifc src_* phases while using multilib |
| 21 |
functions to iterate through implementations. |
| 22 |
3) Discuss further design & next steps (installing mpi-dependent libs to |
| 23 |
certain location, etc) |
| 24 |
|
| 25 |
*Day-by-day breakdown:* |
| 26 |
|
| 27 |
*Weekend/Monday Plans:* |
| 28 |
|
| 29 |
Go through notes of meeting. Properly implement and test things like LFLAGS |
| 30 |
or LD_LIBRARY_PATH and has_version to validate $MPI_TARGETS |
| 31 |
|
| 32 |
Work in progress. Now that I can properly test against hpl, I will work on |
| 33 |
getting libmpi.so properly sourced in LFLAGS. |
| 34 |
|
| 35 |
|
| 36 |
Get mpi-select to a "stable" build for further branching. |
| 37 |
|
| 38 |
Completed. hpl now builds successfully within the overlay. |
| 39 |
|
| 40 |
|
| 41 |
*Tuesday's Plans:* |
| 42 |
|
| 43 |
Eliminate dependence on mpi.eclass by adding mpi-select equivalent |
| 44 |
functionality. |
| 45 |
|
| 46 |
Work in progress. Functions from mpi.eclass have been added, but they need |
| 47 |
to be further tested and expanded upon specifically for mpi-select. |
| 48 |
|
| 49 |
|
| 50 |
Append to LFLAGS based on new /usr/$(get_lib)/mpi implementation location. |
| 51 |
Requires further testing, since upon compiling hpl is currently looking for |
| 52 |
the "normal" libmpi.so. |
| 53 |
|
| 54 |
Debug semantics of mpi-select functions using einfo statements. |
| 55 |
Incomplete. I did not get to this due to homework I have due Wednesday, I |
| 56 |
will continue to work on this. |
| 57 |
|
| 58 |
|
| 59 |
*Wednesday's Plans:* |
| 60 |
|
| 61 |
Debug logical errors in mpi-select_src_* |
| 62 |
|
| 63 |
Somewhat complete. While there have been errors patched, there are still a |
| 64 |
few "command not found"s in the build log, but everything is still stable |
| 65 |
and compiles. |
| 66 |
|
| 67 |
|
| 68 |
Test LFLAGS against different installed implementations. |
| 69 |
|
| 70 |
For hpl in particular, it sets a local variable called "mpicc_path" and I |
| 71 |
have been struggling to set this dynamically. I also think using |
| 72 |
LD_LIBRARY_PATH could potentially be easier |
| 73 |
(correct me if I am wrong), as that is what is supported in the current |
| 74 |
mpi.eclass . |
| 75 |
|
| 76 |
Currently in the overlay this is statically set to |
| 77 |
"/usr/lib64/mpi/mpich-3.2/install/usr/bin/mpicc". Even when I do the _bare |
| 78 |
minimum_ and echo that exact string from a function into the "mpicc_path" |
| 79 |
variable, I get the "make: o: command not found" that I would get if the |
| 80 |
path were incorrect. I have some functions set up that I believe would |
| 81 |
handle this dynamically but I am failing to echo the output correctly... Is |
| 82 |
there something simple I'm missing here?? |
| 83 |
|
| 84 |
|
| 85 |
Currently the functions I've set up request the _latest version_ of mpich |
| 86 |
currently installed, which I assume to be correct for now. I could always |
| 87 |
add a variable later to get a user-specified version of mpich for the mpicc |
| 88 |
building process. |
| 89 |
|
| 90 |
|
| 91 |
Further update functions used in hpl ebuild, refactor as needed. |
| 92 |
|
| 93 |
Work in progress. I now believe I have a very good idea of how functions |
| 94 |
should be set up in the overlay, and will continue to work to try to make |
| 95 |
the function calls as clean and simple as possible from the users' |
| 96 |
perspective. |
| 97 |
|
| 98 |
*Thursday's Plans:* |
| 99 |
|
| 100 |
Fix issues with returning correct values from functions (might need some |
| 101 |
insight on this, as I've tried a lot at this point). |
| 102 |
Incomplete. I will email as needed tomorrow. |
| 103 |
|
| 104 |
Replace mpi_pkg_deplist with mpi-select equivalent function. |
| 105 |
Completed. mpi_dependencies now grabs software by version to install. |
| 106 |
|
| 107 |
Replace all other hpl-2.1 mpi.eclass dependent functions. |
| 108 |
Some things like mpi_root still need some work. |
| 109 |
|
| 110 |
|
| 111 |
*Friday's Plans:* |
| 112 |
|
| 113 |
Get rid of "command not found" errors at runtime. |
| 114 |
|
| 115 |
Nearly completed. Most are gone, but I do need to hunt down some extra |
| 116 |
minor build errors. |
| 117 |
|
| 118 |
|
| 119 |
Add function to verify each MPI_TARGET individually. |
| 120 |
|
| 121 |
Having issues with has_version from flag-o-matic. Getting error: "$FUNCNAME |
| 122 |
calls are not allowed at global scope" even though this is being executed |
| 123 |
from within an eclass function (mpi_dependencies). Seems strange... I'd |
| 124 |
love to see if any of you have any insight. |
| 125 |
|
| 126 |
|
| 127 |
Replace existing mpi.eclass functions. |
| 128 |
|
| 129 |
This was very much connected to the "command not found" buildtime errors. |
| 130 |
This is nearly completed. |
| 131 |
|
| 132 |
|
| 133 |
|
| 134 |
Note: I will be working this weekend, but will be adding that progress to |
| 135 |
next week's progress report. |
| 136 |
|
| 137 |
All the best, |
| 138 |
Michael |
Archives