Re: test script configurations
Thanks Dave. I'm CC'ing the others in the group for their information.
The test script is in good working order right now in SVN[1]. In its current form it just tests an empty set of configure options on the 'default' image, but it does run all four tests (make testing, Intel, OSU, NPB). I need to assemble a list of options that you would like to test. Here's what I am currently going to add:
--enable-g=all --enable-fast --enable-debuginfo --enable-nmpi-as-mpi --enable-mpe --enable-sharedlibs=gcc --disable-sharedlibs --enable-alloca --enable-timer-type=clock_gettime --enable-timer-type=gettimeofday --enable-timer-type=linux86_cycle
Rajeev/Bill, are there any other options that need to be tested?
--with-device=ch3 --with-device=ch3:ssm --with-device=ch3:shm --with-device=ch3:nemesis
--with-device=ch3:dllchan=sock --with-device=ch3:dllchan=ssm ... --with-device=ch3:dllchan=sock,ssm,shm
As for compilers, what other compilers do you usually use? On Linux I'm really only directly familiar with gcc, so I'm not sure what else to use. Is there an intel compiler, maybe something else as well? Where can I find the other compilers?
Other than gcc, Intel compilers are all we have on the breadboard machines.
I'm working on adding non-mpd process manager support right now. Currently the tests only use mpd. You mentioned 'gforker' before, are there any others that should be tested (maybe smpd)?
gforker definitely. For windows, we need smpd (which probably the script doesn't handle right now). For linux, I was mentioning to Rajeev that we should probably drop it. Maybe he can confirm.
[1] https://svn.mcs.anl.gov/repos/radix/projects/mpi/release_testing/
Thanks. -- Pavan -- Pavan Balaji, Mathematics and Computer Science, Argonne National Laboratory Ph: 630.252.3017 http://www.mcs.anl.gov/~balaji
On Jul 31, 2007, at 12:49 PM, Pavan Balaji wrote:
Thanks Dave. I'm CC'ing the others in the group for their information.
The test script is in good working order right now in SVN[1]. In its current form it just tests an empty set of configure options on the 'default' image, but it does run all four tests (make testing, Intel, OSU, NPB). I need to assemble a list of options that you would like to test. Here's what I am currently going to add: --enable-g=all --enable-fast --enable-debuginfo --enable-nmpi-as-mpi --enable-mpe --enable-sharedlibs=gcc --disable-sharedlibs --enable-alloca --enable-timer-type=clock_gettime --enable-timer-type=gettimeofday --enable-timer-type=linux86_cycle
Rajeev/Bill, are there any other options that need to be tested?
--enable-threads=funneled --enable-threads=multiple --enable-strict --enable-strict=posix --disable-cxx --disable-f77 --disable-f90 --disable-weak-symbols These are ones that we either use ourselves or advise users to try. Bill
--with-device=ch3 --with-device=ch3:ssm --with-device=ch3:shm --with-device=ch3:nemesis
--with-device=ch3:dllchan=sock --with-device=ch3:dllchan=ssm ... --with-device=ch3:dllchan=sock,ssm,shm
As for compilers, what other compilers do you usually use? On Linux I'm really only directly familiar with gcc, so I'm not sure what else to use. Is there an intel compiler, maybe something else as well? Where can I find the other compilers?
Other than gcc, Intel compilers are all we have on the breadboard machines.
I'm working on adding non-mpd process manager support right now. Currently the tests only use mpd. You mentioned 'gforker' before, are there any others that should be tested (maybe smpd)?
gforker definitely. For windows, we need smpd (which probably the script doesn't handle right now). For linux, I was mentioning to Rajeev that we should probably drop it. Maybe he can confirm.
[1] https://svn.mcs.anl.gov/repos/radix/projects/mpi/release_testing/
Thanks.
-- Pavan
-- Pavan Balaji, Mathematics and Computer Science, Argonne National Laboratory Ph: 630.252.3017 http://www.mcs.anl.gov/~balaji
--enable-g=all --enable-fast --enable-debuginfo --enable-nmpi-as-mpi --enable-mpe --enable-sharedlibs=gcc --disable-sharedlibs --enable-alloca --enable-timer-type=clock_gettime --enable-timer-type=gettimeofday --enable-timer-type=linux86_cycle
Rajeev/Bill, are there any other options that need to be tested?
--enable-threads=funneled --enable-threads=multiple --enable-strict --enable-strict=posix --disable-cxx --disable-f77 --disable-f90 --disable-weak-symbols
Just remembered --enable-nemesis-dbg-nolocal from another email thread. This is specific to nemesis though. -- Pavan -- Pavan Balaji, Mathematics and Computer Science, Argonne National Laboratory Ph: 630.252.3017 http://www.mcs.anl.gov/~balaji
If we're running on more than one machine --enable-nemesis-dbg-nolocal is not needed. It's just there to force using the network when running on a single machine. -d On 08/02/2007 07:51 PM, Pavan Balaji wrote:
--enable-g=all --enable-fast --enable-debuginfo --enable-nmpi-as-mpi --enable-mpe --enable-sharedlibs=gcc --disable-sharedlibs --enable-alloca --enable-timer-type=clock_gettime --enable-timer-type=gettimeofday --enable-timer-type=linux86_cycle
Rajeev/Bill, are there any other options that need to be tested?
--enable-threads=funneled --enable-threads=multiple --enable-strict --enable-strict=posix --disable-cxx --disable-f77 --disable-f90 --disable-weak-symbols
Just remembered --enable-nemesis-dbg-nolocal from another email thread. This is specific to nemesis though.
-- Pavan
gforker definitely. For windows, we need smpd (which probably the script doesn't handle right now). For linux, I was mentioning to Rajeev that we should probably drop it. Maybe he can confirm.
Some users still use SMPD because MPD doesn't work well with Sun Grid Engine for some reason. (There is info on the web on that.) So we still need to include SMPD, but need not develop it further. Rajeev
participants (4)
-
Darius Buntinas -
Pavan Balaji -
Rajeev Thakur -
William Gropp