<nanmiao11>
Could you try srun to run the mpi job?
<nanmiao11>
You dont have use mpirun ..
<hkaiser>
thanks nanmiao11
<gnikunj[m]>
hkaiser: could it be that slurm is not configured with `--with-pmi`?
<gnikunj[m]>
thanks nanmiao11
<hkaiser>
gnikunj[m]: talk to akheir about this, pls
<hkaiser>
gnikunj[m]: Al just said that he has built slurm without mpi support (if I understood correctly), please work with him to find an apropriate way to launch mpi apps on rostam
<gnikunj[m]>
hkaiser: thanks. will do.
akheir has joined #ste||ar
<nanmiao11>
gnikunj[m] Would you like to have a short zoom meeting with me?
<diehlpk_work>
hkaiser, Which HPX version is needed to used with Blaze?
<diehlpk_work>
Using HPX 1.4.1 I get /usr/include/blaze/math/smp/hpx/DenseVector.h:43:10: fatal error: hpx/include/parallel_for_loop.hpp: No such file or directory
<hkaiser>
diehlpk_work: any should work
<hkaiser>
you might get warnings about obsolete headers
<diehlpk_work>
hkaiser, see the error message above
<hkaiser>
ms[m]: I was thinking more about what we talked about this morning
<hkaiser>
the the range and mdrange stuff might be more appropriate for for_loop than for for_each
<hkaiser>
diehlpk_work: could you post this again, pls?
<diehlpk_work>
hkaiser, I get this strange error /usr/include/blaze/math/smp/hpx/DenseVector.h:43:10: fatal error: hpx/include/parallel_for_loop.hpp: No such file or directory
<shahrzad>
diehlpk_work: do you install hpx?
<diehlpk_work>
shahrzad, Yes
<hkaiser>
your application is responsible for setting up HPX for blaze
<hkaiser>
blaze doesn't know anything about this, even if you tell it to use the HPX backend
<shahrzad>
I got this when I was building blazemark but was including the source hpx files
<shahrzad>
since that header has been moved to another directory after modularizing HPX
<hkaiser>
your application cmake file should do a find_package(HPX) and target_link_libraries(<your-target> HPX::hpx)
<hkaiser>
(with recent HPX versions that is)
<ms[m]>
hkaiser: nod, good point
<ms[m]>
both of them kind of fit
<ms[m]>
the nice thing with the cpos is this kind of thing will be very easy to implement
<ms[m]>
for_loop would be next after for_each in any case, I can have a go at that as well
<diehlpk_work>
hkaiser, Ok, I found the issue, I did all this, but if I use the the hpx in the build dir it does not work
<diehlpk_work>
I have to use the isntalled version of HPX
<ms[m]>
That header has been unchanged for a long time (including 1.5.0)
<diehlpk_work>
ms[m], I found the issue
<hkaiser>
ms[m]: right, thanks!
<diehlpk_work>
I should not use hpx/build/lib/cmake and I have to install HPX first
<hkaiser>
diehlpk_work: find_package() should work from the build-directory as well
<hkaiser>
at least it works for me
<diehlpk_work>
hkaiser, No, the HPX_INCLUDE_DIR is empyt and therefore HPX headers are not found
<hkaiser>
don't use those cmake variables, use the HPX::hpx target instead
<diehlpk_work>
Ok, will do
<hkaiser>
we deprecated using the cmake variables - ms[m], rori: we might want to explicitly mention this in the release notes
<diehlpk_work>
Ok, I will update my code accordingly
<ms[m]>
"hkaiser" (https://matrix.to/#/@freenode_hkaiser:matrix.org): yeah, good point
<ms[m]>
But something else isn't set up correctly in that case, the HPX_LIBRARIES variable should still work (although not recommended) and give the include directories if linked to
<hkaiser>
ms[m]: HPX_INCLUDE_DIR is emtpy
<ms[m]>
the include directories would come via the cmake target in HPX_LIBRARIES
<hkaiser>
ms[m]: sure, I know
hkaiser has quit [Ping timeout: 256 seconds]
<zao>
Does CI still install everything in system directories or is it actually testing library lookup?
<ms[m]>
System... That is indeed something we could change