K-ballo changed the topic of #ste||ar to: STE||AR: Systems Technology, Emergent Parallelism, and Algorithm Research | stellar.cct.lsu.edu | HPX: A cure for performance impaired parallel applications | github.com/STEllAR-GROUP/hpx | Buildbot: http://rostam.cct.lsu.edu/ | Log: http://irclog.cct.lsu.edu/
bita has quit [Ping timeout: 264 seconds]
hkaiser has quit [Quit: bye]
jehelset has joined #ste||ar
<gonidelis[m]>
ms[m]: yt?
<ms[m]>
gonidelis[m]: here
<gonidelis[m]>
hey
<gonidelis[m]>
ms[m]: do we need docs for `Sent` here ?
<hkaiser>
K-ballo: you can mix and match debug and release std libraries on a module by module basis
<k-ballo[m]>
as long as you don't cross boundaries?
<hkaiser>
yes
<k-ballo[m]>
i don't know about that, but doesn't apply to us
<k-ballo[m]>
we have plenty of std:: (and boost::) on interfaces
<hkaiser>
nod
<hkaiser>
you're right...
<gonidelis[m]>
hkaiser: shouldn't this be sth like "merge two sorted ranges \a rng1 \a rng2 etc."
<hkaiser>
ok, what does it say instead?
<pedro_barbosa[m]>
Does anyone know where I can learn more about HPX.Compute? I've looked in the website and on some papers but the information seems to be very limited
<hkaiser>
K-ballo: are you aware of any thread-safety limitations for intrusive_ptr? especially concurrent reading/writing from the same instance would create a race, no?
<hkaiser>
pedro_barbosa[m]: what are you looking for
<hkaiser>
generally, HPX.Compute was an experiment in making CUDA available for parallel loops through the (std) algorithm interface
<pedro_barbosa[m]>
I read in a paper the following "HPX.Compute (Copik & Kaiser, 2017) aims to provide a single-source solution to heterogeneity by automatically generating GPU kernels from C++ code." and was interested in learning more about it as a I might use it in university related work
<k-ballo[m]>
same instance is generally thread unsafe
<hkaiser>
pedro_barbosa[m]: yes, the generate kernels from c++ code part depends on the CUDA compiler, though
<ms[m]>
gnikunj: which kokkos version and which options? it sounds like you've somehow circumvented the check that Kokkos should be built with Kokkos_ENABLE_HPX_ASYNC_DISPATCH=ON, but it's likely that I've messed up somewhere as well
<gnikunj[m]>
I'm using kokkos master and I enable the HPX_ASYNC_DISPATH
<ms[m]>
k-ballo[m], hkaiser "stdlib isn't debug/release compatible" meaning the whole stack should be built with the same build type? we keep the build type check for windows only in https://github.com/STEllAR-GROUP/hpx/pull/5099?
<gnikunj[m]>
ms: Is there anything else that I need to do to make it work?
<ms[m]>
gnikunj: shouldn't need anything else :/ hpx only or hpx+cuda? I'll see if I can reproduce it
<gnikunj[m]>
It's hpx only. No cuda
<diehlpk_work_>
ms[m], Are you helping out with the GSoC application again?
<ms[m]>
diehlpk_work_: ugh, yes, maybe a bit
<ms[m]>
we haven't really had any great new ideas for projects though
<ms[m]>
gnikunj: which commit of Kokkos specifically?
<gnikunj[m]>
ms: 953d7968
<ms[m]>
gnikunj: git pull...
<gnikunj[m]>
ms: didn't get you
<ms[m]>
(apologies for no version check in hpx-kokkos, at the time when I was adding that things were still on kokkos develop)
<ms[m]>
that's a 9 month old commit on master
<gnikunj[m]>
ms: commit 5292da1 hpx-kokkos
<ms[m]>
I think the version where that feature was in kokkos is 3.2
<gnikunj[m]>
<ms[m] "that's a 9 month old commit on m"> lol I just pulled their github repo
<gnikunj[m]>
which repo do you work on? or branch?
<ms[m]>
hpx-kokkos is just rolling on master
<ms[m]>
currently there's nothing in progress on upstream kokkos so their releases should do
<gnikunj[m]>
They recently switched to github from gitlab. It could probably be due to that.
<ms[m]>
gnikunj: note that I just pushed a couple of small commits that you might want to pull
<gnikunj[m]>
<ms[m] "gnikunj: note that I just pushed"> Ok, thanks!
<ms[m]>
kokkos switched? it's been on github for at least the last few years
<gnikunj[m]>
Ohh. It must be something else then, my bad.
<gnikunj[m]>
How is it that I have a 9 month old commit then. Strange. I literally cloned it 2d ago.
<ms[m]>
they might have a (n outdated?) mirror on gitlab?
<gnikunj[m]>
Thanks for the help! I'll try building it and see if it's still failing for me.
<ms[m]>
👍️
<ms[m]>
let me know how it goes
<gnikunj[m]>
👍
<hkaiser>
ms[m]: yah, applications should use the same build type as the library