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/
jehelset has quit [Remote host closed the connection]
hkaiser has joined #ste||ar
hkaiser has quit [Client Quit]
bita has quit [Ping timeout: 260 seconds]
nanmiao111 has joined #ste||ar
hkaiser has joined #ste||ar
jehelset has joined #ste||ar
hkaiser has quit [Quit: bye]
jehelset has quit [Remote host closed the connection]
diehlpk_work has quit [Remote host closed the connection]
nanmiao111 has quit [Ping timeout: 240 seconds]
<ms[m]> gnikunj: fyi, I could reproduce your cuda clang issue
<ms[m]> seems like cray clang is surprisingly saner than vanilla clang in this case...
<gnikunj[m]> Lol
<ms[m]> clang defaults to c++14 when compiling cpp files, but to a lower standard when compiling cu files
<ms[m]> this issue summarizes it https://github.com/kokkos/kokkos/issues/2520
<ms[m]> it's a combination of clang weirdness and cmake weirdness
<ms[m]> we use the cxx_std_XX compile features which add a -std= flag if needed, but cmake decides it's not needed with clang because it's the default (except not always, so it breaks...)
<ms[m]> I'll see if the workaround in kokkos is what we want to do or if there's something else
<ms[m]> for now you'll just have to set CMAKE_CXX_STANDARD explicitly
hkaiser has joined #ste||ar
hkaiser has quit [Quit: bye]
hkaiser has joined #ste||ar
<hkaiser> all: please upvote the release here: https://www.reddit.com/r/cpp/
<ms[m]> hkaiser: thanks for posting that!
<hkaiser> I posted on isocpp.org as well
diehlpk_work has joined #ste||ar
<ms[m]> 👍️
<diehlpk_work> ms[m], hkaiser hpx 1.6.0 is alvailable on the next fedora release
<hkaiser> diehlpk_work: perfect, thanks!
<ms[m]> hkaiser: are you in a room that's above 0 degrees celcius right now? ;)
<ms[m]> thank you diehlpk_work !
<hkaiser> we need to find a workaround for the problem on next fedora, however
<hkaiser> ms[m]: yah, we've moved to a hotel
<diehlpk_work> However, we have some issues on the future fedora release
<ms[m]> I wanted to ask if we could shift the meeting to next week instead
<ms[m]> I'll be away otherwise in two weeks
<ms[m]> what sort of problems?
<hkaiser> ms[m]: absolutely
<ms[m]> perfect, thanks
<hkaiser> ms[m]: they redefined a mcro that was resolving to a constant to a non-const system call
<hkaiser> easy to fix, I believe
<hkaiser> ms[m]: I'll let Katie know to reschedule for next week
<diehlpk_work> Error message is close the the end of this file
<diehlpk_work> 224 | default_stack_size = SIGSTKSZ
<ms[m]> dirty, thanks diehlpk_work
<diehlpk_work> Would be nice to have a patch in the next few days to fix the build there
<hkaiser> diehlpk_work: I can do that
<hkaiser> what fedora version will that be?
<diehlpk_work> However, it is not too critical since it we can ship it with the next fedora release
<diehlpk_work> 35
<hkaiser> k
<diehlpk_work> It compiles in 34 which will be releases in April
<diehlpk_work> 35 is targeted for October or November
<hkaiser> I'll create a PR in a couple of minutes
<ms[m]> I wonder if we even need that, I had the impression we always explicitly set the stack size
<hkaiser> it's the posix fallback implementation for power architectures
<hkaiser> is there a boost context version available for power?
<ms[m]> ah, I just meant the default stack size, not the whole implementation
nanmiao111 has joined #ste||ar
<ms[m]> although I thought we use the boost implementation usually on power as well
<hkaiser> ms[m]: nod, apparently diehlpk_work used the posix fallback instead
<ms[m]> I suppose both might work, and I might also be completely misremembering
<hkaiser> diehlpk_work: here is the list of architectures supported by Boost.Context: https://www.boost.org/doc/libs/1_75_0/libs/context/doc/html/context/architectures.html
<diehlpk_work> Anyways we will run in the same issues on Summit soon or?
<hkaiser> I think power is not in that list
<ms[m]> ppc?
<ms[m]> diehlpk_work: that's the unreleased gcc 11 right?
<hkaiser> diehlpk_work: would you mind trying https://github.com/STEllAR-GROUP/hpx/pull/5182?
<diehlpk_work> Yes
<diehlpk_work> But the gcc will be released before Fedora 35 ships
<diehlpk_work> So there will be plenty of updates
<diehlpk_work> hkaiser, Sure, I will test the patch soon
<hkaiser> ms[m]: worth trying... - might give better performance that the fallback
<hkaiser> diehlpk_work: thanks
<diehlpk_work> On good thing of Fedora is that it keeps us up to the latest unstable gcc :)
<diehlpk_work> Hey gnikunj[m], gonidelis[m] what about the GSoc text you wanted to write?
<diehlpk_work> I could not find it on the mailing list
<diehlpk_work> and we have to submit the application by tomorrow
<k-ballo[m]> i just finished writing an email and... pressed the shortcut keys to run clang-format on it
<k-ballo[m]> we should clang-format everything left so i can turn on format-on-save
<hkaiser> k-ballo[m]: did the email look better afterwards?
<hkaiser> ;-)
<k-ballo[m]> luckily thunderbird did nothing on those key presses
<hkaiser> k-ballo[m]: I expect to move the remaining files into modules over the next month or two, after that this shouldn't be a an issue anymore
<gonidelis[m]> diehlpk_work: sorry for stalling on this
<gonidelis[m]> i will send it now (don't take it for granted. we may as well not put it in there)
<gnikunj[m]> diehlpk_work: I'm working with gonidelis[m] to see if we can add anything in there. As it stands, it already looks good to me.
<gnikunj[m]> hkaiser: progress: got code to work for host codes to return the final result. The code doesn't compile with CUDA HPX kernel but I should be able to make that work too! Btw, Kokkos views ftw :D
<hkaiser> gnikunj[m]: +1
<gonidelis[m]> hkaiser: diehlpk_work did you got it?
<hkaiser> not yet
<diehlpk_work> gonidelis[m], Just approved your post
<gonidelis[m]> get^^
<diehlpk_work> gonidelis[m], https://www.diehlpk.de/blog/cpp-lectures/
<diehlpk_work> I would recommend to use the summer series for the hpx related talks
<gonidelis[m]> hey!!!
<gonidelis[m]> i didn't know these were public!!!!
<gonidelis[m]> these are golden
<diehlpk_work> Please use the link to that play list
<diehlpk_work> maybe mention that the templates one are helpful
<gonidelis[m]> does hkaiser approve on referencing them?
<diehlpk_work> yes, he has to
<gnikunj[m]> diehlpk_work: I had no clue you made the course public!
<gonidelis[m]> <diehlpk_work "maybe mention that the templates"> remember: we don't wanna scare them
<diehlpk_work> No, we want to scare them and only get the best students
<gonidelis[m]> lol
<diehlpk_work> For the irc channel, I would add its name and that we are on the freenode server
<gonidelis[m]> do you wanna add those?
<diehlpk_work> No, please add them
<diehlpk_work> Please add a link to the pull requests
<gonidelis[m]> ok i will
<diehlpk_work> And maybe we should mention that the should try to compile HPX, so they do not lose one week doing that
<gonidelis[m]> yeah that sounds important too indeed
<diehlpk_work> We asked them to implement a simple matrix vector multiplication using HPX and upload it to GitHub with a CMake script
<diehlpk_work> We might want to do that again this year
<diehlpk_work> maybe ## Tips to speed-up familiarization from former successful GSoc students?
<srinivasyadav227> diehlpk_work: is this for students who to participate in GSoC for HPX this year?
<diehlpk_work> That is all suggestions, I have
<gnikunj[m]> diehlpk_work: give me 1h. I'm working on my list as well that we can directly integrate it with gonidelis[m] 's ones
<diehlpk_work> srinivasyadav227, Yes, we want to add some hints to the webpage
<hkaiser> srinivasyadav227: we have now an idea how you could start helping without having to get into things too deeply, if you like
<diehlpk_work> gnikunj[m], Sure, but you now me well enough that I always have something to complain or add :)
<diehlpk_work> *know
<hkaiser> srinivasyadav227: I'm referring to ms[m]' coment here: https://github.com/STEllAR-GROUP/hpx/pull/5176#issuecomment-781142703
<srinivasyadav227> hkaiser: yea sure, that helps me a lot .. currently I am working with examples and trying to understand HPX programming model deeper
<gnikunj[m]> I sure do. But that's how it goes, right. We have an improved version in the end ;)
<hkaiser> convert the existing non-segmented algorithms that are already ported to CPOs (customization point objects) from tag_invoke to tag_fallback_invoke
<hkaiser> srinivasyadav227: ^^
<hkaiser> it's a stupid task alright, but might help you to get into the code base
<srinivasyadav227> hkaiser: okay, if I am not wrong I should make a PR on this before GSoC application starts (March 29) to be a part of GSoC this year with HPX right?
<hkaiser> srinivasyadav227: no guarantees
<srinivasyadav227> hkaiser: yea yea.. I will do it
<hkaiser> I'm assuming you're interested to get involved with HX, so this is a good way to start
<srinivasyadav227> hkaiser: yes yes..this would be great! 😊
<hkaiser> srinivasyadav227: ask gonidelis[m] or gnikunj[m], they both have started that way, have done gsoc, and now they're hooked and still involved
<gnikunj[m]> HPX is the best runtime that nobody understands
<hkaiser> gnikunj[m]: lol
<srinivasyadav227> gnikunj: lol
<hkaiser> don't you scare off people
<gnikunj[m]> that's how zao puts it as well xD
<gnikunj[m]> according to zao it would be: HPX is the best runtime that doesn't compile or run
<hkaiser> gnikunj[m]: zao is special ;-)
<zao> ^_^
<gnikunj[m]> but honestly, HPX is one of the best code bases I've looked at
<zao> I'm always out on the bleeding edge of platforms, some friction is to be expected.
<gnikunj[m]> and I've always been interested in the functioning
<srinivasyadav227> srinivasyadav227: I just started by watching almost all Conference talks of hkaiser and workshops in Youtube to understand HPX, now I got some clarity..and started working on examples.. gnikunj now please don't scare me..lol 😂
<srinivasyadav227> gnikunj: gnikunj zao hkaiser diehlpk_work are the mentors ?
<hkaiser> srinivasyadav227: yah, but there are more
<srinivasyadav227> hkaiser: ok
<diehlpk_work> srinivasyadav227, We have different mentors for different projects
<zao> srinivasyadav227: I'm thankfully completely uninvolved in GSoC apart from being friendly on IRC.
<srinivasyadav227> diehlpk_work: oh..cool!
<gnikunj[m]> zao and I still preach that you join gsoc mentors list xD
<srinivasyadav227> zao: lol..ok..
<diehlpk_work> ms[m], hkaiser Any idea why HPX's Cmake fails on the first run and succeeds on the second run?
<diehlpk_work> On Summit I have to run cmake twice to get my make files generated
jedi1898 has joined #ste||ar
jedi1898 has quit [Client Quit]
gentaco18 has joined #ste||ar
<gentaco18> Hi everyone! I've been going through the gsoc organizations list and found ste||ar and hpx looks like a very interesting, though I must say intimidating project to contribute to. I have experience contributing to C++ codebases (mostly Qt desktop applications) but have very little experience with template metaprogramming and parallel programming (I
<gentaco18> hope to learn both this year, though probably not in time for gsoc). Is it still possible for me to effectively contribute to this project? Could you tell me how you would advice me to proceed? Should I attempt to create some small projects making extensive use of modern C++ and template metaprogramming? (Regardless of whether I have a chance of
<gentaco18> participating in gsoc this year, I'm still interested in contributing outside of it)
<zao> Hi there!
<gonidelis[m]> gentaco18: hey!
<gentaco18> Hello! Sorry for the long wall of text, like I said I'm a bit intimidated by the project :D
<gentaco18> If you don't mind me asking, how did you all learn template metaprogramming? CppCon videos and practice or did you read a book?
<gonidelis[m]> gentaco18: how did you find out that HPX is based on template meta programming if i may? btw your question is right on point ;)
<gonidelis[m]> i was in the exact situation 12 months ago if that helps :)
<gentaco18> I went through the ideas list and looked at some of the code
<gentaco18> Ohh that does help, what did you do? How long did it take you to learn?
gentaco18 has quit [Quit: Connection closed]
<gonidelis[m]> Try to lean by the minimal examples that we offer
<gonidelis[m]> try implementing mini stuff and also read the docs
<gonidelis[m]> plus watch some lectures or talks related to HPX so you can get what our mission is
<gonidelis[m]> diehlpk_work: ;)
hkaiser has quit [Quit: bye]
<gonidelis[m]> ms[m]: what's the difference between `tag_invoke` and `tag_fallback_invoke` ?
hkaiser has joined #ste||ar