aserio 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/
Smasher has quit [Remote host closed the connection]
galabc has quit [Quit: Leaving]
hkaiser has quit [Quit: bye]
diehlpk has quit [Ping timeout: 240 seconds]
jaafar has joined #ste||ar
diehlpk has joined #ste||ar
diehlpk has quit [Remote host closed the connection]
mcopik has quit [Ping timeout: 268 seconds]
K-ballo has quit [Quit: K-ballo]
pdales has joined #ste||ar
parsa has quit [Quit: Zzzzzzzzzzzz]
parsa has joined #ste||ar
parsa has quit [Client Quit]
mcopik has joined #ste||ar
mcopik has quit [Ping timeout: 256 seconds]
parsa has joined #ste||ar
parsa has quit [Quit: Zzzzzzzzzzzz]
nanashi55 has quit [Ping timeout: 264 seconds]
nanashi55 has joined #ste||ar
jaafar has quit [Ping timeout: 248 seconds]
Anushi1998 has joined #ste||ar
mcopik has joined #ste||ar
Anushi1998 has quit [Quit: Leaving]
FjordPrefect has quit [*.net *.split]
FjordPrefect has joined #ste||ar
mcopik has quit [Ping timeout: 256 seconds]
mcopik has joined #ste||ar
hkaiser has joined #ste||ar
hkaiser has quit [Read error: Connection reset by peer]
nikunj has joined #ste||ar
hkaiser has joined #ste||ar
verganz has joined #ste||ar
david_pfander has joined #ste||ar
<simbergm> jbjnr: the updated pycicle with working github status works fine for me, would you mind just wiping your build folders?
<jbjnr> I did that last time you asked
<simbergm> hmm :/
<jbjnr> I am running your patches, but also the merge request from peter doak, so mybe his stuff broke it again ?
<jbjnr> I'll have a quick debug now
<simbergm> ah, maybe
<simbergm> ok, thanks
<simbergm> unrelated, but apparently there's https://github.com/codepanda/pycicl
<simbergm> luckily the names are clearly unique ;)
<jbjnr> wow!
<jbjnr> same name- nearly
Guest47721 has quit [Ping timeout: 260 seconds]
<zao> Some day I'll learn how to spell pycicle.
<zao> I keep associating it with bikes.
<github> [hpx] msimberg opened pull request #3228: WIP: Allow suspension also on static schedulers (master...suspend-static-schedulers) https://git.io/vxvBB
<github> [hpx] Naios opened pull request #3229: Fix sync traversals of containers passed as r-value (master...traverse_rvalue_container) https://git.io/vxvRc
CaptainRubik has joined #ste||ar
<github> [hpx] msimberg pushed 3 new commits to master: https://git.io/vxvuA
<github> hpx/master 6dc95c1 Mikael Simberg: Relax locks for thread suspension to avoid holding locks when yielding
<github> hpx/master 75dd135 Mikael Simberg: Remove redundant function calls when locking in thread pool
<github> hpx/master a297a15 Mikael Simberg: Merge pull request #3213 from msimberg/suspension-locks...
david_pfander has quit [Ping timeout: 265 seconds]
anushi has quit [Ping timeout: 252 seconds]
david_pfander has joined #ste||ar
<hkaiser> jbjnr: could we use hwloc v2 on one of the pycicle builders, please?
<jbjnr> not easily
<jbjnr> I have not added the option to pick different build configurations from a list of choices yet
<hkaiser> ok
<hkaiser> I'll ask Al to add it to one of the buildbots
<heller_> my 32 node job failed :(
<hkaiser> where? on cori?
<heller_> yeah
<heller_> haswell partition
<hkaiser> strange
<heller_> both libfabric and MPI
<heller_> could be my code after all
K-ballo has joined #ste||ar
anushi has joined #ste||ar
<github> [hpx] msimberg opened pull request #3230: Add optional scheduler mode parameter to create_thread_pool function (master...create-thread-pool) https://git.io/vxvMk
eschnett has quit [Quit: eschnett]
<github> [hpx] NK-Nikunj opened pull request #3231: Mini gsoc (master...mini-gsoc) https://git.io/vxvyR
CaptainRubik has quit [Quit: Page closed]
<jbjnr> hkaiser: if we're happy that hwloc 1.x works for us, then I can just change the build on daint to all use hwloc-2 - that's easy
<jbjnr> ^all the builds on daint
<hkaiser> jbjnr: sure, why not?
<hkaiser> circle uses v1, so does appveyor - we would have coverage there
<jbjnr> why not? because if hwloc-2 support is incomplete, then every build on pycicle will give test fails
<hkaiser> jbjnr: wouldn't that be a Good Thing (tm)?
<jbjnr> if hwloc-2 support hasn't been mered to master then ....
<jbjnr> ^merged
<hkaiser> it has, which is the problem
<jbjnr> aha ok
<jbjnr> then I'll upgrade daint hwloc
<verganz> Hello all. Can I ask some questions about legacy projects for GSoC?
<hkaiser> sure
<hkaiser> ask away
<hkaiser> jbjnr: thanks a lot!
<verganz> what is the status of the project Newtonian Physics Sandbox? Is there any beginnings of it?
<simbergm> wait, isn't 3225 the PR to make hwloc 2 work? that's not merged yet
<jbjnr> yes. that was what I'm worried about. Everything fails completely if hwloc-2 isn't merged
<simbergm> yep, I agree
<simbergm> I think it's nice if we can get hwloc 2 support in soon (before release) but I don't think it's urgent in any way
apsknight has joined #ste||ar
eschnett has joined #ste||ar
<hkaiser> simbergm: it has to go into the release, I think
<hkaiser> the release is there to stay for a while, and hwloc v2 will become more commonplace
viraj has joined #ste||ar
<viraj> Hi
<viraj> Can anyone tell how to set HPX_WITH_APEX=ON during compile time?
<zao> Sounds like a CMake -D thing
<heller_> it is
<zao> So like cmake srcdir -DCMAKE_INSTALL_PREFIX=/opt/hpx -DHPX_WITH_APEX=ON -DHPX_WITH_CXX14=ON
<viraj> oh i thought that some pc file needed to change
<jbjnr> viraj: you might find this page useful where some of my settings for the cray are listed https://github.com/biddisco/biddisco.github.io/wiki/daint
<viraj> thank you for the link
apsknight has quit [Quit: apsknight]
viraj has quit [Ping timeout: 260 seconds]
<nikunj> @hkaiser: I was thinking of re-implementing quicksort example since you told me to remove memory_block from hpx. Should that be fine?
<hkaiser> nikunj: sure, try it
HoloIRCUser3 has joined #ste||ar
<nikunj> @hkaiser: ok
HoloIRCUser3 has quit [Client Quit]
viraj has joined #ste||ar
hkaiser has quit [Quit: bye]
aserio has joined #ste||ar
viraj has quit [Ping timeout: 255 seconds]
HoloIRCUser2 has joined #ste||ar
HoloIRCUser2 is now known as Viraj
Viraj has quit [Client Quit]
parsa has joined #ste||ar
<simbergm> heller_: any updates on suspension in allscale?
<heller_> simbergm: seems to have stopped working
<heller_> simbergm: trying to get the distributed version to scale now
<simbergm> ugh :/
<simbergm> everything or suspension?
<simbergm> if I understood you correctly suspension didn't work before either
<github> [hpx] msimberg opened pull request #3232: Clean up suspension tests (master...suspension-tests-cleanup) https://git.io/vxvpj
<jbjnr> simbergm: will I need to do anything to my scheduler for it to work with your suspension stuff?
<simbergm> jbjnr: if you want it to not schedule tasks on suspended threads you need to update schedule_thread and create_thread to take that into account (same as in the local_queue_schedulers)
<simbergm> but pool and runtime suspension will work just like that because they wait for the pools to be empty outside the scheduler
<jbjnr> ok thanks. How crucial is the not scheduling on suspended threads bit? is that somethien extra or part of the core design - I mean is that some feature that lets you suspend threads one at a time?
<jbjnr> I can't see myself ever needing that. I'd suspend a pool, but not an individual thread
<simbergm> yeah, it's more for heller_'s stuff, we (cscs) most probably will never that
<heller_> simbergm: everything, without suspension at first
<simbergm> it's really just to allow suspending individual threads
<simbergm> heller_: ok
<heller_> currently trying to crack the 32 node mark :/
<jbjnr> (we'll have to work on convincing heller that suspending threads is not a good idea, but pools is ok).
<simbergm> I will open a PR with the schedule_thread changes, it's better than nothing at least and once you get to that point we can try to see what more you need
<simbergm> jbjnr: at least you shouldn't have any performance hit from that, if you don't have enable_elasticity enabled the scheduler doesn't care about suspended threads
<jbjnr> heller_: you said that after #3146 you'd work on removing wait_or_add_new - can you briefly tell me how and why it can be removed - where will the things it does get moved to?
<jbjnr> simbergm: ok
<heller_> jbjnr: to nowhere
<simbergm> no more new_tasks?
<heller_> yeah, the idea is to only have one queue, instead of having two as we have it right now
parsa has quit [Quit: Zzzzzzzzzzzz]
<simbergm> then every task gets its stack etc allocated right away, or?
<heller_> no
<heller_> that's what #3146 is about
<simbergm> ah, right, makes sense
<jbjnr> the lazy allocate means we don't need to worry. ok. I understand
<jbjnr> I'll work on that then
<jbjnr> thanks
<github> [hpx] msimberg opened pull request #3233: Make schedule_thread take suspended threads into account (master...fix-schedule_thread) https://git.io/vxfJW
nikunj has quit [Ping timeout: 260 seconds]
hkaiser has joined #ste||ar
<github> [hpx] msimberg opened pull request #3234: WIP: Disable background thread when networking is off (master...no-background-thread-networking-off) https://git.io/vxfIs
dattatreya has joined #ste||ar
dattatreya has quit [Quit: Page closed]
dattatreya has joined #ste||ar
apsknight has joined #ste||ar
dattatreya has quit [Quit: Page closed]
Anushi1998 has joined #ste||ar
<jbjnr> simbergm: my test hangs on shutdown in thread_helpers.cpp::suspoend function. Should I be worried?
<jbjnr> hint : answer = yes!
anushi has quit [Read error: Connection reset by peer]
Anushi1998 has quit [Read error: Connection reset by peer]
anushi has joined #ste||ar
Anushi1998 has joined #ste||ar
anushi_ has joined #ste||ar
anushi has quit [Remote host closed the connection]
<simbergm> jbjnr: uhm, yes, you should be, but I'm not sure I can help you
<simbergm> that's hpx thread suspension
<jbjnr> Are there known 'problems' <cough> with the suspend stuff?
<jbjnr> my custom scheduler?
<jbjnr> hmmm
<simbergm> not known, but probably there are unknown ones
<diehlpk_work> To all our GSoC students: The application for GSoC 2018 is now open
<diehlpk_work> Applications close Tuesday, March 27 at 16:00 UTC
<K-ballo> hkaiser: did you know lenovo laptops have SIM slots?!?
<hkaiser> yah, never used those, though
<K-ballo> I just found out mine... after a summer of having a dummy phone around to operate as hot spot
<hkaiser> heh
hkaiser has quit [Quit: bye]
galabc has joined #ste||ar
<jbjnr> interesting (lenovo)
parsa has joined #ste||ar
galabc has quit [Quit: Leaving]
david_pfander has quit [Ping timeout: 260 seconds]
parsa has quit [Quit: Zzzzzzzzzzzz]
mcopik_ has joined #ste||ar
EverYoung has joined #ste||ar
EverYoung has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
anushi has joined #ste||ar
anushi has quit [Remote host closed the connection]
anushi_ has quit [Ping timeout: 260 seconds]
anushi has joined #ste||ar
<simbergm> jbjnr: which test and which branch? is this one of heller's branches?
<jbjnr> my test on my branch. sorry. gtg now
<simbergm> np
<jbjnr> I'll try other tests
nikunj has joined #ste||ar
aserio has quit [Ping timeout: 245 seconds]
victor_ludorum has joined #ste||ar
EverYoung has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
anushi_ has joined #ste||ar
anushi has quit [Read error: Connection reset by peer]
jakub_golinowski has joined #ste||ar
EverYoun_ has joined #ste||ar
EverYoung has quit [Ping timeout: 276 seconds]
jakub_golinowski has quit [Quit: Ex-Chat]
mcopik_ has quit [Ping timeout: 264 seconds]
apsknight has quit [Read error: Connection reset by peer]
EverYoun_ has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
EverYoun_ has joined #ste||ar
EverYoung has quit [Ping timeout: 245 seconds]
EverYoun_ has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
aserio has joined #ste||ar
anushi_ has quit [Remote host closed the connection]
victor_ludorum has quit [Quit: Page closed]
jaafar has joined #ste||ar
kasprov has joined #ste||ar
gablab has joined #ste||ar
zao has quit [Ping timeout: 240 seconds]
Smasher has joined #ste||ar
jaafar has quit [Quit: Konversation terminated!]
jaafar has joined #ste||ar
gablab has quit [Ping timeout: 240 seconds]
<kasprov> Hello everyone! My name is Mostafa Ibahem. I’m a 3rd year undergraduate student from Egypt and I am very excited to apply for GSoC under STE||AR.
<kasprov> I am interested in the following task :Coroutine-like Interface . Before going into the details of yield ,
anushi_ has joined #ste||ar
<kasprov> would like to have a good understanding of the current indentation process. How can I get started?
zao has joined #ste||ar
kasprov has quit [Quit: Page closed]
kasprov has joined #ste||ar
kasprov has quit [Client Quit]
nikunj has quit [Quit: Page closed]
mcopik has quit [Ping timeout: 268 seconds]
<jbjnr> heller_: just fyi - according to vtune, the #1 hotspot in my code is dgemm - which is expected, the #2 hotspot is wait_or_add_new
<jbjnr> however that's not right, because the runtime spends 30s spinning in startup/init code on 36 threads doing nothing at the start and this dominates the timing cos the work later is only 20s.
<jbjnr> so I need to tell vtun to not record at the start ...
eschnett has quit [Quit: eschnett]
<heller_> Yes, or Zoom in
hkaiser has joined #ste||ar
nikunj has joined #ste||ar
<jbjnr> hotspots doesn't have any kind of zooming
<diehlpk_work> hkaiser, Do you know if gentryx is still around?
diehlpk_work has quit [Quit: Leaving]
kasprov has joined #ste||ar
EverYoun_ has joined #ste||ar
EverYoung has quit [Ping timeout: 245 seconds]
EverYoun_ has quit [Ping timeout: 245 seconds]
EverYoung has joined #ste||ar
Smasher has quit [Remote host closed the connection]
aserio has quit [Quit: aserio]
wash has quit [Ping timeout: 260 seconds]
wash has joined #ste||ar
EverYoung has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
diehlpk has joined #ste||ar
<heller_> jbjnr: in those bars at the bottom, you should be able to select and filter
<heller_> Based on the time
<diehlpk> heller_, hkaiser I have an outline for the heise article and will transfer it from paper to the google doc
<heller_> You should see some form of spikes or jittery, those are the interesting spots to look at
<heller_> diehlpk: ok, I texted gentryx
<diehlpk> heller_, thanks
diehlpk has quit [Ping timeout: 240 seconds]
kasprov has quit [Quit: Page closed]
parsa has joined #ste||ar
EverYoun_ has joined #ste||ar
EverYoung has quit [Ping timeout: 252 seconds]
EverYoun_ has quit [Ping timeout: 240 seconds]
anushi_ has quit [Remote host closed the connection]
parsa has quit [Read error: Connection reset by peer]
anushi has joined #ste||ar
parsa has joined #ste||ar
parsa has quit [Client Quit]
diehlpk_mobile has joined #ste||ar
nikunj has quit [Quit: Page closed]
EverYoung has joined #ste||ar
<jbjnr> heller_: it's rubbish though. to get anything meaningful, I have to add -g to the cxx flags, and combining that with using vtune, the performance drops from 800gflops to 300gflops and the profiling is utterly meaningless.
<jbjnr> the #2 hotspot is always wait_or_add_new
<jbjnr> I added __itt_suspend and __itt_resume around the core of the algorithm to remove the spurious start time wasted etc.
EverYoung has quit [Read error: Connection reset by peer]
EverYoung has joined #ste||ar
diehlpk_mobile has quit [Read error: Connection reset by peer]
diehlpk_mobile has joined #ste||ar
EverYoun_ has joined #ste||ar
EverYoung has quit [Read error: Connection reset by peer]