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/
nanmiao has joined #ste||ar
nanmiao has quit [Quit: Connection closed]
K-ballo has quit [Quit: K-ballo]
hkaiser has quit [Quit: bye]
diehlpk_work has quit [Ping timeout: 276 seconds]
mortenbirkelund has joined #ste||ar
mortenbirkelund has left #ste||ar ["Leaving"]
<gnikunj[m]>
I just realized that we've overengineered Fibonacci to the core in hpx/examples
<ms[m]>
srinivasyadav227, hkaiser, I can change the gcc 11 setup to do c++20
<ms[m]>
it makes even more sense when we require c++17 support
<srinivasyadav227>
<ms[m] "srinivasyadav227, hkaiser, I can"> yes, that could help a lot!
<ms[m]>
since as you say the build is broken with c++20 I'm not going to merge that change separately, but it's better to have it changed together with your branch
<srinivasyadav227>
ms: okay cool! ;-)
hkaiser has joined #ste||ar
<pedro_barbosa[m]>
diehlpk_work: not sure if you noticed but I sent a link to a forum thread with what I was trying to do
<hkaiser>
gnikunj[m]: haha, that's the point of this example ;-)
K-ballo has joined #ste||ar
<hkaiser>
ms[m]: I would like to start making p443r13 executors available for everything in the API (i.e. as an alternative to the existing p443r3 executors)
<hkaiser>
ms[m]: would you have some time at some point to talk about that?
<hkaiser>
for that I was planning to reimplement/overload our executor CPOs (i.e. post, async_execute, etc.) for the new executors, does this sound like a valid aproach?
<ms[m]>
hkaiser: free anytime (if you'd prefer a call), but yes, I think that'd be just fine
<ms[m]>
there's not much to it (I think), one of the questions is if it should be done only for executors, or also schedulers (I think both can make sense)
<ms[m]>
and all the bulk stuff is currently missing so I suppose you'd get post/async_execute/potentially then_execute actually customized and the rest would be synthesized
<ms[m]>
execute is basically post, schedule doesn't have a direct equivalent
<ms[m]>
schedule | transform | make_future is roughly async_execute
<hkaiser>
ms[m]: nod, that's what I thought
<hkaiser>
ms[m]: does when_all have a variant that takes a vector (in addition to the variadic version)?
<ms[m]>
nope (i.e. not in p1897 so I haven't added it), but it would probably make sense
<hkaiser>
ms[m]: might simplify creating the bulk versions
<ms[m]>
yep
<mdiers[m]>
hkaiser: Have you tested #5117 with release build? 😉
<hkaiser>
mdiers[m]: yep, still no luck
<hkaiser>
:/
<mdiers[m]>
What environment/distribution are you using?
<srinivasyadav227>
ok ;-), from past 3-4 hours im banging my head with macros, haha,, they are scary !!
<hkaiser>
they are ... different - indeed
<K-ballo>
you should forget everything you know about the language, and start thinking entirely in terms of token pasting
<K-ballo>
or... if possible.. just don't do macros
<hkaiser>
srinivasyadav227: the boost preprocessor library is a good resource for this kind of stuff
<srinivasyadav227>
<hkaiser "srinivasyadav227: the boost prep"> yes, thanks very much, it just gave some nice idea how things are happening with macros!
<mdiers[m]>
<hkaiser "mdiers: nod - I was thinking tha"> Can -fstack-protector help me? Or do you have another solution for me?
nanmiao has joined #ste||ar
<mdiers[m]>
In my example, I don't see any place where there is particularly much on the stack.
<hkaiser>
mdiers[m]: recursion and continuations can do that to you
diehlpk_work has joined #ste||ar
weilewei has joined #ste||ar
weilewei has quit [Quit: Ping timeout (120 seconds)]
weilewei has joined #ste||ar
<hkaiser>
mdiers[m]: sorry, I'm back now
<hkaiser>
mdiers[m]: to explain, HPX is prone to stack overflows as the HPX threads are run on a relatively small stack segment
<hkaiser>
could you try adding --hpx:ini=hpx.stacks.small_size!=0x10000 (or larger, if needed) to your commend line?
<hkaiser>
note, the value is hexadecimal
weilewei has quit [Quit: Connection closed]
<mdiers[m]>
<hkaiser "could you try adding --hpx:ini=h"> I'm back now. Yes I test it on monday Thanks!
<pedro_barbosa[m]>
is there a way to use nvprof with HPXCL?
<hkaiser>
pedro_barbosa[m]: sure, it's a normal cuda application
<pedro_barbosa[m]>
oh ok, just had to run it with sudo apparently
<pedro_barbosa[m]>
by default the performance counters aren't available to non admin users
<gnikunj[m]>
hkaiser: yt?
<hkaiser>
gnikunj[m]: here
<gnikunj[m]>
could you help me with my slides? I need some feedback
<gnikunj[m]>
I'm confused if I should have more examples and explanations or if I should go towards executors and stuff (emailed you the slides)
<gnikunj[m]>
(you can provide a feedback later down the day. I just want a 2nd opinion on it so that I can add stuff last minute)
<hkaiser>
by when do you need that?
<gnikunj[m]>
by evening today
<gnikunj[m]>
so that I have a couple of hours to make changes
<hkaiser>
gnikunj[m]: looks good, some conclusions might be in order, though
<gnikunj[m]>
yeah, I'm working on those. It's a 1h talk that I wish to make interactive so I have less stuff on slides and want to discuss more through doubts and questions from the audience
<gnikunj[m]>
hkaiser: should I add more examples/elaborate on explaining more?
<hkaiser>
gnikunj[m]: I doubt you will have the tme for more
<gnikunj[m]>
yeah, that's what I felt like too. Should I go ahead and make final cosmetic changes and be done with it then?
<hkaiser>
nod, sounds like a good plan ;-)
<hkaiser>
good luck!
<gnikunj[m]>
thanks! I'll advertise the heck out of HPX :D
<gnikunj[m]>
the greatest library everyone should be using!
<gonidelis[m]>
gnikunj[m]: so you started working towards "the cause"
<gnikunj[m]>
I've been working towards "the cause" before you joined the group ;)
<gonidelis[m]>
namely, deliver HPX to the rookies
<gnikunj[m]>
yes! This will be a mass advertisement this time around :D
<gonidelis[m]>
that's great
<gonidelis[m]>
!
jehelset has joined #ste||ar
nanmiao has quit [Quit: Connection closed]
nanmiao has joined #ste||ar
<gnikunj[m]>
hkaiser: nice panel ;)
<K-ballo>
panel on what?
<gonidelis[m]>
K-ballo the future of boost
<K-ballo>
oh
hkaiser has quit [Read error: Connection reset by peer]