hkaiser 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/ | GSoC: https://github.com/STEllAR-GROUP/hpx/wiki/Google-Summer-of-Code-%28GSoC%29-2020
LiliumAtratum has joined #ste||ar
<LiliumAtratum> @hkaiser, I have provided a more accurate and formally correct example for the hpx issue #4787. Unfortunately I see no option for me to reopen the issue. I would be grateful if you could have another look. Thanks!
<LiliumAtratum> that's the one with the noncommutative but associative operator for the transform_inclusive_scan
LiliumAtratum has quit [Remote host closed the connection]
<K-ballo> just do the string concatenation ones we already have...
kale[m] has quit [Ping timeout: 260 seconds]
kale[m] has joined #ste||ar
LiliumAtratum has joined #ste||ar
LiliumAtratum has quit [Remote host closed the connection]
<weilewei> it seems adding bidirectional ring algorithm in DCA will take only 75% time of original uni-directional runtime (60 ranks for 10 nodes situation)
<weilewei> more cleanup is needed, however...
RostamLog has joined #ste||ar
<hkaiser> weilewei: nice result!
<weilewei> hkaiser thanks!
hkaiser has quit [Quit: bye]
K-ballo1 has joined #ste||ar
K-ballo has quit [Ping timeout: 265 seconds]
K-ballo1 is now known as K-ballo
Yorlik has quit [Ping timeout: 260 seconds]
nanmiao11 has quit [Remote host closed the connection]
kale[m] has quit [Ping timeout: 260 seconds]
kale[m] has joined #ste||ar
bita__ has quit [Ping timeout: 260 seconds]
kale[m] has quit [Ping timeout: 244 seconds]
kale[m] has joined #ste||ar
nikunj97 has joined #ste||ar
Nikunj__ has joined #ste||ar
nikunj97 has quit [Ping timeout: 260 seconds]
kale[m] has quit [Ping timeout: 260 seconds]
kale[m] has joined #ste||ar
kale[m] has quit [Ping timeout: 240 seconds]
kale[m] has joined #ste||ar
kale[m] has quit [Ping timeout: 265 seconds]
kale[m] has joined #ste||ar
kale[m] has quit [Ping timeout: 256 seconds]
Nikunj__ has quit [Read error: Connection reset by peer]
Nikunj__ has joined #ste||ar
kale[m] has joined #ste||ar
nikunj97 has joined #ste||ar
Nikunj__ has quit [Ping timeout: 260 seconds]
hkaiser has joined #ste||ar
kale[m] has quit [Ping timeout: 246 seconds]
kale[m] has joined #ste||ar
kale[m] has quit [Ping timeout: 260 seconds]
kale[m] has joined #ste||ar
kale[m] has quit [Ping timeout: 260 seconds]
kale[m] has joined #ste||ar
<gonidelis[m]> Also as I am still getting familiar with the cluster, why can't I git fetch upstream?
kale[m] has quit [Ping timeout: 260 seconds]
kale[m] has joined #ste||ar
<gonidelis[m]> Also where is my .gitconfig file on the cluster? ... ahh I am confused
<gonidelis[m]> And lastly, does master compile for you guys?
<ms[m]> gonidelis: yes, things are compiling ok at least for me (and ci)
<ms[m]> is that the error you have up there? is that on master with a clean build directory?
<gonidelis[m]> ms[m]: thanks for your response... let me try with a clean build (it should take a while)
<gonidelis[m]> and speaking of "taking a while" do you know why can't I `git fetch upstream` on my rostam git repo?
<gonidelis[m]> this is what I get
<ms[m]> gonidelis: not sure what could be the problem there... does `git remote show upstream` show something useful?
<ms[m]> could be a firewall issue? is it only for upstream? does your the remote of your own fork work fine?
<gonidelis[m]> ahh my bad... i think origin is at stellar which mean I git cloned from stellar instead of my fork
<gonidelis[m]> I am sory
kale[m] has quit [Read error: Connection reset by peer]
kale[m] has joined #ste||ar
<kale[m]> Is highfive a necessary dependency of Phylanx ?
<gonidelis[m]> Do we know what's going on with the `is_deferred_callable` `deferred_result_of`warnings???
<ms[m]> gonidelis: this should fix that https://github.com/STEllAR-GROUP/hpx/pull/4790
kale[m] has quit [Ping timeout: 240 seconds]
kale[m] has joined #ste||ar
kale[m] has quit [Ping timeout: 256 seconds]
kale[m] has joined #ste||ar
<gonidelis[m]> hkaiser: Should `is_sentinel_for` used in the other specializations (segmeneted and detail) except from here?
<gonidelis[m]> I reckon it should....
<hkaiser> kale[m]: it's optional
<hkaiser> gonidelis[m]: we should use it wherever a iterator/sentinel pair is supported
<hkaiser> I don't think we have adapted any of the segmented algorithms however
<hkaiser> ms[m]: btw, the hpx-docs domain seems to work now: https://hpx-docs.stellar-group.org/docs/sphinx/latest/html/index.html
<hkaiser> looks like we need to adapt some of our links, however, the readme redirects to a different directory
<gonidelis[m]> hkaiser: oh ok.. I don't really know what segmented algos do...
<hkaiser> gonidelis[m]: they work with special containers that have their data segmented (partitioned), possibly across more than one HPX locality
<gonidelis[m]> wow great! Thank you :)
Yorlik has joined #ste||ar
<ms[m]> hkaiser: hmm, that's the old domain (when the docs were still the hpx repo...) any chance you can make it point to https://stellar-group.github.io/hpx-docs/branches/master/html/index.html?
<hkaiser> I wouldn't know where I should change that
<hkaiser> all I did was referring stellar-group.github.io to hpx-docs.stellar-group.org
<hkaiser> the fact that the old pages work using the new domain shows that the redirection is working
<hkaiser> ms[m]: where are those directories come from: hpx-docs/branches/?
<ms[m]> where is that set up? I see there's something in the github pages section on the hpx repo... can we do the same on the hpx-docs repo?
<ms[m]> I tried setting it there but get "The CNAME `hpx-docs.stellar-group.org` is already taken."
<ms[m]> the hpx-docs repo
<hkaiser> the redirection is done on the hpx-docs repo and on the site where stellar-group.org is hosted
<ms[m]> I see no custom domain on the hpx-docs repo...
<hkaiser> ahh!
<hkaiser> we have configured it on the wrong repository
<hkaiser> my fault
<ms[m]> hkaiser: amazing! thanks :D
<ms[m]> I'll update the links
<hkaiser> thanks!
<ms[m]> they're nice and short now :)
<hkaiser> https://hpx-docs.stellar-group.org/ works as well ;-)
<gonidelis[m]> hkaiser: https://github.com/STEllAR-GROUP/hpx/pull/4792 sorry for being late
akheir has joined #ste||ar
kale[m] has quit [Ping timeout: 265 seconds]
nanmiao11 has joined #ste||ar
kale[m] has joined #ste||ar
<K-ballo> hkaiser: I don't quite grasp the situation in #4786, too many things intertwined at once.. what's the divergence there?
<hkaiser> K-ballo: we believe the standard is wrong ;-)
<K-ballo> that seems to be only one part of it
<K-ballo> there's like 17 different concerns all at once
<hkaiser> if we could agree on how to interpret the standard everything should fall in place
<hkaiser> the offending piece is here: http://eel.is/c++draft/transform.inclusive.scan#1
<hkaiser> and then 2.3 saying that binary_op() should return something convertible to U
<hkaiser> there are two possible ways to fix this obvious mistake
<K-ballo> two?
<hkaiser> either say that U is decltype(*first) or decltype(unary_op(*first))
<hkaiser> msvc and clang apparently chose the latter
<K-ballo> U is iterator_traits<decltype(first)>::value_type
<K-ballo> the value_type of the iterator
<hkaiser> yes, that would be the first option
<K-ballo> not quite, but ok
<K-ballo> it's not a mistake
<hkaiser> sure
<K-ballo> the value type of the iterator is iterator_traits<Iter>::value_type
<hkaiser> I agree with that, but it says U should be decltype(first), i.e. the iterator type itself
<K-ballo> no no, it says U should be the *value type* of decltype(first)
<hkaiser> grrr
<hkaiser> missed that :/
<K-ballo> me too, subtle
<hkaiser> but then msvc and clang do it wrong
<K-ballo> what do they do?
<hkaiser> they assume U to be unary_op(value_type)
<K-ballo> how is that even observable?
<hkaiser> it wouldn't compile otherwise
<hkaiser> U is the type used for the internal accumulator variable
<K-ballo> how is that observable?
<hkaiser> it's not, but it wouldn't compile if it wasn't
<K-ballo> I don't follow
<hkaiser> if you use the value_type of first as the internal accumulator type, then binary_op has to return something convertible to it
kale[m] has quit [Ping timeout: 240 seconds]
<hkaiser> unary_op as well, btw
<hkaiser> but the code in the comment does not provide a conversion path from Integer to int
<hkaiser> compiles anyways
<hkaiser> the code in the comment can compile only if the accumulator value's type is whatever unary_op returns
<K-ballo> ok
<weilewei> hkaiser meeting today?
<hkaiser> yes
kale[m] has joined #ste||ar
<K-ballo> using the transformed type as accumulator is the only sensible interpretation, to be able to achieve what we wanted with the transform_* variants
<hkaiser> nod
<hkaiser> that's what I've implemented now
<K-ballo> "interpretation" in a very loose sense
<hkaiser> sure, I think we agree
<hkaiser> and msvc/clang agree as well
<K-ballo> https://wandbox.org/permlink/KQjZNICPb4RxHAOO a more approachable example
<K-ballo> hkaiser: I don't see reflected on the issue the fact that the binary predicate was being called twice for each pair
<K-ballo> reportedly, for every invocation of binary_op(x, y) there was also binary_op(y, x)
<hkaiser> K-ballo: who reported that?
<K-ballo> lithium something
<zao> `LiliumAtratum`
<zao> ?
kale[m] has quit [Ping timeout: 240 seconds]
kale[m] has joined #ste||ar
<hkaiser> K-ballo: ok - for the parallel version this makes sense
<hkaiser> the sequential version should invoke binary_op exactly N times
<hkaiser> however the swapped arguments are a bug (fixed on that PR)
<K-ballo> LiliumAtratum, that's it, thanks zao
<K-ballo> let's get them to file an LWG issue
<K-ballo> the constrain was introduced by P0574
bita__ has joined #ste||ar
kale[m] has quit [Ping timeout: 260 seconds]
nikunj97 has quit [Ping timeout: 240 seconds]
<weilewei> If I would like to measure libcds overhead with HPX support, does this code look good? https://gist.github.com/weilewei/e060c069135c37f5dd5d7ba7b0310143
kale[m] has joined #ste||ar
bita__ has quit [Ping timeout: 244 seconds]
nanmiao11 has quit [Ping timeout: 245 seconds]