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/ | GSoD: https://developers.google.com/season-of-docs/
weilewei has joined #ste||ar
K-ballo has quit [Quit: K-ballo]
Yorlik has quit [Read error: Connection reset by peer]
jaafar has joined #ste||ar
<weilewei> zao I finally remembered that the reason you are seeing runtime error is that you need to add hpx_setup_target(main_dca), which I had not pushed to my branch yet
weilewei has quit [Remote host closed the connection]
<zao> That's the problem. It still doesn't do any form of wrapping, with or without hpx_main.hpp included in main_dca.cpp
<zao> Something's hecked up here.
K-ballo has joined #ste||ar
nikunj has joined #ste||ar
<heller> jbjnr_: see pm please
<nikunj> hkaiser: yt?
Yorlik has joined #ste||ar
david_pfander has joined #ste||ar
<heller> clang just iced me :/
<heller> hkaiser: we should really try to seperate the changes from the reformatting. I would like to suggest that if we touch a file, we first reformat it, commit it, and then perform the change
<heller> that way, we can select the commit and see the change immediately instead of trying to figure out what the gist of the change was
david_pfander has quit [Quit: david_pfander]
<hkaiser> heller: I agree - and sorry for not doing it
<hkaiser> I had to rebase and consolidate everything more than once which caused the mess
<heller> not talking about the serialization stuff btw ;)
<hkaiser> its difficult to go back, however - without redoing everything
<heller> the performance regression is what struck me
<heller> hmm
<heller> not sure
<heller> put the changes in a patch
<hkaiser> yah, I agree as well - I can go back and redo that
<heller> revert to master
<heller> apply the formatting
<hkaiser> will do
<heller> commit
<heller> apply the patch
<heller> commit again
<heller> that should do the trick
<hkaiser> heller: better now?
<heller> hkaiser: thanks!
<heller> hkaiser: so the regression happened when 128 bit atomics where present?
<hkaiser> yes, essentially always
<hkaiser> queue is 20slower than deque (in that particular usecase
<hkaiser> 20 percent
<heller> well, the PR still uses the queue with no 128 bit atomics
<hkaiser> sure, deque requires 128bit atomics
<hkaiser> nothing we can do there
<heller> -latomic ...
<heller> no?
<hkaiser> -latomic has no 128bit implementation on some platforms - don't ask why
<hkaiser> that was why we introduced all of this, see #3996
<heller> ok
<heller> stupid ..
<hkaiser> indeed
<heller> in any case: How did #3996 introduce a performance regression then?
<heller> and why is this fixing it?
<heller> I can see how it mit generally improve the situation though
<hkaiser> #3996 switch everything to use queue, its simbergm's fault: https://github.com/STEllAR-GROUP/hpx/pull/3996#discussion_r305695568 ;-)
<heller> hkaiser: sorry, I am not following, your change does the exact same thing, doesn't it?
<hkaiser> my change reintroduces deque
<hkaiser> when 128bit atomics are available
<heller> this, right?
<hkaiser> yes
<heller> ok, here it is gone
<hkaiser> yes
<heller> but here it is marked to be merged in
<heller> that's why I am confused...
<hkaiser> that's the first f several commits, later commits change that
<heller> bummer
<heller> ok
<heller> gtg
weilewei has joined #ste||ar
<heller> hkaiser: we clearly need properly engineered lockfree backends, boost seems not the way to go here ;)
<weilewei> zao then I am not sure what's happening, if you pull my latest branch, then our code should be the same basically.
weilewei has quit [Remote host closed the connection]
weilewei has joined #ste||ar
diehlpk has joined #ste||ar
weilewei has quit [Remote host closed the connection]
diehlpk has quit [Remote host closed the connection]
diehlpk has joined #ste||ar
jbjnr has joined #ste||ar
diehlpk has quit [Ping timeout: 276 seconds]
K-ballo has quit [Quit: K-ballo]
K-ballo has joined #ste||ar
weilewei has joined #ste||ar
jbjnr has quit [Ping timeout: 246 seconds]
weilewei has quit [Remote host closed the connection]
weilewei has joined #ste||ar