<simbergm>
aserio: get circleci running for that repo
<simbergm>
I don't have the permissions
hkaiser has joined #ste||ar
<simbergm>
do you have the rights to do it or add me as an admin?
<aserio>
simbergm: What do you need me to click?
<aserio>
I can add you as an admin.
<aserio>
One sec
<aserio>
simbergm: you have been added as an admin
<aserio>
let me know if you need another button clicked
<simbergm>
aserio: awesome, thanks, I think it worked
RostamLog has joined #ste||ar
RostamLog has joined #ste||ar
RostamLog has joined #ste||ar
aserio1 has joined #ste||ar
aserio has quit [Ping timeout: 252 seconds]
aserio1 has quit [Ping timeout: 252 seconds]
aserio has joined #ste||ar
rori has quit [Quit: bye]
aserio has quit [Ping timeout: 252 seconds]
diehlpk has joined #ste||ar
diehlpk has quit [Ping timeout: 276 seconds]
<parsa>
diehlpk_work: ping
hkaiser has quit [Quit: bye]
diehlpk has joined #ste||ar
aserio has joined #ste||ar
eschnett has quit [Quit: eschnett]
aserio has quit [Ping timeout: 246 seconds]
eschnett has joined #ste||ar
aserio has joined #ste||ar
aserio1 has joined #ste||ar
aserio has quit [Ping timeout: 276 seconds]
aserio1 is now known as aserio
eschnett has quit [Quit: eschnett]
diehlpk has quit [Ping timeout: 276 seconds]
aserio has quit [Ping timeout: 248 seconds]
aserio has joined #ste||ar
aserio has quit [Client Quit]
akheir has quit [Quit: Leaving]
<Yorlik>
I replaced std::thread in a little test app with hpx::async and fut.wait() instead of thread::join(). The result is horribly slow by ~600x. What could be wrong ?
<Yorlik>
The tasks are supposed to run ~4 seconds and they do, but the throughput is just abysmal.
<Yorlik>
launching like this: auto prod_fut = hpx::async( hpx::launch::async, std::ref( prod ) );
<Yorlik>
using prod_fut.wait( ); instead of prod_fut.get( ); locks up.
<Yorlik>
I think I'm probably missing something very basic - I've not been working with HPX for a while.
hkaiser has joined #ste||ar
<Yorlik>
hkaiser: ?
<hkaiser>
Yorlik: here
<Yorlik>
I started HPXifying my testapp for the datastructure we were talking about.
<Yorlik>
Just swapping out std::thread for the workers against hpx::thread slowed it down 600x
<Yorlik>
Using async and futures has the same effect
<Yorlik>
So - even without any task swapping it starts crawling
<Yorlik>
using fut.wait() instead of fut.get() makes it hang always
<Yorlik>
Need a coffee ... sec ..brb
<Yorlik>
back.
<Yorlik>
I don't even know how to debug this tbh.
diehlpk has joined #ste||ar
<diehlpk>
hkaiser, Have you read the letter?
<hkaiser>
diehlpk: what letter
<diehlpk>
The response to the SC committee, see Dirk's e-mail
<hkaiser>
Yorlik: most likely there no work to keep the scheduler busy so the cores start disrupting each other
<hkaiser>
diehlpk: yah, saw that
<diehlpk>
Are you ok, so I will submit?
<hkaiser>
pls go ahead
<Yorlik>
hkaiser: the tasks are running for 4 seconds before they get stopped from the outside by sending a signal
<hkaiser>
how many tasks?
<Yorlik>
4 long rnning tasks
<hkaiser>
how many cores?
<Yorlik>
The workers running around in circles around the ringbuffer
<Yorlik>
4
<hkaiser>
Yorlik: then I don't know what's wrong
<Yorlik>
It was running 600 times faster with the OS threads
<hkaiser>
are you running a debug or a release version?
<Yorlik>
Debug - but realse has the same issue
<hkaiser>
need to see the code in order to do anything
<Yorlik>
(used RelWithDebInfo)
<Yorlik>
I rewplaced the std::thread with this: auto prod_fut = hpx::async( hpx::launch::async, std::ref( prod ) );
<hkaiser>
there is no reason why hpx should be slower if there are just 4 tasks on 4 cores running for a long time
<Yorlik>
prod is a functor
<hkaiser>
nod
<Yorlik>
collecting is done with prod_fut.get() and I also tried .wait()
<Yorlik>
wait locked up
<Yorlik>
Even get sometimes froze
<hkaiser>
shouldn't lock up, there is no reason why it should
<Yorlik>
I'm totally clueless
<hkaiser>
how am I supposed to understand the problem without seeing the code?
<Yorlik>
Would you have time for a screenshare?
<hkaiser>
not right now, sorry
<Yorlik>
I could add you to the repo if you want
<Yorlik>
heller also has access
<Yorlik>
But I shouldn't trigger double work.
<Yorlik>
If it's so weird then maybe I should wait until Thomas has time.
<hkaiser>
as you like it, either way is fine
jbjnr has quit [Read error: Connection reset by peer]
<Yorlik>
Time to properly install my hpx symbols ... :)
<Yorlik>
It looked as if it was hanging in some condition var