weilewei has quit [Remote host closed the connection]
nan11 has quit [Remote host closed the connection]
Amy1 has quit [Ping timeout: 246 seconds]
<Yorlik>
simbergm: I didn't try. Compiling HPX and figuring out combinations with jemalloc on my machine takes quite some time, So I choose the commit, because it was reasonably far behind to potentially work for me. I tested another ~14 days earlier which had other problems.
weilewei has joined #ste||ar
nan11 has joined #ste||ar
wate123_Jun has quit [Remote host closed the connection]
<hkaiser>
Yorlik: don't try things
<hkaiser>
the only thing we need to know what variable is accessed while not being constructed yet
<Yorlik>
I posted the stack trace earlier when it busted.
<hkaiser>
that doesn't show what variable is being accessed
<hkaiser>
to explain: C++ doesn't guarantee the sequence of constructor calls for global variables
<Yorlik>
Dependencies between globals == bad maybe?
<hkaiser>
so if the initialization of one variable accesses another one which has not been initialized yet you see the problem
<hkaiser>
yes
<hkaiser>
there is an easy fix however
<hkaiser>
but for that we need to know what variable actually is uninitialized
<Yorlik>
Lemme rol back and crash it
<hkaiser>
Yorlik: the problem might have shown up as a consequence of us modularizing things which might cause changes in how the globals are being initialize
<hkaiser>
d
<Yorlik>
The sausage in the making ... :)
<hkaiser>
Yorlik: also, I don't think that this is a jemalloc problem
<Yorlik>
Me too - it was just a coincidence it popped up.
<hkaiser>
simply things that have worked before by pure chance start breaking now
<simbergm>
you need HPX_WITH_PAPI=ON (I think there's also APEX_WITH_PAPI but I don't know what APEX does with it)
aalekhnigam has quit [Remote host closed the connection]
<nikunj97>
simbergm, ahh nice! this is what i wanted
<hkaiser>
simbergm: would you mind approving #4532, please?
<hkaiser>
our sponsor wants to see their contributions ;-)
<simbergm>
hkaiser: yep, sorry :P
<simbergm>
done
<simbergm>
and thanks
<simbergm>
does that change in the link we won't be getting rostam back anytime soon, or is it just "for now"?
<simbergm>
I might also update the cscs contribution there (or really, pasc funding)
nikunj has quit [Read error: Connection reset by peer]
nikunj has joined #ste||ar
nikunj has quit [Read error: Connection reset by peer]
nikunj has joined #ste||ar
aalekhnigam has joined #ste||ar
<hkaiser>
simbergm: once we do we can change it back to rostam
<hkaiser>
simbergm: also, I will create a sponsors page on hpx.stellar-group.org and remove the whole section from the readme - so we can refer to it from anywhere (the JOSS paper for instance)
<simbergm>
hkaiser: very nice, thanks!
mukund has joined #ste||ar
nikunj has quit [Ping timeout: 260 seconds]
<mukund>
Hi everyone! I had raised a PR in GitHub regarding documenting the hpxcxx compiler script
<mukund>
I am not able to add examples for using the script
<mukund>
Very new to libraries and binaries and the whole of it
<K-ballo>
hi mukund!
nikunj has joined #ste||ar
<mukund>
So I'm not able to grasp the concepts even
<mukund>
I took too long coz I was a bit busy with other works
<mukund>
I'm sorry for that
nikunj has quit [Ping timeout: 265 seconds]
<heller1>
So you have a specific question?
nikunj has joined #ste||ar
<mukund>
I have no idea on how to build applications and components through hpxcxx and I am unable to grasp what is what
<mukund>
So I am requesting for examples
<heller1>
diehlpk: how is that markup for the joss paper being compiled?
<heller1>
mukund: the simplest application is a single cpp file containing `int main() {}`. I think there are plenty single source applications in the examples directory. Did you give that a try?
<mukund>
heller1: what about components?
<mukund>
also, what other files come under applications? only .cpp files?
<heller1>
Components usually come in the form of shared libraries
<heller1>
Are you familiar on how to do that with plain gcc or clang?
<mukund>
No I'm not, sorry
<mukund>
I'm very new to this
wate123_Jun has joined #ste||ar
<heller1>
Ok, I suggest to go through those basics first
<mukund>
Can you please direct me to some helpful links or something of that sort?
<Amy1>
who knows IRC channel about openmp or parallel computing?
<nikunj97>
well then use the intrinsic to reduce
<hkaiser>
*shiver*
<hkaiser>
but I guess once you despaired enough to use openmp, using intrisics is not much worse anyways...
Nikunj__ has joined #ste||ar
<Amy1>
no
<Amy1>
no
<Amy1>
hello
<Nikunj__>
hkaiser, isn't that a bit extreme :P
<heller1>
No
<hkaiser>
lol
<Nikunj__>
lol
<Amy1>
jifsdji
nikunj97 has quit [Ping timeout: 258 seconds]
<Amy1>
hello
<heller1>
What's up?
<Amy1>
oh, I thought I can't send word just now.
<Amy1>
sorry
<simbergm>
Amy1: the #hpc channel on https://cpplang.now.sh/ might be a better place to ask for advice
<simbergm>
reading the openmp manual is another place you can look
<simbergm>
a third is using vectorization reports from your compiler (I think at least icc and clang give you one, check their manpages for the options)
<Amy1>
simbergm: how to join in
<Amy1>
input my email?
<simbergm>
Amy1: yes
<Amy1>
it shows no perm
<Amy1>
no_perms
aalekhnigam has quit []
<diehlpk_work>
hkaiser, will you join the SC paper meeting
<hkaiser>
diehlpk_work: sorry no
<hkaiser>
Katie will be there
<diehlpk_work>
ok, will write you some small update
bita has joined #ste||ar
bita has quit [Read error: Connection reset by peer]
<heller1>
There are lot of things that you can't express well with openmp
<nikunj97>
heller1, I see
<nikunj97>
diehlpk_mobile, what's orcid in JOSS?
K-ballo has quit [Ping timeout: 256 seconds]
K-ballo1 has joined #ste||ar
rtohid has joined #ste||ar
K-ballo1 is now known as K-ballo
K-ballo has quit [Remote host closed the connection]
K-ballo has joined #ste||ar
<hkaiser>
Yorlik: ok, so register_coalescing_for_action<Action>::instance_ is the culprit
<Yorlik>
So something has changed from that commit that works for me to the current stable that breaks my debug runtime
<heller1>
Make it constexpr constructable?
<hkaiser>
or wrap it into a function as a function-local
<hkaiser>
not sure if it's actually consexpr constructable
<hkaiser>
as a *function-local static* that is
<hkaiser>
Yorlik: let me do that so you can try
<Yorlik>
Allright. It's time for my beaty-nap - I'll be back later and read up.
<Yorlik>
beauty-nap ... or my beard gets even more grey ... ;) ... bbl.
nikunj has quit [Read error: Connection reset by peer]
nikunj has joined #ste||ar
<hkaiser>
Yorlik: you simply have an outdated version of HPX, this problem was fixed a while ago (see #4486, merged 2 weeks ago)
rtohid has quit [Remote host closed the connection]
K-ballo has quit [Remote host closed the connection]
K-ballo has joined #ste||ar
<diehlpk_work>
hkaiser, see pm
gonidelis has quit [Remote host closed the connection]
rtohid has joined #ste||ar
<tiagofg[m]>
Hello everyone, sorry to come back to this topic (channel topic), but just to clarify, when I execute this code, a get() that is waiting in a channel for a set(), all physical CPU cores work at 100%.
<tiagofg[m]>
hkaiser: I found this, like you said, hpx::threads::set_scheduler_mode, but I don't know how to use it, in order to put the get() call executing in passive mode.
<tiagofg[m]>
I am using hpx 1.4.1 version on CentOS 7
<hkaiser>
tiagofg[m]: the scheduler mode has no direct relation to future::get
<hkaiser>
the scheduler_mode can be used to instruct th escheduler to step back faster than normal, putting the scheduling thread to sleep almost right way if no work is available
<tiagofg[m]>
I see, but how can I do that?
<tiagofg[m]>
I've tried put this line of code but I think is more complicated than that.
<zao>
You can hover the truncated status message on AppVeyor to see non-truncated line too.
<zao>
"... non-mergeable pull request"
<bita>
got it
<heller1>
tiago.fg: do you use hpx_init or hpx_main?
mcopik has joined #ste||ar
mcopik has quit [Client Quit]
Hashmi has joined #ste||ar
<tiagofg[m]>
heller: I use hpx_main, but with hpx_init the result is the same
<hkaiser>
tiagofg[m]: could you explain please why the 100% CPU usage bothers you?
bita has quit [Quit: Leaving]
<tiagofg[m]>
sure. When I do a get() in a future, and the future isn't ready, the tread wait for that future to become ready, in idle mode. In the channel case, this does not happen.
<tiagofg[m]>
As an example, if I have a server receiving messages from several client processes, and these clients don't send messages, it gives the illusion that the server is working
<tiagofg[m]>
and there's an unnecessary energy waste
<heller1>
The flag needs to be set before hpx_init is being called
<tiagofg[m]>
heller: If I put "hpx::threads::set_scheduler_mode(hpx::threads::policies::scheduler_mode::fast_idle_mode);" in C++ main function, before "hpx::init();", I got Segmentation fault error :/
<hkaiser>
heller1: not necessarily
<hkaiser>
tiagofg[m]: the 100% CPU is a red herring as the CPU is _always_ running at 100%, it's just that normally the OS tools don't attribute the use to the process in question (it's running 100% in the kernel)
<hkaiser>
tiagofg[m]: so can you explain why do you think having the cpu runing at 100% is a bad thing?
<zao>
hkaiser: Actual utilization affects things like turbo and whatnot, if pedantic :)
<hkaiser>
zao: well, sure - energy might be a concern - but - really?
<hkaiser>
on an HPC system?
<zao>
Might matter a bit if you've got a hybrid code where there's other threads or system processes that could use some time.
<hkaiser>
zao: yes, but in HPX?
<tiagofg[m]>
Ok I see your point
<zao>
Well, you've got different frequencies based on how many cores you're using and what instruction set they're on.
rtohid has left #ste||ar [#ste||ar]
<hkaiser>
zao: I understand
<zao>
On the subject of actual utilization, HPX has some decent counters and instrumentation to figure out how efficient it actually is, right?
<hkaiser>
tiagofg[m]: but I will try to have a look - fast_idle_mode should reduce the use
<hkaiser>
we might have broken something if it doesn't
<hkaiser>
zao: yes, you can look at idle-rates and what not
<tiagofg[m]>
I'm thinking a problem with having the cpu at 100% is the heat produced.
<hkaiser>
tiagofg[m]: yah, as said, it might have some energy impact
<tiagofg[m]>
I already tryed and didn't work for me
<hkaiser>
ok, I'll have a look
<hkaiser>
this doesn't block you, I presume, does it?
<tiagofg[m]>
But thank you a lot for the discussion and support
<tiagofg[m]>
more or less, but isn't a problem for now :)
weilewei has quit [Remote host closed the connection]
Hashmi has quit [Quit: Connection closed for inactivity]