14:40 <Yorlik> Hello hkaiser: YT?

14:57 <Yorlik> Is there some functionality in HPX which helps / which I should use if I want to create global sets and set operations, like merge / intersect, etc.?

14:57 <Yorlik> Basically I want to implement global tagged object search.

15:01 <hkaiser> Yorlik: we have a number of collective operations, barrier, reduce, altoall, scatter/gather, etc.

15:02 <hkaiser> https://github.com/STEllAR-GROUP/hpx/tree/master/libs/full/collectives/include/hpx/collectives

15:08 <Yorlik> hkaiser: thanks: I'll look into these functions/objects.

15:58 <Yorlik> Do we have a comprehensive example how collectives work? Or do I have to dig the source? The only reference I found in the examples was a communicator in the stencil examples.

15:59 <Yorlik> It's basically a framework for scatter-gather semantics - is that correct?

16:00 <hkaiser> Yorlik: the tests should show things

16:01 <hkaiser> also, the semantics are close to the corresponding MPI functions

16:01 <Yorlik> hkaiser: Allright - thank you!

21:48 <gnikunj[m]> hkaiser: started with 3d stencil code from Keita and as usual, it doesn't execute properly :/

22:09 <gonidelis[m]> https://ibb.co/VwxSYqw Any idea how do I get info about the internals of hpx_main using APEX?

22:09 <gonidelis[m]> hkaiser: ^^

22:09 <gnikunj[m]> gonidelis: what part are you interested in?

22:10 <gonidelis[m]> gnikunj: part?

22:10 <gnikunj[m]> when you say "internals of hpx_main" do you mean the call stack before hpx_main or your code within hpx_main?

22:10 <gonidelis[m]> within

22:11 <gonidelis[m]> given that all my program logic is within it

22:11 <gnikunj[m]> is there a specific counter you're looking for?

22:12 <gonidelis[m]> ahh probably now

22:12 <gonidelis[m]> no*

22:12 <gonidelis[m]> what do you mean counter?

22:13 <gnikunj[m]> IIRC using APEX counters require you to include APEX and then use APEX function calls before and the instruction you're interested in for accurate results. HPX does have a list of common counters available as well but they won't tell you the counter results for the new things you define (for instance your own function call)

22:13 <gonidelis[m]> ahh

22:13 <gonidelis[m]> darn ok

22:13 <gonidelis[m]> i need to embed those apex functions then

22:13 <gnikunj[m]> Patrick Diehl would know better know. He's used it more than me. I just used it once for profiling purposes.

22:14 <gonidelis[m]> do you remember where you found these apex functions reference?

22:14 <gnikunj[m]> gonidelis[m]: essentially yes. If you don't want any of this, try using VTune. It returns the results for literally every instruction.

22:15 <gnikunj[m]> gonidelis: http://uo-oaciss.github.io/apex/

22:15 <gonidelis[m]> gnikunj: they literally got nothing in there

22:15 <gonidelis[m]> ^^

22:16 <gnikunj[m]> Aah, then Patrick Diehl is your best bet

23:07 <gonidelis[m]> gnikunj: hkaiser vtune is ridiculous

23:07 <gonidelis[m]> why the heck do even other profilers exist

23:07 <gnikunj[m]> hahahaha VTune is overkill for most purposes, yes

23:07 <gnikunj[m]> but VTune is only for Intel and is expensive :P

23:07 <gnikunj[m]> also, it can't do distributed profiling so applications like Vampir exists

23:08 <gonidelis[m]> i have a ryzen

23:08 <gonidelis[m]> and it works with certain restrictions ;)

23:08 <gnikunj[m]> ok, idk how it worked for you then lol

23:08 <gnikunj[m]> maybe they're limited to x86 and not Intel then

23:09 <gonidelis[m]> yy i ge ta huge error that says "BUY INTEL YOU CHEAPSHOT"

23:09 <gonidelis[m]> but i can skip it and at least check my hotspots

23:09 <gonidelis[m]> gnikunj[m]: exactly

23:09 <gnikunj[m]> nice, glad I could be of help

23:10 <gonidelis[m]> certainly