K-ballo 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/
<weilewei>
"walker" and "accumulator" are annotated in dca applications but they are captured in APEX report properly
<weilewei>
and this function is indeed being executed. Not sure why it is not being captured in APEX
<weilewei>
hkaiser ^^
<hkaiser>
hmmm
<hkaiser>
I think apex grabs the annotation only for hpx tasks
<hkaiser>
weilewei: ^
<weilewei>
ok, then.. what shall I do ?
<weilewei>
shall I wrap it into hpx future?
<hkaiser>
let me have a look at the code
<weilewei>
sure
<hkaiser>
I meant I'll look at the hpx code
<weilewei>
ok
<hkaiser>
weilewei: try putting a annotate_function f("name"); on the stack in a block surrounding the function call, i.e. { annotate_function f("name"); dca_loop_dispatch(parameters, dca_data, concurrency); }
<hkaiser>
not annotate_function, not annotated_function - those are different
<weilewei>
sure, let me try now
<weilewei>
nice! It gets capture now!
nanmiao111 has quit [Quit: Connection closed]
weilewei has quit [Quit: Connection closed]
nanmiao111 has joined #ste||ar
hkaiser has quit [Quit: bye]
jehelset has quit [Remote host closed the connection]
bita has joined #ste||ar
bita has quit [Ping timeout: 264 seconds]
jehelset has joined #ste||ar
parsa[m] has quit [Quit: Idle for 30+ days]
ramses[m] has left #ste||ar ["User left"]
hkaiser has joined #ste||ar
<hkaiser>
ms[m]: yt?
nanmiao11183 has joined #ste||ar
nanmiao111 has quit [Quit: Connection closed]
<ms[m]>
hkaiser: hey, here
<hkaiser>
hey
<hkaiser>
I have a comment/question about the segmented algo stuff
<hkaiser>
I had to mark the non-segmented algorithms with !is_segemented_iterator<>
<hkaiser>
I don't really like that
<hkaiser>
do you have an idea whether this can be avoided?
<ms[m]>
agreed
<hkaiser>
ms[m]: another option would be to specialize the segmented algorithms for the concrete iterator types
<ms[m]>
what does it take for the segmented overloads to be better matches than the non-segmented ones? would we need a base for the segmented iterators?
<hkaiser>
while that's ok (as we have only one iterator type), it's not nice either
<ms[m]>
right
<hkaiser>
an CRTP base for segmented iterators would do as well, I think
<ms[m]>
I'd look to you for solutions on this one :P that is the only way I'd know how to solve it right now, it
<ms[m]>
it's mostly a question of if every iterator type needs its own overload
<hkaiser>
that wouldn't be nice
<hkaiser>
an annoying
<ms[m]>
no, that'd quite terrible, and not really extensible
<hkaiser>
I'll play with the crtp base idea
<ms[m]>
yep
<ms[m]>
yep, sounds reasonable
<ms[m]>
should at least be technically feasible
<ms[m]>
hkaiser: I also wanted to ask you before the release: did you want to do delay the deprecation warnings on the parallel algorithms still one more release or should we leave them in this time (for 1.5.0 we bumped the deprecation warnings so they wouldn't trigger yet)?
<hkaiser>
no, let's trigger them
<hkaiser>
let's get it over with asap
<ms[m]>
👍️
<ms[m]>
the crtp might btw work nicely in the kokkos specializations as well, I currently just overload on the different policies (which is ok, since it's a smaller set than the segmented iterators, but still ugly)
<hkaiser>
right, this is a disadvantage of the tag_dispatch stuff
<hkaiser>
you have to make the overloads more specialized
hkaiser has quit [Quit: bye]
<gonidelis[m]>
any idea where to find what `enable_stealing` is?
<ms[m]>
gonidelis[m]: it refers to task stealing, grepping the schedulers module for that might give you some hits (I don't think there's anything in the docs about it)
<gonidelis[m]>
ms[m]: thanks !
<gonidelis[m]>
ms[m]: plus do you know where can i find info about the `.with()` function used in the exec_policy structs
<gonidelis[m]>
?
<ms[m]>
gonidelis[m]: depends, I suspect the only information you'll find about that inside hpx is also in the code
<gonidelis[m]>
is that a thing outside hpx?
<gonidelis[m]>
a googled execution parameters but i got nothing
<ms[m]>
There might be an older revision of p0443 that mentions it, although I'm not sure
<ms[m]>
It might be an extension that only exists in hpx
<gonidelis[m]>
huh
<gonidelis[m]>
ok
<ms[m]>
hkaiser would know right away ;)
<gonidelis[m]>
yeah i will ask him when he comes back
<ms[m]>
The equivalent of `with` in the later proposals would be properties which is in the latest p0443 (and a separate paper, don't recall the number, alternatively p2220)
<gonidelis[m]>
ms[m]: so if it exists outside hpx, then it should be somewhere in p0443
<ms[m]>
Not necessarily, but it'd be the first place where I'd look