hkaiser changed the topic of #ste||ar to: STE||AR: Systems Technology, Emergent Parallelism, and Algorithm Research | stellar-group.org | HPX: A cure for performance impaired parallel applications | github.com/STEllAR-GROUP/hpx | This channel is logged: irclog.cct.lsu.edu | Everybody: please respond to the documentation survey: https://forms.gle/aCpNkhjSfW55isXGA
<gnikunj[m]>
srinivasyadav227: hey! You want do a quick meeting today?
<srinivasyadav227>
gnikunj: yea i am ready!
<gnikunj[m]>
BTW: I got permissions on the Julich clusters. I can run your benchmarks there on newer Arm PCs.
<gnikunj[m]>
srinivasyadav227: great. Post lunch at 2PM?
<srinivasyadav227>
woah!, thats great! ;-)
<srinivasyadav227>
yup, sure, perfect for me ;-)
<gnikunj[m]>
cool! See you at 2!
<srinivasyadav227>
ok!
<gnikunj[m]>
ms: a quick side note to the above question. I'm working on different locking strategies to compare thread contention and finally decide on integrating the best performing lock back to HPX. I'm writing benchmarks that may use conditional variables hence the question.
ms2 has joined #ste||ar
<ms[m]>
gnikunj: I think the base implementation in detail can take any lock
<gnikunj[m]>
ohh, let me check
<ms[m]>
and my guess for why the non-detail one isn't templated is that std::condition_variable::wait* isn't templated either, it just takes a std::mutex
<gnikunj[m]>
Right, I agree. I was simply wondering if there's an internal implementation in HPX that I can use for the testing. Otherwise, I would've had to implement one myself.
<ms[m]>
gnikunj: btw, I'm unsure if my reply the other day to your conan question came through, so in case it didn't, the answer is no demand nor supply so far
<ms[m]>
now there seems to be at least some demand
<gnikunj[m]>
ms: It didn't come through. I've been asked a couple of times in the last few days so I thought of having one. Spack isn't popular in the general dev community so people on linux usually have to build from source.
<gnikunj[m]>
vcpkg works great on windows. Also, gitlab let's you create repositories as conan package that can be used in any other repository as a dependency. Having HPX as a package there would certainly help with CI/CD pipelines.
<ms[m]>
gnikunj: noone's stopping you ;) it just needs a maintainer
<gnikunj[m]>
ms: I actually looked into it ;-)
<ms[m]>
👍️
<gnikunj[m]>
The only problem here is hwloc (which is required by HPX) is not available on conan. All other dependencies are available, so that complicates things slightly. But, let's see how it goes :D
<ms[m]>
rachitt_shah: ok with me and I hope your friend will be all right
<ms[m]>
gnikunj: see above, ok with you?
<ms[m]>
let's wait for hkaiser to respond as well
<ms[m]>
Kor de Jong: thank you!
<gnikunj[m]>
rachitt_shah: It's fine. I hope your friend has a speedy recovery
<jedi18[m]>
ms: unintialized_copy and uninitialized_fill PRs are ready, the failing tests https://github.com/STEllAR-GROUP/hpx/pull/5402 seem to have been fixed after the rebase right?
<rachitt_shah[m]>
Thank you ms gnikunj I'll update the wiki tonight.
<rachitt_shah[m]>
My friend broke my wrists unfortunately
<ms[m]>
rachitt_shah: yours or your friend's? :S
<ms[m]>
yes please, it'd be nice if you can still update the wiki
<ms[m]>
fyi (and gnikunj) I have a small build working with exhale
<ms[m]>
the results are mixed, but possibly better than before
<ms[m]>
jedi18: yep, looks good now
<ms[m]>
thanks!
<gnikunj[m]>
that's amazing, how does the docs look?
<gnikunj[m]>
ms: ^^
<ms[m]>
gnikunj: I'll upload something a bit later, then you can have a look yourself
<ms[m]>
;)
<gnikunj[m]>
thanks!!
<ms[m]>
jedi18: just merged uninitialized fill and copy, thanks for working on them!
<ms[m]>
it's still uploading... so the css is missing
<ms[m]>
might be done in an hour 🙄
<ms[m]>
that's only the algorithms, futures, and synchronization modules, so things are missing, but gives you an idea of the (default) structure that exhale generates
<ms[m]>
I'll also rebuild with the sphinx_rtd_theme but you'll need to wait a bit for that as well
<gnikunj[m]>
ms: I'll wait for the CSS to load. At the moment, it isn't very pleasing to browse :P
<ms[m]>
yep, it's not massively better with the css either
<ms[m]>
navigation is a bit funky...
<ms[m]>
note: the bullet points are clickable (at least some of them)
<ms[m]>
I think on the sphinx_rtd_theme they actually look expandable
<gnikunj[m]>
the clickable points are taking me to a separate page as well
hkaiser has joined #ste||ar
K-ballo has joined #ste||ar
hkaiser has quit [Quit: Bye!]
<gnikunj[m]>
ms: is there an equivalent to thread_local in HPX? I finally found why my MCS lock was deadlocking. That was because I was using thread_local to implement the MCS lock but thread_local here is based on pthreads and not HPX threads.
<gnikunj[m]>
I see a build time option: HPX_WITH_THREAD_LOCAL_STORAGE:BOOL -> "Enable thread local storage for all HPX threads (default: OFF)". Is it the same thing that I'm asking?
<zao>
gnikunj[m]: It's used for the "tss" thread-specific storage system, which is kind of what you want.
<gnikunj[m]>
zao: Thanks! Let me have a look. It's a bummer that it's off by default. Really curious, why?
<zao>
Don't know, might require support in the context runtime, and may have non-zero cost just being enabled.
<zao>
If I had to guess.
<ms[m]>
that's a pretty good guess (and as good a guess as mine)
<ms[m]>
we can ask hartmut in the meeting
<gnikunj[m]>
Good idea. Will do!
hkaiser has joined #ste||ar
hkaiser has quit [Quit: Bye!]
hkaiser has joined #ste||ar
<srinivasyadav227>
hkaiser: could you please open a ticket on moving util::loop & transform_loop things to parallel::detail , i could start working on it and it would be easy for me creating PRs for the datapar algorithms that i have adapted. thanks! ;-)