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/
<gonidelis[m]>
of course just changing the type won't do the job
<gonidelis[m]>
on its own
<gonidelis[m]>
jedi18: there is further tweaking need to be don in order for the base implementations to work properly. To be honest, that's the most essential part of the adaptation, as you are actually altering fundamental utilities in order to support sentinels
<gonidelis[m]>
So for `std::distance` we have already ipmlemented `ranges::distance` in order to work with sentinels, so you just use that ;)
<gonidelis[m]>
Now it is very likely that you will encounter a bunch of those utilities that don't support sentinels yeet
<gonidelis[m]>
what you have to do is the following to:
<gonidelis[m]>
1. Search in past adaptation PRs if someone has already adapted or provided a workaround solution for the same utility
<gonidelis[m]>
2. if not, it's your task to implement them and put them in your code.
<gonidelis[m]>
Now (2) could be challenging and would probably need some discussion here on IRC or on Friday's meeting on how to do it and where but if you could provide an initial solution that would help
<gonidelis[m]>
(again sorry for the late response)
<gonidelis[m]>
* what you have to do is the following:
<gonidelis[m]>
* Now (2) could be challenging and would probably need some discussion here on IRC or on Friday's meeting on how to do it and where, but if you could provide an initial solution, it would help
<gonidelis[m]>
ps: believe it or not there are certain algos that work out-of-the-box by just changing the Iter to type to a Sent type. If you are lucky, you will probably come across such a case
<jedi18[m]>
Thanks a lot for explaining that!
<jedi18[m]>
and np with the late responses, there's plenty of time xD
<rachitt_shah[m]>
<harsh "Can I contribute by starting by "> Hey Harsh, welcome.
<rachitt_shah[m]>
ms sharing my timeline for patches and the platform migration by tomorrow.
<ms[m]>
rachitt_shah: great, thank you, looking forward to that
<ms[m]>
also thanks for the blog post, looks good to me
<ms[m]>
we'll get it up in the next few days
<rachitt_shah[m]>
Sure. Will share soon, so we could dive into collecting feedback.
<rachitt_shah[m]>
I would need your assistance when I start with the API documentation. Probably will ask Nikunj as well.
<gnikunj[m]>
Yup, let me know when you start with it
hkaiser has joined #ste||ar
<gnikunj[m]>
hkaiser: seems like there is more room to grow wrt contention. 1.2s for 1ms task and 1000 tasks is a great step up over the previous 22s execution time but this can be further lowered to 0.4-0.5s (even lower perhaps).
<gnikunj[m]>
also, ms is looking into the segmentation fault that I was talking about. Seems like it's coming from the HPX backend within Kokkos. He'll look into the issue soon.