08:36 <simbergm> heller: btw, do my responses on the local runtime pr seem sensible to you?

08:36 <heller1> ms: ah sorry, haven't checked

08:37 <heller1> ms: if you think it's the way to go, sure

08:37 <simbergm> no worries, wasn't sure if that or you just didn't bother replying :P

08:37 <heller1> ms: I think I said it a few times that I am not in favor, but I guess it is not my decision ;)

08:37 <simbergm> it's definitely something that people at CSCS will appreciate, and I think it's orthogonal to what you have in mind

08:38 <simbergm> it shouldn't prevent the kind of things you're hoping for (I'm not quite sure how it's different from HPX_WITH_NETWORKING=OFF, but still)

08:38 <heller1> I think it just creates extra work with 'surface level' benefits, but potentially creates more harm in the long run

08:38 <simbergm> yeah, I know you're not in favour and I just wanted to understand why

08:40 <simbergm> mmh, I think for us there are enough benefits to justify it

08:40 <simbergm> the pr is there after all

08:42 <heller1> because I don't get it why you would want such a thing: 1) You don't have to use components or actions or anything if you don't want to 2) With HPX_WITH_NETWORKING=Off we reduced compile times already significantly (I was actually hoping that you would say the the additional changes cut those compile times even more). 3) What about documentation? What do we actually advertise? Will we put a warning sign "Not usable when not

08:42 <heller1> compiled with distributed runtime" everywhere?

08:43 <heller1> For me, it sounds a lot like psychological comforting, because people expressed they don't want to use the distributed stuff, so they should be able to turn it of. But why exactly do they need to turn it of, what's the real reason behind all this?

08:44 <heller1> ahh, you have a compile time comparison

08:46 <heller1> is it really just compile times?

08:47 <simbergm> https://github.com/STEllAR-GROUP/hpx/pull/4091#issuecomment-547491831

08:47 <simbergm> compile time is a big thing

08:47 <simbergm> my long term goal would still be to not have it as a separate compile/cmake option

08:48 <heller1> so the biggest issue people have it compile time? What will happen once we get the performance coutners back in ;)?

08:48 <simbergm> I'd like it to just be enabled, but allow users to just include the parts that they use

08:48 <simbergm> hand in hand with modularization

08:48 <simbergm> this is just part one of many I hope

08:48 <heller1> I completely agree here

08:49 <heller1> but that's also why I am not a big fan here

08:49 <simbergm> if performance counters go back in with a local runtime they would hopefully be much lighter

08:49 <heller1> it's a temporary solution, and you know, temporary solutions tend to stay longest ;)

08:50 <simbergm> yeah, I can't argue with that...

08:50 <simbergm> but also solving all of those problems in one go is not feasible

08:50 <heller1> so it's a very hacky solution to a non problem (at least from my point of view). However, I understand your point.

08:50 <simbergm> I'd like to have something minimal that's already usable

08:53 <heller1> ms: left a comment, guess there are some more fixes from hkaiser that need to be addressed though

08:53 <simbergm> heller: thanks

08:54 <simbergm> yeah, going through those at the moment

08:56 <heller1> ms: would have hit approved if the PR itself was clean

08:58 <simbergm> no worries, I know it still has cleaning up to do (it's the local runtime builder that's ironically failing... it keeps breaking every second rebase because of files moving)

08:59 <simbergm> I was just looking for general comments, just like yours ;)

08:59 <simbergm> so thanks!

10:07 <simbergm> jbjnr: would you mind having a look at https://github.com/STEllAR-GROUP/hpx/pull/4487? there are a few changes on your scheduler

12:55 <Nikunj__> simbergm, just saw the pr

12:56 <Nikunj__> what sort of linking problems are we facing in macos?

12:56 <Nikunj__> does it complain about redefinition of main? or perhaps can't find main itself?

15:09 <diehlpk_work> hkaiser, Can you have a look into the AGAS results in the SC paper?

15:09 <diehlpk_work> They do not look good at all for the short runs

15:10 <hkaiser> diehlpk_work: can do later today, finalizing the proposal right now

15:11 <diehlpk_work> Good luck

15:11 <hkaiser> tks

15:18 <Yorlik> hkaiser: When we were talking about spatial indexing and search: Did you think of a try like tree or a leaf-only tree for storing values, or would it just not matter?

15:18 <Yorlik> trie - like tree

15:18 <hkaiser> Yorlik: sec, in a meeting

15:19 <Yorlik> KK

15:25 <diehlpk_work> Lol, got a rejection letter nearly two years after I submitted my application. Was wondering why they sent me a letter per mail

15:44 <hkaiser> Yorlik: ok, now

15:44 <Yorlik> o/

15:44 <hkaiser> Yorlik: tree for what?

15:45 <hkaiser> the spatial tree?

15:45 <Yorlik> I'm just a bit overwhelmed by the many variations on Quadtrees and wondered what specific properties would be needed for the type of application you used.

15:45 <Yorlik> Yes

15:45 <Yorlik> Like leaf only storage or across the entire tree

15:45 <hkaiser> objects are points, so you can store them in the leafs

15:45 <Yorlik> Alo: Even splitting or adaptive

15:46 <Yorlik> Maybe the splitting is just a detail in the end, just thought I'd ask.

15:47 <Yorlik> My idea would be to start with even splits, like a classical image pyramid

15:47 <hkaiser> Yorlik: that's up to you, I was using a hysteresis to split/join children and a fixed max ceiling for the objects per leaf

15:47 <Yorlik> OK - makes sense. So it's an impl detail only.

15:47 <hkaiser> Yorlik: sure, just remember that half of the tree is empty so at the boudary min==max you need only half of the children

15:48 <hkaiser> well, three quarters, actually

15:48 <Yorlik> I was thinking of making a simple graphical starter application for visualization of the problem

15:48 <Yorlik> Yesterday I used an online graphing tool to visualize certain queries

15:48 <weilewei> hkaiser just quick reminder, meeting w/ Ronnie this afternoon is canceled. If we have any urgent matters, the meeting can be resumed. He updated it on email and calendar

15:48 <hkaiser> Yorlik: we can have another chat where I can try drawing things out

15:49 <hkaiser> weilewei: saw that, thanks

15:49 <Yorlik> E.g.: Is a point inside the interval comes down to: (x-a)*(y-a)<0 ?

15:49 <weilewei> hkaiser ok, thanks

15:49 <hkaiser> weilewei: however next Monday I'm officially off - Katie will send an email

15:49 <weilewei> hkaiser ok, sure

15:51 <weilewei> I am verifying my implementation on gpudirect + main_dca where allows 1 accumulator thread. For multiple threads, lab people suggested not an urgent or necessary steps, but could be next step to look at.

15:52 <weilewei> So this week, I will be working on this. I will sync messages timely between you and Ronnie

15:53 <weilewei> During the DCA++ call today, they are happy about my work so far

15:54 <weilewei> Let me put everything on email for all of us instead.

16:14 <hkaiser> Yorlik: a point inside an interval means that is not outside it ;-)

16:15 <Yorlik> The formula would visualize the space containing the points

16:15 <hkaiser> you can use the tree to exclude areas based on the selection box, what remains holds candidates

16:15 <Yorlik> It was 2 diagonal quadrants of 4 around the point

16:16 <Yorlik> Just exploring the possibilities by playing around a lot.

16:16 <hkaiser> in 1d space, let's say you're interested in finding all objects that overlap a give box min-max, then all objects (points) for which their max is smaller than min, and all objects (points) for which their min is larger than max are definitely not touched by the seclection box

16:16 <hkaiser> all others potentially overlap

16:21 <hkaiser> Yorlik: the same is true for all nodes in the tree, that allows you to explude all subtrees (and all objects stored in those) from any node down that is not touched by the box, possibly brute-forcing every object in other regions of the tree

16:23 <Yorlik> I still wonder what the meaning of subtrees and sibling trees is, when you collabse a cube to a 6D point in a 64ary tree

16:24 <Yorlik> Like - which questionsa could I answer - and what dimensionality is optimal for the questions I have

16:30 <simbergm> Nikunj__: thanks for looking through the init pr! it was indeed working on linux and not on macos

16:31 <simbergm> if the changes made sense to you that's good :)

16:31 <simbergm> heller, hkaiser if I'm a bit unresponsive it's because I'm in bed with a fever...

16:31 <simbergm> no reason to believe it's corona though!

16:50 <nikunj97> simbergm, don't work too hard ;) Take care, get well soon!

17:10 <hkaiser> simbergm: hope you'll feel better soon!

17:11 <hkaiser> Yorlik: do you want to chat again?

17:11 <Yorlik> Hey!

17:12 <Yorlik> Actually that might make sense.

17:12 <hkaiser> now?

17:12 <Yorlik> If you can - great!

17:13 <Yorlik> Same link?

17:15 <hkaiser> yah, same link

17:15 <Yorlik> I think I'm in