07:53 <srinivasyadav227> gonidelis[m]: yes I am there, I am sorry, I had exams, just finished today, resuming now

08:08 <gnikunj[m]> gonidelis[m]: yt?

09:45 <gonidelis[m]> gnikunj[m] hey

09:46 <gonidelis[m]> freenode_srinivasyadav227[m] hey... I was going to ask you about a pending PR ment for gsoc students but i think jedi18 has already made one

09:55 <srinivasyadav227> gonidelis[m]: I just started 2 hrs back (delay because of my college internals exams) , I think today I will make one PR soon, working on it now 🙂

09:56 <gnikunj[m]> gonidelis[m] I updated gsod project. Would you like to mentor it with me?

09:57 <gnikunj[m]> (The one we were talking about in yesterday's meeting)

11:15 <ms[m]> srinivasyadav227: there are plenty of other algorithms that need the tag_invoke to tag_fallback_invoke conversion, you can also have a go at one

11:15 <ms[m]> just coordinate with jedi18[m] that you don't do the same ones

11:15 <srinivasyadav227> ms: yes, im working on copy.hpp now

11:16 <ms[m]> 👍️

11:16 <srinivasyadav227> :)

11:21 <srinivasyadav227> I forked hpx, and created a seperate branch for copy algorithm called segmented_copy, now tests that are running on my circle.ci are only partial or complete? if partial should I create a PR against my master, just want to make sure if tests are passing on my circle.ci before creating a PR against hpx master

12:59 <ms[m]> srinivasyadav227: thanks for the PR! I just want to make sure: relying on ci to check that everything is ok is a fair approach, but part of the idea of getting you applicants to open PRs is that you first try building hpx locally, so if you haven't done that I highly recommend doing that (not implying you haven't, just wasn't clear from your messages if you have)

13:12 <srinivasyadav227> ms[m]: I have build hpx locally on 3 machines (Nvidia dgx station, normal linux machine and Mac OS), then went through docs, hkaiser cpp con talks, cscs workshop videos and worked with few examples(one week back), I was busy with my exams so was not active in between, so today I forked the repo and created branches and PRs against my master and HPX master. :)

13:13 <hkaiser> srinivasyadav227: bravo, you mean it!

13:13 <ms[m]> srinivasyadav227: perfect, was just checking :)

13:14 <srinivasyadav227> hkaiser: sorry, I didn't understand..

13:14 <hkaiser> srinivasyadav227: thanks for the PR

13:15 <srinivasyadav227> hkaiser: oh, got it..thanks, 😀

13:20 <jedi18[m]1> ms[m]: Just wanted to clarify, by retest you mean I should run the tests which are failing in the CI locally?

13:21 <ms[m]> jedi18[m]: no, don't worry, that wasn't for you :D that's for jenkins, another ci system that we use

13:21 <ms[m]> same on srinivasyadav227's pr

13:21 <jedi18[m]1> Ah right ok

13:21 <ms[m]> you two are just not on our whitelist (yet?) to have builds run automatically, so we trigger them manually

13:21 <srinivasyadav227> when I created PR against my master branch, only one WORK or one platform or set of jobs were running, but PR against HPX master, number of platforms were running, ? so is this default, is there any way I can change so that tests against all platforms happen for my circle.ci account or against my PR

13:23 <hkaiser> you have to create a PR for our test system to kick in

13:23 <srinivasyadav227> hkaiser: oh ok, so if some other person also creates PR, then do they run in parallel?

13:24 <hkaiser> jedi18[m]1, srinivasyadav227, please make sure you really understand the changes you proposed - please ask questions, if necessary

13:24 <hkaiser> srinivasyadav227: yes

13:24 <srinivasyadav227> hkaiser: ok :)

13:29 <jedi18[m]1> Thanks! Yeah tbh it's bugging me a lot that I don't understand most of the changes in the PR, I'm going through https://www.youtube.com/watch?v=T_bijOA1jts&feature=emb_logo which will hopefully help

13:32 <hkaiser> jedi18[m]1: https://www.bfgroup.xyz/duck_invoke/ is a nice introduction to tag_invoke, I'd suggest you study the code, tag_fallback_invoke is an extension of that idea that adds another layer of abstraction, allowing additional compile time criteria to be used to decide which overload is chosen

13:33 <gnikunj[m]> ms: yt?

13:34 <srinivasyadav227> jedi18: thanks for sharing the video, helps a lot!

13:36 <hkaiser> there is also wg21.link/p1895 for the bone dry factology

13:38 <jedi18[m]1> hkaiser: Thanks a lot! I'll read up on those

13:40 <ms[m]> gnikunj: type away, I'll be back in a bit

13:40 <hkaiser> ms[m]: kokkos call today?

13:41 <gnikunj[m]> ms: I was wondering if it's possible to have something like this working: https://gist.github.com/NK-Nikunj/99c8f5598eae5434b45be5c5a73ffe9a taking inspiration from (https://github.com/STEllAR-GROUP/hpx-kokkos/blob/master/src/hpx/kokkos/executors.hpp#L72)

13:42 <ms[m]> hkaiser: not planned, but also not not planned

13:42 <hkaiser> ok

13:42 <ms[m]> we can definitely do one, it'll merge with the hpx meeting anyway

13:42 <hkaiser> I might be late for that

13:42 <gnikunj[m]> ms: if you have a call? Can I join as well?

13:42 <ms[m]> but I'm not off by an hour am I?

13:42 <ms[m]> gnikunj: absolutely

13:42 <hkaiser> no, it's still early

13:42 <ms[m]> same number just 30-ish minutes before the hpx call

13:42 <gnikunj[m]> thanks!

13:43 <ms[m]> i.e. in about an hour?

13:43 <hkaiser> yes

14:21 <ms[m]> gnikunj: I assume that doesn't compile if you're asking? what does the compiler tell you?

14:22 <gnikunj[m]> it compiles but it doesn't run as expected. If you make it device only, then it won't compile saying: reference to device function 'foo' in host function

14:24 <ms[m]> right, that's ok

14:24 <ms[m]> how does it not run as expected?

14:24 <gnikunj[m]> I mean not as expected for me :D

14:24 <gnikunj[m]> it is the expected CUDA behavior

14:24 <gnikunj[m]> how do I pass an invocable that I can later invoke within device?

14:26 <ms[m]> either make it a lambda or a struct with a call operator, I think function pointers are tricky exactly because of what you said (not sure if impossible, but at least not the recommended way)

14:26 <gnikunj[m]> aah! That should work yes. Let me try that.

14:28 <gnikunj[m]> ms: are you having a meeting today too?

14:28 <ms[m]> kokkos and hpx, yes

14:30 <srinivasyadav227> <gnikunj[m] "it compiles but it doesn't run a"> yes, in CUDA if a function is declared with __device__ only other __device__ functions can call it (and couple of modifications depending on compute capability), hope this is related only :)

18:10 <ms[m]> did you get the invitation for the meeting with christian?

18:12 <gnikunj[m]> srinivasyadav227 yes, I'm aware of it. I wasn't trying to invoke a device function from host either. It was about forwarding device function from host so that I could invoke it within another device function.

18:13 <gnikunj[m]> ms I haven't received the invitation. Can I attend the meeting as well?

18:14 <ms[m]> gnikunj: sorry, brainfart, will add you

18:14 <ms[m]> thanks for telling me

18:14 <gnikunj[m]> Thanks!

19:04 <ms[m]> gnikunj see pm

22:01 <bita> hkaiser, will you join the meeting?

22:01 <hkaiser> yep

22:47 <hkaiser> bita: https://gist.github.com/hkaiser/7aca6309a0eeb94bc2276e15fbda22cf

22:48 <hkaiser> I directly modified the vcall example to use LLVM instead of CUDA

22:49 <hkaiser> bita: also, in the enoki-python project, you'll need to edit the library python39_d.lib to python39.lib (except if you have the debug Python libraries installed

22:49 <hkaiser> but, as said, the example reports a heap corruption

23:10 <bita> Thank you hkaiser