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/
bita has joined #ste||ar
K-ballo has quit [Quit: K-ballo]
hkaiser has quit [Quit: bye]
bita has quit [Ping timeout: 265 seconds]
bita has joined #ste||ar
bita has quit [Ping timeout: 240 seconds]
peltonp1 has quit [Quit: Konversation terminated!]
<hkaiser>
error generation is always slow for compilers
<gonidelis[m]>
the problem is that when i use the "proper" iter_sent which declares the `==` `!=` operators independently of the iterator/sentinel types it fials
<gonidelis[m]>
fails
<hkaiser>
how does it fail?
<gonidelis[m]>
test fail. I tried running the executable but it hanged. I will wait a couple of more minutes
<hkaiser>
so it compiles?
<gonidelis[m]>
yeah
<hkaiser>
why does it fail?
<hkaiser>
have you used a debugger to find out?
<gonidelis[m]>
hm... you suggest I go with gdb?
<hkaiser>
you're working with VSCode, right?
<gonidelis[m]>
yes
<gonidelis[m]>
but you said the debugger is only good on visual studio, i think?
<hkaiser>
well, then it shouldn't be hard to use gdb, it's built in
<hkaiser>
I said VS is the best debugger, VSCode is usable as well
<gonidelis[m]>
huh
<gonidelis[m]>
ok
<gonidelis[m]>
let me see
<gonidelis[m]>
hkaiser: ahhh does this work on rostam? do i need any permitions or sth?
<gonidelis[m]>
i am a little bit confused on how to do that remotely. i mean should i touch my build dir ?
<hkaiser>
gonidelis[m]: should work remotely
<gonidelis[m]>
hkaiser: the debugger should run on my end or on rostam?
<zao>
Some part of the debugger needs to run on the remote host, whether it's a full debugger instance or a debug server stub.
<zao>
Depends a bit on how you set your environment up.
<zao>
If you use the "Remote: SSH" extension for VS Code, it should probably mostly work as you expect, might have to set some additional paths and env if you normally use modules for compilers/libraries.
<gonidelis[m]>
zao: yes i do. thanks ;)
<zao>
I tend to load the modules, dump out the environment variables and jam into `launch.json` or whatever Code calls it these days.
<gonidelis[m]>
`launch.json` on my machinge, right?
<hkaiser>
srinivasyadav227: I'd like to pull ms[m] in on those questions
<hkaiser>
he installed vc on circleci
<srinivasyadav227>
hkaiser: ok ;-)
hkaiser has quit [Quit: bye]
<ms[m]>
srinivasyadav227: in the end I installed whichever version of vc was available in the repositories on ubuntu (which is 1.3.X), if we really require 1.4.1 I install that as well
<ms[m]>
is there a problem with 1.3.X?
hkaiser has joined #ste||ar
<srinivasyadav227>
ms: i got some internal issues(some long error messages related to intrinsics) in 1.3 while debugging in stellargroup/buildenv docker container, so tried 1.4.1 release branch and this is working, (1.4.x should be fine)
<K-ballo>
gonidelis[m]: are the relational operators used? they seem meaningless
<gonidelis[m]>
iter_sent is used in every iterator sentinel test we want to create on our parallel algorithms module
<gonidelis[m]>
do you mean they semm useless in general, or that they seem useless for the count_3646 test?
<gonidelis[m]>
K-ballo:
jehelset has joined #ste||ar
<jedi18[m]>
Can https://github.com/STEllAR-GROUP/hpx/issues/5219 not be implemented until revision 15 of the paper? It's included as one of the ideas for GSoC so I was wondering how hard it would be to implement
<gonidelis[m]>
!!!!!!! hkaiser darn it!!!! the operator overloading re-definition created huge overhead! I found that while installing VS on my Windows VM :p I will keep both the VM and the VS though in order to debug locally next time. Remote debugging seemed a littrle fraustrating
<gnikunj[m]>
gonidelis[m]: one piece of advise. Keep scripts for doing performance runs. Do NOT do it remotely, just send in a batch script or take an active screen session.
<gnikunj[m]>
you should analyze the results remotely though, which is both quicker and easier to the eyes
<gnikunj[m]>
but why is the operator overloading creating a huge overhead? shouldn't be much imo.
<gonidelis[m]>
gnikunj[m]: i am not doing any performance runs
<gnikunj[m]>
aah, just remote debugging?
<gonidelis[m]>
gnikunj[m]: but plz explain what do you mean because I will certainly do ;p
<gonidelis[m]>
gnikunj[m]: yeah I was just fixing a broken pr
<gonidelis[m]>
actually what do you mean active screen session
<gnikunj[m]>
do it locally (unless it involves more than 1 node)
<gonidelis[m]>
i get the batch thing
<gnikunj[m]>
"screen" is a piece of software just like tmux is. It grants you a terminal that you can detach from. When you detach from a screen session, the session is still running and you do not loose whatever's executing within the session.
<gnikunj[m]>
You can later attach to the screen session using the session id
<gonidelis[m]>
huh i think that's what i am doing using `srun -p medua -N 1 --pty /bin/bash -l`, isn't that right?
<gonidelis[m]>
or is the opposite of what you are suggesting? ;p
<gnikunj[m]>
this way, you can initiate a screen session, execute your script and then detach from it. Let's say after 2d you feel that the performance runs are complete, return to the session and you will be greeted to the session again. You can then choose to close (if the execution is complete) or detach again (if your runs will take more time).
<gnikunj[m]>
srun only gives you an active node medusa (since you provide -p medusa). What I'm suggesting is do: `screen srun -p medusa -N 1 --pty /bin/bash -l`
<gnikunj[m]>
this time you will be created to a screen session, to detach from the session you can do `ctrl+a + d`
<gonidelis[m]>
ahh ok! oh alright!
<gnikunj[m]>
it's a neat way of handling live sessions and you do not want to run batch scripts
<gonidelis[m]>
so you suggest running my performance tests like that, instead of just leaving the terminal open
diehlpk_work has joined #ste||ar
<gnikunj[m]>
yes! You should never keep your terminal open :P
<gonidelis[m]>
so again, why do that? because of the timings?
<gonidelis[m]>
gnikunj[m]: why?
<gnikunj[m]>
because it's more convenient
<gonidelis[m]>
i could either log out from rostam right?
<gonidelis[m]>
even^^
<gnikunj[m]>
<gonidelis[m] "gnikunj[m]: why?"> you won't be able to log out in this case (from rostam)
<gnikunj[m]>
<gonidelis[m] "i could either log out from rost"> yes, that's the beauty of it. Since your screen session is running in background, logging out of rostam won't kill the screen session.
<gonidelis[m]>
hah ok... that sounds much convinient
<gnikunj[m]>
nod
<gonidelis[m]>
what I used to do is send the batch and then keeping an eye on the out.log file
<gnikunj[m]>
yeah, that's another way of handling it as well
<gonidelis[m]>
hm... ok thanks!
<zao>
If your task is self-contained and can be run fully automatically, prepare a batch job script and submit it, waiting for it to complete by looking at "scontrol show job" or squeue output, and having the log file for output.
<gonidelis[m]>
zao: what's a case where a task is not self-contained?
<zao>
If you need interactive access, start an interactive session as you've done before, but in a multiplexer like tmux or screen to be able to disconnect (voluntarily or unvoluntarily) from the login node.
<zao>
Note that running things interactively means that you're hogging the resources for the duration of your session, even when you're not doing anything.
<zao>
You also rely on the login node being up and working, while a batch job will run fine while the login node might be rebooting.
<gonidelis[m]>
zao: yeah that's a thing i did was considering
<zao>
You'd use an interactive session when you're doing multiple short things in sequence with human interpretation/interaction inbetween, like iterating on some work.
<gonidelis[m]>
and I always kept an eye if rostam got too crowdy in case i created overhead
<zao>
While things like perf or parameter sweeps you can prepare standalone scripts for and throw them all at the system.
<gonidelis[m]>
zao: hah ok makes sense now
<gnikunj[m]>
zao: right
<gonidelis[m]>
i used the interactive sessions for the adaptation game
<gonidelis[m]>
which is an exhaustive iteration of make -j, ctest commands
<gonidelis[m]>
so that makes sense
<gonidelis[m]>
i get the perf case though
<zao>
Having the process explicitly written down helps with trying to reproduce it later too ^_^
<hkaiser>
srinivasyadav227: I think I found a nice way to decouple the dapar code from the rest of it now
<hkaiser>
at least I'm happy with it ;-)
<hkaiser>
I've done that in the context of a different branch, but you could use the same technique to decouple the rest of it
<srinivasyadav227>
hkaiser: yes sure, i will do the rest, once datapar ci is fixed,
<srinivasyadav227>
hkaiser: instead of template specialisations, now it uses tag_fallback_invoke and function overload right?
<hkaiser>
srinivasyadav227: yes
<hkaiser>
that allows to remove the is_vectorpack_policy from the base implementation
<srinivasyadav227>
hkaiser: thanks! nice idea! that should help me a lot seperating others! ;-)
<hkaiser>
srinivasyadav227: you will have to understand how tag_invoke works, however
<hkaiser>
jedi18[m]: done
<srinivasyadav227>
yes, now i have enough clarity with it, (separating segmented algorithm PR helped me with this)
<jedi18[m]>
@freenode_hkaiser:matrix.org thanks!
<K-ballo>
gonidelis[m]: I mean in general, this iterator claims to be a forward iterator, it should not implement relational ops
<K-ballo>
only random access iterators have those
<K-ballo>
your new == and != overloads are global, and have to perform sfinae for every single == call out there in the tests.. it's bad, but it shouldn't be noticeable bad
jehelset has quit [Remote host closed the connection]
<ms[m]>
they've slightly changed the application this year though
<ms[m]>
they want a single project proposal with a budget set by us
<diehlpk_work>
ms[m], Thanks
<ms[m]>
would you mind having a look? I just put something for the budget right now, they recommend 5-15000 and it can be for other things than just paying the writer as well
<diehlpk_work>
Sure
<ms[m]>
gonidelis[m]: gnikunj could you two also have a look at the project proposal link above? I've tried to combine some of your ideas for the project ideas page
<diehlpk_work>
ms[m], Thange you as the primary admin, since you did all the work
<ms[m]>
feel free to edit as you see fit
<diehlpk_work>
I am happy to be your backup
<ms[m]>
diehlpk_work: yeah, I already put myself on the actual form, will still update that page
<gnikunj[m]>
ms: sure. Will do.
<ms[m]>
thanks!
hkaiser has joined #ste||ar
<diehlpk_work>
ms[m], It reads good
<gnikunj[m]>
+1
<diehlpk_work>
I do not understand the 5000 estimate
<ms[m]>
it's a placeholder for now, I'm just not quite sure what to put there instead
<ms[m]>
it's different from before in that we essentially apply for a grant from google to pay a technical writer, and we get to choose how much we might need
<diehlpk_work>
I like to have Giant inflatable unicorn (15 foot version)
<ms[m]>
please add that to the budget
<ms[m]>
;)
<diehlpk_work>
ms[m], We should add some stickers and t-shirts as swag
<diehlpk_work>
We can give swag to people helping us in proof-read the new content
<ms[m]>
diehlpk_work: yep, I'll add that as well
<diehlpk_work>
Are the volunteer stipends the same as the mentor stipends?
<diehlpk_work>
I ordered stikcers for octo-tiger last year and it was around $80
<diehlpk_work>
hkaiser, might now how much the t-shirts were
<diehlpk_work>
ms[m], Technical Writers made a median salary of $72,850 in 2019. The best-paid 25 percent made $93,590 that year, while the lowest-paid 25 percent made $56,570.
<diehlpk_work>
They might want that we calculate the salary according to some numbers
<diehlpk_work>
The average salary for a Technical Writer is $33.67 per hour in United States.
<diehlpk_work>
So we might could add some base value to the budget
<diehlpk_work>
I think salary and swag would be enough for us
shahrzad has joined #ste||ar
<hkaiser>
diehlpk_work: don't remember, something around $15-20
<diehlpk_work>
Ok, we can add 120 for t-shirts + 80 for the stickers
<gnikunj[m]>
hkaiser can we have a meeting tomorrow before we have the resiliency meeting? I'd like to take you through what I've been doing. I'm currently trying to get everything compiled on rostam so that I can do some performance analysis on it. If the results are out before tomorrow's meeting, I'll share the graphs too.
nanmiao has joined #ste||ar
<hkaiser>
gnikunj[m]: yah, I could meet at 11 my time for 30 minutes
<gnikunj[m]>
Sounds good. Should be good enough.
<gnikunj[m]>
Should I ask Katie to arrange the meeting?
<gnikunj[m]>
hkaiser ^^
<ms[m]>
hkaiser, do you have any comments on #5172, #5237, or #5244?
<hkaiser>
ms[m]: sorry for being sluggish - will lokk
<hkaiser>
look
<hkaiser>
gnikunj[m]: yes, pls
<ms[m]>
hkaiser: no worries and thank you!
nanmiao has quit [Quit: Connection closed]
jehelset has quit [Remote host closed the connection]
bita has joined #ste||ar
jehelset has joined #ste||ar
hkaiser has quit [Read error: Connection reset by peer]
hkaiser has joined #ste||ar
bita has quit [Read error: Connection reset by peer]