aserio 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/
EverYoung has joined #ste||ar
EverYoung has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
EverYoung has quit [Ping timeout: 246 seconds]
EverYoung has joined #ste||ar
rod_t has left #ste||ar [#ste||ar]
vamatya has quit [Ping timeout: 240 seconds]
EverYoung has quit [Ping timeout: 240 seconds]
K-ballo has quit [Quit: K-ballo]
EverYoung has joined #ste||ar
EverYoung has quit [Ping timeout: 240 seconds]
EverYoung has joined #ste||ar
EverYoung has quit [Ping timeout: 246 seconds]
mcopik_ has quit [Ping timeout: 246 seconds]
parsa has joined #ste||ar
parsa has quit [Client Quit]
hkaiser has quit [Quit: bye]
parsa has joined #ste||ar
vamatya has joined #ste||ar
parsa has quit [Quit: Zzzzzzzzzzzz]
parsa has joined #ste||ar
parsa has quit [Client Quit]
vamatya has quit [Ping timeout: 240 seconds]
EverYoung has joined #ste||ar
EverYoung has quit [Ping timeout: 246 seconds]
AnujSharma has joined #ste||ar
david_pfander has joined #ste||ar
EverYoung has joined #ste||ar
Matombo has joined #ste||ar
EverYoung has quit [Ping timeout: 246 seconds]
bikineev has joined #ste||ar
bikineev has quit [Remote host closed the connection]
bikineev has joined #ste||ar
bikineev has quit [Remote host closed the connection]
<hkaiser>
jbjnr: very nice - you're getting there!
<jbjnr>
thanks for counter stuff. I used them ages ago and wasn't sure if there was an even easier way of getting the numbers.
<jbjnr>
yes, getting very close now with the matrix work. Soo we will be ahead I think,.
<hkaiser>
sure
<jbjnr>
^soon
<hkaiser>
:D
<hkaiser>
you're da'man!
<jbjnr>
for the papi counters, I can't just query any counter can I - it needs to be one that was initialized at start yes? (I believe papi onkly allows a limited number to be used at once)
<hkaiser>
jbjnr: I thik you can use any counter supported by the hardware - there is just a limit on how many you can have active at a time
<hkaiser>
he creation of the counter initializes everything
<jbjnr>
so if I exceed the limit - then one just gets invalidated I guess? I would not want to use 15 simulataneously if only 12 are allowed and I wonder how I should police that
<hkaiser>
not sure - you might not be able to create more
<hkaiser>
this is limited by the hardware
<heller>
use papi_avail
<heller>
to check which are there
<heller>
for the number of counters, you'd need to check the docs
<heller>
just a little too many slides for 30 minutes :/
<heller>
yes, I went to the dinner
<hkaiser>
yah, I thought that as well - you can shorten the first half a bit
<heller>
yes
<hkaiser>
heller: the important thing to highlight in your talk is (if you asked me) that AMTs (sorry for the term) are a viable alternative to classic programming models
<heller>
agreed
<heller>
I think everyone got that by now
<hkaiser>
HPX just happens to be th eonly one which is currentl available to demonstrate that
<heller>
well, there is always the MPI + X model
<heller>
with X being kokkos
<hkaiser>
forces fork-join on you
<hkaiser>
on all levels
<heller>
which is way more popular in the DoE circles
EverYoung has joined #ste||ar
<heller>
ok, good to know
<heller>
they don't have futures?
<hkaiser>
shrug - I didn't say you should highlight anything others might do ;)
<hkaiser>
heller: not that I know of
<hkaiser>
try combining async computing with mpi - doesn't work
<heller>
yes
<heller>
How does the term "A standard C++ AMT" sound to you?
<hkaiser>
shrug - anything works if it works to get people understand what you want to say
<hkaiser>
the term AMT is used in DoE circles, so you can't go wrong using it
EverYoung has quit [Ping timeout: 246 seconds]
<heller>
ok
<jbjnr>
heller: can we watch your talk online? I see dial in numbers, but not a linky link kind of link thingy
<heller>
jbjnr: it's using zoom
<hkaiser>
jaafar: it's zoom, so yes
<heller>
where you can watch it
<hkaiser>
jbjnr: ^^
<heller>
zoom meeting id: Meeting ID: 468 190 374
<heller>
works quite nicely on linux
<heller>
I was asked to upload my PPT file, they looked quite confused when I told them I only got tex sources
<heller>
I'm too stupid for powerpoint :/
<hkaiser>
makes your 'hero-programmer' image just so much more believable ;)
<hkaiser>
the trick with PPT is to use a MOUSE ;)
<jbjnr>
I got silence with zoom
<hkaiser>
too early
<jbjnr>
ok
<hkaiser>
it's 5:43 over there
<jbjnr>
I go for tea now then
<hkaiser>
AM
<jbjnr>
no video either?
<heller>
jbjnr: there will be video
<jbjnr>
ok
<heller>
hkaiser: I prefer the term coder ninja
<jbjnr>
5:43 - I see. I though 2pm or something
<hkaiser>
tex junkie
<jbjnr>
"sad loser"
<hkaiser>
lol
parsa has joined #ste||ar
parsa has quit [Client Quit]
<heller>
jbjnr: it is 5:45 now
<heller>
jbjnr: the talk will be 14:20 california time
<hkaiser>
that in order to get the algorithms performant one would need something like hpx underneath
<heller>
;)
<jbjnr>
hkaiser: I'd like to add scheduler_policies to async/task launch - so that we can guide the scheduler when we add tasks- similar to using high priority or low, but it needs to be more customizable for each different algorithm. (so cholesky might use one set of scheduler params, but QR decomposition another). have you thought about this at all. I don't really want to provide another async...
<jbjnr>
...overload, but I might need to.
<heller>
-"something like"
<hkaiser>
jbjnr: executors?
<jbjnr>
with(scheduler_policy(...)) is what I'm thinking of now
<hkaiser>
now would be a good time to get the executor story straight...
<jbjnr>
so yes, executors, but new param types
<heller>
btw, can someone have a look at #2891 please?
<jbjnr>
this would go beyond the current standard
<hkaiser>
jbjnr: sure, the parameters are beyond the standard anyways
<jbjnr>
for the linear algebra we need a way of telling the scheduler about the tasks with extra hints etc.
<hkaiser>
jbjnr: I have a half finished parameters revamp here on my disk, can't get it done...
<jbjnr>
anything interesting that I might liketo experiment with - this is next on my todo list (urgetnt for cholesky), so I'm happy to help
<hkaiser>
the whole thing is currently a big mess in hpx - some parts have been redone, some not
<jbjnr>
indeed
<hkaiser>
jbjnr: ok, let me raise the priority of the parameters refactoring
<hkaiser>
will try to do it asap
<jbjnr>
you can dump a half finished branch on me if you want.
<hkaiser>
ok, will have a look
<jbjnr>
but I'd naturally prefer you to finish what you're doing first :)
<github>
hpx/master fc4769a John Biddiscombe: replace boost::format with hpx::util::format
<jbjnr>
<cough> pushed inspect fix to master <cough>
aserio has quit [Quit: aserio]
<heller>
jbjnr: good luck!
<jbjnr>
with?
aserio has joined #ste||ar
pagrubel has joined #ste||ar
<heller>
jbjnr: not being yelled at by hartmut :P
pagrubel is now known as pat[[w]]
<pat[[w]]>
aserio: I guess there's no meeting until next week?
eschnett has quit [Quit: eschnett]
StefanLSU has joined #ste||ar
pree has joined #ste||ar
vamatya has joined #ste||ar
EverYoung has joined #ste||ar
StefanLSU has quit [Quit: StefanLSU]
rod_t has joined #ste||ar
EverYoung has quit [Ping timeout: 252 seconds]
pat[[w]] has quit [Quit: Leaving]
mcopik has joined #ste||ar
hkaiser has joined #ste||ar
eschnett has joined #ste||ar
bikineev has joined #ste||ar
EverYoung has joined #ste||ar
EverYoun_ has joined #ste||ar
EverYoung has quit [Ping timeout: 248 seconds]
EverYoun_ has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
pree has quit [Read error: Connection reset by peer]
jaafar_ has joined #ste||ar
jaafar has quit [Ping timeout: 246 seconds]
pradeepisro49 has joined #ste||ar
pree has joined #ste||ar
pree has quit [Quit: AaBbCc]
david_pfander has quit [Ping timeout: 240 seconds]
harsh has joined #ste||ar
<harsh>
Hey
<harsh>
I want to know is Stellar is interested in Algorithmic Graph theory related projects?
<harsh>
In parallel setting
<harsh>
Like Modular decomposition in Parallel
<harsh>
Anyone there?
<heller>
hey harsh
<harsh>
Hey
<harsh>
Could you please see my query and response
<heller>
we welcome any project
<harsh>
Even for GSoC-2018?
<heller>
that is, we try to help all of our users
<heller>
sure
<heller>
that's still some time away ;)
<harsh>
Ya I agree
<heller>
you have to be a bit more specific on what you want to work on though
<heller>
so what do you have in mind?
EverYoung has quit [Ping timeout: 246 seconds]
EverYoung has joined #ste||ar
<harsh>
I will implement 3 algorithms in parallel, Modular Decomposition to get connected components from Graph G, Ear Decomposition and Getting Biconnected components from Graph G.
<heller>
ok, what about graph data structures?
<harsh>
It will be CSR
<harsh>
Graph can be Directed and Undirected.
<heller>
sure, do you have references for the parallel algorithms or will you try to develop them yourself from scratch?
<harsh>
Mostly from scratch
parsa has quit [Quit: Zzzzzzzzzzzz]
<heller>
ok, what's your background?
<harsh>
I am an MS by Research student at IIIT-Hyderabad
<heller>
ok, how would you rate your C++ knowledge (1 is worst, 5 is best)?
<harsh>
My domain is Algorithmic Graph Theory in Parallel setting and recently I am published one paper(An Efficient Ear Decomposition Algorithm)
<heller>
cool
<harsh>
The complete experiments I have done in C++
<harsh>
From past 3 years I am working in C++.
<harsh>
Rating: 3.2 as it's been 10 months in Cuda
<harsh>
I am not expert in C++ but above I am Avg
<harsh>
Are you there?
<heller>
you know, sometimes people are doing other stuff
<heller>
just because I am not answering within a second, doesn't mean I disappeared
<harsh>
Ya sorry for that
<heller>
so trying to push for an answer is a little rude
<heller>
anyways
<heller>
I suggest for you to familiarize with HPX, try different things and see how it fits for your algorithms
<heller>
and try to come up with a proposal as early as possible
<heller>
so we can discuss various things
<harsh>
Thanks for the suggestion, sure I will come up with an proposal.
<harsh>
Where I need to share it if it's done?
<heller>
depends on you ;)
EverYoun_ has joined #ste||ar
<harsh>
is there any stellar Google group where we can discuss.
<harsh>
?
EverYoung has quit [Ping timeout: 246 seconds]
<heller>
harsh: there is a mailing list, yes
<heller>
harsh: hpx-users@stellar.cct.lsu.edu
<harsh>
Thanks I will share the proposal soon
<harsh>
Last question
<harsh>
Is there any condition to contribute or solve some patch/issues?
<harsh>
for submitting proposal?
aserio has quit [Ping timeout: 246 seconds]
bikineev has quit [Ping timeout: 240 seconds]
<heller>
harsh: if we know you, it will be better. we don't have hard rules
<harsh>
Thanks Heller for helping me.
<harsh>
Have a good day
<harsh>
Bye, I will try to complete the proposal as soon as possible and share it on the mailing list
<heller>
you are welcome
<heller>
try to familirize yourself with HPX first
Matombo has quit [Remote host closed the connection]
<harsh>
Yes I will start from it first
StefanLSU has joined #ste||ar
twwright has quit [Quit: twwright]
EverYoun_ has quit [Remote host closed the connection]
parsa has joined #ste||ar
StefanLSU has quit [Quit: StefanLSU]
EverYoung has joined #ste||ar
StefanLSU has joined #ste||ar
mbremer has joined #ste||ar
StefanLSU has quit [Quit: StefanLSU]
twwright has joined #ste||ar
<heller>
mbremer: hey
<heller>
mbremer: I couldn't get the blas stuff to work ...