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/
diehlpk_work has quit [Remote host closed the connection]
Coldblackice has quit [Ping timeout: 260 seconds]
hkaiser has quit [Quit: bye]
nanmiao has quit [Ping timeout: 240 seconds]
jehelset has quit [Remote host closed the connection]
AbanoubAsaad[m] has quit [Quit: Idle for 30+ days]
parsa| has joined #ste||ar
parsa has quit [Ping timeout: 245 seconds]
parsa| is now known as parsa
<pedro_barbosa[m]>
Is there a way in HPXCL to copy an array, from the GPU to the CPU, to a specific position of another array?
<pedro_barbosa[m]>
Example, I have array X and I split it in A and B, I send them both to different GPUs and do whatever I want in both of them, then I want to copy it back to the CPU in the right order, A fills the first 50% of array X and B starts filling array X after the middle
<pedro_barbosa[m]>
Currently I'm copying both of them to different arrays and then joining it together, it works just fine but was wondering if there's a way of doing it while copying it from the GPU
<ms[m]>
pedro_barbosa: I don't know about hpxcl specifically, but if it has an interface to just copy from pointer to pointer, there's no problem to copy into pointer + offset
<pedro_barbosa[m]>
I tried replacing 0 with res+offset, but was getting an error
<pedro_barbosa[m]>
* I tried replacing 0 with res+offset, but was getting an error, but I think those values are the positions for the buffer and not the res array on the CPU
<ms[m]>
pedro_barbosa: ok, I'm not familiar with those functions, sorry
<ms[m]>
better wait for diehlpk
<pedro_barbosa[m]>
No problem, thanks
K-ballo has joined #ste||ar
hkaiser has joined #ste||ar
CynthiaPeter[m] has quit [Ping timeout: 276 seconds]
RosheenNaeem[m] has quit [Ping timeout: 245 seconds]
rainmaker6[m] has quit [Ping timeout: 245 seconds]
prahitha[m] has quit [Ping timeout: 245 seconds]
CynthiaPeter[m] has joined #ste||ar
RosheenNaeem[m] has joined #ste||ar
rainmaker6[m] has joined #ste||ar
prahitha[m] has joined #ste||ar
<jedi18[m]>
@gonidelis:matrix.org and @freenode_hkaiser:matrix.org Thank you for choosing to be my mentors, glad to have you both as my mentors :D
<jedi18[m]>
I'll begin with unique/unique_copy first as mentioned in my proposal?
<srinivasyadav227>
hkaiser: hi, i have tried building by rebasing against 5338, but it failed with same error
<hkaiser>
jedi18[m]: most welcome, I'm looking forward to working with you over the summer (and beyond)
<hkaiser>
srinivasyadav227: ok, thanks - I have no idea what's wrong, so any help you can provide is much appreciated
<jedi18[m]>
Also, are there are any resources I could look into for understanding how segmented algorithms are implemented? If I could understand it maybe I could have another go at fixing that minmax PR bug
<srinivasyadav227>
hkaiser: ok :)
<hkaiser>
jedi18[m]: nothing I'm aware of - the code is the best resource here :/
<jedi18[m]>
@freenode_hkaiser:matrix.org oh ok, I'll try to understand the code then
<hkaiser>
jedi18[m]: but I'd be happy to meet with you to answer questions
<hkaiser>
jedi18[m]: we should start organizing weekly meetings or so anyways
<jedi18[m]>
@freenode_hkaiser:matrix.org Sure! Maybe once I've looked into the code and collect a few questions to ask
<jedi18[m]>
oh ok sure
<jedi18[m]>
What time is suitable for you?
<hkaiser>
jedi18[m]: I'll ask Katie (kmbailey@lsu.edu) to find a mutually agrreeable time for Giannis, yourself and myself
<hkaiser>
she helps me organizing things over here
<hkaiser>
jedi18[m]: what email address should I use for you, btw?
<jedi18[m]>
Oh ok sounds good, thanks
<jedi18[m]>
akhiljnair.188@gmail.com
<jedi18[m]>
Or targetakhil@gmail.com which is the gsoc email (don't ask me why I named it that, made it back when I was a small kid who for some reason came up with that address and I've been stuck with it as my main email :D)
<jedi18[m]>
Every email is forwarded between both the emails so any would do
<hkaiser>
jedi18[m]: ok - what timezone are you in?
<jedi18[m]>
IST
<gnikunj[m]>
hkaiser +5.5h
<hkaiser>
+1
<rachitt_shah[m]>
Can I join in as well?
<rachitt_shah[m]>
Would a good idea to start with the community. As it is, we would need to setup a weekly/biweekly meeting for GSoD as well.
<hkaiser>
rachitt_shah[m]: might be best to organize a separate meeting with you to be able to focus on the specifics
<hkaiser>
I think either ms[m] or Patrick will get that going
<hkaiser>
in general, we have the biweekly HPX meeting (Thursdays 9am CDT) for general project discussions - everybody is welcome to join that one
<hkaiser>
the next HPX meeting is planned for May 27
<ms[m]>
rachitt_shah: hkaiser yep, we'll have the gsod meeting this thursday and I was thinking we could make that a biweekly meeting (alternating with the hpx meetings, which are also biweekly)
<ms[m]>
and then we can always schedule additional meetings on top of that
<hkaiser>
ms[m]: ok, perfect
<hkaiser>
ms[m]: btw, I think #5332 is good to go now
<jedi18[m]>
Should I let Katie know that I stay up late till 2am so she could factor that into the meeting timings?
<ms[m]>
hkaiser: ok, great! I'll have another quick look through it but it looked good on my first pass anyway, so it's unlikely I'll have more comments
<hkaiser>
did my explanation make any sense?
<hkaiser>
jedi18[m]: she will likely get in contact, but pls feel free to email her
<ms[m]>
hkaiser: yeah, I think so (it was the part about skipping one level of dispatch that wasn't obvious from the changes, but with the explanation it makes sense)
<hkaiser>
ok
<hkaiser>
removing that dispatching was the goal
<jedi18[m]>
@freenode_hkaiser:matrix.org oh ok I'll wait for her to contact me then, thanks
jaafar_ has joined #ste||ar
jaafar has quit [Remote host closed the connection]
<srinivasyadav227>
gnikunj: and Patrick Diehl thank you for being my mentors for GSoC. Looking forward to have a great summer :)
hkaiser has quit [Ping timeout: 240 seconds]
<gnikunj[m]>
srinivasyadav227: congratulations on being selected. Congratulations jedi18[m] as well!
<gnikunj[m]>
srinivasyadav227: would you like to setup a call every week to discuss on the project?
<srinivasyadav227>
yes, sure, i wanted to ask the same, about when and how to discuss abt project
hkaiser has joined #ste||ar
<gnikunj[m]>
Let's do it this Thursday evening at 7:30PM?
<srinivasyadav227>
IST?
<gnikunj[m]>
hkaiser may want to join on in as well. Yes, IST.
<gnikunj[m]>
that's 9AM for hkaiser
<srinivasyadav227>
gnikunj: okay, that works for me :), i usually stay up to 12 AM - 1 AM IST
<gnikunj[m]>
that's good. Unfortunately, I operate mostly on CST timeline as well. So, I won't be able to help you much during the day according to IST.
<gnikunj[m]>
Let's make this project a success ;)
<srinivasyadav227>
<gnikunj[m] "that's good. Unfortunately, I op"> yea, no prob :),
<srinivasyadav227>
yes, we will :)
<srinivasyadav227>
is that time okay for Patrick Diehl?
<gnikunj[m]>
I have asked if it's for him. Let's see.
<hkaiser>
srinivasyadav227: we'll have to wait for him to respond
<srinivasyadav227>
ok sure
<mdiers[m]>
hkaiser: Regarding #5117: changing the stacksize from default to huge only reduced the probability of the crash.
<hkaiser>
mdiers[m]: ok, thanks - so it is somehow related
diehlpk_work has joined #ste||ar
<hkaiser>
gnikunj[m]: btw, we'll cancel the meeting with Keita today (I have no power)
<gnikunj[m]>
yes, I read the email. We have some good performance improvements to report btw ;)
<hkaiser>
do you want to compose an email I could send instead of the meeting?
<hkaiser>
nice, btw!
<gnikunj[m]>
sure, let me send it to you. I don't have official results (in terms of graphs) because I was initially unable to get things going on rostam. But from what I can see, there's close to 10x difference in performance. So things that takes 20s previously takes 2.5-3s now.
<hkaiser>
nice
nanmiao has joined #ste||ar
<gnikunj[m]>
hkaiser: emailed you. If you want anything changed, let me know.
HaimantikaMitra[ has quit [Quit: Idle for 30+ days]
RosheenNaeem[m] has quit [Quit: Idle for 30+ days]
<hkaiser>
gnikunj[m]: thanks
<gonidelis[m]>
K-ballo: i remember a debate we had about C vs C++. and you were like "C++ is C plus more, minus some very specific stuff". so i started believing C is quite limiting and I was thinking that few people use it. then i came across this
<K-ballo>
it's been around forever, the world is written on top of C
<K-ballo>
its simple, in that there's only a handful of basic materials to build with
<K-ballo>
also, the tiobe index is in general meh
<gonidelis[m]>
it's based on web searches ;p
<gonidelis[m]>
who the hell googles C?
<K-ballo>
it's based on plain mentions
<K-ballo>
also, the hardest a language is to use, the more you'd expect it to be search for, no?
<gonidelis[m]>
which shouldn't be a lot 🤣
<gonidelis[m]>
huh!
<gonidelis[m]>
that sounds reasonable indeed
<K-ballo>
the tiobe index is interesting on its own, but it's hardly meaningful of something else
<gonidelis[m]>
yy... i am just ivoking some conversation here. for sure I wouldn't build my career on tiobe ;p
<gonidelis[m]>
i just though C was obsolete
<gonidelis[m]>
K-ballo: do you know C?
<K-ballo>
far from it, it's probably one of the few language (the only one maybe?) that will never be obsolete
<K-ballo>
sure, I know C
<gonidelis[m]>
i mean like proficiently
<gonidelis[m]>
like the guru stuff you are doing with C++
<K-ballo>
there's very little to know and I know it well, I don't have the years of experience on it
<gonidelis[m]>
if yes, how different is from C++? i mean, people think it's little different. I am presently going down a path where I think C is as far from C++ as is from Java
<gonidelis[m]>
"very little" sound encouraging
<K-ballo>
90% of valid C is also valid C++ with the same semantics
<K-ballo>
there are some minor divergencies, and then there are C things that C++ doesn't support
<K-ballo>
but the common C/C++ subset is only the core basic of C++ itself
<K-ballo>
that 90% of C represents maybe 10% of C++?
<K-ballo>
probably less
<K-ballo>
and most if not all of C's standard library you wouldn't want to know/use when in C++ anyhow
<hkaiser>
I think the main difference between C and C++ is not in the syntax or the semantics of certain constructs
<gonidelis[m]>
isn't the a second part on this sentence ? ;p
<gonidelis[m]>
there^^
<hkaiser>
is there? ;-)
<gonidelis[m]>
i am dangled right now
<gonidelis[m]>
i know what isn't the difference
<gonidelis[m]>
but i don't see what is
<hkaiser>
it's the mindset that let's programmers use the one or the other
<K-ballo>
the goals
<hkaiser>
yes, that too
<K-ballo>
they target entirely different mindsets
<K-ballo>
C tries to be a super simple (as in few features) ultra portable high level assembly
<K-ballo>
C++ tries to be a super effective high-level design tool
<K-ballo>
most C++ features are geared towards design
<K-ballo>
most C features are geared towards talking to the cpu
<gonidelis[m]>
What hkaiser said ^^
<gonidelis[m]>
oh! also the last two sentences od K-ballo feel right
<gonidelis[m]>
of*
<gonidelis[m]>
that's why I was telling that they are different beyond the semantics similarities. they drive towards opposite directions
<K-ballo>
I wouldn't say opposite, C++ is based in C after all, same direction but way further goals
<K-ballo>
C is as far from C++ as it is from Java, almost, yeah
<gonidelis[m]>
\o/
<K-ballo>
the difference is C++ was created to be C compatible, Java wasn't
<K-ballo>
Java dropped a number of things, tweaked others, etc
<K-ballo>
C++ wanted valid C code to be valid C++ code, so it stick to it
<hkaiser>
what about: the main difference between C and C++ is that C++ has destructors ;-)
<K-ballo>
what will happen when C gets destructors?
<gonidelis[m]>
lol
<gonidelis[m]>
hahahah
<gonidelis[m]>
can i destroy things with assembly?
<gonidelis[m]>
should ^^
<K-ballo>
you can destroy things with C too
<K-ballo>
destructor is the compiler "scheduling" those destructions for you
<gonidelis[m]>
yeah sure...it's just that it would never be my initial intention
<K-ballo>
?
<gonidelis[m]>
i won't schedule the destruction, as you said
jehelset has joined #ste||ar
k-ballo[m] has quit [Ping timeout: 245 seconds]
sestro[m] has quit [Ping timeout: 250 seconds]
rainmaker6[m] has quit [Ping timeout: 258 seconds]
srinivasyadav227 has quit [Ping timeout: 258 seconds]
mdiers[m] has quit [Ping timeout: 258 seconds]
rachitt_shah[m] has quit [Ping timeout: 245 seconds]
CynthiaPeter[m] has quit [Ping timeout: 258 seconds]