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/
zbyerly has quit [Ping timeout: 240 seconds]
zbyerly has joined #ste||ar
zbyerly has quit [Remote host closed the connection]
zbyerly has joined #ste||ar
bikineev has quit [Remote host closed the connection]
zbyerly has quit [Ping timeout: 255 seconds]
EverYoung has quit [Ping timeout: 276 seconds]
mcopik has quit [Ping timeout: 240 seconds]
K-ballo has quit [Quit: K-ballo]
eschnett has joined #ste||ar
<github>
[hpx] hkaiser closed pull request #2784: Error with reduce and transform reduce fixed (master...master) https://git.io/v73Bn
<diehlpk_work>
hkaiser, I use export HPXRUN_ARGS="--hpx:bind=none" as in the HPXCL example and -t 1 to run the tests on circle-ci.
<hkaiser>
diehlpk_work: ok, interesting
<diehlpk_work>
hkaiser, It works in my locasl machine
<diehlpk_work>
I will use the latest version from master on my local machine to verify. Mine is two days old
<heller>
aserio: sorry, can't make it today :(
bibek_desktop has joined #ste||ar
bibek_desktop has quit [Client Quit]
zbyerly_ has quit [Ping timeout: 246 seconds]
taeguk has joined #ste||ar
EverYoung has joined #ste||ar
aserio has quit [Ping timeout: 246 seconds]
Matombo has quit [Ping timeout: 240 seconds]
EverYoung has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
mcopik has joined #ste||ar
Matombo has joined #ste||ar
pree_ has joined #ste||ar
aserio has joined #ste||ar
mars0000 has joined #ste||ar
Kiril_ has quit [Quit: Page closed]
Matombo has quit [Remote host closed the connection]
<hkaiser>
jbjnr: yt?
<github>
[hpx] hkaiser pushed 1 new commit to resource_partitioner_jb: https://git.io/v7ZoB
<github>
hpx/resource_partitioner_jb 2a3aaaf Hartmut Kaiser: Add test for thread related perf counters
Matombo has joined #ste||ar
Matombo has quit [Remote host closed the connection]
pree_ has quit [Read error: Connection reset by peer]
eschnett has quit [Quit: eschnett]
pree_ has joined #ste||ar
david_pfander has quit [Ping timeout: 258 seconds]
taeguk has quit [Quit: Page closed]
pree_ has quit [Read error: Connection reset by peer]
vamatya has joined #ste||ar
pree_ has joined #ste||ar
aserio has quit [Ping timeout: 258 seconds]
bibek_desktop has joined #ste||ar
<diehlpk_work>
hkaiser, Any sugesstions for my bug?
vamatya has quit [Ping timeout: 248 seconds]
pree_ has quit [Ping timeout: 276 seconds]
vamatya has joined #ste||ar
zbyerly_ has joined #ste||ar
pree_ has joined #ste||ar
mars0000 has quit [Quit: mars0000]
hkaiser has quit [Ping timeout: 246 seconds]
taeguk has joined #ste||ar
<taeguk>
Coroutine is available in HPX?
<taeguk>
I found some codes about coroutine. But, there is only one example or unittest for coroutine in HPX.
<K-ballo>
that's an ambiguous question..
zbyerly_ has quit [Ping timeout: 246 seconds]
<taeguk>
K-ballo: what do you mean? As I think, there are codes for only core implementations of coroutine. Is the problem lack of keywords like await and yield?
<K-ballo>
taeguk: I can't tell what your question is, and I can imagine several different things you might be asking by it
hkaiser has joined #ste||ar
<heller>
taeguk: in other words: what do you want to do with that information you are asking for
<taeguk>
heller: This is just curiosity.
<taeguk>
I just want to know the progress of coroutine in HPX.
<heller>
Ok. So the answer is: we support the C++ coroutine TS
pree__ has joined #ste||ar
<K-ballo>
as can be seen in the example and the unit test
bibek_desktop has quit [Read error: Connection reset by peer]
Reazul has joined #ste||ar
<Reazul>
Hello, I had a few question about running a simple example in distributed environment
<Reazul>
I see that HPX supports running on top of MPI. What command line do we use? I was trying to modify the allgather example as a first step and print in all the ranks
<heller>
Use the same as if you'd start a regular hybrid mpi application
<Reazul>
but I am failing, also I am confused about how the data transfer takes place
<Reazul>
HPX takes care of the communication , right?
<heller>
What is your error? How would I reproduce it?
<heller>
Yes, the runtime does the communication
<heller>
The underlying message passing layer is considered an implementation detail
<Reazul>
so my understanding so far, we create a server component like in the all gather example
<Reazul>
and run it with mpirun
<heller>
Ok, that will create a component and then exit
<heller>
I guess
<Reazul>
I see, I think I did not load the componenet with the ini file that is why!
<heller>
Could be
<heller>
I don't have enough information to fully understand the problem you are having
<Reazul>
Right. I understand. I will share the full code in a bit
<Reazul>
I am getting this error
<Reazul>
requesting a different number of localities (2) than have been assigned by SLURM (6), the application might not run properly.
<hkaiser>
Reazul: normally you don't need ini files to load components
<Reazul>
I am trying to have my own component
<hkaiser>
Reazul: yah, just ignore that slurm warning
<Reazul>
like trying the all gather as my app not a part of HPX
<hkaiser>
Reazul: I have no idea what you're doing nor what your problem is
<hkaiser>
sorry
<Reazul>
I am trying to come up with a very simple chain example as shown here http://imgur.com/a/FT7rZ
<hkaiser>
ok
<heller>
Reazul: openmp?
<heller>
Sorry, openmpi
<heller>
If yes, you can safely ignore the warning
<Reazul>
yes I am using openmpi
<heller>
Ok, then just ignore it
<Reazul>
I looked at allgather example to comeup with it, not so simple(I am not used to c++) so it is extra difficult
taeguk has quit [Quit: Page closed]
<heller>
So, for that loop, you could ignore components as a start
<Reazul>
I wanted to but I could not figure out how the data part could be handled with out the server component
<heller>
Ok, where you able to implement it without any data?
<Reazul>
No I did not try that. I am even sure how to get the different localities without the component.
<heller>
Just a simple test that prints out where the function is executed or so?
<Reazul>
yeah some thing like that as a first step
<heller>
there is hpx::find_all_localities()
<heller>
This will give you the list
<Reazul>
I see, what is the block type in allgather?
<heller>
Then you need to look into how to get plain actions going
<Reazul>
Right
<heller>
Which example exactly are you looking at?
<Reazul>
simple, I want to launch task in different localities, and create dependencies, and want to transfer data among the tasls
<heller>
one first step would be to first get rid of the notion of different localities, and just think about how to create those components and to get the flow going
<heller>
the distribution, that is where the data is actually distributed can be done after that first step just as well
<hkaiser>
Reazul: I'd say first write a C++ cocde (without HPX) which does what you want, assuming that the objects will be distributed later
<Reazul>
Well the notion of task does not exist if I am trying to do it with out HPX
<Reazul>
the whole purpose is to use HPX :)
<Reazul>
I have tried previously creating embarrasibgly parallel tasks using HPX
<hkaiser>
Reazul: then write an hpx code which is not distributed
<Reazul>
but has dependencies among tasks. I see. Let me try that.
<hkaiser>
do everything local, create three instances of an object, then use hpx::async to call the member of those objects in the order you want
<hkaiser>
transforming this to a distributed code is a matter of some simple rules to apply
<heller>
almost ;)
<hkaiser>
heller: it's the easiest way to get to grips what hpx does
<heller>
sure
<hkaiser>
or how to use it for distributed things
<Reazul>
Ok, Let me try that. Thanks :)
<heller>
just saying that it might sound simpler as it could eventually turn out
<heller>
but yes, what hkaiser suggests, will be a good first step
patg[[w]] has joined #ste||ar
<Reazul>
What is the best way to create a dependency? one way is we have a future and wait on the future.get(). Is there a way where we can pass a future when spawning a task and the runtime makes sure the task runs after that future is valid?
<heller>
yes, with .then
<Reazul>
ok
<heller>
but it is more like the other way around, you specify a continuation that is ran after the future is ready
<heller>
or you use dataflow
<Reazul>
ok
<Reazul>
any example showing how to use the dataflow?
<hkaiser>
Reazul: get the main meat implemented before using dataflow, really
<hkaiser>
use future::get() for now
<hkaiser>
there is plenty of work to be done before even starting to think about asynchronoy, continuations, etc.
<hkaiser>
even more as you don't need dataflow for a simple linear chain of dependencies
bibek_desktop has joined #ste||ar
<hkaiser>
future::then() is the perfect tool for this
bibek_desktop has quit [Quit: Leaving]
mars0000 has joined #ste||ar
<github>
[hpx] diehlpk opened pull request #2786: Update CMake Minimum Version to 3.3.2 (refs #2565) (master...master) https://git.io/v7ns2
pree__ has quit [Ping timeout: 276 seconds]
<jbjnr>
hkaiser: here now
<hkaiser>
jbjnr: I have implemented the perf counters now
<hkaiser>
is there anything missing in order to reproduce the old functionality?
<jbjnr>
Not sure I follow you
<jbjnr>
I don't think we're missing anything else. the perf counters was the main thing ...
<hkaiser>
jbjnr: the rp branch
<jbjnr>
however - I have found a strange bug I'm looking into that makes my stuf run oddly on daint
<hkaiser>
jbjnr: I would like to merge that as soon as we have the functionality reproduced which we've had before
<jbjnr>
great
<jbjnr>
let me fix this bug, then I think I'm happy
<jbjnr>
another day ot two at most
<hkaiser>
what was missing for this except for the perf counters
<jbjnr>
^^
<hkaiser>
k
<jbjnr>
nothig else AFAICR
<hkaiser>
ok
<jbjnr>
if I memnerber anyhting, I'll ping you
<hkaiser>
let's get it merged in then, once you're happy
<hkaiser>
thanks
<hkaiser>
things diverge too much otherwise
<jbjnr>
frankly - I'd be delighted if you merge it now, but I know there a few things fishy somewhere ...
<hkaiser>
we will not find out without using it
<hkaiser>
jbjnr: could you create a list of things you know about, pls?
<hkaiser>
anyways, gtg
hkaiser has quit [Quit: bye]
pree_ has joined #ste||ar
patg[[w]] has quit [Quit: Leaving]
pree_ has quit [Ping timeout: 240 seconds]
pree_ has joined #ste||ar
Matombo has joined #ste||ar
wash has joined #ste||ar
<wash>
aserio: ping
<wash>
aserio: working on the ASPLOS paper updates - I was just wondering, though, how you imported from the other git repo? I assume just copy/paste?
<aserio>
yep
<aserio>
why?
<wash>
aserio: I guess I was a little hesistant to loose the revision history b/c we had so many commits on it/so many different authors, but I think it's fine
<wash>
aserio: btw send me your photos from the wedding!
<aserio>
I think I sent you a dropbox link
<wash>
aserio: oh
<wash>
aserio: follow up - do you take notes from the OB telecons? If so where would I be able to find last weeks notes?
<aserio>
I can send them to you
<wash>
also
<wash>
Each reference must specify
<wash>
all
<wash>
authors (no
<wash>
et al.
<wash>
err that pasted bad
<aserio>
yep
<wash>
Each reference must specify all authors (no et al.). <- that's gonna be a challenge for us, we used that trick to save space :p
hkaiser has joined #ste||ar
<aserio>
The references do not cout count
<wash>
aserio: I'm reading the information on their formatting requirements right now... have you looked at that yourself?
<aserio>
yep
<wash>
aserio: Unless I misunderstand things, they have special templates for their conference, e.g. it's not just one of the standard tempaltes?
<wash>
or actually, it sounds like they don't require a specific template, they just have a set of guidelines you have to meet.
<aserio>
wash keep reading
<wash>
aserio: they have a suggested package, yah
<aserio>
If you are reading the Paper formatting document it specifically says in paragraph 2 that they accept SIGPLAN style/class
<aserio>
this is the class of the document
<wash>
aserio: hrm... can you link me to what you are looking at?
<wash>
aserio: yah - it does indicate in the first paragraph that it's recommended but not required to use their formatting though.
<wash>
I guess let me try it and see how much work it looks like
<aserio>
wash: if you want to do it go for it
<aserio>
I am not going to stop you :)
<wash>
:p
<wash>
aserio: btw, I think we need to comment out the acknowledgements section for the submission, due to the double blind req
<aserio>
Fine by me
<wash>
hrm
<wash>
aserio: at one point I had a .txt file or spreadsheet with the list of authors and all their info (country, town, etc) that the GB submission wanted... do you have any idea where that might be?
<wash>
ah I found it in my backups. Never saved it
<aserio>
like the one here: \operation_bell\SC17 ?
pree_ has quit [Ping timeout: 246 seconds]
<wash>
that's the older one, I could've sworn I made one while doing the submission. checking my machine...
<aserio>
hkaiser: yt?
<hkaiser>
here
mars0000 has quit [Quit: mars0000]
<aserio>
your using checkpoint = checkpoint<> trick didn't work
<aserio>
the compiler didn't like that it had the same name
<aserio>
hkaiser: did you do this with async?
<hkaiser>
aserio: you can't use the same name for this
aserio has quit [Quit: aserio]
<wash>
Hrrrrm
<diehlpk_work>
wash and parsa[[w]] Could you may do the evaluation?
<parsa[[w]]>
yeah
<wash>
yep, doing that now, will send a mail soon
parsa[[w]] is now known as parsa[w]
<github>
[hpx] hkaiser closed pull request #2656: Reduce MAX_TERMINATED_THREADS default, improve memory use on manycore… (master...terminated_threads) https://git.io/vHCmV
RostamLog_ has joined #ste||ar
parsa[w] has joined #ste||ar
wash has quit [Remote host closed the connection]
diehlpk_work has quit [Ping timeout: 247 seconds]
diehlpk_work has joined #ste||ar
RostamLog has quit [Ping timeout: 246 seconds]
hkaiser has joined #ste||ar
K-ballo has joined #ste||ar
Matombo has quit [Remote host closed the connection]
hkaiser has quit [Read error: Connection reset by peer]