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/
<github>
[hpx] hkaiser force-pushed cmaks_cxx17_support from 5336ff6 to 9274768: https://git.io/vHiRF
<david_pfander>
jbjnr: ^ might also be interesting to you, my hpx cmake failed, because libfabric wasn't detected, even though I didn't intentionally enable it
<zao>
Optimistic detection?
<zao>
Did you start from a clean directory after the failure?
<zao>
Should default to OFF as far as I see in the source, how did it fail?
* zao
jiggles back to work
Matombo has joined #ste||ar
bikineev has joined #ste||ar
<heller_>
david_pfander: jbjnr: I am going to sign us up now. thanks again for the text
<github>
[hpx] hkaiser deleted uninitialized_default_construct at 0d33dcf: https://git.io/vHP21
<david_pfander>
Vir: Hi, I'm currently converting Vc1 stuff into Vc2 "datapar". Do you have documentation for the datapar API? And one specific question: How do I initialize a datapar variable with a scalar? (background: avx512 on KNL in the OctoTiger project)
<Vir>
david_pfander: wg21.link/p0214 is probably the best documentation at this point. `datapar<int> x = 1;`
<zao>
Does that scatter it to all components or just the first?
<Vir>
all
<david_pfander>
Vir, zao: Might be a stupid mistake on my side, checking...
<zao>
ah, // implicit broadcast constructor
eschnett has quit [Quit: eschnett]
eschnett has joined #ste||ar
<jbjnr>
did I miss anything on the first 5 minutes of the call?
<hkaiser_>
jbjnr: not really
<hkaiser_>
nothing happened
<jbjnr>
hackathon is now a thing it seems. if kevein wants to join, we can fit hime in later
<jbjnr>
^ too many ee's in that
<hkaiser_>
jbjnr: I'll mention it to him, will probably talk to him tomorrow
<jbjnr>
great. Even if he doesn't know until the last minute, it will be probably doable,
<jbjnr>
hkaiser_: did you read anything of what I wrote so far?
<hkaiser_>
jbjnr: sorry no - didn't get to it yet
<jbjnr>
ok. I'd wait a bit longer until it's more complete then
<hkaiser_>
k
<hkaiser_>
let me know
<jbjnr>
tomorrow sometime
<david_pfander>
Vir, zao: So the init was a really stupid mistake on my part. Is store == memstore? Does that work with masks, memstore(pointer, mask)?
<Vir>
david_pfander: yes memstore. Masked stores go like this: `where(mask, v).memstore(addr)`
akheir has joined #ste||ar
<david_pfander>
Vir: thanks, that's very helpful!
hkaiser_ has quit [Quit: bye]
ajaivgeorge has joined #ste||ar
EverYoung has joined #ste||ar
EverYoung has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
eschnett has quit [Quit: eschnett]
<david_pfander>
Vir: element_aligned is for unaligned instructions, right?
<Vir>
david_pfander: yes
<Vir>
the committee didn't want to suggest that arbitrary alignment is allowed - therefore the difference between vector and element alignment
eschnett has joined #ste||ar
jakemp has joined #ste||ar
bikineev has joined #ste||ar
ajaivgeorge has quit [Read error: Connection reset by peer]
ajaivgeorge has joined #ste||ar
hkaiser has joined #ste||ar
<david_pfander>
Vir: makes sense, it just a tiny bit unintuitive if you encounter it for the first time :)
<Vir>
I agree. Especially if you already know about SIMD unaligned vs. aligned differences
pree has quit [Ping timeout: 246 seconds]
<david_pfander>
heller_: Are you building with jemalloc on tave, I'm getting "<jemalloc>: Error in dlsym(RTLD_NEXT, "pthread_create")"?
<david_pfander>
(I'm currently rebuilding my build scripts, as they were on scratch and got deleted...)
vamatya has joined #ste||ar
eschnett has quit [Quit: eschnett]
pree has joined #ste||ar
hkaiser has quit [Quit: bye]
eschnett has joined #ste||ar
denis_blank has joined #ste||ar
<diehlpk_work>
Hi, I have a linker problem with templates which I do not undertand. Can someone have a look?
<diehlpk_work>
I got this error: undefined reference to `material::Elastic::Elastic(IO::deck::PD*, problem::Quasistatic<material::Elastic>*, bool)'
ajaivgeorge has quit [Quit: ajaivgeorge]
EverYoun_ has joined #ste||ar
EverYoun_ has quit [Remote host closed the connection]
EverYoun_ has joined #ste||ar
<david_pfander>
diehlpk_work: Hi Patrick, you only posted the header file, that alone of course will give you a undefined reference? Classical Problems: Forgot to define it, cpp not added to build system,...
EverYoung has quit [Ping timeout: 260 seconds]
<david_pfander>
it = the constructor
hkaiser has joined #ste||ar
bikineev has quit [Ping timeout: 258 seconds]
pree has quit [Ping timeout: 246 seconds]
<diehlpk_work>
david_pfander, I added the important part of the corresponding cpp file
<diehlpk_work>
And also [ 30%] Building CXX object src/material/CMakeFiles/libMaterial.dir/Elastic.cpp.o
<david_pfander>
diehlpk_work: thats not the same in the header: problem::Quasistaticmaterial::Elastic* (might be a formatting issue)
<david_pfander>
diehlpk_work: unfortunately, I have to go :( cu
pree has joined #ste||ar
david_pfander has quit [Ping timeout: 245 seconds]
<hkaiser>
aserio: should be fine now
<github>
[hpx] hkaiser pushed 1 new commit to serialization_access_data: https://git.io/vHPb5
<mcopik>
hkaiser: out is std::vector, in is hpx::partitioned_vector
<hkaiser>
mcopik: well, I would think this makes sense after all
<hkaiser>
hmm, not sure
<hkaiser>
does it work with a segemnted out as well?
<mcopik>
hkaiser: do you see an example of use case where a transform on non-segmented dest iterator make sense?
<mcopik>
hkaiser: yes
<mcopik>
I mean, you have to apply transform on N local partitions and then copy N-1 of them to a location of iterator, right?
<hkaiser>
I wouldn't expect it to work as if 'in' is distributed you most likely want to have 'out' being distributed as well, but who knows
<pree>
Why id_type & hpx::id_type are used interchangbly in default_distribution_policy. I got some nasty errors ( no matching call for functions ) ?
<hkaiser>
mcopik: I think the same would be true for a straight copy
<pree>
It think it's my namespace mistake
<hkaiser>
in could be segmented, out could be not
<mcopik>
hkaiser: you can always do transform inplace and then gather
<hkaiser>
preethe default_distribution_policy lives in namespace hpx, so both are valid (even if it's not nice that they are intermixed)
<hkaiser>
mcopik: is that something you'd like not to work?
<hkaiser>
pree: ^^
<pree>
Okay hkaiser . I think I'm writing some mistakes .I want to correct it
<github>
[hpx] hkaiser pushed 1 new commit to cmaks_cxx17_support: https://git.io/vHXJW
<github>
hpx/cmaks_cxx17_support df5f069 Hartmut Kaiser: Adding possibility to enforce a given C++xx mode (e.g. -DHPX_WITH_CXX14=On)
<mcopik>
hkaiser: I'd be fine if it's expected to not work if out is not segmented because I don't see a significant benefit from it
<mcopik>
I'm not against but I'm not convinced it's useful
aserio has quit [Ping timeout: 260 seconds]
<hkaiser>
mcopik: I understand, but why constrain things if not absolutely necessary?
bikineev has quit [Remote host closed the connection]
<mcopik>
hkaiser: ok, so we'll have two versions of transform
<hkaiser>
will we?
bikineev has joined #ste||ar
<hkaiser>
what did we do in parallel::copy, do we support non-segmented iterators as destination?
<hkaiser>
I think we should do the same for transform
aserio has joined #ste||ar
<mcopik>
hkaiser: I'd have to try it, but AFAIK from reading the code, segmented transfer works only if (in, out) are both segmented or non-segmented
<mcopik>
there's no case for (segmented, non-segmented)
<hkaiser>
nod
<hkaiser>
frankly, I never have thought of mixing segmented/non-segmented iterators
<hkaiser>
could be useful at times, but let's leave that out of consideration for now
<mcopik>
hkaiser: me neither
mcopik has left #ste||ar ["Leaving"]
mcopik has joined #ste||ar
<mcopik>
hkaiser: it wasn't until I looked at the scan and realized I have never even considered this case
<hkaiser>
right
eschnett has quit [Quit: eschnett]
shoshijak has quit [Ping timeout: 255 seconds]
pree has quit [Ping timeout: 246 seconds]
pree has joined #ste||ar
<mcopik>
has anyone built HPX with gcc 7.1? I'm getting a lot of warnings in hpx/util/jenkins_hash.hpp (-Wimplicit-fallthrough) and tests/unit/threads/stack_check does not compile at all
<mcopik>
-Werror=maybe-uninitialized
<mcopik>
should I create an issue for that?
<hkaiser>
mcopik: yes please, better yet, fix it ;)
<mcopik>
hkaiser: will do
<hkaiser>
mcopik: best would be to add a HPX_FALLTHROUGH macro and add a feature check for [[fallthrough]]
<hkaiser>
or simply suppress the warning for this file ;)
<mcopik>
hkaiser: HPX_FALLTHROUGH would expand to [[fallthrough]], if supported by the compiler, or to nothing?
<jbjnr>
https://pasteboard.co/gdS8bIljp.png hkaiser heller_ this is the latest run of the osu test, compared againt the cray mpi osu_bibw test. Should I include the cray result , or does it make our stuff look too bad?
<hkaiser>
jbjnr: what's the window size?
<jbjnr>
I think 100 in that example, but I can use smaller
<hkaiser>
just asking
<jbjnr>
I forget what I used now...
<jbjnr>
good that you asked
<hkaiser>
I think showing the original osu test results is fine, although these are apples and bananas
<jbjnr>
in my version of the osu tests, the window size does not make so much difference
<jbjnr>
(I have modified it quite a bit, the version in hpx repo is a bit dodgy IMHO)
<hkaiser>
it is indeed
<hkaiser>
you might want to update it with your version
<jbjnr>
I will be doing a PR soon.
<hkaiser>
great!
<jbjnr>
I have a ton of changes, but I need to put them onto one brach independent of other changes
<jbjnr>
well,apart from the rma use that's new
<hkaiser>
I hope you'll be able to do that at some point
<hkaiser>
jaafar: nice results!
<hkaiser>
jbjnr: ^^
<jbjnr>
we're actually getting pretty close to raw MPI now. still some work to do - once we are able to bypass parcels altogether ....
<jbjnr>
who the hell is jaafar ? I read that as jar-jar at first and was confused ...
<jbjnr>
sorry to jaafar - not meaning to be rude ...
EverYoung has joined #ste||ar
<K-ballo>
I choose to read it as jafar, the villain from aladdin :P
<jbjnr>
I need to read more fairy tales and less sci-fi
<zao>
Jar-Jar Bings, Microsoft stooge :P
<jbjnr>
:?
<K-ballo>
Jeff is a long time Boost and Spirit people
<K-ballo>
recently he has been playing with HPX implementation of the parallel algorithms
<K-ballo>
person?
<jbjnr>
is it hpx::async(func, locality or hpx::async(locality, func - I've forgotten
<hkaiser>
async(action, id, args...)
<hkaiser>
or async<action_type>(id, args...)
<jbjnr>
thanks
EverYoung has quit [Remote host closed the connection]
josef__k has joined #ste||ar
josef_k has quit [Ping timeout: 260 seconds]
EverYoung has joined #ste||ar
josef__k has quit [Ping timeout: 240 seconds]
EverYoung has quit [Remote host closed the connection]
EverYoung has joined #ste||ar
EverYoung has quit [Remote host closed the connection]