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/
<K-ballo>
heller: maybe we should consider -fsyntax-only, caveats and all, but only for circle ci
Vir has quit [Ping timeout: 240 seconds]
EverYoung has joined #ste||ar
Vir has joined #ste||ar
EverYoung has quit [Ping timeout: 255 seconds]
<github>
[hpx] hkaiser created enforce_function_initialization (+1 new commit): https://git.io/vdiP6
<github>
hpx/enforce_function_initialization 9a01810 Hartmut Kaiser: Make sure the function vtables are initialized before use
<github>
[hpx] hkaiser opened pull request #2945: Make sure the function vtables are initialized before use (master...enforce_function_initialization) https://git.io/vdiPi
<github>
[hpx] hkaiser opened pull request #2946: Adding explicit feature test for thread_local (master...fixing_2137) https://git.io/vdiPX
hkaiser has quit [Quit: bye]
K-ballo has quit [Quit: K-ballo]
pagrubel has joined #ste||ar
pagrubel has quit [Ping timeout: 255 seconds]
<github>
[hpx] sithhell force-pushed circle_2 from 39080e1 to a0cdcfc: https://git.io/vdKfd
<github>
hpx/circle_2 a0cdcfc Thomas Heller: Switching to CircleCI 2.0...
<github>
[hpx] sithhell closed pull request #2944: Attempting to solve affinity problems on CircleCI (master...fixing_affinity_issue) https://git.io/vdifI
<github>
[hpx] sithhell pushed 1 new commit to fix_this_thread_executor: https://git.io/vdiQm
<github>
hpx/fix_this_thread_executor a71173b Thomas Heller: Merge branch 'master' into fix_this_thread_executor
EverYoung has joined #ste||ar
EverYoung has quit [Ping timeout: 255 seconds]
david_pfander has joined #ste||ar
<jbjnr>
heller: yt? I am getting an error at runtime that is "RTTI symbol not found for class 'hpx::lcos::local::detail::task_object<void, hpx::util::detail::deferred<void (*(int&&, bool&&, char const*&&))(unsigned long, bool, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)>, hpx::lcos::detail::task_base<void> >'"
<jbjnr>
any idea who might be looking for the RTTI and not finding it?
<heller>
hmmm
<heller>
not from the top of my head
<heller>
what does the debugger say?
<jbjnr>
it seems to be coming from somewhere deep inside async, but I didn't find the exact place where it happens. No actual error, just a message emitted by the runtime
<jbjnr>
(or by libstdc+
<jbjnr>
+ more likely)
<jbjnr>
ooh. maybe not. I found something. I'll be back ina mo
<heller>
ok
<jbjnr>
it's coming from the deereferencing of a boost::intrusive_ptr
<jbjnr>
according to google, it's caused by gdb not finsding a symbol ...
<zao>
Seems like I gain time, even if I'm swapping :)
EverYoung has joined #ste||ar
<heller>
zao: yay!~
<heller>
jbjnr: ahh. you can ignore that, gdb sometimes does this, I don't know where this comes from
EverYoung has quit [Ping timeout: 246 seconds]
<zao>
heller: Installed an additional 16G of memory yesterday after running AMD's tests.
<heller>
zao: wwee!
<zao>
Now 8c/16t w/ 32G of mem.
<msimberg>
what do i need for setting the number of threads from an hpx.ini file?
<msimberg>
--hpx:ini=hpx.os_threads=n works for me but if i do hpx:config=hpx.ini it doesn't
<msimberg>
and hpx.ini contains
<msimberg>
[hpx]
<msimberg>
os_threads = 1
<heller>
hmmm
<heller>
should work
<heller>
what does the application print when adding a --hpx:dump-config?
pree has joined #ste||ar
<zao>
I forget if I asked this before, but is there any way to build all the test binaries without running the test suite at the end?
<jbjnr>
make tests.unit tests.regression
<jbjnr>
tests.performance
<jbjnr>
make help | grep tests. top three
<zao>
Ah, great!
pree has quit [Read error: Connection reset by peer]
david_pfander has quit [Remote host closed the connection]
david_pfander has joined #ste||ar
pree has joined #ste||ar
<msimberg>
@heller dump-config output is consistent, i.e. it still shows os_threads = 4 when i use the ini file, and with the other methods i can change it
<heller>
msimberg: looks like a bug
<heller>
I assume your machine has 4 cores?
<msimberg>
if i put a made up config value in the ini file it shows up in dump-config so it's reading the file
<msimberg>
yeah, 4 hyperthreads
<msimberg>
also tried just using hpx.os_threads directly in the ini file, but it makes no difference
<msimberg>
unless it's something one of you can fix very easily i could have a go at fixing it, would be a good exercise
<jbjnr>
msimberg: Are you using a simple (hello world) example, or does the example use the resource partitioner -cos looking at the code, the ini file won't be parsed at the right time of the RP is being used.
<jbjnr>
the changes we made to use RP miht have broken this - the threads setup is done first, before the rest of init
<msimberg>
this was with 1d_stencil_4_parallel
<jbjnr>
ok
<msimberg>
which is not using the rp
<jbjnr>
I'm testing hello_world right now
<jbjnr>
it is broken
<jbjnr>
file an issue
<jbjnr>
we probably broke it with the changes to setup when RP was introduced
<msimberg>
ok
<msimberg>
thanks for testing
<jbjnr>
need to move the reading of the ini file to before the thread binding stuff. what a pain.
pree has quit [Read error: Connection reset by peer]
<msimberg>
@jbjnr i also tried the simple_resource_partitioner example
<msimberg>
if i run it with simple_resource_partitioner --use-pools --pool-threads=3 it segfaults
<jbjnr>
the RP example would definitely fail, because we only parse the command line for threads before calling init
<msimberg>
i.e. there's only one thread left for the main pool
<jbjnr>
it shouldn't segfault.
<msimberg>
if i do --pool-threads=4 it correctly tells me that the main pool is empty
<jbjnr>
any idea what the stacktrace is?
pree has joined #ste||ar
<msimberg>
no stacktrace, can try in debug
<jbjnr>
don't worry. I'm testing that stuff hre anyway, once my build is working again, I'll try it
<msimberg>
ok, thanks
<heller>
everything is using the RP
<heller>
so my guess is: the ini file gets parsed and the values set. After that, the RP runs and overwrites that value
<heller>
something along those lines
<heller>
would have to dig as well
<jbjnr>
no. The RP only checks the basic command line options (threads/bind etc) in a quick parse, then it sets up stuff and calls the old init. The old init reads the ini file
<jbjnr>
it's is all happening too late.
<jbjnr>
we'lll need to add a parse of the ini file to the pre-RP init phase
<jbjnr>
it'll only tak hartmut 5 mins to fix it.
K-ballo has joined #ste||ar
<heller>
aha
<heller>
alrighty. I have cmake and python scripts now that matches build options with available software
<heller>
and assigns workers etc.
hkaiser has joined #ste||ar
zbyerly_ has joined #ste||ar
<jbjnr>
". We are delighted to inform you that your submission entitled HPX -- A open source lib for Parallelism and Concurrency has been accepted! We have attached the reviews below. "
<zbyerly_>
jbjnr, lol i was about to say the same thing!
<zbyerly_>
wooop woop
<jbjnr>
they have low standards indeed!
<zbyerly_>
jbjnr, that's cool with me
<jbjnr>
zbyerly will you be presenting it?
<zbyerly_>
that's the plan, afaik
<zbyerly_>
heller asked me to
<jbjnr>
you'll be the only one there probably!
<jbjnr>
I think the reviewers expectations were much higher. for an open source workshop I thought the material in the paper was about right. Not too much detail, just an overview of the project.
<zbyerly_>
we got 0, 0, 2
<heller>
2 borderline paper, ouch
<zbyerly_>
""If selected for a talk, I would like to suggest the authors consider offering a modest demonstration of their software library."
<zbyerly_>
Hello world from locality 0!
<heller>
lol
<zao>
We need to show that HPX works? Oh gods.
<zbyerly_>
review 1: "The writeup to me looked like a summary of HPX."
<zbyerly_>
uh.. yup
<heller>
that kind of was the intention ;)
<zbyerly_>
they dinged us about our sloppy paper for sure
<jbjnr>
well it's a shit workshop so they got what they deserved
<zbyerly_>
jbjnr, you are so negative!
<jbjnr>
absolutley.
<jbjnr>
"Singing the Blues isn't about making yourself feel better, it's about making everyone else feel as bad as you"
<zbyerly_>
jbjnr, yeah life sucks and then you die
<heller>
diehlpk: the first thing i'd check is if it happens with an empty hpx_main. And then slowly enable piece by piece until i get the failure again, you can then pretty easily isolate the problem
<github>
hpx/master d52b477 Denis Blank: Implement an API for asynchronous pack traversal...
<github>
hpx/master 0139b9d Denis Blank: Prepare traverse_pack_async for using it inside when_all
<github>
hpx/master e9e8878 Denis Blank: Inherit the visitor rather than embedding it as a class member...
<hkaiser>
K-ballo: ^^
<hkaiser>
async pack has been merged
<K-ballo>
alright.. time to dive into the dataflow overload set
<K-ballo>
fun.
jaafar has joined #ste||ar
eschnett has quit [Quit: eschnett]
<aserio>
hkaiser: what does nil{} evaluate to?
<hkaiser>
nil{}
<hkaiser>
nil, nada, nothing
<hkaiser>
phylanx::execution_tree::valid(nil{}) will be false
<aserio>
I am trying to pass something to test_generate_tree
<aserio>
I tried to create an instance of nil to compare it to the result but it did not like that
<hkaiser>
use phylanx::execution_tree::valid instead
<zao>
`nil`? Sounds like someone doesn't plan on working under a Obj-C++ compiler :)
<hkaiser>
lol
<github>
[hpx] biddisco force-pushed alloc_membind from 489d456 to 6524d7c: https://git.io/vdnl9
<github>
hpx/alloc_membind 86cb4cd John Biddiscombe: Add hwloc alloc_membind support and nodeset bitmap type...
<github>
hpx/alloc_membind d6807b7 John Biddiscombe: Move hwloc_bitmap_ptr to threads namespace
<github>
hpx/alloc_membind e7a7f1d John Biddiscombe: Add ostream operator for hwloc bitmap printout
<jbjnr>
does circleci use the new workflows stuff now for PRs, or is it still using the old one
<hkaiser>
still old, I think
<hkaiser>
heller: has not created a PR for this yet
aserio has quit [Quit: aserio]
rod_t has left #ste||ar [#ste||ar]
jaafar has quit [Ping timeout: 248 seconds]
Aalice has joined #ste||ar
<Aalice>
I was wanting to build a NERSC module for 1.0.0. Before I downloaded 0.9.99.tar.gz with wget. Is there a new way to do that now? I don't see hpx_1.0.0.tar.gz under stellar.cct.lsu.edu/files
<Aalice>
In particular, I'm trying to update the directions we have under:
<Aalice>
I'm wondering if I still need to swap out for older modules of gcc, etc. , but I'll see what happens. And there is no longer a specific knl branch is there?
<hkaiser>
Aalice: no special knl branch, just a special cmake toolchain file