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/
bikineev has quit [Remote host closed the connection]
eschnett has quit [Ping timeout: 255 seconds]
eschnett has joined #ste||ar
hkaiser has quit [Quit: bye]
parsa has quit [Quit: Zzzzzzzzzzzz]
K-ballo has quit [Quit: K-ballo]
ajaivgeorge has quit [Quit: ajaivgeorge]
mars0000 has joined #ste||ar
vamatya has quit [Ping timeout: 260 seconds]
pree has joined #ste||ar
pree has quit [Quit: AaBbCc]
pree has joined #ste||ar
patg has quit [Quit: This computer has gone to sleep]
parsa has joined #ste||ar
zbyerly_ has quit [Ping timeout: 268 seconds]
parsa has quit [Quit: Zzzzzzzzzzzz]
pree has quit [Quit: AaBbCc]
pree has joined #ste||ar
pree has quit [Read error: Connection reset by peer]
<heller>
Finally no guessing if it was a segfault or not
<heller>
Thanks
<hkaiser>
most welcome, thanks Chris for this
<hkaiser>
to*
<heller>
Yeah
<heller>
What's the consequence for the gsoc project?
<hkaiser>
heller: no idea, have not heard from ajaivgeorge in a while
<hkaiser>
he wanted to look into automatic stack increase, iirc
<heller>
Ok
<zao>
soc is soon over, isn't it?
<hkaiser>
zao: another month or so
<ajaivgeorge>
hkaiser: i am working on parallel version of the segmented find end. lots of interesting issues with that. I am not sure about why you mentioned automatic stack increase.
<zao>
abresting (sic?)
<hkaiser>
ajaivgeorge: sorry, I meant ABresting
<hkaiser>
:D
<hkaiser>
I know you're working hard on the segmented algorithms
<hkaiser>
ajaivgeorge: do you think you could fix the existing segemented transform_reduce as you go?
<ajaivgeorge>
ok cool :D, with regard to the PR, i have updated it with the reduce/transform_reduce fix. There are merge errors so should I pull from hpx/master and fix it in the PR will it be handled in the merge?
<ajaivgeorge>
Have already fixed transform_reduce.
<hkaiser>
ajaivgeorge: yes, merge with master locally and (force-) push to your branch, that will update th ePR
<hkaiser>
ajaivgeorge: excellent!
<ajaivgeorge>
ok, i will do that. Another PR with find/find_if/find_if_not is ready and I have a working sequential find_end and find_first_of. Parallel version still eludes me.
<hkaiser>
nice
<hkaiser>
ajaivgeorge: the problem here are search sequences which possibly straddle partition boundaries, no?
<ajaivgeorge>
yep, so me and marcin came up with a custom return which returns if a suffix or prefix or whole sequence or any two of them are present.
<hkaiser>
ok
<hkaiser>
sounds like a good startin gpoint
<ajaivgeorge>
yea, so i have implemented a base version of that and now am navigating through seg faults. :P
<hkaiser>
lol
<hkaiser>
good luck!
<hkaiser>
ajaivgeorge: marcin will definitely be able to help, if needed
<ajaivgeorge>
yep i will get back to him tomorrow, once i have ironed out as many errors as i can through valgrind and gdb. cya, then. Just wanted to update you.
<hkaiser>
ajaivgeorge: thanks, pls respond to that email I sent this morning
<ajaivgeorge>
oh just saw it. I will send you a report by Tuesday.
<hkaiser>
thanks
<zao>
i wonder if i can stabilize my ryzen as a conpile machine, not enough RAM to link tests :P
parsa has joined #ste||ar
parsa has quit [Quit: *yawn*]
patg has joined #ste||ar
pree has quit [Read error: Connection reset by peer]
vamatya has joined #ste||ar
<ABresting>
hkaiser:
<hkaiser>
here
<ABresting>
hkaiser: stuck with co-routine implementation via libsigsegv
<hkaiser>
ABresting: ok
<hkaiser>
what's the problem?
<ABresting>
and yea increasing stack size wasomething that can be done
<ABresting>
problem is when thread calls itself enough times that process memory gets overflow and for process thats segmentation fault as the thread is trying to read that memory
<ABresting>
so the problem is libsigsegv is quite not good enough for this use case
<hkaiser>
ABresting: please elaborate, I'm not sure I understand...
<hkaiser>
ABresting: have you seen the latest changes adding stack overflow detection to HPX?
<ABresting>
ok so assume a process has 30 kb memory, and thread generating inside takes 10 kb each thread so if thread calls itself multiple times them same amount 10kb is allocated, when auch allocations exceeds the process memory then process indicates segmentation fault not stack overflow
<hkaiser>
ABresting: sure, I think that's expected
<hkaiser>
nothing we can do about this
<ABresting>
yes I have seen it amd I could jave made this earlier but was specifically told me not to do any derived work
<ABresting>
but thats the challenge isnt it?
<hkaiser>
ABresting: I don't think the changes to HPX comprise 'derived work'
<ABresting>
that every use case should justify
<hkaiser>
from GPL that is
<ABresting>
haha I understand that, but 'derived work' is something that my mentor not expecting from me.
<ABresting>
lil tensed about the problem at hand
vamatya has quit [Ping timeout: 240 seconds]
<ABresting>
but I am gonna try it for next two days, if till then its not solved then as you said 'nothing we can do about this'. sad life
<ABresting>
then I will need your help expertise in the matter.
<ABresting>
new change in HPX is solving my use case?
<hkaiser>
ABresting: you tell me - I don't know
mars0000 has joined #ste||ar
<ABresting>
ok let me check the test case chris wrote
ajaivgeorge has quit [Ping timeout: 240 seconds]
ajaivgeorge has joined #ste||ar
<ajaivgeorge>
setting Localities=some number in the CMakeLists for partitioned vector tests simulates multiple nodes right?
mcopik has joined #ste||ar
<hkaiser>
ajaivgeorge: only when the tests are run
<ajaivgeorge>
when they are run using ctest, right?
<hkaiser>
right
<ajaivgeorge>
I am confused because my base implementation which is supposed to work only when the sequence is entirely within in a segment is passing for all test cases. :P
<hkaiser>
lol
<hkaiser>
ajaivgeorge: run it manually
<ajaivgeorge>
same result
<ajaivgeorge>
i have put a whole bunch of printfs in my code which tell me that the data is in a single segment. In the test i am using find all localities and then using that in the constructor of the partitioned vector, exactly as in test for inclusive/exclusive scan
hkaiser has quit [Read error: Connection reset by peer]
hkaiser has joined #ste||ar
<hkaiser>
ajaivgeorge: difficult to tell from here
mcopik has quit [Ping timeout: 260 seconds]
<ajaivgeorge>
Found the issue. There was an issue with the CMakeLists that I was using (launch parameters was being reset inside the foreach($test), so localities was 1), I will rerun all the tests including the ones in the PR.