hkaiser 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/
<zao>
Heaven knows where khuck runs their buildbot.
<hkaiser>
zao: on his machines
<zao>
Handy :)
<zao>
Enough sleuthing from my side, back to writing presentation material I don't want to make :)
<heller__>
good luck
aserio has joined #ste||ar
<aserio>
heller__: give me a second
<heller__>
aserio: got it
<heller__>
now failing differntly at runtime ;)
shahrzad has joined #ste||ar
shahrzad_ has joined #ste||ar
<aserio>
heller__: if you have further questions feel free to ask shahrzad
<heller__>
shahrzad: hey, I was wondering how you created the movielens files. They don't seem to match with the cvs files to be found on the internets
<shahrzad_>
Hi, I was just emailing you about that
<heller__>
great
<shahrzad_>
so I used a code to create the current dataset out of the original dataset to use it as the input of the algorithm
<heller__>
did you ever run into illegal instruction errors?
<shahrzad_>
no
<heller__>
hmm
<shahrzad_>
are you using openblas?
<shahrzad_>
the openblas on Rostam is compiled for Haswell architecture
<shahrzad_>
if you run it on Marwin it will fail
<heller__>
I am running on my own machine
<heller__>
but yeah ... I guess it has something to do with blas/lapack
<K-ballo>
zao: what will you be presenting?
<zao>
K-ballo: One of my colleagues volunteered me for an internal 1h seminar on stuff.
<zao>
Last week I talked about containers and singularity to assorted site staff. This friday I'm talking about the module system and how we build software.
<heller__>
shahrzad: btw, the paths don't expand '~'
<heller__>
yup: 0x00007ffff3614004 in dgemm_oncopy () from /opt/apps/spack/opt/spack/linux-debian8-x86_64/gcc-8.1.0/openblas-0.3.0-s7hm7f4sy3xcfpieqwnxmpdvzioninrf/lib/libopenblas.so.0
<heller__>
shahrzad: I guess the first number in the output is the runtime?
mbremer has joined #ste||ar
<heller__>
hkaiser: is there any other benchmark I might want to look into?
<heller__>
so yes, it is my openblas installation ...
<heller__>
great
<hkaiser>
heller__: we have 3 or 4 algorithms written up using Python, PhySL, and C++ for comparisions
<heller__>
ok
<heller__>
hkaiser: I wanted to look into the issue with the set_thread_lco_description
<hkaiser>
ok
<hkaiser>
good move
<heller__>
So I wanted to reproduce Kevin's numbers ...
<heller__>
900 seconds is just too long for a quick turnaround though
<heller__>
something like 10 seconds might be better there ;)
<hkaiser>
you can use als with almost any data size
<heller__>
I don't even know what als stands for :/
<hkaiser>
ALS == alternating least square
<heller__>
ok
<heller__>
how do I alternate the problem size?
<hkaiser>
make it smaller ;-)
<hkaiser>
use less columns or rows
<hkaiser>
(see command line options)
<heller__>
ok, that's something to start with
<hkaiser>
heller__: shahrzad can help with every question here, it's her code
aserio has quit [Quit: aserio]
<heller__>
hkaiser: the move to clang7 should be good now
<heller__>
except for the usual...
Anushi1998 has quit [Ping timeout: 246 seconds]
Anushi1998 has joined #ste||ar
<heller__>
hkaiser: which test do you usually run for profiling?
Anushi1998 has quit [Ping timeout: 244 seconds]
Anushi1998 has joined #ste||ar
Anushi1998 has quit [Ping timeout: 252 seconds]
Anushi1998 has joined #ste||ar
<hkaiser>
heller__: Kevin runs the ALS and LRA algorithms on a nightly bases and collects perf data