K-ballo 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/
nanmiao has joined #ste||ar
hkaiser has quit [Quit: bye]
nanmiao has quit [Quit: Connection closed]
K-ballo has quit [Quit: K-ballo]
diehlpk_work has quit [Remote host closed the connection]
bita has quit [Ping timeout: 268 seconds]
bita has joined #ste||ar
bita has quit [Ping timeout: 264 seconds]
<ms[m]>
k-ballo[m]: debug size down due to the iostreams headers, or due to some changes you have in progress? or just compared to last time you checked?
<ms[m]>
pedro_barbosa: the device vectors that you're capturing by reference do point to memory that's allocated on the gpu, but the vector instance itself is still allocated on the host and using a reference to a host variable on the gpu is not allowed
<ms[m]>
in general I wouldn't recommend implementing matrix-matrix multiplication like that on the gpu because we don't really have the facilities to implement that efficiently but if you just want to try something out you'd need to capture iterators to the device vector by value
shubh_ has joined #ste||ar
<gonidelis[m]>
ms[m]: do you know how could i retrieve the build type from my build dir?
<ms[m]>
gonidelis[m]: grep 'CMAKE_BUILD_TYPE' <build_dir>/CMakeCache.txt? or ccmake <build_dir>?
shubh_ has quit [Quit: Leaving]
<peltonp1>
I have a question about practical usage of hpx in stencil code
<peltonp1>
Say that I have a 9 point stencil and the operation is a sum of the values in the points. What is the order of magnitude for the optimal "grain"/partition size on a typical 128 core cluster node?
<peltonp1>
In the examples the partitions (=stripes) are created such that only one of the spatial dimensions in the box-like domain is split which makes handling the boundaries quite easy. However, a more general approach would be to allow the partitions to have a box like shape i.e. boundary exchange happens in all directions.
<peltonp1>
How are the partitions created in, for instance, octo-tiger or does it even use the partition type approach?
<peltonp1>
I guess what I'm asking is that has anyone experienced any problems in assuming a one-dimensional splitting for a n-dimensional domain when the stencil width increases?
<gonidelis[m]>
ms[m]: you have any build flags to suggest for performance analysis runs?
<gonidelis[m]>
ms[m]: thanks btw for the built_type one :)
K-ballo has joined #ste||ar
hkaiser has joined #ste||ar
hkaiser has quit [Quit: bye]
hkaiser has joined #ste||ar
<srinivasyadav227>
hkaiser: could you please review #5254
nanmiao has joined #ste||ar
<hkaiser>
srinivasyadav227: ok
<srinivasyadav227>
hkaiser: do we have any specific file format for GSoC proposal template?