verne.freenode.net changed the topic of #mlpack to: http://www.mlpack.org/ -- We don't respond instantly... but we will respond. Give it a few minutes. Or hours. -- Channel logs: http://www.mlpack.org/irc/
tsathoggua has joined #mlpack
tsathoggua has quit [Quit: Konversation terminated!]
Mathnerd314 has quit [Ping timeout: 258 seconds]
tsathoggua has joined #mlpack
tsathoggua has quit [Quit: Konversation terminated!]
mentekid has joined #mlpack
mentekid has quit [Ping timeout: 276 seconds]
mentekid has joined #mlpack
< mentekid>
rcurtin, zoq: sorry for disappearing suddenly yesterday, I was out and had to leave.
< mentekid>
It is true that a very simple parallelization can happen only with #pragmas and no openMP functions - for example processing more than one query at a time is embarassingly parallel (3 lines of openMP really)
< mentekid>
But if we want more fine-grained control, we need to use fancy openMP functions - environment variables can for example set the maximum number of threads and allow/disallow nested parallelism (a thread spawning other threads spawning other threads and so on) but we can't control the CPU utilization with them
< mentekid>
What I could do, using only #pragmas, is write very simple directives that are "blind" in the sense they don't know how many threads they could spawn
< mentekid>
so if the user sets OMP_NUM_THREADS=8 and OMP_NESTED=FALSE, then 8 queries will be processed at once
< mentekid>
then when OMP_NESTED=TRUE the ReturnIndicesFromTable part will be parallelized as well, without tracking how many threads are used
lozhnikov_ has joined #mlpack
kwikadi_ has joined #mlpack
kwikadi has quit [Ping timeout: 258 seconds]
keonkim has quit [Ping timeout: 258 seconds]
lozhnikov has quit [Ping timeout: 258 seconds]
kwikadi_ is now known as kwikadi
lozhnikov_ is now known as lozhnikov
keonkim has joined #mlpack
Mathnerd314 has joined #mlpack
marcosirc has joined #mlpack
mentekid has quit [Remote host closed the connection]
< rcurtin>
lozhnikov: I will see if I can merge that PR tonight but it is more likely to be Monday
< rcurtin>
this karting championship is more time consuming than I expected
< rcurtin>
I just need to drive like Argentina plays and I think I will be fine :)
< zoq>
wow, nice first and second place in round 3 :)
< lozhnikov>
rcurtin: Congratulations, great results!. It's ok, I am working on the vantage point trees
< zoq>
mentekid: I agree, using #pragmas is a very clean and easy way to add parallelization. Do you think, that in your case, using OpenMP functions could improve the runtime compared with #pragmas?
tsathoggua has joined #mlpack
tsathoggua has quit [Client Quit]
mentekid has joined #mlpack
nilay has joined #mlpack
mentekid has quit [Ping timeout: 244 seconds]
nilay has quit [Ping timeout: 250 seconds]
< zoq>
nilay: Can you take a look at the comments in https://github.com/mlpack/mlpack/pull/703 and https://github.com/mlpack/mlpack/pull/696? Ignore the comments about possible performance improvements for now. We like, to merge the code but there are some things we should fix first. Also, it looks like you can easily fix the Travis build error: