Skip to content

R package with multi-threaded C

2 messages · Guillaume Chapron, Dirk Eddelbuettel

Original
Guillaume Chapron
# 
Hello,

I am trying to develop a R package that calls a multi-threaded C function and I am struggling on how to do this properly. The C program I originally wrote (prior to thinking about building it into a R package) uses Grand Central Dispatch library (for multi-threading on a Mac) and GSL (for RNG). Importantly, each thread has its own seed (the program runs Monte Carlo simulations). 

Now that I am looking at modifying this to be included in a R package, I am not sure how to proceed with having a parallel and portable C code using the RNG seed from R. In the book R Programming for Bioinformatics by Gentleman (2008), it says p. 201:

"The decision to have these functions (GetRNGstate and SetRNGstate) manipulate a global variable, .Random.seed, is slightly unfortunate as it makes it somewhat more difficult to manage several different random number streams simultaneously."

Has anyone been able to do this or is there a package implementing something similar? I could always have the C code single-threaded but then simulations are going to be quite long so I am really keen on a multi-thread approach.

Thanks!

--
Guillaume Chapron, PhD

Grims? Wildlife Research Station
Swedish University of Agricultural Sciences
SE - 73091 Riddarhyttan, Sweden
Dirk Eddelbuettel
#
On 9 May 2015 at 20:38, Guillaume Chapron wrote:
| Hello,
| 
| I am trying to develop a R package that calls a multi-threaded C function and I am struggling on how to do this properly. The C program I originally wrote (prior to thinking about building it into a R package) uses Grand Central Dispatch library (for multi-threading on a Mac) and GSL (for RNG). Importantly, each thread has its own seed (the program runs Monte Carlo simulations). 
| 
| Now that I am looking at modifying this to be included in a R package, I am not sure how to proceed with having a parallel and portable C code using the RNG seed from R. In the book R Programming for Bioinformatics by Gentleman (2008), it says p. 201:
| 
| "The decision to have these functions (GetRNGstate and SetRNGstate) manipulate a global variable, .Random.seed, is slightly unfortunate as it makes it somewhat more difficult to manage several different random number streams simultaneously."
| 
| Has anyone been able to do this or is there a package implementing something similar? I could always have the C code single-threaded but then simulations are going to be quite long so I am really keen on a multi-thread approach.

Yes and yes. People have done / are doing this a lot.  

You correctly identified the issue with the RNGs -- so you need stream-aware
and/or multithreading-ready RNGs.  Which exist on CRAN in a few packages --
have a look at the Task View for HPC where I think I mention some.

Dirk