I’m working on a Mac so ymmv, but I’ve been running benchmarks of vanilla R 
from cran vs. recompiles with different versions of gcc and R, and I see a 
speedup of 20-30% vs the cran binary after recompiling with gnu 4.9.  Its 
quite distinct.  

Each jump in gcc revision (4.7-4.8, 4.8-4.9) seems to improve benchmark 
performance by around 10% over the previous generation.  

So there may be advantages to recompiling yourself.

On the other hand, there may be potential issues compiling some packages.  
Some of the benchmarking I’ve been doing is 3.1.1 vs. pqR (distinctly faster 
than 3.1.1, even without “helper threads”), which is based on R 2.15.0, 
which in turn forces you to use older versions of some packages, some of which 
have language-compatibility issues with recent compilers.  The version of Rcpp 
that accepts R 2.15.0, for example, won’t compile against gcc 4.9.  You may 
find that the current versions of other packages have similar issues with 
recent gnu compilers, I can’t say.  

I’m surprised, frankly, that there isn’t more discussion of this, 
considering the size of the data and complexity of the problems people are 
feeding into R, and the possibility of essentially “free” performance 
boosts.

-- 
Amos Elberg
Sent with Airmail

From: Rguy <r...@123mail.org>
Reply: Rguy <r...@123mail.org>>
Date: August 6, 2014 at 2:47:47 AM
To: r-help@r-project.org <r-help@r-project.org>>
Subject:  [R] Old g++ in Rtools  

I recently downloaded Rtools. I see the g++ version is  
gcc version 4.6.3 20111208 (prerelease) (GCC)  

I also recently downloaded MinGW. Its version of g++ is  
gcc version 4.8.1 (GCC)  

I believe that later versions of g++ provide better support for C++11.  
Why does Rtools provide a version considerably older than the latest?  
Any plans to update the version?  
Is it bad practice to compile with a later version when interfacing with R?  

[[alternative HTML version deleted]]  

______________________________________________  
R-help@r-project.org mailing list  
https://stat.ethz.ch/mailman/listinfo/r-help  
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html  
and provide commented, minimal, self-contained, reproducible code.  

        [[alternative HTML version deleted]]

______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

Reply via email to