R devel - Oct 2001 - R Performance on P-4 RE: P-4

Is this still an issue to this community?  If so, and results, I may pick
this up.  Intel does have C++ and Fortran compilers.

You might see a compiler among the list under the links
http://cedar.intel.com/cgi-bin/ids.dll/topic.jsp?catCode=BON.
Evaluation licenses are available at 
http://developer.intel.com/software/products/compilers/

If they are promising, perhaps binaries could at least be built using Intel
compilers.  

But first, are there any R tasks/workloads that we might comparably
benchmark?

Colin



-----Original Message-----
From: Prof Brian D Ripley [mailto:ripley@stats.ox.ac.uk]
Sent: Wednesday, April 25, 2001 10:58 PM
To: Andrew Schuh
Cc: r-devel@stat.math.ethz.ch
Subject: Re: [Rd] P-4


On Wed, 25 Apr 2001, Andrew Schuh wrote:
> I just got a new machine with a Intel 1.3 P4 chip in it.  I am aware that
> many (most) software companies have not optimized their code specifically
> for the chip.  Furthermore, I have found that this machine is even slower,
> on many apps, than a 8xx P3 we have here.  I enjoy using R and was
> wondering if there are any plans to create a version that accommodates
this
> chip.
Under what OS?

AFAIK the only thing one can do is to find a compiler that optimizes for
the P4, but I don't know of one, nor if it actually helps.  As far as I am
aware under both Linux and Windows the binary versions of R are not
even optimized for an i686 (Pentium Pro/II/III), but I find that under
Windows (and VC++) using the optimizations for i686 makes only negligible
differences.

-- 
Brian D. Ripley,                  ripley@stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.
-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._.
_._
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._

"Cunningham, Colin A" <colin.a.cunningham@intel.com> writes:
> Is this still an issue to this community?  If so, and results, I may pick
> this up.  Intel does have C++ and Fortran compilers.
> 
> You might see a compiler among the list under the links
> http://cedar.intel.com/cgi-bin/ids.dll/topic.jsp?catCode=BON.
> Evaluation licenses are available at 
> http://developer.intel.com/software/products/compilers/
> 
> If they are promising, perhaps binaries could at least be built using Intel
> compilers.  
> 
> But first, are there any R tasks/workloads that we might comparably
> benchmark?
An easily evaluated benchmark of R performance on Unix/Linux systems
is obtained by running "make check" after building from the sources.
I know there is a similar check done for compilations under Windows
but I don't know how it is run.

"make check" runs every example in every help page in every package
that is shipped with R.  The base package contains the most important
and most commonly used functions.  If everything is successful the
last few lines of the file ./tests/Examples/base-Ex.Rout should read
like

 > cat("Time elapsed: ", proc.time() - get("ptime", env =
.CheckExEnv),"\n")
 Time elapsed: 84.22 1.9 105.61 0.82 0.61
 > dev.off(); quit('no')
 null device
           1
The times, in seconds, are the user time, the system time, and
the elapsed (wall clock) time for the parent process, and elapsed and
system times for child processes.  The user time for the parent
process is probably the most important.

The run shown above is for the r-devel sources on a Linux system with
a 1.2 GHz Athlon and 1 GB of PC-133 memory.  The compilers were gcc
and g77 from the 2.95.4 compiler suite.  This compiled version does
use atlas BLAS but that should not make much of a difference on this
benchmark.

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To:
r-devel-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._