R SIG Debian - Jul 2011 - revolution-mkl package not functioning correctly with R 2.13

It appears that the revoluton-mkl package, available via the multiverse 
Ubuntu repository with the purpose of adding multi-threaded numeric 
libraries to R, is not function correctly with R 2.13 (at least the 
version found on CRAN).  Testing with "R-benchmark-25.R" (found at 
http://r.research.att.com/benchmarks/), I get the following error, after 
installing the revolution-mkl package (4 core, AMD processor, 11.04 
64-bit Ubuntu):

 > source("R-benchmark-25.R")
Loading required package: Matrix
Loading required package: lattice

Attaching package: 'Matrix'

The following object(s) are masked from 'package:base':

     det

Loading required package: SuppDists


    R Benchmark 2.5
    ==============Number of times each test is run__________________________:  3

    I. Matrix calculation
    ---------------------
Creation, transp., deformation of a 2500x2500 matrix (sec):  1.237
2400x2400 normal distributed random matrix ^1000____ (sec): 
0.861333333333333
Sorting of 7,000,000 random values__________________ (sec): 
1.09133333333333
2800x2800 cross-product matrix (b = a' * a)_________ (sec): 
1.55633333333333
Error in solve(crossprod(a), crossprod(a, b)) :
   the leading minor of order 1271 is not positive definite
Timing stopped at: 2.55 0.04 0.729

I was also unable to exit from R cleanly.  I tried R-benchmark-25.R 
without the SuppDists package, and the same error occurred.

In a virtual machine, I rolled back R to version 2.12.1, the version 
found in the main Ubuntu repository, and R-benchmark-25.R worked without 
error.  So it appears that rolling back R to 2.12.1 may be a solution (R 
2.12 is no longer available as a deb on CRAN).

I have not tested the GOTOBlas2 libraries.  Ian Fellows mentions issues 
on the Mac side of things 
(http://groups.google.com/group/deducer/browse_thread/thread/8c9a8af0465f9b9b), 
although I am not sure if this applicable as Ubuntu 11.04 is based on 
gcc 4.5.

If anyone has the revolution-mkl package working with R 2.13 or has 
tested the GOTOBlas2 libraries, please let me know.

Michael

-- 
Dr. Michael A. Rutter
School of Science
Penn State Erie, The Behrend College
Station Road
Erie, PA 16563
http://math.bd.psu.edu/faculty/rutter

On 21 July 2011 at 18:59, Michael Rutter wrote:
| It appears that the revoluton-mkl package, available via the multiverse 
| Ubuntu repository with the purpose of adding multi-threaded numeric 
| libraries to R, is not function correctly with R 2.13 (at least the 
| version found on CRAN).  Testing with "R-benchmark-25.R" (found at 

See r-devel today where Simon (regarding another matter) confirms that R
2.13.0 started to default to using Open MP --- and I understand MKL does too
(and you use eg the same env var to select number of threads etc) so I can
see this leading to a conflict.

Hope this helps,  Dirk

-- 
Gauss once played himself in a zero-sum game and won $50.
                      -- #11 at http://www.gaussfacts.com

On 07/23/2011 08:21 PM, Dirk Eddelbuettel wrote:
>
> On 21 July 2011 at 18:59, Michael Rutter wrote:
> | It appears that the revoluton-mkl package, available via the multiverse
> | Ubuntu repository with the purpose of adding multi-threaded numeric
> | libraries to R, is not function correctly with R 2.13 (at least the
> | version found on CRAN).  Testing with "R-benchmark-25.R" (found
at
>
> See r-devel today where Simon (regarding another matter) confirms that R
> 2.13.0 started to default to using Open MP --- and I understand MKL does
too
> (and you use eg the same env var to select number of threads etc) so I can
> see this leading to a conflict.
>
> Hope this helps,  Dirk
>
The suggestion by Simon to compile with the "--disable-openmp" solves 
the problem, and the revolution-mkl package works properly with R 2.13.1.

As mentioned in the R 2.13.0 release notes, Open MP integration is 
experimental and nothing uses it, but it will in the future.  Therefore, 
should we disable Open MP (for now) on the Ubuntu builds in order to 
have compatibility with currently available packages or is this 
something Revolution Computing needs to look into (assuming they are 
still supporting the revolution-mkl package)?

Michael
-- 
Dr. Michael A. Rutter
School of Science
Penn State Erie, The Behrend College
Station Road
Erie, PA 16563
http://math.bd.psu.edu/faculty/rutter