[Rcpp-devel] [Lme4-authors] got a minute to answer dumb questions about throw/catch in Rcpp?
Dirk Eddelbuettel
edd at debian.org
Wed Jan 5 02:39:42 CET 2011
On 4 January 2011 at 14:20, Ben Bolker wrote:
| On 11-01-04 01:35 PM, Douglas Bates wrote:
[...]
| > Well, it's the version of the accelerated BLAS/Lapack code that is
| > more at fault than Lapack itself. As this occurs on Mac OS X and not
| > on other operating systems one test would be to build a OS X version
| > of R that used the reference BLAS and the reference Lapack code
| > shipped with R (i.e. use the --without-blas and --without-lapack
| > flags) and see if the problem persists.
|
| Just to clarify:
|
| this is *not* happening on MacOS, it's happening on Ubuntu 10.10
| (32-bit).
|
| Just to make things even murkier, I have tried building R-devel
| (2.13.0) both with BLAS and LAPACK and without and ... it seems to work
| both ways. I will dig out R 2.12.1 source code and try compiling with
| and without (my current (CRAN Ubuntu repository) version has
|
| bolker at ubuntu-10:~$ R CMD config LAPACK_LIBS
| - -llapack
| bolker at ubuntu-10:~$ R CMD config BLAS_LIBS
| - -lblas
|
| i.e. system libraries)
Actually, not really. This is sometimes a suble point of confusion.
On Debian and Ubuntu, -llapack -lblas still leaves you with the choice of
swapping the actual LAPACK and BLAS implementations. That happens to be a
massive feature and not a bug. It is setup in such a way that different
library packages can provide different _implementations_ of the standardised
LAPACK/BLAS _interfaces_. So you could be running any on of basic
(unoptimised), Atlas, Goto, or MKL.
[ And that happens to be the setup I describe and use in the BLAS/GPU
benchmarking package/paper/vignette in my CRAN package 'gcbd'; I also tried
to explain the underlying issue in the paper/vignette in some detail. ]
Doug actually touches upon that above. If you have test script that fails
for you on Ubuntu, then I could try to take a look and swap LAPACK/BLAS
libraries to see if it makes a difference.
Dirk
--
Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com
More information about the Rcpp-devel
mailing list