[Rcpp-devel] RcppArmadillo BLAS/LAPACK force number of threads to 1
Dirk Eddelbuettel
edd at debian.org
Wed Aug 13 06:11:21 CEST 2014
Hi Scott,
On 13 August 2014 at 13:52, Scott Ritchie wrote:
| Thanks Dirk,
|
| Since the original question has been answered (no there isn't a way to control
| this through Armadillo or Rcpp, and the problem is unrelated), I'll go bug the
One of the issue is that you start frim R when using Rcpp -- so you always
get R's BLAS/LAPACK combination.
| r-sig-hpc mailing list if I can't find a working answer.
|
| Thanks for the additional pointers,
Sorry for the goose chase. This should be better documented -- maybe we'll
just write up a short Gallery post once done.
I should have a better pointer. There was also something by Simon U on this;
as I (very vaguely) recall the parallel may have needed a fix or something.
Cheers, Dirk
| Cheers,
|
| Scott
|
|
| On 12 August 2014 21:24, Dirk Eddelbuettel <edd at debian.org> wrote:
|
|
| On 12 August 2014 at 17:09, Scott Ritchie wrote:
| | Well that's frustrating.
| |
| | I've tried the OpenMPController package without success, and tried to
| | explicitly disable BLAS in the Armadillo headers and removing the linking
| code
| | in the Makevars files.
|
| That is IMHO the wrong approach. You want BLAS, you just want to control
| it.
| And as I tried to explain, Rcpp and Armadillo really have nothing to do
| with it.
|
| This has all been discussed before, and AFAICT, also solved. I think one
| (R)
| package did something wrong on intialization -- but AFAICR this was fixed a
| while back. Also check the r-sig-hpc archives on that.
|
| | If my reading of this thread: http://lists.r-forge.r-project.org/
| pipermail/
| | rcpp-devel/2012-June/003888.html is correct, I should be able to switch
| to
| | RcppEigen since it uses its own routines.
|
| You certainly can.
|
| I really do not think you have to.
|
| Dirk
|
| --
| http://dirk.eddelbuettel.com | @eddelbuettel | edd at debian.org
|
|
--
http://dirk.eddelbuettel.com | @eddelbuettel | edd at debian.org
More information about the Rcpp-devel
mailing list