[Rcpp-devel] R & RcppArmadillo decomposition disagreement
edd at debian.org
Fri May 31 13:19:26 CEST 2013
On 31 May 2013 at 02:05, Colin Rundel wrote:
| > reference blas (packages libblas3 and liblapack)
| > atlas (package libatlas3-base or a tuned variant)
| > open-blas (package libopenblas-base)
| I had been playing around with update-alternatives to swap between the three, but just in case this was causing issues I've used apt to purge all three and do a clean install for each one.
| However, the end result is the same for all three circumstances - both of the arma chol calls fail but the R chol works and none of the eigen results match. Additionally, I am regularly seeing circumstances where one of the arma eigen calls runs, I can see activity across all 4 cores when using openblas or atlas, and then a hang where one core stays pegged at 100% but nothing appears to happen until I kill the process.
Ok, then it seems like we should prepare an email / test case for Conrad.
| I am still at a loss as to why R is working fine while the arma calls fail.
For certain operations R does _not_ go to lapack but uses its own. I can
never remember if chol() was one of them -- but this suggests it. As I
mentioned in my earlier email you probably really have to follow the chol()
call all the way down (in the sources).
Dirk Eddelbuettel | edd at debian.org | http://dirk.eddelbuettel.com
More information about the Rcpp-devel