On Jun 18, 2013, at 21:49 , peter dalgaard wrote: > > On Jun 18, 2013, at 21:23 , Berend Hasselman wrote: > >> >> So it seems that the blocked algorithm is the cause of the error and that >> using the (possibly slow) unblocked algorithm gives the correct result. > > Thanks Berend, > > The finger seems to be pointing to the internal libRlapack/libRblas. The > apparent pattern is that things work when R is linked against external > libraries and not when the internal ones are used. So it could be time to > start looking for differences between R's lapack module and the original > LAPACK code.
Now done and no (relevant) differences found. Hmmm.... There could be compiler issues. It could also be that the internal LAPACK uses a newer version than the external libs and a bug was introduced in between them. One option could be to bypass R by coding Robin's example in pure Fortran and see whether issues persist when linked against libRlapack, vecLib, and the relevant subset of current LAPACK sources. (The bogus 1.0000 eigenvalue in the 33x33 variant of Robin's example should make it rather easy to spot whether things work or not.) -- Peter Dalgaard, Professor, Center for Statistics, Copenhagen Business School Solbjerg Plads 3, 2000 Frederiksberg, Denmark Phone: (+45)38153501 Email: pd....@cbs.dk Priv: pda...@gmail.com ______________________________________________ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
