<Snippage>
On Jun 30, 2009, at 2:12 PM, Marc Schwartz wrote:
Well....barring a correction from someone with more low level
insight, I would tend to suspect that the difference in our findings
may be attributed to some interaction of hardware, OS, compiler and
BLAS, perhaps weighted more to the latter two. When I build from
source, I do use the Apple vecLib BLAS (the default), as opposed to
the R BLAS.
To distill this down to the key issue, all else being equal
(within .Machine$double.eps), I suspect that this leads us back to
rounding as being the source of Fabrizio's observations.
FWIW, I re-built R here so that it would use the R provided BLAS and
LAPACK rather than the Apple provided versions.
Bottom line, no difference in my results here. So there is something
more subtle going on here relative to either rounding and/or perhaps
precision, which may now point less to the BLAS/LAPACK combination and
perhaps more to other compiler optimizations or OS issues.
Perhaps a member of R Core could offer some insights?
Thanks,
Marc
______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
