[R] Segfaults of eigen with blas-atlas at x86_64-pc-linux-gnu systems

Juergen Rose rose at rz.uni-potsdam.de
Mon Feb 21 14:30:43 CET 2011


Am Montag, den 21.02.2011, 11:25 +0100 schrieb Karl Ove Hufthammer:
> Juergen Rose wrote:
> 
> >> eigen(D)
> > 
> > *** caught segfault ***
> > address (nil), cause 'unknown'
> > 
> > Traceback:
> > 1: .Call("La_rs", x, only.values, PACKAGE = "base")
> > 2: eigen(D)
> > 
> > All systems are Gentoo systems, i.e. R-2.12.1 and blas-atlas-3.9.23-r4
> > is installed by compiling the sources. Recompiling R and blas-atlas did
> > not solve the issue. This issue seems the reason that example(svm)
> > creates segfaults, too.
> 
> FWIW, this does *not* crash R on my system, running 2.12.2 RC with 
> GotoBLAS2.

Thanks Karl and Brian,

under gentoo I have the possibility to switch (by eselect) between some
different blas versions, if there are installed. Usually I have
blas-atlas and blas-reference installed. For blas-atlas it is still
possible to switch between pure blas-atlas and blas-atlas-threads.
I tried at some systems, which are now available for me, to reproduce
these segfaults with eigen in R. I got the following.


                    blas version        eigen(dist(iris[,-5]))

inspiron (32-bit)   blas-reference      OK
i686-pc-linux-gnu

cheetah (32-bit)    blas-atlas          OK
i686-pc-linux-gnu

tiger (32-bit)      blas-atlas          OK
i686-pc-linux-gnu

moose (64-bit)      blas-atlas-threads  Segfault
x86_64-pc-linux-gnu
                    blas-atlas          Segfault

                    blas-reference      OK


orca (64-bit)       blas-atlas-threads  Segfault
x86_64-pc-linux-gnu
                    blas-atlas          Segfault

                    blas-reference      OK


R was everytime recompiled after switching to an other blas version.
R is on every computer R-2.12.1,  blas-atlas is blas-atlas-3.9.23-r4,
blas-reference is blas-reference-20070226-r2.
It would be rather interesting for me, if anybody with similar
blas-atlas version is able to run eigen(dist(iris[,-5])) without
segfaults under R.

What are the differences between blas-reference, blas-atlas and
blas-goto. And what will happen with the other blas depended packages
as 

apbs, gromacs, grass, arpack, ccp4-libs and umfpack,

if I switch to blas-goto? Do you have some insight?

Regards



More information about the R-help mailing list