BLAS-LAPACK-ATLAS-R installation with multithreading? - r

Question

I am trying to setup ATLAS, BLAS, LAPACK, and R on an OpenSuSE 12.2 server with an 8 core AMD athlon FX-8320. I have turned off the CPU frequency scaling using cpufreq and a script cfu from paste.opensuse.org/92162247 made by James McDaniel.
So following the instructions, here http://math-atlas.sourceforge.net/atlas_install/node45.html I didn't need to install gcc (that I know yet) I have gcc (SUSE Linux) 4.7.1 20120723 installed, g++, and gfortran the same version as gcc.
The instructions are to make a bogus configuration of ATLAS to determine the compiler flags to use. ATLAS is configured in the instructions as '../configure -b 64 -D c -DPentiumCPS=2200 -Fa alg -fPIC'. I modify this to what I think will work for my setup ' ../configure -b 64 -t 8 -Fa alg -fpic -A x86_64 -D c -DpentiumCPS=3500 -Si latune 1 --prefix=/usr/local/lib64 --with-netlib-lapack-tarfile=/opt/maths/lapack-3.4.2.tgz' I get these errors.
OS configured as Linux (1)
Assembly configured as GAS_x8664 (2)
Vector ISA Extension configured as AVXMAC (3,504)
ERROR: enum fam=6, chip=8, mach=0
make[3]: *** [atlas_run] Error 44
make[2]: *** [IRunArchInfo_x86] Error 2
/bin/sh: line 1: 13849 Segmentation fault ./xarchinfo_linux -a > config0.out
make[3]: *** [atlas_run] Error 139
make[2]: *** [IRunArchInfo_linux] Error 2
Architecture configured as UNKNOWNx86 (35)
ERROR: enum fam=6, chip=8, mach=0
make[3]: *** [atlas_run] Error 44
make[2]: *** [IRunArchInfo_x86] Error 2
Clock rate configured as 3500Mhz
Parallel make command configured as '$(MAKE) -j 8'
ERROR: enum fam=6, chip=8, mach=0
make[3]: *** [atlas_run] Error 44
make[2]: *** [IRunArchInfo_x86] Error 2
Cannot detect CPU throttling.
Atlas makes a Make.inc and if I fgrep F77 and F77 flags I get 'gfortran-4.7' AND '-O -mavx -fpic -m64'. Maybe the -mavx flag handles the AVXMAC error? Do I need -mavx? I will proceed with my instructions...

Answer 1

(Not a real answer, but difficult to read as a comment):
A few more considerations on Dirk's recommendation to use OpenBLAS.
I also use OpenBLAS (on Ubuntu and CentOS systems).
One consideration that is important for me on the 8 and 12 core machines is that OpenBLAS allows dynamic setting of the number of threads to use, which ATLAS doesn't. In Order to set the number of threads dynamicall from within R, I use Simon Fuller's OpenBLASThreads.
OpenBLAS was pretty straightforward to build on both systems.
Binary .deb packages are available, but last time I checked them, they had the maximum possible number of threads set to 2. However, this is going to change. Till now, I also build the library from source on Ubuntu machines with > 2 cores.