2
Notes on using Atlas libs with GNU Octave and GNU R
6
As of the Debian releases 2.1.34-6 (for GNU Octave) and 1.3.0-3 (for GNU R),
7
both Octave and R can be used with Atlas, the Automatically Tuned Linear
8
Algebra Software, in order to obtain much faster linear algebra operations.
10
To make use of Atlas, Debian users need to install the Atlas libraries for
11
their given cpu architecture. Concretely, one of
13
atlas2-base - Automatically Tuned Linear Algebra Software
14
atlas2-p3 - Automatically Tuned Linear Algebra Software
15
atlas2-p4 - Automatically Tuned Linear Algebra Software
16
atlas2-athlon - Automatically Tuned Linear Algebra Software
18
must be installed. Here, 'base' provides generic libraries which run on all
19
platforms whereas 'p3', 'p4' and 'athlon' stand for the Pentium III and IV as
20
well as the AMD Athlon, respectively. The actual libraries are installed in
21
/usr/lib/atlas (in the case of 'base') and in /usr/lib/$arch/atlas for the
22
cpu-specific versions. Here $arch stands for the cpu code used by the kernel
23
and shown in /proc/cpuinfo.
25
The Atlas libraries can be loaded dynamically instead of the (non-optimised)
26
blas libraries against which both Octave and R are compiled.
28
Section III below briefly describes how Atlas libraries can be compiled for
29
your specific machine to further optimise performance.
31
II. Using the Atlas libraries
33
II.A New default behaviour with automatic loading of the Atlas libraries
35
In order to have the libraries loaded at run-time, the location needs to be
36
communicated to the dynamic linker/loader. As of the Debian release
37
libc6_2.2.4-5 of the glibc library, a patch to ldconfig automates the use of
38
the Atlas library. If an Atlas package is installed, and correctly registered
39
in /etc/ld.so.conf as done by its postinst script, ldconfig will
40
automatically load the Atlas' version of the Blas instead of the (slower)
43
The following text is hence only relevant for systems which have not yet
44
upgraded to libc6_2.2.4-5 or later.
46
II.B Old behaviour requiring LD_LIBRARY_PATH for Octave
48
For Octave, use the variable LD_LIBRARY_PATH. On a computer with the
51
$ LD_LIBRARY_PATH=/usr/lib/atlas octave2.1 -q
52
octave2.1:1> X=randn(1000,1000);t=cputime();Y=X'*X;cputime-t
55
$ edd@homebud:~> octave2.1 -q
56
octave2.1:1> X=randn(1000,1000);t=cputime();Y=X'*X;cputime-t
59
For R version 1.3.0-4, the R_LD_LIBRARY_PATH variable has to be used, and its
60
value needs to be copied out of /usr/bin/R (or edited therein). For R version
61
1.3.1 or later this is done automatically in the R startup shell script. For
62
an Athlon machine, and with the explicit definition which is no longer needed
63
as of R 1.3.1, the example becomes
65
$ R_LD_LIBRARY_PATH=/usr/lib/R/bin:/usr/local/lib:/usr/X11R6/lib:/usr/lib/3dnow/atlas:/usr/lib:/usr/X11R6/lib:/usr/lib/gcc-lib/i386-linux/2.95.4:. R --vanilla -q
66
> mm <- matrix(rnorm(10^6), ncol = 10^3)
67
> system.time(crossprod(mm))
68
[1] 2.38 0.04 2.84 0.00 0.00
71
> mm <- matrix(rnorm(10^6), ncol = 10^3)
72
> system.time(crossprod(mm))
73
[1] 28.28 0.08 33.54 0.00 0.00
76
Running such a small example is highly recommded to ascertain that the
77
libraries are indeed found, and to "prove" that the speed gain is real (and
78
significant) for problems of at least a medium size as the 1000x1000 examples
81
Note that the example use "/usr/lib/atlas" for the atlas2-base package;
82
Athlon users should employ "/usr/lib/3dnow/atlas", Pentium III users should
83
employ "/usr/lib/xmm/atlas" and Pentium IV users should employ
86
Lastly, it should be pointed out that it is probably worthwhile to locally
87
compile, and thereby optimise, the Atlas libraries if at least a moderately
88
intensive load is expected. This is described in the next section.
90
III. Locally compiling the Atlas libraries
92
The Debian Atlas packages have been setup to allow for local recompilation of
93
the Atlas libraries. This way the behaviour will be tuned exactly to the
94
specific CPU rather than the broader class of CPUs. It has been reported that
95
this can increase performance by a further 12% on the examples above.
96
Detailed instructions are in /usr/share/doc/atlas2-base/README.debian.gz but
97
the process is essentially the following [ courtesy of Doug Bates ]
99
apt-get source atlas2-base
101
fakeroot debian/rules/custom
102
# wait for a *very* long time
103
dpkg -i ../atlas2-base*.deb
108
The Atlas packages have a very detailed README.Debian file which should be
109
consulted; it also details local recompilation. Sources and documentation for
110
Atlas are at http://www.netlib.org/atlas.
115
Camm Maguire developed the scheme of overloading Atlas over the default blas
116
libraries and deserves all the credit. Many thanks to John Eaton for helping
117
debug some errors in the initial setup, and to Doug Bates for work on the R
118
package. Special thanks to Ben Collins for providing a patched ldconfig as
119
part of the libc6 package.
123
-- Dirk Eddelbuettel <edd@debian.org> Tue, 21 Aug 2001 21:37:15 -0500
126
-- Dirk Eddelbuettel <edd@debian.org> Sun, 11 Nov 2001 11:03:19 -0600