From help-octave-request at bevo dot che dot wisc dot edu Mon Apr 21 23:24:31 2003
Subject: Octave on Solaris -- success
From: "Dmitri A. Sergatskov" <dmitri at unm dot edu>
To: help-octave <help-octave at bevo dot che dot wisc dot edu>
Date: Mon, 21 Apr 2003 22:12:27 -0600

So, after some trials and tribulations I managed to
compile both octave-2.1.46 and octave-forge (cvs snapshot
on Feb 04 2003).

Here is the report for those who's interested.

I used gcc 3.2.2 with -O3 -mcpu=ultrasparc -mtune=ultrasparc.

I had compiled atlas3.4.1 prior to that and copied all resulting
libraries into /usr/local/lib.

I have pretty much entire GNU toolchain installed in /usr/local
and /usr/local/bin is the first directory in the $PATH.

For octave I used
./configure --enable-shared --enable-rpath --with-fastblas --with-readline=gnu
(last two may be redundant). After I fixed my g++ compiler installation, it
compiled fine.

octave-forge was somewhat more troublesome.

I did not have TeX installed on this computer. It was fine with
octave proper, but main/comm in octave-forge bombed. I had to
comment out all references to docs in its Makefile to get it
through.

./configure  put -R/usr/openwin/lib into Makeconf and that upset mkoctfile
(I got an error "mkoctfile: unknown flag -R/usr/openwin/lib" compiling main/plot).
I deleted it from Makeconf and it compiled fine after that.

Here resulting benchmarks (ftp://fly.cnuce.cnr.it/pub/benchmark.m modified
to normalize to SparcStation 20 50MHz x 2 CPU no atlas):

First on Ultra-10 UltraSparc-IIi 440 MHz (500MB RAM)

octave:1> benchmark
Octave benchmark version bm 1.10
Speed of octave 2.1.46 on sparc-sun-solaris2.8 relative to Sun SparcStation 2x20/50

Matrix inversion (LAPACK)         5.64 +/- 2.7% (105 runs)
Schur decomposition (LAPACK)     16.45 +/- 0.5% (63 runs)
Differential equation (LSODE)     5.45 +/- 2.6% (42 runs)
Fourier transforms (FFTPACK)      7.97 +/- 4.9% (168 runs)
for loop                          5.06 +/- 1.6% (77 runs)
-- Performance index (bm 1.10): 7.3

For comparison the same benchmark on RedHat Linux 9 Athlon XP 2100+ (1737 MHz)
with 133 MHZ SDRAM (1 GB) (same atlas, gcc 3.2.2 -O2 -march=athlon-xp):
octave:1> benchmark
Octave benchmark version bm 1.10
Speed of octave 2.1.46 on i686-pc-linux-gnu relative to Sun SparcStation 2x20/50

Matrix inversion (LAPACK)        19.58 +/- 0.5% (385 runs)
Schur decomposition (LAPACK)     76.89 +/- 0.3% (287 runs)
Differential equation (LSODE)    43.90 +/- 0.3% (322 runs)
Fourier transforms (FFTPACK)     11.86 +/- 0.4% (217 runs)
for loop                         41.36 +/- 0.8% (602 runs)
-- Performance index (bm 1.10): 32


It is somewhat surprising that loop performance is so poor on Sparc, but
besides that all numbers seem to make sense.

Sincerely,

Dmitri.



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------