Sorry, my employer's contracted personnel (from IBM) switched me into RTF
mode. Please excuse the dual posting modes.
----- Original Message -----
From: Tim Prince
To: [log in to unmask]
Sent: Saturday, January 13, 2001 10:28 PM
Subject: scatteri benchmark from polyhedron site
I'm wondering whether there is merit in compilers attempting to optimize the
style of code contained in the scatteri.f90 file from www.polyhedron.com .
In case anyone is interested, I've attached a version where the polynomials
are written Horner fashion rather than using separate terms with all
exponents typed double, and atan() is replaced with atan2().
I've submitted performance bug reports against one of the commercial
compilers which falls down on others of these benchmarks, by allocating and
deallocating a temporary array twice per use in an inner loop, and failing
to recognize common subexpressions involving exponentiation . In that
regard, how many compilers optimize X/Y**1.5, by invoking Y**(-1.5), sqrt()
or other appropriate substitution?
Tim Prince
[log in to unmask]
|