Robert Harley (Robert.Harley@inria.fr)
Tue, 6 Apr 1999 19:47:15 +0200 (MET DST)
Greg Lindahl (lindahl@cs.virginia.edu) wrote:
>Well, that's the first in-the-press announcement I've seen of this. It
>exceeds what the manager of the GEM compiler group says, so it's
>probably not exactly true, but hey...
It's the top story today at Compaq's "newsroom":
http://www.compaq.com/newsroom/pr/1999/pr060499a.html
Scroll down to this section:
Linux and Tru64 UNIX Compatibility; Compaq Support
For instance they claim a factor of two for Fortran:
------------------------------------------------------------------------------
Compaq has now incorporated Linux into its enterprise UNIX
strategy, emphasizing its intentions to increase Linux and Tru64
UNIX application and product compatibility. [...]
Further establishing AlphaServers as the high performance leader
in Linux systems, and in direct response to demand from High
Performance Technical Computing customers, Compaq is also
porting its Tru64 UNIX Alpha Fortan and C compilers along with
the associated runtime libraries to Linux (Alpha) later this year.
Additional compilers/development tools will be considered as
more requirements are identified. Early testing has shown that the
code generated by Compaq Fortran for Linux on AlphaServers will
perform floating point calculations at least twice as fast as code
generated by the g77 compiler for Linux on AlphaServers.
Within the next 90 days, Compaq Services will introduce customer
support services for the Linux operating system running on
Compaq's Alpha- and Intel-based servers. [...]
------------------------------------------------------------------------------
Looking good! =:-)
Although there may be a factor of two gained on floating-point Fortran
programs compared to g77, for the sort of integer work in C that I do
the gain is typically more like 0.9 to 1.3 times compared to gcc/egcs.
People shouldn't lock on to the "two times" meme and claim that
Compaq's compilers are universally twice as fast.
Sometimes twice as fast: sure.
Always twice as fast: no way.
There's a lot of variation. The bottom line is: benchmark your own code!
Bye,
Rob.
-- To unsubscribe: send e-mail to axp-list-request@redhat.com with 'unsubscribe' as the subject. Do not send it to axp-list@redhat.com
This archive was generated by hypermail 2.0b3 on Tue Apr 06 1999 - 11:00:06 PDT