Re: Itanium2@900MHz slower than alpha@666MHz ?

[David Mosberger <davidm@napali.hpl.hp.com>]

This probably won't answer your question, but you may be better off
asking such questions on linux-ia64@vger.kernel.org (since it's
unlikely to be related to Debian itself).  In general, you'll want to
watch out for non-finite arithmetic (including denormal arithmetic),
but since you're comparing it to Alpha, that's probably not likely the
source of the problem.  Just to be on the safe side, you may want to
double-check the output of dmesg.  If you're getting lots of fpswa
traps, then denormal arithmetic may be the issue.  Compiling with -ftz
might help then (it will cause denormals to get flushed to zero and
hence you'd have the fpswa trap, at the cost of some loss of
precision).

	--david

>>>>> On Tue, 21 Oct 2003 01:01:46 +0200, Ionut Georgescu <george@physik.tu-cottbus.de> said:

  Ionut> Hi, I am puzzled about the speed of a zx2000 workstation with
  Ionut> a 900Mhz CPU.  According to the SPECfp2000 benchmarks, this
  Ionut> workstation should be about twice as fast as a DS10 alpha
  Ionut> workstation.

  Ionut> The test program is only doing some Fourier transforms
  Ionut> (fftw2) and some matrix multiplications. Both fftw2 and the
  Ionut> program have been compiled with ecc -O2 -ipo -limf. ecc is
  Ionut> Version 7.1, Build 20030307.

  Ionut> With or without optimization, the program runs about 25%
  Ionut> slower on the zx2000. Are there any compiler related issues ?
  Ionut> The specfp numbers were all obtained with ecc 6.0, which I
  Ionut> don't have.

  Ionut> Both the alpha and the Itanium2 run Debian stable and kernel
  Ionut> 2.4.20.