Re: [boost] Re: N-Dimensional array library submission...

From: Ross Smith (r-smith_at_[hidden])
Date: 2001-12-21 17:55:39

Toon Knapen wrote:
> 
> Beman Dawes wrote:
> >
> > Just out of curiosity, was this a debug or an optimized build?
> >
> > Do you know if the compiler involved normally has a low abstraction
> > penalty?
> 
> Sorry, vital info was missing : I used gcc-3.0.2 with option -O4.
> (gcc-2.95.3 won't compile multi_array).

GCC 3.0.x releases have a flawed inlinming strategy that leads to a
significant template penalty. 3.0.2 is better than its immediate
predecessors, but it still leaves something to be desired in that area.
(A real fix is expected in 3.1; due out in April, or you could try the
current snapshots from that branch.)

Also, -O4 is an undocumented option; I have no idea what (if anything)
it does, or whether it would be expected to produce better code than the
documented optimisation options. The GCC experts generally agree that
-O2 should be considered the normal "best optimisation" setting, and
higher options (-O3 and the undocumented higher values) should only be
used when you've analysed the generated code and know that you need
specific nonstandard optimisations.

-- 
Ross Smith ...................................... Auckland, New Zealand
r-smith_at_[hidden] ......................... http://storm.net.nz/~ross/
  "We need a new cosmology. New gods. New sacraments. Another drink."
                                                       -- Patti Smith