|
Boost : |
From: Richard Newman (richard_at_[hidden])
Date: 2004-11-15 20:23:39
David Abrahams wrote:
> Richard Newman <richard_at_[hidden]> writes:
>
>> Thank you, Dave, for replying. I will check out the link.
>>
>> Do you have specific advice/experience for Linux's g++ that might
>> improve or explain what we're seeing here?
>
> I'm sure others can do better than this, but:
>
> 1. Optimization quality has varied widely over recent compiler
> versions
>
> 2. The compiler has *many* optimization options so it's easy to
> overlook some (e.g. inlining limits)
>
> 3. g++ has never been known as a compiler with extremely low
> abstraction penalty.
>
> HTH,
>
I guess this is why we need to test early and often.
Point one above though is a little scary. So we make a change for
performance based on our observations from a particular compiler version
only to learn that we lose the improvement (or worse, it degrades beyond
the original code) when we upgrade the compiler. We'll have to test our
compiler upgrades before we rely on them; one more reason to not just
upgrade willy-nilly.
Thank you all very much for your advice.
Kind regards,
Richard Newman
Crowley Davis Research
richard_at_[hidden] (take out the nospam. to email me directly)
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk