Re: GCC 4.2.0 Status Report (2007-02-19)

[Robert Dewar <dewar at adacore dot com>]

Gabriel Dos Reis wrote:
> Andi Kleen <andi@firstfloor.org> writes:
>
> | "Kaveh R. GHAZI" <ghazi@caip.rutgers.edu> writes:
> | > 
> | > And we don't want to arm our detractors with bad SPEC numbers.  I can just
> | > imagine the FUD spreading... we've got to fix it or backout.

So what if gcc is a bit behind some other compiler on SPEC? This
is only one of many considerations. The number of people for whom
this is a deciding factor is very small. There will always be
individual instances of compilers for specific architectures
that do better in code generation than gcc in some cases. gcc
pays an inevitable price for generality and ease of
retargetting. The amazing achievment of gcc is that this
price is remarkably small in most cases.

Of course we do want to "fix" this if we can, and we want the
best possible performance. But for most people, reliability
and correct code generation is by *far* more important than
a few percent in efficiency.

So if a bug in code generation is found, the first priority
is to fix it, and performance degradation is not an argument
against doing the fix (it is an argument for subsequently
looking for a better fix that retrieves the performance,
but performance can never stand in the way of correctness).