| Re: [AD] gcc 3.x workaround. |
[ Thread Index |
Date Index
| More lists.liballeg.org/allegro-developers Archives
]
On Friday 30 November 2001 16:31, Eric Botcazou wrote:
> > i have removed all instances of -fomit-frame-pointer optimalisations from
> > the makefiles in preparation for the stable release. please don't touch
> > this.
>
> I disagree: the x86 platform is well-known for its scarcity of registers,
> so having one less register to use may have significant performance hits.
i don't think the performance hit will be of any particular significance.
it's more important that allegro works.
> Moreover, I don't see why we would do that for the Mingw32 port, the BeOS
> port and the QNX port since AFAIK they haven't yet seen the first
> incarnation of the gcc 3.0.x series.
mingw has a gcc 3.x release. but it's considered to be alpha. i don't know
about the other ports.
> > that goes for the djgpp port as well!
>
> Not very fair, given that the DJGPP workers (especially Andris Pavenis)
> filed a GCC bug report for us (PR 4763 contains the code generated for
> ustrzcpy() from src/unicode.c), then backported the fix and re-released the
> gcc 3.0.2 packages.
that's very nice, but how many users will be running the re-released 3.0.2 or
3.0.3 when allegro 4.0 comes out? certainly not everybody. (!) which is the
exact reason i removed this optimalisation. and not everybody read the docs
each time there's a new allegro release unless they have to.
> The fix has been applied to the 3.0 branch on November 10, so it is highly
> likely to be in the official 3.0.3 release.
>
> > at a later date this optimalisation can be put back and replaced by a FAQ
> > entry explaining that gcc 3.0 >< 3.0.3 has an optimiser which generates
> > invalid code for some optimalisations used by allegro.
>
> docs/build/djgpp.txt already states that gcc 3.0.x (0 <= x <= 1) can't
> properly compile Allegro.
not enough. this affects all ports which has (or will soon get) a gcc 3.x
compiler. it's better to be safe than sorry.
> And it would be IMHO more clever to add a little test in the configure
> script under Linux/Unix in order to disable -fomit-frame-pointer if needed.
i agree. but that's a unix-only solution.
--
Sincerely Henrik Stokseth.
-----------------------------------------------------------------------
E-mail: hstokset@xxxxxxxxxx Homepage: http://hstokset.n3.net