[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: using non-standard optflags (-O3, in particular)



On Fri, 2007-02-09 at 11:41 -0500, Simon Perreault wrote:
> Ralf Corsepius wrote:
> > On Fri, 2007-02-09 at 10:59 -0500, Simon Perreault wrote:
> >> Ralf Corsepius wrote:
> >>> So, the only results of recommendations to trust when it comes to
> >>> packaging binaries for a distro is those who are deeply familiar with
> >>> the guts of the OS, in case of Fedora, RH's GCC, glibc and kernel
> >>> developers.
> >> This is BS.
> > Beg your pardon? 
> 
> You don't need that level of familiarity to make good recommendations. 
> You make it sound like Red Hat has some kind of old wise man and only he 
> knows enough, and he only speaks once a year on summer solstice.
I am not affiliated with RH.

> >>  Sure, people have thought about the defaults, but it doesn't 
> >> mean that the packager doesn't know what he's doing either.
> > Rest assured: In 99% of all cases they don't know.
> 
> Ok now, you say 99% of packagers can't package? BS.
Well, this sentence was ill formulated, I meant 99% of all maintainers
who want their packages to use -O3 don't know what they are doing.

> >>  Some 
> >> software, particularly numerical computation stuff, is built for being 
> >> optimized properly. There are some extreme cases where using -O2 instead 
> >> of -O3 simply makes a piece of software useless (take Blitz++ for example).
> > In other words: Crappy non-portable SW,
> 
> The particular example I gave is a marvel of software engineering, 
> useful to many scientists working with big arrays.
I am well aware what Blitz++ is. This doesn't change anything about the
fact that a package which needs "-O3 to be useful", is plain broken -
period.

A package that is reported to "gain a 10% performance boost" on a
particular HW environment is something completely different.

> > It is - It renders debugging impossible on many systems,
> 
> Not much worse than -O2 already is.

There are reasons why certain optimizations are in -O3 and why they are
not in -O2, e.g. because some of them do not provide sufficient
performance gains, some of them because they are too architecture
specific, some of them because they are unsafe, some of them because
they are considered unstable/experimental.

> > strict-aliasing
> > silently kills SW on some targets
> 
> You know this is enabled at -O2 right?
I know, but there is more to it.

> > it might trigger exotic
> > target-specific bugs etc. etc. 
> 
> BS, and you know it.
Great, then you must be using a different GCC than I am. I have seen -O3
triggering target specific bugs on many occasions.

> > As part of the distro you can't to compromise between different
> > trade-offs.
> 
> WTF? You think Fedora isn't about compromise?

On a distro you must compromise: Exactly why using -O3 is a bad thing
for a distro.

Ralf



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]