Re: Win32 version of XFree86

[Mumit Khan <khan@xraylith.wisc.EDU>]

Chris Faylor <cgf@cygnus.com> writes:
> 
> I think that recent benchmarks have shown that gcc is at least
> approaching MSVC speeds in most regards and surpassing it in a
> few areas.

In one area however, gcc trails behind the aforementioned compilers
quite badly -- floating point. My code, which is numerically intensive,
shows about 15-35% degradation using gcc. For integer code, it's still
5-20% slower than MSVC 6.x.

One big problem is the way x86 machine description is written in GCC,
and that's the primary bottleneck. For good news, see below.

> I'm not a gcc engineer, however, and I can't give you complete details.
> I can say that the PII optimizations in our Code Fusion product are
> rumored to make linux run as much as 5% faster.

Yes. Cygnus has donated the new ia32 backend and my numerical code shows
as good a performance as any of the commercial offerings. However, it'll
be a while before this is integrated into FSF gcc (gcc-3.0 perhaps).

Intel is not a good comparison; it's built for benchmarking, and chock
full of bugs. MW5 is excellent for both C and C++. MSVC C is excellent, 
but C++ is quite behind when it comes to standard conformance (I can't
use it on even the simplest C++ code making heavy use of templates for
example).

This is a topic for gcc lists, not xfree, so I'll rest here.

Regards,
Mumit