GCC quality (was Re: AMD-64 optimizations, some (new) code)

[David T. Ashley]

> I have had more problems with 3.4.x than with 3.3.x.  But with
> the track record of gcc, it is not safe to use any gcc release.
> You need to have a good test suite for anything compiled with
> gcc.  The gcc release that have caused me most problem is 4.0.0.
> It seems 4.0.1 is better, but I haven't been bothered to try it
> much.
>
> The gmp testsuite has great coverage.  That is part of the reason
> why gmp trips so often on gcc bugs.  Another part is that few
> other programs expect fully correct integer arithmetic behaviour.

Y'all are losing me on these statements.  What is a "typical" gcc bug that
affects gmp?

> Another part is that few
> other programs expect fully correct integer arithmetic behaviour.

As I recall gmp, it phrases larger integer operations in terms of "limb"
operations.  "Limb" arithmetic is so basic -- I think most of it is
supported directly by the machine instruction set.  It is hard to imagine
bugs.  I mean, in most cases the compiler should just be processing flags
set by machine instructions ... and that comes down to digital logic ...

a)What is a typical "bug" you've found?

b)Was this with gcc for the x86 platform or for other platforms?

Thanks, Dave.