Re: [RFC] Merge criteria for tree-ssa

[Robert Dewar <dewar at gnat dot com>]

Andrew Pinski wrote:

> Actually it is the more functional HPUX IA64 support and not really 
> GNU/Linux IA64 support
> so it is a not really a problem at all.  Remember that GCC is GNU's 
> compiler first then
> a free source one.

Sure, but users are still affected. The question was why not tell F77
users "just use an old compiler". You are saying that we can also tell
them "buzz off, this is our compiler, too bad if it does not support
Fortran on your proprietary system so well". Yes, of course we can, 
there is no contractual obligation here, but that does not mean it is 
unreasonable to suggest how to improve Fortran support, and the original 
message in this thread was actually very constructive in terms of 
suggesting testing regimes for improving that support.

Also, although of course GNU systems are a primary goal, the support
of a wide range of proprietary systems is a very important secondary
goal, and is important for the general success of the gnu project.
We don't go out of our way to support such systems, but we also try
not to take decisions that negatively affects such support if we can.

If tree-ssa has the effect of negative impact on such systems, that's
not a fatal flaw, but it is certainly a negative, and means that the
promised positives have to stack up a bit higher.

> Also IA64 is doomed to fail as Intel and HP seems like they are moving 
> away from it into x86_64 support.

What you are referring to here is two rather separate developments.
HP has announced that they will sell Opeteron servers, and Intel
has indicated that they will provide some 64-bit follow on to the
x86, but it would be premature to say that either move means that
they are moving *away* from the ia64, which still has a lot of support.
Instead they are simply offering a wider range of products in response
to customer demand.

The ia64 still has plenty of steam left in terms of existing interest.
Note for example that ia64 is the natural follow on for SGI customers,
and HP customers in general are still quite focused on the ia64.
Whether that materializes into major sales remains to be seen. But
SGI for example is certainly betting the store on ia64, and at least
the market seems to be reasonably happy with this (SGI stock has risen
four fold in the last few months).

It's interesting that we (AdaCore) have had FAR more interest in an ia64
GNAT port than an Opteron port so far. We have a substantial
contract with HP to implement GNAT on ia64/OpenVMS, and we may
well be announcing one or more ia64/GNU/Linux contracts in the
near future. So far there is much less interest in the Opteron
for Ada, but of course that may change.

But to write off the ia64 in the gcc world based on
your guess that it is "doomed to fail" would be premature. There
are very substantial resources being devoted to ensuring its
success (inside and outside the gcc world).

> Thanks,
> Andrew Pinski