optimizing egcs for alpha

optimizing egcs for alpha

[Jiann-Ming Su]

---------- Forwarded message ----------
Date: Wed, 24 Mar 1999 13:35:49 -0800
From: spink@pa.dec.com
Reply-To: axp-list@redhat.com
To: axp-list@redhat.com
Cc: spink@pa.dec.com
Subject: Re: Compaq sees Linux as selling Alpha chips 
Resent-Date: 24 Mar 1999 21:49:41 -0000
Resent-From: axp-list@redhat.com
Resent-cc: recipient list not shown: ;

> 
> >>> Jiann-Ming Su said:
> > 
> > If Compaq wants to make real noise with Linux, help optimize gcc/egcs for
> > the Alpha processor.  And put the updates right back into gcc/egcs unlike
> > what Intel is doing with Cygnus and the PII/PIII optimizations.
> 

Well, part of the problem is that both gcc and to some extent egcs are broken 
from a plug ability perspective.  The basic optimization model of gcc has 
serious problems.  What you are asking for is essentially for Compaq to 
rewrite gcc from scratch and then release the source code, and while I think 
it would be cool, it just isn't going to happen.  If gcc/egcs had a more 
modern design(like, shudder, MSVC) then it would be relatively easy to slot in 
the GEM back end.  


Aaron Spink
not speaking for Compaq



-- 
To unsubscribe: send e-mail to axp-list-request@redhat.com with
'unsubscribe' as the subject.  Do not send it to axp-list@redhat.com

Re: optimizing egcs for alpha

[Richard Henderson]

On Wed, Mar 24, 1999 at 04:16:31PM -0600, Jiann-Ming Su forwarded:
> > If gcc/egcs had a more modern design(like, shudder, MSVC) then
> > it would be relatively easy to slot in the GEM back end.  
> > 
> > Aaron Spink
> > not speaking for Compaq

Erum, but we don't _want_ to just "slot in" the GEM back end,
even if the design of gcc allowed it.


r~

Re: optimizing egcs for alpha

[Jiann-Ming Su]

On Thu, 25 Mar 1999, Richard Henderson wrote:

> On Wed, Mar 24, 1999 at 04:16:31PM -0600, Jiann-Ming Su forwarded:
> > > If gcc/egcs had a more modern design(like, shudder, MSVC) then
> > > it would be relatively easy to slot in the GEM back end.  
> > > 
> > > Aaron Spink
> > > not speaking for Compaq
> 
> Erum, but we don't _want_ to just "slot in" the GEM back end,
> even if the design of gcc allowed it.
> 

Ok, I'll bite.  Why not? 

Jiann-Ming Su                   "People seldom do what they believe in. They do
js1@microwave.ph.msstate.edu    what is convenient, then repent." --Bob Dylan

Re: optimizing egcs for alpha

[Richard Henderson]

On Fri, Mar 26, 1999 at 01:37:13PM -0600, Jiann-Ming Su wrote:
> > Erum, but we don't _want_ to just "slot in" the GEM back end,
> > even if the design of gcc allowed it.
> 
> Ok, I'll bite.  Why not? 

5 gets you 10 is isn't flexible enough to handle the breadth of
targets gcc supports.


r~

Re: optimizing egcs for alpha

[Jiann-Ming Su]

On Thu, 25 Mar 1999, Richard Henderson wrote:

> On Wed, Mar 24, 1999 at 04:16:31PM -0600, Jiann-Ming Su forwarded:
> > > If gcc/egcs had a more modern design(like, shudder, MSVC) then
> > > it would be relatively easy to slot in the GEM back end.  
> > > 
> > > Aaron Spink
> > > not speaking for Compaq
> 
> Erum, but we don't _want_ to just "slot in" the GEM back end,
> even if the design of gcc allowed it.
> 

Ok, I'll bite.  Why not? 

Jiann-Ming Su                   "People seldom do what they believe in. They do
js1@microwave.ph.msstate.edu    what is convenient, then repent." --Bob Dylan

Re: optimizing egcs for alpha

[Richard Henderson]

On Fri, Mar 26, 1999 at 01:37:13PM -0600, Jiann-Ming Su wrote:
> > Erum, but we don't _want_ to just "slot in" the GEM back end,
> > even if the design of gcc allowed it.
> 
> Ok, I'll bite.  Why not? 

5 gets you 10 is isn't flexible enough to handle the breadth of
targets gcc supports.


r~

optimizing egcs for alpha

[Jiann-Ming Su]

---------- Forwarded message ----------
Date: Wed, 24 Mar 1999 13:35:49 -0800
From: spink@pa.dec.com
Reply-To: axp-list@redhat.com
To: axp-list@redhat.com
Cc: spink@pa.dec.com
Subject: Re: Compaq sees Linux as selling Alpha chips 
Resent-Date: 24 Mar 1999 21:49:41 -0000
Resent-From: axp-list@redhat.com
Resent-cc: recipient list not shown: ;

> 
> >>> Jiann-Ming Su said:
> > 
> > If Compaq wants to make real noise with Linux, help optimize gcc/egcs for
> > the Alpha processor.  And put the updates right back into gcc/egcs unlike
> > what Intel is doing with Cygnus and the PII/PIII optimizations.
> 

Well, part of the problem is that both gcc and to some extent egcs are broken 
from a plug ability perspective.  The basic optimization model of gcc has 
serious problems.  What you are asking for is essentially for Compaq to 
rewrite gcc from scratch and then release the source code, and while I think 
it would be cool, it just isn't going to happen.  If gcc/egcs had a more 
modern design(like, shudder, MSVC) then it would be relatively easy to slot in 
the GEM back end.  


Aaron Spink
not speaking for Compaq



-- 
To unsubscribe: send e-mail to axp-list-request@redhat.com with
'unsubscribe' as the subject.  Do not send it to axp-list@redhat.com

Re: optimizing egcs for alpha

[Richard Henderson]

On Wed, Mar 24, 1999 at 04:16:31PM -0600, Jiann-Ming Su forwarded:
> > If gcc/egcs had a more modern design(like, shudder, MSVC) then
> > it would be relatively easy to slot in the GEM back end.  
> > 
> > Aaron Spink
> > not speaking for Compaq

Erum, but we don't _want_ to just "slot in" the GEM back end,
even if the design of gcc allowed it.


r~