GVN-PRE sucks -- well, for SPECCFP2000 mgrid anyway

[Andrew Pinski]

On Aug 15, 2004, at 5:05 PM, Mark Mitchell wrote:

> Daniel Berlin wrote:
>
>> Yeah.
>> I'll see if i can come up with some simple heuristics that help here, 
>> without hurting anything else too much.
>
> I wonder if it's better to restrain PRE, or let it go ahead, and then 
> have the register allocator move computations back into the loop.
>
> I've copied Kenny here to see if he happens to know of any magic 
> bullets for this kind of thing.
>
> (Kenny, the issue is that PRE is pulling so much stuff out of loops 
> that we get register pressure, and hence bad code.)

Just so I got my say and not get Danny into too much trouble but I will 
note that
NAG F95 with gcc 3.3 as the compiler gets around the same score as 
gfortran does.
This is on a dual 2.5GHz G5.


-- Pinski