[Gc] gcc prefetch accidentally disabled

Boehm, Hans hans.boehm at hp.com
Thu Jul 20 17:08:05 PDT 2006


Thanks.  That's interesting.

I have no idea about the 10% difference.  Performance is very sensitive
to the mark loop, and I have seen significant differences there between
compilers.  On X86, I would guess it's a difference in the register
allocator and the generated spill code.  But that's only a guess.

Hans 

> -----Original Message-----
> From: gc-bounces at napali.hpl.hp.com 
> [mailto:gc-bounces at napali.hpl.hp.com] On Behalf Of Peter Wang
> Sent: Wednesday, July 19, 2006 11:54 PM
> To: Boehm, Hans
> Cc: gc at napali.hpl.hp.com
> Subject: Re: [Gc] gcc prefetch accidentally disabled
> 
> On 2006-07-19, Hans Boehm <Hans.Boehm at hp.com> wrote:
> > Thanks.  I checked the patch into 7.0 CVS.  If there is a 
> 6.9 release 
> > I'll include it there, too.
> > 
> > Any idea whether there is a significant performance impact? 
>  It seems 
> > to vary greatly across platforms.
> > 
> > Hans
> 
> It seems to make the Mercury compiler ~4% faster (in a 
> particular configuration).  This is with gcc 3.3.5.
> 
> BTW, when the collector is compiled with gcc 3.4.4 it seems 
> to be around 10% slower (nothing to do with the prefetching 
> change, and the rest of Mercury compiled with gcc 3.4.4).  
> Have you seen that before?
> 
> Peter
> 
> _______________________________________________
> Gc mailing list
> Gc at linux.hpl.hp.com
> http://www.hpl.hp.com/hosted/linux/mail-archives/gc/
> 



More information about the Gc mailing list