[RFC] Run copyrename one extra time before out-of-ssa

Andrew Pinski pinskia@physics.uc.edu
Sun Jan 23 02:00:00 GMT 2005


On Jan 22, 2005, at 8:42 PM, Steven Bosscher wrote:

> - For SPECfp there is a 30% improvement for mgrid (!), and 6% for mesa.
>   The rest is in the noise (within 0.4%).
>
> Good enough?


For PPC, a inner loop of mgrid (I don't know if this is a major inner 
loop or
not because I was just looking at the asm), we improve a lot.
We go from:
  	lfd f0,0(r2)
  	lfd f13,8(r2)
  	addi r2,r2,8
To:
  	fmr f0,f12
  	lfdu f12,8(r9)

So we changed one load to a move so SMS actually worked in this case 
with
your patch which is cool.

-- Pinski



More information about the Gcc-patches mailing list