Effect of -fno-reorder-blocks?

[Jan Hubicka]

> In message <200211182045.gAIKjin2021714@ignucius.axis.se>, Hans-Peter Nilsson w
> rites:
>  >> Date: Mon, 18 Nov 2002 15:23:08 -0500
>  >> From: Diego Novillo <dnovillo@redhat.com>
>  >
>  >> > FWIW, I see only that change in the SPEC95 results.  Nothing
>  >> >
>  >> Since I'm not tracking mainline with the SPEC2000 runs I didn't
>  >> think they would interest you.  I can add it for tonight's run if
>  >> you'd like.
>  >
>  >Thanks but I don't think that's necessary.  I think the SPEC95
>  >results are conclusive: the effect of -f(no-)reorder-blocks is
>  >minimal for this machine (PIII).
> I'd expect reorder-blocks to have the most effect on targets that do static
> prediction of branches (such at BTFNT).  These schemes still show up in
> embedded systems (and of course older desktops like the PA1.1 based 
> targets).
It also should show up on targets that are bound by instruction decoding
as jumps usually stall the process a lot.  THis includes P3 and Athlon.
On Athlon it helps, I am quite puzzled it does not on P3.

Honza
> 
> jeff
>