PATCH: Optimize IA64 brl to br
H. J. Lu
hjl@lucon.org
Thu Jan 29 17:07:00 GMT 2004
On Wed, Jan 28, 2004 at 04:13:39PM -0800, Zack Weinberg wrote:
> "H. J. Lu" <hjl@lucon.org> writes:
>
> > The IA64 compiler may generate brl. But it may be replaced by br
> > during the relax finalize pass. This patch implements it.
>
> Is this replacement actually an improvement? It doesn't shrink the
> code, or execute anything in the now-freed-up slot, so the only way
> it could be is if br is faster than brl, and this I find unlikely.
I was told that br is faster than brl. It has something to do with
branch prediction on hardware.
H.J.
More information about the Binutils
mailing list