RFA: improvement to if-conversion
Joern RENNECKE
joern.rennecke@st.com
Wed Jan 12 21:19:00 GMT 2005
I've finally found some time to finish the iinitial coding for
if-conversion / cross-jump merge.
> > I think full register liveness information would be harder to keep
> > up-to-date as if conversion progresses.
>
> I think that global_live_at_start/end are kept up-to-date. Leastwise,
> I'd be interested to know how badly they're off. We re-run life info
> at the end, but that's to get death notes inserted properly.
It turns out that it's actually much worse. We don't have any life info
in the first
if-conversion pass at all. A lot of loop optimizations only make sense
after
if-conversion has been properly performed, so I don't think only relying
on the
second if-conversion pass is much good. I suppose that means that I'll
have to
run flow before if-conversion then.
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: cmove-diff-20050112
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20050112/607a241a/attachment.ksh>
More information about the Gcc-patches
mailing list