This is the mail archive of the
java@gcc.gnu.org
mailing list for the Java project.
RE: [cygnus.project.sid] branch probability hinting
- To: "Boehm, Hans" <hans_boehm at hp dot com>
- Subject: RE: [cygnus.project.sid] branch probability hinting
- From: Jeff Sturm <jsturm at one-point dot com>
- Date: Wed, 7 Mar 2001 01:26:57 -0500 (EST)
- cc: Tom Tromey <tromey at redhat dot com>, "'Andrew Haley'" <aph at redhat dot com>, Java Discuss List <java at gcc dot gnu dot org>
On Tue, 6 Mar 2001, Boehm, Hans wrote:
> I know very little about the current gcc back end. I'll believe that a
> 50-50 hint currently doesn't help. But it would be nice to design this
> facility so that the right information could be supplied.
I agree. It appears that the current backend does support a range of
probabilities, among which __builtin_expect can only select
PROB_VERY_LIKELY or PROB_VERY_UNLIKELY. Perhaps this should be mentioned
on the GCC developers list. I don't think many backend hackers read the
java list.
There is also a comment that the register allocator ought to consider
basic block execution counts. Since a new experimental allocator is now
in development, there is a chance that branch probabilities can soon
become more useful.
Jeff