This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
[tree-ssa]: I've got LICM done, but it requires Zdenek's changes
- From: Daniel Berlin <dberlin at dberlin dot org>
- To: dnovillo at redhat dot com, gcc at gcc dot gnu dot org
- Cc: Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>
- Date: Sun, 10 Aug 2003 09:00:36 -0400 (EDT)
- Subject: [tree-ssa]: I've got LICM done, but it requires Zdenek's changes
In order to be able to move variables to the correct scopes, I needed
Zdenek's changes so that we know what scope a variable is in.
Zdenek, i would suggest, rather than keep get_common_scope and
move_var_to_scope static to tree-ssa-copyprop, that they be external
functions in the next iteration of your patch.
LICM uses them unmodified.
I ran a full SPEC on the forward substitution changes, and I get 1% better
on some tests, no regressions on any tests. Not bad for something that
does absolutely nothing in the way of removal or movement.
I'm SPEC'ing fwdsubst + LICM right now.
Then i'll SPEC Zdenek's changes vs an unpatched branch to make sure we
haven't regressed.
LICM hoists over 10000 statements during a bootstrap of gcc, so it's
promising.
It currently doesn't hoist loads or stores.