Hello,

This brings down compile time for fwprop by avoiding CFG changes in
loop_optimizer_init(). The CFG changes it could make would invalidate
the fast dominance queries, so that walk_dominator_tree had to work
with the slow queries -- rather painful on a CFG with an almost flat
dominator tree.

With the n=10000 test case, we go from unpatched:

 forward prop            :  39.13 (13%) usr

to patched:

 forward prop            :  13.60 ( 5%) usr

That's still not a reasonable amount of time for such a simple pass,
but it's better than before...

To prevent that mistake in the future, I've add an assert in dominance.c.

OK for trunk?

Ciao!
Steven

Attachment: PR56113_3.diff
Description: Binary data

Reply via email to