This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Patches waiting for review
- To: "Zack Weinberg" <zackw at stanford dot edu>
- Subject: Re: Patches waiting for review
- From: Jason Merrill <jason at redhat dot com>
- Date: 03 Mar 2001 13:17:18 +0000
- Cc: gcc-patches at gcc dot gnu dot org
- References: <20010227171026.J27567@wolery.stanford.edu>
>>>>> "Zack" == Zack Weinberg <zackw@stanford.edu> writes:
> http://gcc.gnu.org/ml/gcc-patches/2001-02/msg00350.html
> "Re: c++/1687: Extreme compile time regression from 2.95.2"
> Tue, 6 Feb 2001
> kills exponential time consumption in tree inliner.
You mention a possible problem with a CALL_EXPR that's used in two
different places; this does happen with EH, when a destructor call is
expanded both in the body and as an exception cleanup.
I think we want a variant that only adds to the hash table things for which
tree node sharing really means only the one instance, such as decls, types
and SAVE_EXPRs (modulo UNSAVE_EXPR).
Another possibility would be to notice that scanning a particular node
didn't do us any good, and add it to the hash table.
Thoughts?
Jason