This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: PATCH to remove unnecessary NOTE_INSN_DELETED
- To: Jeffrey A Law <law at cygnus dot com>
- Subject: Re: PATCH to remove unnecessary NOTE_INSN_DELETED
- From: Bernd Schmidt <bernds at cygnus dot co dot uk>
- Date: Thu, 4 Nov 1999 18:10:22 +0000 (GMT)
- cc: Mark Mitchell <mark at codesourcery dot com>, kenner at vlsi1 dot ultra dot nyu dot edu, gcc-patches at gcc dot gnu dot org
> > OK, that sounds plausible to me. I'm still concerned about the extra
> > word per instruction; profiling showed that about 35% of all RTL comes
> > from insn/jump_insn usage. On the other hand, inlining on trees will
> > me no RTL is saved across functions, so RTL usage will probably become
> > a non-concern altogether; instead, we'll need to worry about tree
> > usage.
> Well, we burn an extra word per insn, but we lose the per-pass uid<->map map
> which was 1 word * max_uid. It ought to be a slight improvement.
>
> Any per-pass tables currently based on uids, we convert them to luids.
> (for example INSN_PRIORITY, INSN_COST, INSN_UNIT, etc in haifa).
Just FYI, I started to look into this. I've hit one stumbling block so far
(expressions of the form uid_luid[REGNO_FIRST_UID (xxx)], but I guess that
can be worked around somehow.
Bernd