This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: regclass.c: scan_one_insn special casing
- From: Eric Christopher <echristo at redhat dot com>
- To: Richard Kenner <kenner at vlsi1 dot ultra dot nyu dot edu>
- Cc: gcc at gcc dot gnu dot org
- Date: Sat, 27 Mar 2004 16:32:39 -0800
- Subject: Re: regclass.c: scan_one_insn special casing
- References: <10403272355.AA22136@vlsi1.ultra.nyu.edu>
> So that you aren't penalizing every call to scan_one_insn with this
> code that is almost never called.
>
> But that's a small cost compared to the maintenance issue of having
> that define_split in every port.
Which only runs if it is cost effective for the port, instead of your
code which checks every insn at least once. Generic algorithms in the
middle end and port specifics in the backend.
>
> Now, you could certainly claim that there's some reason that this
> transformation can't occur (experimental isn't that interesting here)
> and that the code is therefore obsolete. But I haven't seen that claim.
Of course, you can pull it around and make someone justify that the code
should exist - especially given the compile time penalty. A testcase
that proves it can be called and is in the testsuite as an example.
-eric
--
Eric Christopher <echristo@redhat.com>