This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: block merging
- To: kenner at vlsi1 dot ultra dot nyu dot edu (Richard Kenner)
- Subject: Re: block merging
- From: Jeffrey A Law <law at cygnus dot com>
- Date: Tue, 21 Sep 1999 03:30:56 -0600
- cc: gcc at gcc dot gnu dot org
- Reply-To: law at cygnus dot com
In message <9909202133.AA15682@vlsi1.ultra.nyu.edu>you write:
> Although it's a major conceptual change, it may well be that the right
> approach is to weaken the test and have a BARRIER get deleted even though
> it's separated by one or more NOTEs from the JUMP_INSN (or CALL_INSN).
> I think this is more robust under the principle of "be liberal and what you
> expect and conservative in what you generate".
What rth & I have been discussing for the long term is to basically
redesign jump.c to work on the cfg. In that model BARRIERs basically go away
because they are not associated with any basic block (they appear between them
which is part of the design problem).
Stan Cox is (possibly unknowingly :-) doing much of the infrastructure for this
project by implementing block reordering. He's got to have the ability to
unlink the insns between blocks, scramble the blocks and use the block/edge
structures to relink/update the insns when he's done (and in the short term
recreate BARRIERS).
jeff