This is the mail archive of the
mailing list for the GCC project.
Re: Quadratic behavior due to tree-cfg.c:tree_redirect_edge_and_branch
- From: Steven Bosscher <stevenb at suse dot de>
- To: gcc at gcc dot gnu dot org
- Date: Sun, 19 Sep 2004 23:12:38 +0200
- Subject: Re: Quadratic behavior due to tree-cfg.c:tree_redirect_edge_and_branch
- Organization: SUSE Labs
- References: <email@example.com>
On Sunday 19 September 2004 22:32, Steven Bosscher wrote:
> Note that at -O2 or better we always split critical edges for PRE, so
> this behavior does not only show up in artificial test cases, it happens
> in real code too. It looks like we really need a reverse mapping from
> edges to case labels for each SWITCH_EXPR :-/
Or (to stir up an old discussion ;-) not have labels at all and
not code the CFG in the IL.
This is actually more attractive now because we have the tree CFG
even at -O0, and we preserve the CFG over expand, so there is in
fact no reason to have non-user labels for anything as soon as the
tree CFG is constructed...