This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa] tree-ssa vs. fold
- From: Jeff Sturm <jsturm at one-point dot com>
- To: Richard Henderson <rth at redhat dot com>
- Cc: gcc at gcc dot gnu dot org
- Date: Sat, 28 Jun 2003 01:25:56 -0400 (EDT)
- Subject: Re: [tree-ssa] tree-ssa vs. fold
Going back to some older email...
On Sat, 7 Jun 2003, Richard Henderson wrote:
> On Fri, Jun 06, 2003 at 06:18:17PM -0400, Jeff Sturm wrote:
> > d) Fix fold() not to save_expr.
> >
> > Is d) the only real choice? Am I missing something easy?
>
> While (d) is certainly desirable, there is an even more
> short-term solution that would probably benefit mainline
> as well: a SAVE_EXPR shouldn't have been created for a
> simple non-volatile DECL.
That sounds nice, though in this specific example it seems a little bit
perverse to teach save_expr about SSA_NAME nodes. At this stage the
DECLs have already been rewritten for SSA.
It also may not solve the whole problem if there are other ways fold() can
return a non-GIMPLE tree other than SAVE_EXPR.
I can't think of anything short of factoring out the bits of fold()
suitable for GIMPLE. Seems as though this routine is trying to be too
many things at once.
Jeff