This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: loop patches mark 11a


On Fri, Jun 19, 1998 at 12:25:20AM -0600, Jeffrey A Law wrote:
> Previously I mentioned that they were still mis-compiling some of the
> spec92 suite on the PA.  I finally got around to looking at this a little
> closer and have something you might be able to investigate.

Got it.  When examining movables for giv relations, we may not use
(or rather stop on) movables that match another, as they will not be
reemitted.

I'll post mark 12 momentarilly with this bit in:

                    /* If we match another movable, we must use that, as
                       this one is going away.  */
                    if (m->match)
                      return simplify_giv_expr (m->match->set_dest, benefit);

which yields

 (insn 704 703 708 (set (reg:SI 449)
         (plus:SI (reg:SI 452)
-            (reg:SI 185))) -1 (nil)
+            (reg:SI 112))) -1 (nil)
     (nil))

in that test case.


r~


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]