[PATCH][4.5] Move gimplifier predicates

Richard Guenther richard.guenther@gmail.com
Sun Dec 7 12:14:00 GMT 2008


On Sun, Dec 7, 2008 at 1:52 AM, Andrew Pinski <pinskia@gmail.com> wrote:
> On Sat, Dec 6, 2008 at 9:10 AM, Richard Guenther <rguenther@suse.de> wrote:
>> So the following removes DECL_GIMPLE_FORMAL_TEMP_P and instead fixes
>> the long-standing post-modify fixme and moves the EH related split of
>> the assignment and call into EH lowering which is IMHO where it belongs.
>> It fixes some more stuff noticed in intermediate fallouts as well.
>> It does not remove the temporary decl sharing with poor mans CSE.
>
> One of the FIXME you will be removing is PR 23401.  Do we get better
> gimplification for the testcase mentioned in the bug report now?  It
> might be a good idea to put the testcase in a form for the testsuite
> to make sure the gimplifier does not regress either.

Yes, we now gimplify the testcase to

ffff (int i)
{
  int D.1246;
  int t1;
  int i1;
  int i2;

  t1 = 0;
  i1 = g.ff[t1].i;
  i2 = g.ff[i].i;
  D.1246 = i1 + i2;
  return D.1246;
}

I'll add a testcase to make sure we don't regress.

Richard.



More information about the Gcc-patches mailing list