Re: sched-deps speedup

[Richard Henderson <rth at redhat dot com>]

On Thu, Jan 11, 2001 at 05:54:01PM +0000, Richard Earnshaw wrote:
> I'm not sure how best to fix it.  I think that somehow we need to make 
> merge_trees find the common values and split the decision graph at a lower 
> layer.

Actually, merge_trees _does_ find the common values, but then notices

  /* Check that their subnodes are at the same position, as any one set
     of sibling decisions must be at the same position.  */
  if (d1->success.first
      && d2->success.first
      && strcmp (d1->success.first->position, d2->success.first->position))
    return 0;

this because the {code=set} test from the first node is at a
different position than the {c_test} test from the second node.

I think the proper solution is to modify the switch generating
code to notice this case.  I'm working on a fix.


r~