From: Rob MacLachlan <Rob_MacL<achlan@ra...> - 2002-08-09 16:14:52
In CMUCL there are two limits on reoptimization efforts in
IR1-OPTIMIZE-UNTIL-DONE. Possibly one of these came unglued, or maybe a
new limit is needed. Without these limits there were cases where
optimization would run for an arbitrarily large number of iterations.
In one case, the optimizer was effectively proving that a fixnum loop index
was in fact non-negative, but taking MOST-POSITIVE-FIXNUM steps to do so.