fix opt/2960

[Ulrich Weigand]

Richard Henderson wrote:

>This analysis is incorrect.

Doh.  Sorry for the confusion ...

>So the schedule looks ok to me.  Perhaps it's the doloop optimization
>going wrong...  Try -fno-branch-count-reg.

Yes, as David already said, the problem is that the
loop is actually executed 6 times; that is the loop
*exit test* is executed 6 times, but the BIV increment
only 5 times.

The doloop code actually has code to recognize this
particular situation; check the ??? comment in 
doloop_modify_runtime.  However, this code is used
only when the iteration count is not known at 
compile time; when it is (like here), no such check 
is performed.

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  weigand@informatik.uni-erlangen.de