Topic: Odd optimizer bug

I stumbled across an odd optimizer bug. I just though that I'd flag it here.
I was updating "Calpak" a collection of Calendar routines by John Burkardt
and I found that it generated a lot on nonsense at optimization level -O3
but worked perfectly with no optimizations. After a lot to sleuthing, I
tracked it down to this line:       CALL Y_ASTRONOMICAL_TO_COMMON(Y , Y)
the left "Y" is the input and the right "Y" is the output. I am not sure exactly what
the optimizer was doing, but at -O3, it was returning just random junk.
doing this, fixed it;
temp=y
CALL Y_ASTRONOMICAL_TO_COMMON(Y , temp)
y=temp
So, if you are getting inexplicable results in your code, check that you don't have the same error.

  • jeff
  • Administrator
  • Offline

Re: Odd optimizer bug

I'm not overly surprised about this issue.  Just glancing at the routine, the arguments aren't marked with intent() attributes, so the optimizer is taking some liberties since the arguments are, in fact, passed by reference.

Jeff Armstrong
Approximatrix, LLC