Possible Buglet in gfortran Release 143193 (and 4.3.1 and 4.3.2)

[Thomas Koenig]

On Sat, 2009-01-24 at 09:16 -0600, Tom Browder wrote:

> In that regard, I'll start another subject on some numerical
> differences I've found I hope someone can address.

I'll give you some stock answers before I even know the question :-)

- If you use -ffast-math, leave it off and see if the problem persists.

- For i386, try "-ffloat-store" or "-march=native -mfpmath=sse2" (if you
  have a system which supports SSE2) to make sure you don't run into the
  extended precision problem.

- If your code is standard-conforming f77, running it through ftncheck
  tends to spot a lot of mistakes.

- You can also use g77 and compare its output against gfortran and your
  other compiler.

	Thomas