Silverfrost Forums

Welcome to our forums

Numerical difference Silverfrost v8 vs Salford 2.54

6 Mar 2019 4:09 #23324

You don't need a completely standard way of doing anything once you locked yourself into FTNxx by using Clearwin+ - you just need a way that works, and tells you explicitly just in case SF changes the rules.

The best way of doing this is not through a compiler setting or command line option, but with an OPTIONS directive in every file.

      OPTIONS (DREAL)

If you try to compile this source code with a different compiler, it will baulk at the line.

You might even help someone in future by adding a comment:

      OPTIONS (DREAL) ! This makes all REAL variables DOUBLE PRECISION in FTN95 (non-standard)

Who knows, you might even add comments to further describe the issue solved:

      ! The right answers are not obtained with low precision REALs

For mecej4, I suspect that the round-off error was cumulative, as the results shown in the graph slowly diverge. It's that, rather than the 'big error' at the end of the procedure that count.

Eddie

Please login to reply.