Bugs

CM Fortran is a new product and it still has problems. Example. Two arrays which were aligned on the virtual machine in the main program became unaligned in subroutines which used them. This had a catastrophic effect on timing (worse than 100 times slower). While this and other problems are recognized by Thinking Machines, I suspect that the compiler will always be somewhat fragile, and will repay simple coding. Error tracebacks seem to be well handled, but it would be nice to see how the code is really mapped on to the machine in some simple, non-techie way. One of the attractions of the Convex fortran compiler is the way that it makes suggestions. I'd like to see the same sort of friendly intercourse available on the CM-2

 

elliptic
Figure 1 An elliptic impulse from a layered, isotropic medium