need zero / x equals zero

[jim_cox]

Apparently IFort thinks zero / x = NaN

Our program logic expects zero / x = zero

How do we persuade IFort - or is writing our own exception handler the only only option ?

advTHANKXance

Jimmy

[jimdempseyatthecove]

I was lucky. The original code was written in the mid 1980's and had to run with the memory restrictions of that time and with the language constraints of F77. This code had untold number of revisions and extensions over the years (~25). This code was very well written and solid but had inherited all the legacies of each revision (and requirement of being backwards compatible). Redoing the data layout to use modules and allocatables/pointers was a big chore. Using the FPP to trick the compiler to compile old code using new data layout saved an immense amount of work and untold hours of debugging (expected to be introduced due to editing snafus). I will likely revisit this code when Larrabee comes out. I would like (need) to get another 40x if possible (the problems I want to solve still take weeks and months to compute).

If you would like some consultation on this I am available (as I suppose others on this forum are as well).

Jim Dempsey

[jim_cox]

Quoting - jimdempseyatthecove

(the problems I want to solve still take weeks and months to compute).

Jim, we are in the "tens of minutes" category at present

So an increase in speed would be nice - but really we need to increase scale as well.

An increase in scale has some "interesting" implications for the data storage - especially when backwards compatible is a requirement.

Hopefully a speed incraese can be gained simply by using the hardware to best advantage.

But we will leave all that for later - For now I'm concentrating on just getting the existing essentials working.

Cheers

Jim

=mjc=
.