I want to build main program with ifort, and CUDA subroutine with pgf.
To port whole source code to pgf is quite burden.
The problem is, I have to pass some derived type arrays to the CUDA subroutine.
Does this work? If so, how can I do it?
I saw a similar topic in this forum, but that topic didn’t involve passing derived type variables.
Sorry, won’t work, can’t work.
Every Fortran compiler represents the new data types and
argument descriptors in their own way.
Fortran in the past (pgf77) passed all arguments as addresses,
and interfacing two different fortran objects was very easy.
If a future Fortran language standard ‘standardizes’ the internal format
of data types and argument descriptors, this would be possible.