Suppose you declare a static array in the scope of a CUDA Fortran kernel. Is that array kept in register (as scalars are)?
I’ve tried it out, but the compiler’s ptxinfo is giving me some conflicting info…
Hi Ron,
Register allocation is handled by the back-end CUDA compiler so not something we directly control. Though, my assumption is that static arrays would be kept in registers and in a small test case, I do see that my static arrays are being put into registers.
Could explain a bit more about what you’re seeing?
-Mat