Advisory: Cannot tell what pointer points to, assuming global memory space

Accelerated Computing CUDA CUDA Programming and Performance
#1

I get the mentioned warning when I compile
I pass in a pointer to pointers to floats.

I tried the search, but got nothing. Kinda hard to believe I’m the first one who gets this message?

Can I do anything about it?

#2

It is a common message. This forum search is dumb. Use google and say “site:http://forums.nvidia.com” to have a consummate search.

This has been discussed many times.

Simply put – The compiler is assuming a global memory access at the indicated location in your source code. Because the compiler cannot know whether a pointer points to shared memory/global memory at run-time. There are no pointer-qualifiers in CUDA. So, the compiler is just making you aware that – it is assuming a global memory indirection at that location in your source code.

#3

And basically there nothing I can do about it? Warnings always makes my skin crawl :P

Well, thanks for taking the time

#4

Yeah true. You can’t do anything about those warnings (probably there is something to disable it…)

Atleast, I know in VC++, there are some pragmas to disable warnings (especially when you have C++ classes exported via DLLs)

#5

Disabling the warnings isn’t a good practice. :rolleyes:

I remove them by initializing my pointers manually.

#6

How do you do this when it’s a parameter to a device function that’s giving the warning? I’d like to get the warnings to go away too.