nvcc doesn't like __shared__ incomplete type not allowed.

Accelerated Computing CUDA CUDA Programming and Performance
1

I have one shared array just like the programming guide says to do, but when I compile it barfs. Is there an option I’m missing or what? This seems to be something pretty big for it to be a compiler bug, so I have to be missing something

error: incomplete type is not allowed
attribute((shared)) RayStruct shared;

Any help is appreciated.

::SOLVED::
Needed to add extern keyword.

2

Was it an Cpp or C file then? because I don’t think you need to do that in .cu

Next time it’s a good idea to tell where this happens.

3

@Ninkazu,

It would be a better idea to add a reply and say “SOLVED” instead of editing the original note. External Media

4

I have an addendum to this problem. Now that I have everything else working, when I try to set a value in shared memory, the execution stops and I get a message “Aborted (core dumped)”. Anyone else get this problem?
Really shared[0].orig[0] = 0.f is enough to make it abort (orig is a float[3] in RayStruct).

5

I’m kinda lost, maybe it is better to post some parts of your code, because this is something we cannot do allot about.

6

Are you allocating your shared memory in the kernel call? I.e. in kernel<<<blocks, threads, shared>>>() is shared set to the correct number of bytes for the amount of shared memory you will use in each block?

7

That did the trick. Thanks.