Shared memory and multiple blocks

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi Everyone,

When I use shared memory within a kernel, is the shared memory variable created for each block of threads?

For example:

__global__ void SOMEKERNEL(double *a, double *b)

{

     __shared__ double c[512];

}

Does each block get c[512]? So if I had 10 blocks, would there be 10 copies of c[512]? Each of which I could load with data from global memory…? I’m trying to implement the reduction code from the SDK in one of my kernels, but my program is crashing and I want to make sure I’ve understood how using shared memory works.

Thanks in advance to any help!

2

Yes, each running block gets its own set of shared memory variables.

3

Great, thanks!