cuda - Is there a way of setting default value for shared memory array?

Question

Consider the following code:

__global__ void kernel(int *something) {
extern __shared__ int shared_array[];     

// Some operations on shared_array here.

}

Is it possible to set whole shared_array to some value - e.g. 0 - without explicitly addressing each cell in some thread?

Answer 1

You can efficiently initialize shared arrays in parallel like this

// if SHARED_SIZE == blockDim.x, eliminate this loop
for (int i = threadIdx.x; i < SHARED_SIZE; i += blockDim.x) 
    shared_array[i] = INITIAL_VALUE;
__syncthreads();