This seems like an obvious issue, but Google turns up nothing interesting. Is it legal to use <code>memset</code> in a CUDA kernel such as: <pre class="prettyprint"><code>__device__ void myKernel() { int array[10]; memset(array, 0, sizeof(array)); // ...etc... } </code></pre> (I know <code>int array[10] = {0};</code> is probably better, but this is just an example of a more complicated case.)

Yes, as described in Appendix B of the programming manual, <code>memset</code>, as well as <code>memcpy</code>, <code>malloc</code>, and <code>free</code> (the latter two only on Compute capability >= 2.0 devices) are supported in device code.

Using memset in a CUDA kernel

This seems like an obvious issue, but Google turns up nothing interesting. Is it legal to use memset in a CUDA kernel such as:

__device__ void myKernel()
{
    int array[10];
    memset(array, 0, sizeof(array));
    // ...etc...
}

(I know int array[10] = {0}; is probably better, but this is just an example of a more complicated case.)

Yes, as described in Appendix B of the programming manual, memset, as well as memcpy, malloc, and free (the latter two only on Compute capability >= 2.0 devices) are supported in device code.