Can you create threads outside __ global __?

C开发者_如何转开发an I create threads in _device _ like:

__device__ float func() 
{
  int idx = threadIdx.x + blockIdx.x * blockDim.x;

  // do stuff

  return some_float;
}

Or can you only make threads in _global _ kernels?

int idx = threadIdx.x + blockIdx.x * blockDim.x;

Short answer: The above line of code is perfectly valid in __device__ functions.

However it does not "create" threads. It simply computes an index in idx for the current thread, using that thread's values of threadIdx.x, blockIdx.x and blockDim.x.

The only way to create threads in CUDA is to launch a kernel using the <<<>>> syntax to specify the number and grouping of threads using block and grid dimensions:

int blockSize = 128;
int gridSize = (N + blockSize - 1) / blockSize;

myKernel<<<gridSize, blockSize>>>();