__global__ void Cuda_CommCuda_PackCommVel_Kernel(int* sendlist, int n, int maxlistlength, int iswap, X_CFLOAT dx, X_CFLOAT dy, X_CFLOAT dz, void* buffer)
{
int i = (blockIdx.x * gridDim.y + blockIdx.y) * blockDim.x + threadIdx.x;
__global__ void Cuda_CommCuda_PackComm_Self_Kernel(int* sendlist, int n, int maxlistlength, int iswap, X_CFLOAT dx, X_CFLOAT dy, X_CFLOAT dz, int first)
{
int i = (blockIdx.x * gridDim.y + blockIdx.y) * blockDim.x + threadIdx.x;
__global__ void Cuda_CommCuda_PackCommVel_Self_Kernel(int* sendlist, int n, int maxlistlength, int iswap, X_CFLOAT dx, X_CFLOAT dy, X_CFLOAT dz, int first)
{
int i = (blockIdx.x * gridDim.y + blockIdx.y) * blockDim.x + threadIdx.x;