dynamically global memory allocation in __global__ or __device__ function?

cudaMalloc allocates global memory in host function.
I don’t know a way to allocate global memory dynamically in global or device function.
Is there a way?

NO, you cannot allocate device memory in kernel function

Thank you