You can compile the different vector sum binaries as follows:

nvcc vector_sum_gpu.cu -o vecto_sum_gpu
nvcc vector_sum_gpu_v2.cu -o vecto_sum_gpu
nvcc vector_sum_gpu_v3.cu -o vecto_sum_gpu

To run:

./vector_sum_gpu <array length>

These three different vector sum files add two vectors together with multiple levels of correctness and usage of threads and blocks.  vector_sum_gpu can only add two arrays of length up to 512.  vector_sum_gpu_v2 has the same limitation, but it uses a 2 dimensional allocation of threads (it sets the block dimensions).  vector_sum_gpu_v3 will work for arrays of any length by using multiple kernel calls.