I’m not sure what your use-case is but consider that in CUDA, structures of arrays are preferable to arrays of structures.
By this I mean, instead of one array of structures, you have to arrays that you can use to represent the structure. Basically, take the x and y and make them their own array.
Fortunately, two ints are relatively small and CUDA has built-in vectorized types that you can use to load them in efficiently but in general, I think it’s advisable in CUDA to separate things into their own arrays.