chat cuda gpu

the chatgpt trick for writing kernels is unironically amazing, recently took some pytorch code, said please write this into C then slapped on some OMP macros. Here we go a custom vectorized kernel