Row modifying scan kernel in CUDA, SYCL or another GPU language

Question

Assume we have M x N array A. I would like to modify it by running a kernel with M independent GPU threads, each executin