Hi, I'm unsure how to implement the ops::Div division operator

Hi, I've spent some time reading the code,
In this project, there is no direct implementation towards the division operator.

But I found there is a similar CPU implementation in src/cpu/kernels.cc

    template <CpuIsa ISA, typename T>
    void rcp(const T* x, T* y, dim_t size) {
      vectorized_unary_transform<ISA>(x, y, size, Vec<T, ISA>::rcp);
    }

which is defined in src/cpu/vec.h:

      static inline value_type rcp(value_type a) {
        return static_cast<T>(1) / a;
      }

However, I can not find any similar implementation for GPU operation. Did I miss anything? Or, if there's no such implementation in the GPU, what's the recommended way to produce it?

Sincerely,

Start by comparing the CPU reciprocal implementation in src/cpu/kernels.cc and src/cpu/vec.h with the project's GPU kernel sources, which are not identified in the issue. Determine where ops::Div is expected to be handled and what GPU backends must support it; done requires a confirmed implementation path or documented recommendation.