Ascend / pytorch

Ascend PyTorch adapter (torch_npu). Mirror of https://gitee.com/ascend/pytorch
https://ascend.github.io/docs/
Other
259 stars 15 forks source link

请问op_pulgin的最终实现在哪? #30

Closed lms-mt closed 7 months ago

lms-mt commented 7 months ago

例如,以relu为例,会调用:


at::Tensor& relu_out_npu_nocheck(at::Tensor& result, const at::Tensor& self) {
  at_npu::native::OpCommand cmd;
  cmd.Name("Relu")
      .Input(self)
      .Output(result)
      .Run();

  return result;
}

那么这个Relu cmd在哪注册的?我又在哪能看到它的kernel实现呢?

yunyiyun commented 7 months ago

底层kernel实现暂未开放