using “memcpyDeviceToHost” to get the value but it runs slowly

Question

I change the kernel script "FakeQuantWithMinMaxVarsFunctor" below.It compiles with GPU pass,but it excuted very very slowly and crashed finally. How do I modify th