b autocast_mode.h:23       if cast is support
b Dispatcher.h:548
b autocast_mode.cpp:174    do cast
p dispatchKeySet.highestPriorityTypeId()  $4 = c10::DispatchKey::AutocastCUDA
p kernel  unboxed_kernel_func_ = 0x7fffbc8d7792 <c10::impl::wrap_kernel_functor_unboxed_
          <c10::impl::detail::WrapFunctionIntoFunctor_<c10::CompileTimeFunctionPointer
          <at::Tensor(c10::DispatchKeySet, const at::Tensor&, const at::Tensor&), 
          torch::autograd::VariableType::(anonymous namespace)::mm>, at::Tensor, 
          c10::guts::typelist::typelist<c10::DispatchKeySet, const at::Tensor&, 
          const at::Tensor&> >, at::Tensor(c10::DispatchKeySet, const at::Tensor&, const at::Tensor&)>
          ::call(c10::OperatorKernel *, c10::DispatchKeySet, const at::Tensor &, const at::Tensor &)>}


# /home/colin/develop/pytorch/aten/src/ATen/core/boxing/KernelFunction_impl.h:57
# break pytorch/build/aten/src/ATen/Operators_1.cpp:3641
# break Dispatcher.cpp:75

# autocast_mode.cpp:193
# Dispatcher.h:548