# How to pass Python device and dtype to C++ **URL:** \ **Category:** C++\ **Created:** [February 13, 2020, 10:02am UTC](https://discuss.pytorch.org/t/how-to-pass-python-device-and-dtype-to-c/69577 "2020-02-13T10:02:06Z")\ **Posts on this page:** 8\ **Page:** 1
**Author:** ![G.M](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/g.m/32/19266_2.png) [@G.M](https://discuss.pytorch.org/u/G.M)\ **Post date:** [February 13, 2020, 10:02am UTC](https://discuss.pytorch.org/t/how-to-pass-python-device-and-dtype-to-c/69577/1 "2020-02-13T10:02:06Z")
I’m trying to implement a C++ function that has the arguments of a Device and Dtype and will be called in Python. But I got error trying to achieve this. Here’s my C++ code: ```auto #define tensor torch::Tensorv tensor fun1(tensor x,torch::Device device,torch::Dtype dtype) { auto opt = torch::TensorOptions(device).dtype(dtype); return x.to(opt); } ``` Here’s the Python part: ```auto a=torch.tensor([1]) fun1(a,torch.device('cpu'),torch.float) ``` Here’s the error: ```auto fun1(a,torch.device('cpu'),torch.float) TypeError: fun1(): incompatible function arguments. The following argument types are supported: 1. (arg0: at::Tensor, arg1: c10::Device, arg2: c10::ScalarType) -> at::Tensor Invoked with: tensor([1]), device(type='cpu'), torch.float32 ``` ---
**Author:** ![tom](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/tom/32/3162_2.png) [@tom](https://discuss.pytorch.org/u/tom)\ **Post date:** [February 14, 2020, 9:07pm UTC](https://discuss.pytorch.org/t/how-to-pass-python-device-and-dtype-to-c/69577/2 "2020-02-14T21:07:54Z")
I don’t think this works directly right now. If you have compiled torch yourself, you see that in torch/csrc/autograd/generated/python\_torch\_functions.cpp the factory functions take the Python objects and unpacks them. The PyBind converters are only present for IntArrayRef and Tensor ([toch/csrc/utils/pybind.h](https://github.com/pytorch/pytorch/blob/master/torch/csrc/utils/pybind.h)). You could try to register your own for torch::Device and torch::Dtype – or submit that PR (1.5 years ago, it was only Tensors). A standard workaround could be to just pass a (potentially empty) tensor and take the device and dtype of that. Best regards Thomas ---
**Author:** ![G.M](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/g.m/32/19266_2.png) [@G.M](https://discuss.pytorch.org/u/G.M)\ **Post date:** [February 15, 2020, 3:13am UTC](https://discuss.pytorch.org/t/how-to-pass-python-device-and-dtype-to-c/69577/3 "2020-02-15T03:13:28Z")
Ok, thank you very much. I’ll try that. ---
**Author:** ![G.M](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/g.m/32/19266_2.png) [@G.M](https://discuss.pytorch.org/u/G.M)\ **Post date:** [February 19, 2020, 12:50pm UTC](https://discuss.pytorch.org/t/how-to-pass-python-device-and-dtype-to-c/69577/4 "2020-02-19T12:50:33Z")
Actually, I found a way to pass dtype to C++. In TouchScript(JIT), `torch.dtype` type are treated as python `int`, which corressponds to what happens in C++, as the dtypes are all `ScalarType`. So I can just create an C++ function that takes in an integer and make it an `ScalarType` to pass the dtype. But this method can only be called in TouchScript because only in TouchScript are dtypes treated as python `int`. ---
**Author:** ![kaihsinwu](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/kaihsinwu/32/36742_2.png) [@kaihsinwu](https://discuss.pytorch.org/u/kaihsinwu)\ **Post date:** [September 3, 2020, 12:02am UTC](https://discuss.pytorch.org/t/how-to-pass-python-device-and-dtype-to-c/69577/5 "2020-09-03T00:02:07Z")
It’s just an follow-up. I recently encounter the similar situation. The workaround solution that I did is to use the str() to cast the torch.dtype into a C++ string, and in the python end write a wrapper around it: Here is the C++ part: ```auto torch::Tensor func1(const vector &shape, const std::string &dtype){ // convert the string -> torch::ScalarType here. auto options = torch.TensorOptions(); if(dtype=="torch.float32"){ options.dtype(torch::kFloat32); } // ... extend this return torch::Tensor(shape,options); } ``` Here is the pybind wrapper part : ```auto PYBIND11_MODULE(cytnx,m){ m.def("c_func1",&func1); } ``` Here is the python module in the directory structure: cytnx/ |- \_\_ init \_\_.py |- [the dynamic linking file compiled from pybind] Inside \_\_ init \_\_.py: ```auto from .cytnx import * ## this is the API expose to user def func1(shape, dtype): # calling the wrapper from pybind return c_func1(shape,str(dtype),str(device)) ``` I would hope there will be a direct support over this in the future ---
**Author:** ![kaihsinwu](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/kaihsinwu/32/36742_2.png) [@kaihsinwu](https://discuss.pytorch.org/u/kaihsinwu)\ **Post date:** [September 7, 2020, 3:38am UTC](https://discuss.pytorch.org/t/how-to-pass-python-device-and-dtype-to-c/69577/6 "2020-09-07T03:38:23Z")
Another way to do this, which I realized later-on is you can pass the py::object, and then cast to the torch::Dtype (torch::ScalarType): ```auto PYBIND11_MODULE(cytnx,m){ m.def("func1",[](py::object dtype){ torch::ScalarType type = torch::python::detail::py_object_to_dtype(dtype); }); } ``` ---
**Author:** ![tom](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/tom/32/3162_2.png) [@tom](https://discuss.pytorch.org/u/tom)\ **Post date:** [September 7, 2020, 6:44am UTC](https://discuss.pytorch.org/t/how-to-pass-python-device-and-dtype-to-c/69577/7 "2020-09-07T06:44:00Z")
Thank you for sharing! ---
**Author:** ![esoha-nvidia](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/esoha-nvidia/32/68502_2.png) [@esoha-nvidia](https://discuss.pytorch.org/u/esoha-nvidia)\ **Post date:** [March 15, 2024, 4:26pm UTC](https://discuss.pytorch.org/t/how-to-pass-python-device-and-dtype-to-c/69577/8 "2024-03-15T16:26:47Z")
If you cast the object to a dtype, you should first check that the conversion is valid. if (THPDtype\_Check(obj.ptr())) { torch::ScalarType scalar\_type = reinterpret\_cast\(obj.ptr())-\>scalar\_type; }