Program Listing for File wrapped_tensor.cpp

Return to documentation for file (morpheus/_lib/src/objects/wrapped_tensor.cpp)

Copy
Copied!
            

#include "morpheus/objects/wrapped_tensor.hpp" #include "morpheus/objects/tensor_object.hpp" #include <pybind11/cast.h> #include <pybind11/pytypes.h> #include <array>// needed for make_tuple #include <cstdint>// for uintptr_t #include <vector>// get_shape & get_stride return vectors namespace morpheus { /****** Component public implementations *******************/ /****** TensorObject****************************************/ /****** TensorObjectInterfaceProxy *************************/ pybind11::dict TensorObjectInterfaceProxy::cuda_array_interface(TensorObject &self) { pybind11::dict array_interface; pybind11::list shape_list; for (auto &idx : self.get_shape()) { shape_list.append(idx); } pybind11::list stride_list; for (auto &idx : self.get_stride()) { stride_list.append(idx * self.dtype_size()); } // pybind11::list shape_list = pybind11::cast(self.get_shape()); pybind11::int_ stream_val = 1; // See https://numba.readthedocs.io/en/stable/cuda/cuda_array_interface.html // if (self.get_stream().is_default()) // { // stream_val = 1; // } // else if (self.get_stream().is_per_thread_default()) // { // stream_val = 2; // } // else // { // // Custom stream. Return value // stream_val = (int64_t)self.get_stream().value(); // } array_interface["shape"] = pybind11::cast<pybind11::tuple>(shape_list); array_interface["typestr"] = self.get_numpy_typestr(); array_interface["stream"] = stream_val; array_interface["version"] = 3; if (self.is_compact() || self.get_stride().empty()) { array_interface["strides"] = pybind11::none(); } else { array_interface["strides"] = pybind11::cast<pybind11::tuple>(stride_list); } array_interface["data"] = pybind11::make_tuple((uintptr_t)self.data(), false); return array_interface; } } // namespace morpheus

© Copyright 2023, NVIDIA. Last updated on Feb 3, 2023.