Template Class device_tensor

• Defined in File device_tensor.h

Class Documentation

template<typename T, uint32_t NumDims>
class device_tensor

A cuda device tensor buffer object that references a tensor in cuda device memory buffer for manipulation by the host.

Note

See https://www.quantstart.com/articles/Matrix-Matrix-Multiplication-on-the-GPU-with-Nvidia-CUDA/

Note

See https://forums.developer.nvidia.com/t/guide-cudamalloc3d-and-cudaarrays/23421

Template Parameters:

• T – the tensor data type

• NumDims – number of tensor dimensions

Public Functions

inline device_tensor(Eigen::DSizes<Eigen::DenseIndex, NumDims> shape, const T *data = nullptr)

inline device_tensor(size_t sizeDim0, size_t sizeDim1, size_t sizeDim2, const T *data = nullptr)

inline device_tensor(size_t sizeDim0, size_t sizeDim1, size_t sizeDim2, size_t sizeDim3, const T *data = nullptr)

inline device_tensor(const Eigen::Tensor<T, NumDims> &tensor)

inline device_tensor(device_tensor &&other)

Copy Constructor.

Parameters:

other –

inline device_tensor &operator=(device_tensor &&other) noexcept

inline ~device_tensor()

__host__ inline T *Get()

Gets the raw pointer to device buffer memory.

Returns:

T*

__host__ inline const T *Get() const

Gets the raw pointer to device buffer memory.

Returns:

T const*

__host__ inline Eigen::DenseIndex GetDimensionSize(int dim) const

__host__ inline Eigen::DSizes<Eigen::DenseIndex, NumDims> GetDimensions() const

__host__ inline size_t GetCount() const

Gets the total number of elements in the tensor.

Returns:

host

__host__ inline size_t GetSize() const

Gets the total number of elements in the tensor.

Returns:

host

__host__ inline size_t GetByteSize() const

Gets the total number of bytes in the memory buffer.

Returns:

host

__host__ inline void SetDataSync(const T *data)

Performs a synchronous copy of data into the device buffer.

Parameters:

data –

Returns:

host

__host__ inline void SetDataAsync(const T *data)

Set the Data asyncronously from host memory.

Parameters:

data –

Returns:

host

__host__ inline void SetDataAsync(const device_tensor<T, NumDims> &data)

__host__ inline void ToHost(T *out) const

__host__ inline void ToHost(std::vector<T> &out) const

__host__ inline void ToHost(Eigen::Tensor<T, NumDims> &out) const

__host__ inline void ToHostAsync(T *out) const