* @brief Checks if the current layout ends with a specified layout. * @param test The layout to check against. * @return true if the current

Converts a hierarchical NVTX range tree with parent-child relationship into a flat tree by adding the names of parent nodes in-front of all t

Generate data as numpy array Args: shape (tuple or list): Data shape dtype (numpy dtype): Data type (e.g. np.uint8) max_r

* @brief Retrieves the depth index in the tensor layout. * @return Depth index or -1 if not found. */

* @brief Check if the layout includes a batch dimension. * * @return true if the layout includes a batch dimension, false otherwise. */

* @brief Check if the layout corresponds to an image. * * @return true if the layout corresponds to an image, false otherwise. */

* @brief Checks if the tensor layout is row-major. * @return true if row-major, false otherwise. */

* @brief Returns the number of samples in the tensor. If the tensor is not batched, the number of samples is 1. (i.e Nhwc) * * @return The n

Returns whether the layout ends with a given test layout. * * If test layout is empty, it's always considered to be found at the end. * * @param [

Returns the index of the dimension in the layout given its label. * * @param [in] layout Layout where the label is to be searched. * @param [in] di

Returns the layout label at the given index. * * @param [in] layout Layout to be queried * @param [in] idx Index of the label * * @returns If @p

Returns the number of dimensions of the tensor layout * * @param [in] layout Layout to be queried * * @returns Number of dimensions. */

Returns whether the layout starts with a given test layout. * * If test layout is empty, it's always considered to be found at the start. * * @par

HACK needed by gcc-7.3/7.4 or else it'll complain that it can't inline fprintf in a function with attributes like the ones below.

Gets a CUDA event that signifies that the payload is ready. */

Replaces the managed handle and destroys the previously owned resource, if the handle was not null. * * @param handle The handle to be ma

Auxiliary function to get tensor access for input tensor in

Convert host data to a CUDA buffer Args: host_data (numpy array): Host data Returns: CudaBuffer: The converted CUDA buffer

Convert a data type into one supported by torch Args: data_type (numpy dtype): Original data type Returns: dtype: A data typ

* Convert a vector of this class to an stl vector. * * @return STL std::vector. */

Split a batch tensor into a list of individual tensors using zero-copy. Args: batch_tensor: The batch tensor to split. Returns: