Crate ndrs 

Re-exports

pub use backend::cpu;

pub use backend::cuda;

pub use builtin_kernels::cpu as cpu_kernels;

pub use builtin_kernels::cuda as cuda_kernels;

pub use device::Device;

pub use dtype::BinaryOpFn;

pub use dtype::BinaryOpKind;

pub use dtype::DTYPE_FLOAT32;

pub use dtype::DTYPE_INT32;

pub use dtype::DType;

pub use dtype::TypeInfo;

pub use dtype::allocate_dtype;

pub use dtype::get_add_op;

pub use dtype::get_binary_op;

pub use dtype::register_binary_op;

pub use dtype::register_dtype;

pub use tensor::ArcTensor;

pub use tensor::RcTensor;

pub use tensor::Tensor;

pub use view::ArcTensorView;

pub use view::RcTensorView;

pub use view::SliceInfo;

pub use view::TensorViewOps;

pub use view::broadcast_shapes;

Modules

backend

builtin_kernels

device

dtype

kernel

macros

stream

tensor

view

view_ops

Macros

impl_add_for_view

加法操作宏（Add, AddAssign）

impl_broadcast_to

形状操作方法宏（每个宏生成一个方法定义）

impl_concat_split

impl_contiguous

impl_device_transfer

impl_elementwise_kernel_for_view

impl_fill

impl_matmul

impl_matmul_into

矩阵乘法方法宏

impl_slice

impl_strided_copy_to

拷贝与连续化方法宏

impl_transpose

launch

启动 RawKernel 的宏，自动添加参数。 用法: launch!(kernel, stream, cfg, arg1, arg2, ...);

s

切片宏实现

tensor