Struct Matrix

pub struct Matrix<C: CubeType> { /* private fields */ }

A matrix represent a 2D grid of numbers.

They can either be in a row major or a column major format.

Implementations

impl<C: CubePrimitive> Matrix<C>

pub unsafe fn uninitialized( ident: MatrixIdent, m: u32, n: u32, k: u32, layout: MatrixLayout, ) -> Self

Create a new uninitialized matrix that is going to be used in the matrix-multiply and accumulate function.

Safety

Must be initialized with load or fill before use. Using it without initialization is undefined behaviour on CUDA, and completely invalid on Vulkan.

You have to declare the shape used for the execution. The shape of the current matrix is determined using the MatrixIdent.

• MatrixIdent::A Shape => (M, K)
• MatrixIdent::B Shape => (K, N)
• MatrixIdent::Accumulator Shape => (M, N)

Not all shapes are supported, and the permitted shapes depend on the element type.

Refer to nvidia documentation.

pub fn from_value( ident: MatrixIdent, m: u32, n: u32, k: u32, layout: MatrixLayout, value: C, ) -> Self

Create a new matrix that is going to be used in the matrix-multiply and accumulate function and is filled with value.

You have to declare the shape used for the execution. The shape of the current matrix is determined using the MatrixIdent.

• MatrixIdent::A Shape => (M, K)
• MatrixIdent::B Shape => (K, N)
• MatrixIdent::Accumulator Shape => (M, N)

Not all shapes are supported, and the permitted shapes depend on the element type.

Refer to nvidia documentation.

pub fn from_slice( ident: MatrixIdent, m: u32, n: u32, k: u32, layout: MatrixLayout, value: &Slice<'_, C>, stride: u32, ) -> Self

Create a new matrix that is going to be used in the matrix-multiply and accumulate function and is loaded from value with stride.

You have to declare the shape used for the execution. The shape of the current matrix is determined using the MatrixIdent.

• MatrixIdent::A Shape => (M, K)
• MatrixIdent::B Shape => (K, N)
• MatrixIdent::Accumulator Shape => (M, N)

Not all shapes are supported, and the permitted shapes depend on the element type.

Refer to nvidia documentation.

pub fn __expand_uninitialized( context: &mut CubeContext, ident: MatrixIdent, m: ExpandElementTyped<u32>, n: ExpandElementTyped<u32>, k: ExpandElementTyped<u32>, layout: MatrixLayout, ) -> MatrixExpand

pub fn __expand_from_value( context: &mut CubeContext, ident: MatrixIdent, m: ExpandElementTyped<u32>, n: ExpandElementTyped<u32>, k: ExpandElementTyped<u32>, layout: MatrixLayout, value: ExpandElementTyped<C>, ) -> MatrixExpand

pub fn __expand_from_slice( context: &mut CubeContext, ident: MatrixIdent, m: ExpandElementTyped<u32>, n: ExpandElementTyped<u32>, k: ExpandElementTyped<u32>, layout: MatrixLayout, value: ExpandElementTyped<Slice<'static, C>>, stride: ExpandElementTyped<u32>, ) -> MatrixExpand

Trait Implementations

impl<C: Clone + CubeType> Clone for Matrix<C>

fn clone(&self) -> Matrix<C>

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<C: CubeType> CubeType for Matrix<C>

type ExpandType = MatrixExpand

fn init(context: &mut CubeContext, expand: Self::ExpandType) -> Self::ExpandType

Wrapper around the init method, necessary to type inference.

impl<C: CubeType> IntoRuntime for Matrix<C>

fn __expand_runtime_method(self, _context: &mut CubeContext) -> MatrixExpand

fn runtime(self) -> Self

Make sure a type is actually expanded into its runtime expand type.

impl<C: Copy + CubeType> Copy for Matrix<C>