Function dfdx::tensor_ops::matmul

pub fn matmul<A, B, C>(a: A, b: &B) -> <A as MatMulTyping<B>>::C where
    A: Tensor<Dtype = f32> + MatMulTyping<B, C = C>,
    B: 'static + Tensor<Dtype = f32> + Clone,
    C: Tensor<Dtype = f32, Tape = A::Tape>,
    A::Array: Transpose,
    B::Array: Transpose,
    C::Array: Transpose,
    A::Device: MatMulOp<A::Array, B::Array, C::Array>, 

Matrix multiplication. This also supports batched matrix multiplication, and broadcasted matrix multiplication.

Examples

1. Normal matmul

let x: Tensor2D<3, 2> = Tensor2D::zeros();
let y: Tensor2D<2, 4> = Tensor2D::zeros();
let result: Tensor2D<3, 4> = matmul(x, &y);

2. Batched matmul

let x: Tensor3D<10, 3, 2> = Tensor3D::zeros();
let y: Tensor3D<10, 2, 4> = Tensor3D::zeros();
let result: Tensor3D<10, 3, 4> = matmul(x, &y);

3. Broadcasted matmul

let x: Tensor3D<10, 3, 2> = Tensor3D::zeros();
let y: Tensor2D<2, 4> = Tensor2D::zeros();
let result: Tensor3D<10, 3, 4> = matmul(x, &y);