Function try_launch_copy_perpendicular 

pub fn try_launch_copy_perpendicular(
    client: &Client,
    input: &TensorBinding,
    output: &TensorBinding,
    dtype: ElemType,
) -> bool

Copies input into output with the perpendicular kernel when its blocks can be vectorized, returning whether it did.

Without vectorization a block is a single element and the kernel is a strided copy, so the caller keeps its general path for that case. input and output must have the same shape.