Struct PairContraction 

pub struct PairContraction<A> {
    lhs_simplification: Option<SimplificationMethodAndOutput<A>>,
    rhs_simplification: Option<SimplificationMethodAndOutput<A>>,
    method: PairMethod,
    op: Box<dyn PairContractor<A>>,
    simplified_einsum_string: String,
}

Holds a Boxed PairContractor trait object and two Option<Box>ed simplifications for the LHS and RHS tensors.

For example, the contraction ijk,kj->jk will currently be performed as follows:

1. Simplify the LHS with the contraction ijk->jk
2. Don’t simplify the RHS
3. Use HadamardProductGeneral to compute jk,kj->jk

A second example is the contraction iij,jkk->ik:

1. Simplify the LHS with the contraction iij->ij
2. Simplify the RHS with the contraction jkk->jk
3. Use TensordotGeneral to compute ij,jk->ik

Since the axis lengths aren’t known until runtime, and the actual einsum string may not be either, it is generally not possible to know at compile time which specific PairContractor will be used to perform a given contraction, or even which contractions will be performed; the optimizer could choose a different order.

Fields

lhs_simplification: Option<SimplificationMethodAndOutput<A>>

rhs_simplification: Option<SimplificationMethodAndOutput<A>>

method: PairMethod

op: Box<dyn PairContractor<A>>

simplified_einsum_string: String

Implementations

impl<A> PairContraction<A>

pub fn new(sc: &SizedContraction) -> Self

Trait Implementations

impl<A> Debug for PairContraction<A>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<A> PairContractor<A> for PairContraction<A>

fn contract_pair<'a, 'b, 'c, 'd>( &self, lhs: &'b ArrayViewD<'a, A>, rhs: &'d ArrayViewD<'c, A>, ) -> ArrayD<A>

where A: Clone + LinalgScalar, 'a: 'b, 'c: 'd,

fn contract_and_assign_pair<'a, 'b, 'c, 'd, 'e, 'f>( &self, lhs: &'b ArrayViewD<'a, A>, rhs: &'d ArrayViewD<'c, A>, out: &'f mut ArrayViewMutD<'e, A>, )

where A: Clone + LinalgScalar, 'a: 'b, 'c: 'd, 'e: 'f,