Add Field::sum_of_products method
#80
Conversation
str4d
force-pushed
the
79-sum-of-products
branch
from
9d9b11f to
8ef6be8
Compare
|
Force-pushed to remove the |
str4d
force-pushed
the
79-sum-of-products
branch
from
8ef6be8 to
be49a7c
Compare
| /// above. Implementations of `Field` should override this to use more efficient | ||
| /// methods that take advantage of their internal representation, such as interleaving | ||
| /// or sharing modular reductions. | ||
| fn sum_of_products<'a, I: IntoIterator<Item = (&'a Self, &'a Self)> + Clone>(pairs: I) -> Self { |
There was a problem hiding this comment.
I originally made this an iterator of &(a, b), and I tested this against the array impl and saw little-to-no performance difference. I then changed it to an iterator of (&a, &b) because I thought it would be more flexible, but doing that significantly hurts performance (I presume because we're calling .clone() on an owned tuple that contains references, rather than on a reference).
There was a problem hiding this comment.
Yeah, even using nicer tricks for constructing the input arrays, having this API take (&a, &b) causes bls12_381 pairings to be over 5% slower compared to arrays or &(a, b).
There was a problem hiding this comment.
Tried two iterators instead:
fn sum_of_products<'a>(
a: impl IntoIterator<Item = &'a Self> + Clone,
b: impl IntoIterator<Item = &'a Self> + Clone,
) -> Self
and then a.clone().into_iter().zip(b.clone().into_iter()); this is similarly slower than arrays.
str4d
force-pushed
the
79-sum-of-products
branch
from
be49a7c to
77fb7f0
Compare
Closes #79.