feat: support U384 in scalar arithmetic #1341
Conversation
There was a problem hiding this comment.
This isn't quite right, it should be U352 according to NIST FIPS 186-5
|
The integer size needs to be a multiple of the limb size. 352-bits is 32-bits * 11, however it isn't evenly divisible by 64-bits |
|
Right, so U352 per FIPS but i checked the crypto-bigint source - no U352 available, only U192, U224, U256, U384, U448, U512, etc. U384 is the closestr we've got and it actually divides evenly into both 32-bit (12 limbs) and 64-bit (6 limbs). Should U384 be fine for us or am i missing something? |
|
It is fine, was my bad. Please refrain from explicit pinging for just a regular answer. Thanks. |
p256/src/arithmetic/scalar.rs
Outdated
| type Bytes = [u8; 48]; | ||
|
|
||
| fn reduce(w: U384) -> Self { | ||
| // Convert U384 to U512 for use with barrett_reduce, which expects 512 bits total |
There was a problem hiding this comment.
Perhaps this should be composed in terms of a Reduce<U512> impl.
If you do that, you can also use u384.concat(&U128::ZERO) to obtain a U512.
There was a problem hiding this comment.
Would address #1042 as well.
|
One complication here is I'd like to attempt to migrate to a generic field element implementation (#1311) which wouldn't support bespoke reduce impls like this, and would only support generic zero and nonzero reductions for a modulo-sized integer and a double sized "wide" integer. I can potentially merge this for now, I'm just saying I can't guarantee it will stick around, and it would be up to the caller to concat zeros to make a U512, then reduce. |
Makes sense! This won't fit the generic architecture. U384 -> U512 conversion is cleaner. I'm cool with merging this as-is. When the generic implementation lands, users can just do the zero-padding and use standard reduction I guess |
Co-authored-by: Tony Arcieri <[email protected]>
Co-authored-by: Tony Arcieri <[email protected]>
Description
Implement Reduce and ReduceNonZero traits for U384 in Scalar, update tests to cover new cases.
Fixed #1199