Previously, create_proof() committed to polynomials or evaluations and then performed batch normalization. This approach was inefficient on the CPU (though acceptable on the GPU, as ProjectivePoint was returned) because commit() returned PointXYZZ, which then needed conversion to ProjectivePoint and finally to AffinePoint. This process has now been optimized by directly converting PointXYZZ to AffinePoint, thus improving performance.
Description
Previously,
create_proof()
committed to polynomials or evaluations and then performed batch normalization. This approach was inefficient on the CPU (though acceptable on the GPU, asProjectivePoint
was returned) becausecommit()
returnedPointXYZZ
, which then needed conversion toProjectivePoint
and finally toAffinePoint
. This process has now been optimized by directly convertingPointXYZZ
toAffinePoint
, thus improving performance.