chore: simplify add and sub #84
Conversation
| let new_transcript = | ||
| JTranscript { x3: p2.x, y3: p2.y, z3: B::one(), lambda_numerator: B::zero() }; | ||
| result = (p2, new_transcript); | ||
| } else if (rhs_infinity & !lhs_infinity) { | ||
| result = (self, JTranscript::new()); | ||
| let new_transcript = | ||
| JTranscript { x3: self.x, y3: self.y, z3: B::one(), lambda_numerator: B::zero() }; | ||
| result = (self, new_transcript); |
There was a problem hiding this comment.
This breaks down if the z coordinate of the inputs are not 1.
|
I'm surprised that there isn't a diff on the brillig circuit sizes being shown on this PR now. We should be removing a significant number of brillig opcodes. |
|
Looking at this PR vs https://noir-lang.github.io/noir_bigcurve/dev/bench/, we're cutting the number of brillig opcodes for an add by ~10% but this isn't getting a comment being posted on this PR. |
|
10% is pretty chunky though! We should follow this up with |
We set alert-threshold to be 101%, so a comment will only get posted if performance regresses :( |
|
Ahh, ok. I saw noir_bigcurve/.github/workflows/benchmark.yml Lines 104 to 109 in 54de613 |
Description
remove validation with transcript for add and sub if program is unconstrained
Problem*
Resolve #79
Summary*
Additional Context
PR Checklist*
cargo fmton default settings.