Skip to content

Use batch normalization in CondAdd::init #71

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
davxy merged 4 commits into from
Feb 19, 2026
Merged

Use batch normalization in CondAdd::init #71

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
50b4ab0
Use batch normalization in CondAdd::init to avoid per-step field inve…
davxy Feb 19, 2026
6a32e75
Update summary files
davxy Feb 19, 2026
97f9642
Prevent shifting
davxy Feb 19, 2026
05c963d
Bump plonk-common and ring-proof versions
davxy Feb 19, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion w3f-plonk-common/Cargo.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
[package]
name = "w3f-plonk-common"
version = "0.0.5"
version = "0.0.6"
edition = "2021"
authors = ["Sergey Vasilyev <swasilyev@gmail.com>"]
license = "MIT/Apache-2.0"
Expand Down
36 changes: 18 additions & 18 deletions w3f-plonk-common/benches/SUMMARY.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,51 +10,51 @@ Machine: AMD Ryzen Threadripper 3970X (64 logical cores), 62 GiB RAM, Arch Linux

| Domain Size | Hiding | Non-Hiding |
|-------------|-----------|------------|
| 512 | 884 us | 865 us |
| 1024 | 1.90 ms | 1.89 ms |
| 4096 | 8.79 ms | 8.85 ms |
| 16384 | 44.2 ms | 44.1 ms |
| 512 | 883 us | 869 us |
| 1024 | 1.89 ms | 1.89 ms |
| 4096 | 9.52 ms | 9.71 ms |
| 16384 | 43.3 ms | 45.3 ms |

Hiding vs non-hiding makes no measurable difference. Scales roughly linearly with domain size.

## Field Column Construction

| Domain Size | private_column | public_column | shifted_4x |
|-------------|----------------|---------------|------------|
| 512 | 455 us | 445 us | 2.31 us |
| 1024 | 982 us | 981 us | 4.62 us |
| 4096 | 4.76 ms | 4.68 ms | 22.8 us |
| 512 | 419 us | 418 us | 2.14 us |
| 1024 | 1.02 ms | 1.14 ms | 4.29 us |
| 4096 | 4.45 ms | 4.92 ms | 18.7 us |

Column construction is dominated by FFT (interpolation + 4x evaluation). `shifted_4x` is a cheap rotate+copy.
Column construction is dominated by FFT (interpolation + 4x evaluation). `shifted_4x` clones and rotates the 4x evaluations.

## Booleanity Gadget

Constraint evaluation in 4x domain.

| Domain Size | constraints |
|-------------|-------------|
| 512 | 45.1 us |
| 1024 | 90.9 us |
| 4096 | 384 us |
| 512 | 48.3 us |
| 1024 | 96.8 us |
| 4096 | 412 us |

Single constraint `b(1-b)`. Linear scaling.

## Inner Product Gadget

| Domain Size | init | constraints | constraints_linearized |
|-------------|---------|-------------|------------------------|
| 512 | 1.65 ms | 100 us | 9.73 us |
| 1024 | 3.20 ms | 210 us | 19.6 us |
| 4096 | 13.8 ms | 942 us | 94.2 us |
| 512 | 1.36 ms | 108 us | 10.2 us |
| 1024 | 2.98 ms | 223 us | 20.9 us |
| 4096 | 14.2 ms | 922 us | 81.8 us |

Init includes column construction (2 FFTs). Constraints are evaluated pointwise in 4x domain. Linearization is a single polynomial scalar multiplication.

## TE Conditional Addition Gadget

| Domain Size | init | constraints | constraints_linearized |
|-------------|----------|-------------|------------------------|
| 512 | 3.78 ms | 857 us | 75.9 us |
| 1024 | 8.03 ms | 1.72 ms | 162 us |
| 4096 | 35.2 ms | 13.9 ms | 669 us |
| 512 | 2.39 ms | 913 us | 81.1 us |
| 1024 | 5.20 ms | 1.83 ms | 160 us |
| 4096 | 25.5 ms | 11.3 ms | 642 us |

Init includes EC conditional additions (sequential scan) plus column construction. Constraint evaluation is the most expensive gadget due to the degree-4 EC addition formulas. Linearization remains cheap.
Init includes EC conditional additions (batch-normalized) plus column construction. Constraint evaluation is the most expensive gadget due to the degree-4 EC addition formulas. Linearization remains cheap.
16 changes: 10 additions & 6 deletions w3f-plonk-common/src/gadgets/ec/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,17 +73,21 @@ where
assert_eq!(bitmask.bits.len(), domain.capacity - 1);
// assert_eq!(points.points.len(), domain.capacity - 1); //TODO
let not_last = domain.not_last_row.clone();
let acc = bitmask
let mut projective_acc = seed.into_group();
let projective_points: Vec<_> = bitmask
.bits
.iter()
.zip(points.points.iter())
.scan(seed, |acc, (&b, point)| {
.map(|(&b, point)| {
if b {
*acc = (*acc + point).into_affine();
projective_acc += point;
}
Some(*acc)
});
let acc: Vec<_> = ark_std::iter::once(seed).chain(acc).collect();
projective_acc
})
.collect();
let mut acc = Vec::with_capacity(projective_points.len() + 1);
acc.push(seed);
acc.extend(P::Group::normalize_batch(&projective_points));
let init_plus_result = acc.last().unwrap();
let result = init_plus_result.into_group() - seed.into_group();
let result = result.into_affine();
Expand Down
2 changes: 1 addition & 1 deletion w3f-ring-proof/Cargo.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
[package]
name = "w3f-ring-proof"
version = "0.0.5"
version = "0.0.6"
edition = "2021"
authors = ["Sergey Vasilyev <swasilyev@gmail.com>"]
license = "MIT/Apache-2.0"
Expand Down
26 changes: 13 additions & 13 deletions w3f-ring-proof/benches/SUMMARY.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,46 +10,46 @@ Machine: AMD Ryzen Threadripper 3970X (64 logical cores), 62 GiB RAM, Arch Linux

| Domain Size | Time |
|-------------|----------|
| 512 | 61.9 ms |
| 1024 | 86.1 ms |
| 512 | 49.7 ms |
| 1024 | 78.1 ms |

Includes KZG trusted setup (`3 * domain_size` degree) and domain/PIOP parameter construction.

## Indexing (Fixed Column Commitments)

| Domain Size | Time |
|-------------|----------|
| 512 | 48.0 ms |
| 1024 | 92.0 ms |
| 512 | 43.3 ms |
| 1024 | 73.8 ms |

Commits the ring key columns and selector polynomial using KZG. Full keyset (max capacity).

## Proving

| Domain Size | Time |
|-------------|-----------|
| 512 | 159 ms |
| 1024 | 289 ms |
| 512 | 158 ms |
| 1024 | 276 ms |

Single proof generation. Includes witness generation (conditional additions, inner product accumulation) and PLONK prover (constraint evaluation, quotient polynomial, KZG commitments and openings).

## Single Verification

| Domain Size | Time |
|-------------|----------|
| 512 | 3.63 ms |
| 1024 | 3.36 ms |
| 512 | 3.21 ms |
| 1024 | 3.08 ms |

Single proof verification. Dominated by pairing checks. Near-constant with domain size as the verifier works with evaluations, not full polynomials.

## Batch Verification (domain_size = 1024)

| Batch Size | Sequential | KZG Accumulator | Speedup |
|------------|------------|-----------------|---------|
| 1 | 3.10 ms | 3.10 ms | 1.0x |
| 4 | 14.0 ms | 5.29 ms | 2.6x |
| 16 | 49.8 ms | 11.3 ms | 4.4x |
| 32 | 99.6 ms | 19.8 ms | 5.0x |
| 1 | 3.33 ms | 3.08 ms | 1.1x |
| 4 | 13.2 ms | 5.64 ms | 2.3x |
| 16 | 52.8 ms | 12.0 ms | 4.4x |
| 32 | 106 ms | 19.8 ms | 5.4x |

Sequential verification scales linearly (one pairing check per proof). KZG accumulator batches all pairing equations into a single check via MSM, giving sub-linear scaling.

Expand All @@ -59,4 +59,4 @@ Sequential verification scales linearly (one pairing check per proof). KZG accum
|------------|---------|
| Compressed | 592 bytes |

Serialization time: ~771 ns.
Serialization time: ~770 ns.
Loading