SIMD Barycentric Evaluation and Test

[Gali-StarkWare]

No description provided.

[reviewable-StarkWare]

This change is 

[Gali-StarkWare]

• Replace OODS Eval Method #1074 : 2 dependent PRs (#1096 , #1111 )
• Barycentric Evaluation Bench #1073
• Parallel Barycentric and Weights #1072
• Save Barycentric Weights in Bit Reverse Order #1071
• Barycentric Evaluation in Poly Ops #1070
• SIMD Barycentric Evaluation and Test #1069 👈 (View in Graphite)
• CPU Barycentric Evaluation and Test #1068
• gali/barycentric

This stack of pull requests is managed by Graphite. Learn more about stacking.

[shaharsamocha7]

Reviewable status: 0 of 1 files reviewed, 1 unresolved discussion

---

crates/prover/src/core/poly/circle/evaluation.rs line 249 at r1 (raw file):

// TODO(Gali): Remove.
#[allow(dead_code)]
fn simd_weights(

This function should have unit test

Code quote:

fn simd_weights(

[github-actions]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 2.

Benchmark suite	Current: c76f10e	Previous: 615de1f	Ratio
iffts/simd ifft/27	588941361 ns/iter (± 7171756)	277997930 ns/iter (± 8211035)	2.12
iffts/simd ifft/28	1209210657 ns/iter (± 15563836)	574305772 ns/iter (± 32121353)	2.11

This comment was automatically generated by workflow using github-action-benchmark.

CC: @shaharsamocha7

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 98.38710% with 2 lines in your changes missing coverage. Please review.

Project coverage is 92.73%. Comparing base (ad1902f) to head (c5056c4).
Report is 1 commits behind head on gali/barycentric.

Files with missing lines	Patch %	Lines
crates/prover/src/core/poly/circle/evaluation.rs	98.38%	2 Missing ⚠️

Additional details and impacted files

@@                 Coverage Diff                  @@
##           gali/barycentric    #1069      +/-   ##
====================================================
+ Coverage             92.68%   92.73%   +0.04%     
====================================================
  Files                   106      106              
  Lines                 14374    14497     +123     
  Branches              14374    14497     +123     
====================================================
+ Hits                  13323    13444     +121     
- Misses                  970      972       +2     
  Partials                 81       81              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[shaharsamocha7]

Reviewed all commit messages.
Reviewable status: 0 of 1 files reviewed, 3 unresolved discussions (waiting on @Gali-StarkWare)

---

crates/prover/src/core/poly/circle/evaluation.rs line 249 at r1 (raw file):

Previously, shaharsamocha7 wrote…

This function should have unit test

I want a test that compares this with the cpu weights
Also, try to simplify the code as we did in the cpu pr

---

crates/prover/src/core/poly/circle/evaluation.rs line 321 at r4 (raw file):

    weights: &Col<SimdBackend, SecureField>,
) -> SecureField {
    let evals = evals.clone().bit_reverse();

Is this going to be removed?
If so, please have a TODO on that

Code quote:

let evals = evals.clone().bit_reverse();

---

crates/prover/src/core/poly/circle/evaluation.rs line 326 at r4 (raw file):

            acc + (weights.at(i) * evals.values.at(i))
        });
    }

could we discard this usecase?

Code quote:

    if evals.domain.size() < N_LANES {
        return (0..evals.domain.size()).fold(SecureField::zero(), |acc, i| {
            acc + (weights.at(i) * evals.values.at(i))
        });
    }

[Gali-StarkWare]

Reviewable status: 0 of 1 files reviewed, 3 unresolved discussions (waiting on @Gali-StarkWare and @shaharsamocha7)

---

crates/prover/src/core/poly/circle/evaluation.rs line 249 at r1 (raw file):

Previously, shaharsamocha7 wrote…

I want a test that compares this with the cpu weights
Also, try to simplify the code as we did in the cpu pr

Added the test. Do you want to split the si_i optimization (using only 2 values because it alternates) to another pr?

---

crates/prover/src/core/poly/circle/evaluation.rs line 321 at r4 (raw file):

Previously, shaharsamocha7 wrote…

Is this going to be removed?
If so, please have a TODO on that

I have a TODO in the weights function to change the order, do we need another one here? Same for the CPU function?

---

crates/prover/src/core/poly/circle/evaluation.rs line 326 at r4 (raw file):

Previously, shaharsamocha7 wrote…

could we discard this usecase?

Done.

[shaharsamocha7]

Reviewed 1 of 1 files at r6, all commit messages.
Reviewable status: all files reviewed, 3 unresolved discussions (waiting on @Gali-StarkWare)

---

crates/prover/src/core/poly/circle/evaluation.rs line 238 at r6 (raw file):

/// Computes the weights for Barycentric Lagrange interpolation for point `p` on `coset`.
/// `p` must not be in the domain. For more information, see [`barycentric_weights`].
fn simd_barycentric_weights(

Shouldn't this code be under simd folder somewhere?
I'm not sure

Code quote:

fn simd_barycentric_weights(

---

crates/prover/src/core/poly/circle/evaluation.rs line 281 at r6 (raw file):

    // TODO(Gali): Change weights order to bit-reverse order.
    // S_i(i) is invariant under G_(n−1) and alternate under J

Can you clarify this comment?
It should explain why the second part is multiplied by -1, but I think that comment is a bit implicit.

Code quote:

  // S_i(i) is invariant under G_(n−1) and alternate under J

---

crates/prover/src/core/poly/circle/evaluation.rs line 292 at r6 (raw file):

        .collect();

    weights

Does this pass compiler?

Suggestion:

    (0..weights_vec_len)
        .map(|i| {
            if i < weights_vec_len / 2 {
                vi_p_inverse[i] * si_0_vn_p
            } else {
                vi_p_inverse[i] * -si_0_vn_p
            }
        })
        .collect()

[Gali-StarkWare]

Reviewable status: 0 of 1 files reviewed, 2 unresolved discussions (waiting on @shaharsamocha7)

---

crates/prover/src/core/poly/circle/evaluation.rs line 238 at r6 (raw file):

Previously, shaharsamocha7 wrote…

Shouldn't this code be under simd folder somewhere?
I'm not sure

Next pr is moving it to simd

---

crates/prover/src/core/poly/circle/evaluation.rs line 292 at r6 (raw file):

Previously, shaharsamocha7 wrote…

Does this pass compiler?

Yes, it's like that because in a following parallelism pr I need to do it that way. Would you rather I change it to your suggestion and define weights there?

[shaharsamocha7]

Reviewed 1 of 1 files at r7, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @Gali-StarkWare)

[Gali-StarkWare]

Reviewed 1 of 1 files at r7.
Reviewable status: 0 of 1 files reviewed, all discussions resolved (waiting on @shaharsamocha7)

[shaharsamocha7]

Reviewed 1 of 1 files at r9, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @Gali-StarkWare)

[Gali-StarkWare]

Reviewed all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @Gali-StarkWare)