Groth.jl/docs/src/assets/prove_full_msm_tuning_2026_05_11.json at master · 0xpantera/Groth.jl · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
{
  "summary": "Prover-shaped MSM tuning follow-through: prove_full now combines H and L into one G1 MSM because the Groth16 C element only uses H + L.",
  "baseline_run_id": "2026-05-11_133047",
  "candidate_run_id": "2026-05-11_165756",
  "benchmark_profile": "stage8",
  "julia_version": "1.12.6",
  "threadpools": {
    "default": 24,
    "interactive": 1
  },
  "cpu": "znver5",
  "fixtures": {
    "sum_of_products_small": {
      "constraints": 3,
      "public_inputs": 6,
      "domain_size": 16,
      "median_ms": {
        "prove_full": 10.943,
        "compute_h_total": 0.233678,
        "h_msm": 5.458,
        "l_msm": 0.054632,
        "h_l_msm": 5.619,
        "final_c": 2.133
      },
      "msm_selection": {
        "h_query_g1": "pippenger_w3",
        "l_query_g1": "straus",
        "h_l_query_g1": "pippenger_w3"
      }
    },
    "generated_24_constraints": {
      "constraints": 24,
      "public_inputs": 8,
      "domain_size": 32,
      "median_ms": {
        "prove_full": 26.643,
        "compute_h_total": 1.481,
        "h_msm": 8.726,
        "l_msm": 3.871,
        "h_l_msm": 11.029,
        "final_c": 2.14
      },
      "baseline_median_ms": {
        "prove_full": 28.636,
        "compute_h_total": 1.445,
        "h_msm": 8.925,
        "l_msm": 3.906,
        "final_c": 2.256
      },
      "relative_to_baseline": {
        "prove_full": "-6.96%",
        "final_c": "-5.14%"
      },
      "msm_selection": {
        "a_query_g1": "pippenger_w3",
        "a_b1_query_g1": "pippenger_pair_w2",
        "b_query_g1": "pippenger_w3",
        "b_query_g2": "pippenger_w2",
        "h_query_g1": "pippenger_w3",
        "l_query_g1": "pippenger_w3",
        "h_l_query_g1": "pippenger_w5"
      }
    }
  },
  "notes": [
    "H and L are still reported separately as diagnostic phases; prove_full uses the combined H+L path.",
    "The generated fixture is the preferred continuity signal for prover-shaped tuning because the small fixture is intentionally domain-rounded for correctness coverage.",
    "The comparison script reported no regressions above the configured 20% threshold."
  ]
}