Recursive verifier: inline trivial functions - #553
Merged
Conversation
Member
Author
|
!benchmark aiur-recursive fresh |
Aiur compiles every non-inlined function to its own circuit, so the single-op base-field wrappers (gl_add, gl_sub, gl_mul, gl_sq) each cost a circuit for what is one native op. Drop them and call the operation directly at the call sites; delete the unused ones outright (gl_neg, gl_is_zero, gl_eq) along with the constant fns (gl_zero/one/two/seven), whose uses are literal constants now. Regenerated aiur_multi_stark.rs (225 -> 214 circuits).
Drop gl_div (rewritten as a * @gl_inverse(b) at its single call site) and switch every call of gl_inverse, gl_val, gl_lt_p and gl_to_bytes to the inline form (@fn), so their bodies splice into the callers and they stop being reachable through constrained call edges: committed circuits go 204 -> 201. gl_val keeps a circuit only because bytecode dedup merges it with limb_to_field, which is still called normally. Regenerated aiur_multi_stark.rs.
The previous commit's rewrite missed the two nested calls in gl_to_bytes(gl_val(x)) (canon_lanes, rows_pop), which alone kept the gl_val circuit constrained. With those fixed, limb_to_field still pins the circuit (bytecode dedup merges the two), so inline-call its sites too. Measured on bench-recursive-verifier --execute-only (factorial(5), q=100, RAYON_NUM_THREADS=1), inlining limb_to_field on top of the nested-call fix is a strict win: the gl_val circuit disappears (212 -> 211 circuits), and its callers get NARROWER, not wider -- a call's channel columns cost more than the spliced 8-term byte fold (lanes_to_gl 31 -> 28, digest_eq 99 -> 83). Total committed width 11454 -> 11421, total FFT cost 7.9829e9 -> 7.9800e9 (-0.035%). Regenerated aiur_multi_stark.rs.
Switch every call of eg_add/eg_sub/eg_neg/eg_mul/eg_inverse/eg_div/eg_eq to the inline form (@fn), including eg_div's internal eg_mul/eg_inverse. All seven circuits disappear (constrained fns 201 -> 193; the extra one is a dedup-class merge among rewritten callers), and the hot callers get narrower, same as with limb_to_field: ro_fold 56 -> 48 at height 13502, rollin 39 -> 35, from_ext_basis 18 -> 10; only trivial-height circuits widen (trace_selectors 38 -> 52 at height 3). bench-recursive-verifier --execute-only (factorial(5), q=100, RAYON_NUM_THREADS=1): circuits 211 -> 204, total width 11421 -> 11239, total FFT cost 7.9800e9 -> 7.9407e9 (-0.49%). Regenerated aiur_multi_stark.rs.
gabriel-barrett
force-pushed
the
multistark-inline-gl-wrappers
branch
from
August 12, 2026 15:40
761a6a6 to
43c89ef
Compare
|
| constant | recursive-prove-time (main) | recursive-prove-time (PR) | Δ% | recursive-peak-ram (main) | recursive-peak-ram (PR) | Δ% | recursive-proof-size (main) | recursive-proof-size (PR) | Δ% | recursive-verify-time (main) | recursive-verify-time (PR) | Δ% | recursive-execute-time (main) | recursive-execute-time (PR) | Δ% | recursive-fft-cost (main) | recursive-fft-cost (PR) | Δ% | prove-time (main) | prove-time (PR) | Δ% | proof-size (main) | proof-size (PR) | Δ% | verify-time (main) | verify-time (PR) | Δ% | peak-ram (main) | peak-ram (PR) | Δ% |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Nat.add_comm |
29.234 s | 26.376 s | -9.8% (1.11× faster) 🟢 | 84.87 GiB | 81.19 GiB | -4.3% 🟢 | 5.60 MiB | 5.48 MiB | -2.1% | 35.4 ms | 33.0 ms | -6.8% (1.07× faster) 🟢 | 5.260 s | 4.132 s | -21.4% (1.27× faster) 🟢 | 157.17B | 148.08B | -5.8% (1.06× fewer) 🟢 | 1.057 s | 1.045 s | -1.1% | 9.11 MiB | 9.11 MiB | +0.0% | 57.4 ms | 57.0 ms | -0.8% | 3.45 GiB | 4.42 GiB | +28.0% (1.28× larger) |
Member
Author
|
!benchmark aiur fresh |
arthurpaulino
approved these changes
Aug 12, 2026
|
| constant | prove-time (main) | prove-time (PR) | Δ% | throughput (const/s) (main) | throughput (const/s) (PR) | Δ% | peak-ram (main) | peak-ram (PR) | Δ% | execute-time (main) | execute-time (PR) | Δ% | verify-time (main) | verify-time (PR) | Δ% | proof-size (main) | proof-size (PR) | Δ% | fft-cost (main) | fft-cost (PR) | Δ% |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
ByteArray.utf8DecodeChar?_utf8EncodeChar_append |
39.179 s | 39.363 s | +0.5% | 68.690 | 68.360 | -0.5% | 79.87 GiB | 79.91 GiB | +0.0% | 9.825 s | 9.868 s | +0.4% | 144.5 ms | 138.2 ms | -4.4% 🟢 | 22.03 MiB | 22.03 MiB | +0.0% | 151.82B | 151.82B | +0.0% |
Array.extract_append |
39.182 s | 38.043 s | -2.9% | 39.350 | 40.530 | +3.0% | 77.06 GiB | 77.01 GiB | -0.1% | 9.658 s | 9.573 s | -0.9% | 145.5 ms | 143.8 ms | -1.1% | 21.93 MiB | 21.93 MiB | +0.0% | 141.66B | 141.66B | +0.0% |
Char.ofOrdinal_le_of_le |
29.721 s | 29.397 s | -1.1% | 89.030 | 90.010 | +1.1% | 61.93 GiB | 61.93 GiB | -0.0% | 6.840 s | 6.599 s | -3.5% 🟢 | 139.5 ms | 131.6 ms | -5.7% (1.06× faster) 🟢 | 21.98 MiB | 21.98 MiB | +0.0% | 108.99B | 108.99B | +0.0% |
Vector.extract_append._proof_2 |
21.618 s | 21.648 s | +0.1% | 60.230 | 60.150 | -0.1% | 41.40 GiB | 41.38 GiB | -0.0% | 5.289 s | 5.415 s | +2.4% | 131.4 ms | 134.7 ms | +2.5% | 21.64 MiB | 21.64 MiB | +0.0% | 81.14B | 81.14B | +0.0% |
_private.Init.Data.Range.Polymorphic.SInt.0.Int64.instRxcHasSize_eq |
18.092 s | 18.188 s | +0.5% | 100.150 | 99.620 | -0.5% | 38.16 GiB | 38.21 GiB | +0.1% | 3.642 s | 3.662 s | +0.5% | 135.6 ms | 140.5 ms | +3.6% |
21.82 MiB | 21.82 MiB | +0.0% | 60.56B | 60.56B | +0.0% |
String.split |
17.445 s | 16.959 s | -2.8% | 101.180 | 104.070 | +2.9% | 36.08 GiB | 36.10 GiB | +0.1% | 3.416 s | 3.343 s | -2.1% | 135.4 ms | 131.7 ms | -2.8% | 22.02 MiB | 22.02 MiB | +0.0% | 55.29B | 55.29B | +0.0% |
List.mergeSort |
12.141 s | 12.329 s | +1.5% | 119.260 | 117.450 | -1.5% | 25.42 GiB | 25.42 GiB | -0.0% | 2.378 s | 2.421 s | +1.8% | 132.3 ms | 133.9 ms | +1.2% | 21.87 MiB | 21.87 MiB | +0.0% | 40.03B | 40.03B | +0.0% |
Vector.append |
4.000 s | 4.010 s | +0.3% | 121.240 | 120.940 | -0.2% | 7.29 GiB | 7.30 GiB | +0.1% | 680.6 ms | 670.0 ms | -1.6% | 124.5 ms | 123.6 ms | -0.7% | 20.47 MiB | 20.47 MiB | +0.0% | 8.83B | 8.83B | +0.0% |
Nat.gcd_comm |
3.254 s | 3.323 s | +2.1% | 119.860 | 117.350 | -2.1% | 6.17 GiB | 5.89 GiB | -4.6% 🟢 | 525.8 ms | 535.3 ms | +1.8% | 122.1 ms | 119.2 ms | -2.4% | 20.24 MiB | 20.24 MiB | +0.0% | 5.93B | 5.93B | +0.0% |
String.append |
2.434 s | 2.410 s | -1.0% | 124.900 | 126.140 | +1.0% | 4.61 GiB | 4.64 GiB | +0.6% | 404.7 ms | 401.2 ms | -0.9% | 118.2 ms | 117.4 ms | -0.7% | 19.51 MiB | 19.51 MiB | +0.0% | 3.27B | 3.27B | +0.0% |
Int.gcd |
1.977 s | 2.004 s | +1.4% | 105.220 | 103.770 | -1.4% | 5.46 GiB | 4.55 GiB | -16.7% (1.20× smaller) 🟢 | 344.3 ms | 347.1 ms | +0.8% | 125.5 ms | 124.3 ms | -1.0% | 19.07 MiB | 19.07 MiB | +0.0% | 2.10B | 2.10B | +0.0% |
Nat.sub_le_of_le_add |
1.825 s | 1.818 s | -0.4% | 93.130 | 93.510 | +0.4% | 5.45 GiB | 5.43 GiB | -0.4% | 337.6 ms | 338.0 ms | +0.1% | 114.3 ms | 113.2 ms | -0.9% | 19.42 MiB | 19.42 MiB | +0.0% | 1.79B | 1.79B | +0.0% |
Nat.add_comm |
1.033 s | 1.050 s | +1.7% | 40.670 | 39.990 | -1.7% | 3.92 GiB | 3.59 GiB | -8.4% (1.09× smaller) 🟢 | 247.9 ms | 250.9 ms | +1.2% | 107.2 ms | 105.5 ms | -1.6% | 17.60 MiB | 17.60 MiB | +0.0% | 292.23M | 292.23M | +0.0% |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.