Skip to content

EVM: optimize mkOpIxMap for concrete bytecode - #1048

Open
elopez wants to merge 2 commits into
mainfrom
perf-contract-mk
Open

elopez wants to merge 2 commits into
mainfrom
perf-contract-mk

Conversation

@elopez

@elopez elopez commented Mar 9, 2026 •

Copy link
Copy Markdown
Collaborator

Description

Replace the accumulated monadic action chain (m >> write >> write >> ...)
with direct ST writes during ByteString iteration. The old approach
created O(n) closures for the >> chain; the new approach writes directly
to the mutable vector with zero closure allocation.

Checklist

  • tested locally
  • added automated tests
  • updated the docs
  • updated the changelog

@elopez elopez changed the title Perf contract mk EVM: optimize mkCodeOps, mkOpIxMap for concrete case Mar 9, 2026
@msooseth msooseth changed the title EVM: optimize mkCodeOps, mkOpIxMap for concrete case [DRAFT] EVM: optimize mkCodeOps, mkOpIxMap for concrete case Jun 8, 2026
The zero fetcher built an HTTP session and a solver group for every
query, even with RPC disabled and zero solver workers. CREATE now
queries unknown destination accounts, so contract-creation benchmarks
repeated this setup thousands of times and timed out. On macOS, TLS
initialization spawns security to read certificates, leaving the
benchmark waiting for local subprocesses despite making no RPC calls.

Remove zero and migrate its concrete callers to noRpcConcreteFetcher.
Add concreteOracle for concrete runs that can fetch RPC state. Both
share dispatch with oracle; a typed SMT handler makes solver access
unreachable for concrete queries while symbolic queries use the
caller's existing SolverGroup. This keeps account, storage, FFI and
environment behavior in one place without per-query resource setup.

Use the concrete fetcher for CLI execution, test setup and
counterexample replay, removing the zero-worker solver group from
exec. Make sessions optional through the CLI and unit-test options and
create them only when RPC is configured. Equivalence checking also
skips its unused session. RPC runs retain shared sessions and caches,
and symbolic runs retain their caller-owned solver groups.
@elopez elopez changed the title [DRAFT] EVM: optimize mkCodeOps, mkOpIxMap for concrete case EVM: optimize mkOpIxMap for concrete bytecode Nov 14, 2026
@elopez

elopez commented Nov 14, 2026

Copy link
Copy Markdown
Collaborator Author

Some nice results in contractCreationMem and a bit better in contractCreation as well

bench-perf results
All
  loop
    2:     OK
      17.1 μs ± 849 ns, 134 KB allocated,  13 B  copied,  46 MB peak memory,       same as baseline
    4:     OK
      26.8 μs ± 987 ns, 219 KB allocated,  37 B  copied,  46 MB peak memory,       same as baseline
    8:     OK
      47.7 μs ± 4.0 μs, 390 KB allocated,  68 B  copied,  46 MB peak memory,       same as baseline
    16:    OK
      89.4 μs ± 7.1 μs, 732 KB allocated, 129 B  copied,  46 MB peak memory,       same as baseline
    32:    OK
      172  μs ±  15 μs, 1.4 MB allocated, 249 B  copied,  46 MB peak memory,       same as baseline
    64:    OK
      337  μs ±  28 μs, 2.7 MB allocated, 497 B  copied,  46 MB peak memory,       same as baseline
    128:   OK
      670  μs ±  53 μs, 5.4 MB allocated, 989 B  copied,  46 MB peak memory,       same as baseline
    256:   OK
      1.33 ms ± 128 μs,  11 MB allocated, 1.9 KB copied,  46 MB peak memory,       same as baseline
    512:   OK
      2.63 ms ± 143 μs,  21 MB allocated, 3.8 KB copied,  46 MB peak memory,       same as baseline
    1024:  OK
      5.20 ms ± 472 μs,  43 MB allocated, 7.6 KB copied,  46 MB peak memory,       same as baseline
    2048:  OK
      10.4 ms ± 450 μs,  85 MB allocated,  14 KB copied,  46 MB peak memory,       same as baseline
    4096:  OK
      20.5 ms ± 1.7 ms, 171 MB allocated,  23 KB copied,  46 MB peak memory,       same as baseline
    8192:  OK
      40.8 ms ± 3.6 ms, 342 MB allocated,  42 KB copied,  46 MB peak memory,       same as baseline
    16384: OK
      81.8 ms ± 7.3 ms, 684 MB allocated,  79 KB copied,  46 MB peak memory,       same as baseline
  primes
    2:     OK
      36.7 μs ± 3.4 μs, 268 KB allocated,  45 B  copied,  46 MB peak memory, 17% less than baseline
    4:     OK
      68.6 μs ± 6.7 μs, 485 KB allocated,  88 B  copied,  46 MB peak memory, 10% less than baseline
    8:     OK
      124  μs ± 8.6 μs, 905 KB allocated, 172 B  copied,  46 MB peak memory,  7% less than baseline
    16:    OK
      269  μs ±  21 μs, 1.9 MB allocated, 415 B  copied,  46 MB peak memory,       same as baseline
    32:    OK
      613  μs ±  60 μs, 4.5 MB allocated, 1.1 KB copied,  46 MB peak memory,       same as baseline
    64:    OK
      1.38 ms ±  67 μs,  10 MB allocated, 2.8 KB copied,  46 MB peak memory,       same as baseline
    128:   OK
      3.35 ms ± 220 μs,  24 MB allocated, 8.2 KB copied,  46 MB peak memory,       same as baseline
    256:   OK
      7.81 ms ± 570 μs,  58 MB allocated,  27 KB copied,  46 MB peak memory,       same as baseline
    512:   OK
      18.5 ms ± 899 μs, 141 MB allocated,  71 KB copied,  46 MB peak memory,       same as baseline
    1024:  OK
      46.2 ms ± 3.9 ms, 344 MB allocated, 168 KB copied,  46 MB peak memory,       same as baseline
    2048:  OK
      110  ms ± 9.8 ms, 849 MB allocated, 392 KB copied,  46 MB peak memory,       same as baseline
    4096:  OK
      277  ms ±  14 ms, 2.1 GB allocated, 910 KB copied,  46 MB peak memory,       same as baseline
    8192:  OK
      689  ms ±  18 ms, 5.2 GB allocated, 2.0 MB copied,  46 MB peak memory,       same as baseline
    16384: OK
      1.775 s ± 158 ms,  13 GB allocated, 4.7 MB copied,  46 MB peak memory,       same as baseline
  hashes
    2:     OK
      24.1 μs ± 2.1 μs, 172 KB allocated,  26 B  copied,  46 MB peak memory,       same as baseline
    4:     OK
      42.8 μs ± 3.7 μs, 297 KB allocated,  48 B  copied,  46 MB peak memory,       same as baseline
    8:     OK
      79.4 μs ± 3.6 μs, 547 KB allocated,  92 B  copied,  46 MB peak memory,       same as baseline
    16:    OK
      153  μs ±  13 μs, 1.0 MB allocated, 187 B  copied,  46 MB peak memory,       same as baseline
    32:    OK
      301  μs ±  28 μs, 2.0 MB allocated, 364 B  copied,  46 MB peak memory,       same as baseline
    64:    OK
      604  μs ±  41 μs, 4.0 MB allocated, 713 B  copied,  46 MB peak memory,       same as baseline
    128:   OK
      1.24 ms ± 107 μs, 7.9 MB allocated, 1.4 KB copied,  46 MB peak memory,       same as baseline
    256:   OK
      2.36 ms ± 112 μs,  16 MB allocated, 2.8 KB copied,  46 MB peak memory,       same as baseline
    512:   OK
      4.70 ms ± 253 μs,  31 MB allocated, 5.6 KB copied,  46 MB peak memory,       same as baseline
    1024:  OK
      9.34 ms ± 857 μs,  63 MB allocated,  11 KB copied,  46 MB peak memory,       same as baseline
    2048:  OK
      18.8 ms ± 1.7 ms, 125 MB allocated,  18 KB copied,  46 MB peak memory,       same as baseline
    4096:  OK
      37.4 ms ± 3.6 ms, 250 MB allocated, 229 KB copied,  51 MB peak memory,       same as baseline
    8192:  OK
      74.9 ms ± 6.6 ms, 501 MB allocated, 550 KB copied,  56 MB peak memory,       same as baseline
    16384: OK
      149  ms ± 8.7 ms, 1.0 GB allocated, 1.5 MB copied,  59 MB peak memory,       same as baseline
  hashmem
    2:     OK
      34.0 μs ± 2.2 μs, 222 KB allocated,  36 B  copied,  59 MB peak memory,       same as baseline
    4:     OK
      61.6 μs ± 3.7 μs, 392 KB allocated,  68 B  copied,  59 MB peak memory,       same as baseline
    8:     OK
      115  μs ± 7.0 μs, 733 KB allocated, 142 B  copied,  59 MB peak memory,       same as baseline
    16:    OK
      223  μs ±  15 μs, 1.4 MB allocated, 313 B  copied,  59 MB peak memory,       same as baseline
    32:    OK
      441  μs ±  32 μs, 2.7 MB allocated, 779 B  copied,  59 MB peak memory,       same as baseline
    64:    OK
      883  μs ±  53 μs, 5.4 MB allocated, 1.9 KB copied,  59 MB peak memory,       same as baseline
    128:   OK
      1.76 ms ± 154 μs,  11 MB allocated, 6.8 KB copied,  59 MB peak memory,       same as baseline
    256:   OK
      3.52 ms ± 227 μs,  21 MB allocated,  24 KB copied,  59 MB peak memory,       same as baseline
    512:   OK
      7.13 ms ± 502 μs,  43 MB allocated,  88 KB copied,  59 MB peak memory,       same as baseline
    1024:  OK
      14.8 ms ± 929 μs,  86 MB allocated, 369 KB copied,  59 MB peak memory,       same as baseline
    2048:  OK
      30.0 ms ± 2.1 ms, 172 MB allocated, 1.3 MB copied,  59 MB peak memory,       same as baseline
    4096:  OK
      61.0 ms ± 4.2 ms, 344 MB allocated, 3.5 MB copied,  59 MB peak memory,       same as baseline
    8192:  OK
      125  ms ± 8.5 ms, 689 MB allocated, 8.9 MB copied,  59 MB peak memory,       same as baseline
    16384: OK
      258  ms ±  18 ms, 1.3 GB allocated,  21 MB copied,  59 MB peak memory,       same as baseline
  balanceTransfer
    2:     OK
      21.0 μs ± 1.9 μs, 136 KB allocated,  31 B  copied,  59 MB peak memory,       same as baseline
    4:     OK
      32.9 μs ± 2.6 μs, 216 KB allocated,  69 B  copied,  59 MB peak memory,       same as baseline
    8:     OK
      56.6 μs ± 4.0 μs, 378 KB allocated, 204 B  copied,  59 MB peak memory,       same as baseline
    16:    OK
      104  μs ± 7.3 μs, 700 KB allocated, 662 B  copied,  59 MB peak memory,       same as baseline
    32:    OK
      200  μs ±  15 μs, 1.3 MB allocated, 2.3 KB copied,  59 MB peak memory,       same as baseline
    64:    OK
      390  μs ±  27 μs, 2.6 MB allocated, 8.7 KB copied,  59 MB peak memory,       same as baseline
    128:   OK
      780  μs ±  59 μs, 5.1 MB allocated,  32 KB copied,  59 MB peak memory,       same as baseline
    256:   OK
      1.58 ms ± 125 μs,  10 MB allocated, 126 KB copied,  59 MB peak memory,       same as baseline
    512:   OK
      3.25 ms ± 215 μs,  20 MB allocated, 517 KB copied,  59 MB peak memory,       same as baseline
    1024:  OK
      6.91 ms ± 543 μs,  40 MB allocated, 2.1 MB copied,  59 MB peak memory,       same as baseline
    2048:  OK
      16.3 ms ± 948 μs,  81 MB allocated, 9.1 MB copied,  60 MB peak memory,       same as baseline
    4096:  OK
      36.0 ms ± 3.5 ms, 161 MB allocated,  25 MB copied,  79 MB peak memory,       same as baseline
    8192:  OK
      76.8 ms ± 6.9 ms, 322 MB allocated,  61 MB copied, 106 MB peak memory,       same as baseline
    16384: OK
      156  ms ± 9.7 ms, 645 MB allocated, 131 MB copied, 184 MB peak memory,       same as baseline
  funcCall
    2:     OK
      20.9 μs ± 1.8 μs, 161 KB allocated,  24 B  copied, 184 MB peak memory,  9% less than baseline
    4:     OK
      33.9 μs ± 1.8 μs, 268 KB allocated,  43 B  copied, 184 MB peak memory,  7% less than baseline
    8:     OK
      60.6 μs ± 3.4 μs, 483 KB allocated,  80 B  copied, 184 MB peak memory,       same as baseline
    16:    OK
      114  μs ± 7.0 μs, 911 KB allocated, 155 B  copied, 184 MB peak memory,       same as baseline
    32:    OK
      218  μs ±  14 μs, 1.7 MB allocated, 303 B  copied, 184 MB peak memory,       same as baseline
    64:    OK
      427  μs ±  31 μs, 3.4 MB allocated, 600 B  copied, 184 MB peak memory,       same as baseline
    128:   OK
      847  μs ±  57 μs, 6.8 MB allocated, 1.2 KB copied, 184 MB peak memory,       same as baseline
    256:   OK
      1.70 ms ± 110 μs,  13 MB allocated, 2.3 KB copied, 184 MB peak memory,       same as baseline
    512:   OK
      3.37 ms ± 212 μs,  27 MB allocated, 4.6 KB copied, 184 MB peak memory,       same as baseline
    1024:  OK
      6.72 ms ± 514 μs,  54 MB allocated, 9.3 KB copied, 184 MB peak memory,       same as baseline
    2048:  OK
      13.3 ms ± 959 μs, 107 MB allocated,  15 KB copied, 184 MB peak memory,       same as baseline
    4096:  OK
      26.8 ms ± 2.2 ms, 214 MB allocated,  27 KB copied, 184 MB peak memory,       same as baseline
    8192:  OK
      52.7 ms ± 4.2 ms, 429 MB allocated,  49 KB copied, 184 MB peak memory,       same as baseline
    16384: OK
      105  ms ± 7.4 ms, 857 MB allocated,  94 KB copied, 184 MB peak memory,       same as baseline
  contractCreation
    2:     OK
      39.5 μs ± 3.4 μs, 233 KB allocated,  50 B  copied, 184 MB peak memory,  9% less than baseline
    4:     OK
      73.6 μs ± 6.8 μs, 421 KB allocated, 135 B  copied, 184 MB peak memory,  9% less than baseline
    8:     OK
      141  μs ±  14 μs, 796 KB allocated, 516 B  copied, 184 MB peak memory,       same as baseline
    16:    OK
      278  μs ±  26 μs, 1.5 MB allocated, 1.4 KB copied, 184 MB peak memory,       same as baseline
    32:    OK
      552  μs ±  29 μs, 3.0 MB allocated, 5.3 KB copied, 184 MB peak memory,       same as baseline
    64:    OK
      1.11 ms ±  55 μs, 6.0 MB allocated,  21 KB copied, 184 MB peak memory,       same as baseline
    128:   OK
      2.23 ms ± 118 μs,  12 MB allocated,  86 KB copied, 184 MB peak memory,       same as baseline
    256:   OK
      4.62 ms ± 430 μs,  24 MB allocated, 360 KB copied, 184 MB peak memory,       same as baseline
    512:   OK
      10.1 ms ± 855 μs,  48 MB allocated, 1.7 MB copied, 184 MB peak memory,       same as baseline
    1024:  OK
      24.0 ms ± 2.0 ms,  96 MB allocated, 8.1 MB copied, 184 MB peak memory,       same as baseline
    2048:  OK
      50.1 ms ± 4.2 ms, 194 MB allocated,  20 MB copied, 184 MB peak memory,       same as baseline
    4096:  OK
      111  ms ± 7.1 ms, 389 MB allocated,  51 MB copied, 191 MB peak memory,       same as baseline
    8192:  OK
      229  ms ±  12 ms, 782 MB allocated, 112 MB copied, 379 MB peak memory,       same as baseline
    16384: OK
      493  ms ±  46 ms, 1.5 GB allocated, 264 MB copied, 506 MB peak memory,       same as baseline
  contractCreationMem
    2:     OK
      106  μs ± 7.2 μs, 590 KB allocated, 158 B  copied, 506 MB peak memory, 26% less than baseline
    4:     OK
      207  μs ±  15 μs, 1.1 MB allocated, 454 B  copied, 506 MB peak memory, 21% less than baseline
    8:     OK
      410  μs ±  27 μs, 2.2 MB allocated, 1.5 KB copied, 506 MB peak memory, 19% less than baseline
    16:    OK
      823  μs ±  53 μs, 4.3 MB allocated, 5.3 KB copied, 506 MB peak memory, 17% less than baseline
    32:    OK
      1.65 ms ± 110 μs, 8.5 MB allocated,  21 KB copied, 506 MB peak memory, 17% less than baseline
    64:    OK
      3.35 ms ± 258 μs,  17 MB allocated,  86 KB copied, 506 MB peak memory, 15% less than baseline
    128:   OK
      6.88 ms ± 372 μs,  34 MB allocated, 360 KB copied, 506 MB peak memory, 16% less than baseline
    256:   OK
      15.2 ms ± 1.2 ms,  68 MB allocated, 2.5 MB copied, 506 MB peak memory, 14% less than baseline
    512:   OK
      31.8 ms ± 2.0 ms, 136 MB allocated, 6.7 MB copied, 506 MB peak memory, 11% less than baseline
    1024:  OK
      67.3 ms ± 3.4 ms, 274 MB allocated,  17 MB copied, 506 MB peak memory, 14% less than baseline
    2048:  OK
      138  ms ± 8.3 ms, 549 MB allocated,  38 MB copied, 506 MB peak memory, 13% less than baseline
    4096:  OK
      283  ms ± 7.2 ms, 1.1 GB allocated,  82 MB copied, 506 MB peak memory, 13% less than baseline
    8192:  OK
      596  ms ±  59 ms, 2.2 GB allocated, 194 MB copied, 571 MB peak memory, 13% less than baseline
    16384: OK
      1.221 s ±  96 ms, 4.3 GB allocated, 403 MB copied, 1.1 GB peak memory, 10% less than baseline
  arrayCreationMem
    2:     OK
      77.6 μs ± 7.6 μs, 525 KB allocated,  80 B  copied, 1.1 GB peak memory, 10% less than baseline
    4:     OK
      262  μs ±  15 μs, 1.7 MB allocated, 270 B  copied, 1.1 GB peak memory,  7% less than baseline
    8:     OK
      977  μs ±  56 μs, 6.4 MB allocated, 1.0 KB copied, 1.1 GB peak memory,       same as baseline
    16:    OK
      3.74 ms ± 213 μs,  25 MB allocated, 3.9 KB copied, 1.1 GB peak memory,       same as baseline
    32:    OK
      14.7 ms ± 840 μs,  97 MB allocated,  82 KB copied, 1.1 GB peak memory,       same as baseline
    64:    OK
      58.9 ms ± 3.5 ms, 384 MB allocated, 1.5 MB copied, 1.1 GB peak memory,       same as baseline
    128:   OK
      232  ms ±  14 ms, 1.5 GB allocated, 2.9 MB copied, 1.1 GB peak memory,       same as baseline
    256:   OK
      931  ms ±  27 ms, 6.0 GB allocated, 7.2 MB copied, 1.1 GB peak memory,       same as baseline
    512:   OK
      3.684 s ±  53 ms,  24 GB allocated,  11 MB copied, 1.1 GB peak memory,       same as baseline
  mapStorage
    2:     OK
      38.5 μs ± 3.4 μs, 254 KB allocated,  44 B  copied, 1.1 GB peak memory,       same as baseline
    4:     OK
      70.7 μs ± 7.0 μs, 462 KB allocated,  77 B  copied, 1.1 GB peak memory,       same as baseline
    8:     OK
      137  μs ± 7.2 μs, 880 KB allocated, 171 B  copied, 1.1 GB peak memory,       same as baseline
    16:    OK
      269  μs ±  16 μs, 1.7 MB allocated, 430 B  copied, 1.1 GB peak memory,       same as baseline
    32:    OK
      536  μs ±  53 μs, 3.3 MB allocated, 1.3 KB copied, 1.1 GB peak memory,       same as baseline
    64:    OK
      1.06 ms ±  55 μs, 6.6 MB allocated, 4.0 KB copied, 1.1 GB peak memory,       same as baseline
    128:   OK
      2.13 ms ± 212 μs,  13 MB allocated,  14 KB copied, 1.1 GB peak memory,       same as baseline
    256:   OK
      4.32 ms ± 340 μs,  26 MB allocated,  53 KB copied, 1.1 GB peak memory,       same as baseline
    512:   OK
      8.68 ms ± 455 μs,  53 MB allocated, 221 KB copied, 1.1 GB peak memory,       same as baseline
    1024:  OK
      17.8 ms ± 841 μs, 106 MB allocated, 873 KB copied, 1.1 GB peak memory,       same as baseline
    2048:  OK
      36.3 ms ± 3.4 ms, 212 MB allocated, 2.5 MB copied, 1.1 GB peak memory,       same as baseline
    4096:  OK
      77.9 ms ± 6.3 ms, 424 MB allocated, 6.3 MB copied, 1.1 GB peak memory,       same as baseline
    8192:  OK
      154  ms ±  12 ms, 850 MB allocated,  16 MB copied, 1.1 GB peak memory,       same as baseline
    16384: OK
      312  ms ±  15 ms, 1.7 GB allocated,  35 MB copied, 1.1 GB peak memory,       same as baseline
  swapOperations
    2:     OK
      86.5 μs ± 6.7 μs, 652 KB allocated, 106 B  copied, 1.1 GB peak memory, 12% less than baseline
    4:     OK
      108  μs ± 8.7 μs, 819 KB allocated, 136 B  copied, 1.1 GB peak memory,  9% less than baseline
    8:     OK
      149  μs ±  13 μs, 1.1 MB allocated, 157 B  copied, 1.1 GB peak memory,       same as baseline
    16:    OK
      227  μs ±  18 μs, 1.8 MB allocated, 312 B  copied, 1.1 GB peak memory,       same as baseline
    32:    OK
      387  μs ±  30 μs, 3.1 MB allocated, 560 B  copied, 1.1 GB peak memory,       same as baseline
    64:    OK
      708  μs ±  64 μs, 5.7 MB allocated, 1.0 KB copied, 1.1 GB peak memory,       same as baseline
    128:   OK
      1.33 ms ± 120 μs,  11 MB allocated, 1.9 KB copied, 1.1 GB peak memory,       same as baseline
    256:   OK
      2.55 ms ± 139 μs,  21 MB allocated, 3.7 KB copied, 1.1 GB peak memory,       same as baseline
    512:   OK
      5.02 ms ± 437 μs,  42 MB allocated, 7.4 KB copied, 1.1 GB peak memory,       same as baseline
    1024:  OK
      10.1 ms ± 466 μs,  84 MB allocated,  23 KB copied, 1.1 GB peak memory,       same as baseline
    2048:  OK
      20.0 ms ± 1.3 ms, 168 MB allocated,  20 KB copied, 1.1 GB peak memory,       same as baseline
    4096:  OK
      39.2 ms ± 3.6 ms, 335 MB allocated,  35 KB copied, 1.1 GB peak memory,       same as baseline
    8192:  OK
      78.8 ms ± 6.9 ms, 669 MB allocated,  64 KB copied, 1.1 GB peak memory,       same as baseline
    16384: OK
      157  ms ±  14 ms, 1.3 GB allocated, 121 KB copied, 1.1 GB peak memory,       same as baseline

All 149 tests passed (292.86s)

@elopez
elopez force-pushed the perf-contract-mk branch 2 times, most recently from a799d41 to 8f0f99a Compare November 14, 2026 03:04
Replace the accumulated monadic action chain (m >> write >> write >> ...)
with direct ST writes during ByteString iteration. The old approach
created O(n) closures for the >> chain; the new approach writes directly
to the mutable vector with zero closure allocation.
@elopez
elopez marked this pull request as ready for review November 14, 2026 03:46
@elopez

elopez commented Nov 14, 2026

Copy link
Copy Markdown
Collaborator Author

(first commit is #1261 )

@blishko

blishko commented Nov 16, 2026

Copy link
Copy Markdown
Collaborator

#1261 is now merged, this can be rebased.

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants