Optimize GetUncappedShare with integer big.Int math (+ tests/benchmarks)

ZlyDevMayarequested to merge
zoly/get-uncapped-share-v2-bigint into develop

Summary

This MR replaces GetUncappedShare’s decimal-based implementation with an integer-only big.Int implementation, while keeping the old path as GetUncappedShareV1 for comparison/testing.

Motivation

The previous implementation used LegacyDec conversions and multiple decimal operations, which:

  • added significant allocation/CPU overhead
  • introduced intermediate precision/rounding behavior differences

The new implementation computes share directly in integer arithmetic with final round-half-up behavior.

Changes

  • Added new fast path:
    • GetUncappedShare
    • integer formula: (allocation*part + floor(total/2)) / total
  • Kept prior behavior as:
    • GetUncappedShareV1(...)
  • Added/updated tests:
    • V1 vs V2 behavior checks
    • overflow-safe large number coverage
    • deterministic example case showing expected delta
  • Added benchmarks for small and large input profiles.
  • Updated affected mocknet test expectations where rounding output changed by small integer amounts.

Benchmark

Command: go test ./common -run '^$' -bench '^Benchmark_GetUncappedShare_' -benchmem

Environment: linux, amd64, AMD Ryzen 9 8945HS

Raw results

Benchmark ns/op B/op allocs/op
V1 Small 1450.0 1008 39
V2 Small 194.9 192 13
V1 Large 1669.0 1192 41
V2 Large 214.8 232 13

Ratio comparison (V1 vs V2)

Category Latency speedup Memory reduction Allocation reduction
Small 7.44x faster (-86.6% ns/op) 5.25x less (-81.0% B/op) 3.00x less (-66.7% allocs/op)
Large 7.77x faster (-87.1% ns/op) 5.14x less (-80.5% B/op) 3.15x less (-68.3% allocs/op)

Notes

Behavior is deterministic but may differ by small integer amounts in edge rounding cases versus V1 Tests were updated accordingly.

Merge request reports