bench

Benchmark reports

Committed performance measurements for pbr-cpp-memory-pool. One file per release × host combination, with the spelling v<X.Y.Z>-<os>-<compiler>-<arch>.md per ADR-0014 §7.

Index

Release	Host	Report
v0.2.0	Windows / MSVC / x64	v0.2.0-windows-msvc-x64.md

Contributors who run the benchmark on other hosts (Linux / GCC or Clang, macOS / Apple Clang) can append a row to this table and commit the corresponding report in the same PR. The methodology (scenarios, iterations, repeats, anti-optimization barriers, statistical summary) is binding across hosts; only the host disclosure block and the numbers themselves vary.

How to produce a new report

1. Build with the bench preset on a quiet host (no other interactive workloads):

   cmake --preset bench
   cmake --build --preset bench

2. Run the canonical benchmark and capture the output:

   ./build/bench/src/bench/cpp/it/d4np/memorypool/pool_vs_malloc_bench > raw.txt

3. Wrap raw.txt in a Markdown report matching the shape of the existing reports — Host / Run configuration / Raw output / Results / Observations / How to reproduce sections. The maintainer's v0.2.0-windows-msvc-x64.md is the reference template.

4. Add a row to the Index table above pointing at the new file. In the same PR, refresh the README Performance headline if the new report represents the canonical host for the current release.

Contract

The contract that binds every file in this directory — what to measure, how to time it, what to report — is fixed by ADR-0014. The operational quick-start lives in src/bench/cpp/it/d4np/memorypool/README.md.