fix: Instrument the KeepRunningBatch() loops

[chfast]

Fixes #54.

The start_benchmark()/end_benchmark() hooks were attached to the StateIterator paths only, so the benchmarks looping with while (state.KeepRunningBatch(n)) were executed but never measured. They are moved to State::StartKeepRunning()/State::FinishKeepRunning(), which both the range-based and the KeepRunning*() loops go through, so both loop styles are now instrumented. The call order is preserved: start_benchmark() runs before the rest of StartKeepRunning() and measurement_start() after it, measurement_stop() runs before the rest of FinishKeepRunning() and end_benchmark() after it — the range-based loop therefore behaves exactly as before.

Verified with evmone, whose 8 KeepRunningBatch() benchmarks are reported for the first time with this change and whose range-based ones are unaffected:

Measured: ::{anonymous}::bench_ecrecovery::precompile[PrecompileId::ecrecover, evmone]
Measured: ::{anonymous}::bench_ecpairing::precompile[PrecompileId::ecpairing, evmmax_cpp]
Measured: ::{anonymous}::bench_kzg::precompile[PrecompileId::point_evaluation, evmone_blst]
…

[greptile-apps]

Greptile Summary

The PR relocates CodSpeed analysis hooks into Google Benchmark’s common State lifecycle so KeepRunning() and KeepRunningBatch() loops are measured without changing the range-based loop contract.

• Moves benchmark-start and measurement-start hooks into State::StartKeepRunning().
• Moves measurement-stop and benchmark-end hooks into State::FinishKeepRunning().
• Removes the equivalent range-iterator-specific hooks.

Confidence Score: 5/5

The PR appears safe to merge, with supported benchmark loop styles now sharing a paired instrumentation lifecycle.

The relocated hooks retain their previous ordering for range-based loops and add the same start/stop coverage to KeepRunning and KeepRunningBatch without changing public interfaces or overlapping the separate wall-time build path.

Important Files Changed

Filename	Overview
google_benchmark/include/benchmark/benchmark.h	Removes range-iterator-specific instrumentation while retaining calls into the shared State lifecycle used by all supported loop styles.
google_benchmark/src/benchmark.cc	Adds paired CodSpeed analysis hooks to StartKeepRunning and FinishKeepRunning with the range-loop ordering preserved.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Benchmark loop begins] --> B{Loop style}
  B -->|Range-based loop| C[State::end]
  B -->|KeepRunning or KeepRunningBatch| D[KeepRunningInternal]
  C --> E[StartKeepRunning]
  D --> E
  E --> F[Start CodSpeed benchmark and measurement]
  F --> G[Execute benchmark iterations]
  G --> H[FinishKeepRunning]
  H --> I[Stop measurement and end CodSpeed benchmark]

Loading

_{Reviews (1): Last reviewed commit: "fix: Instrument the KeepRunningBatch() l..." | Re-trigger Greptile}