Add sm_pool utility and cache-locality batch benchmarks

[gabewillen]

Summary

• add boost::sml::utility::sm_pool for indexed event dispatch over pooled storage
• add batch APIs to sm_pool (process_event_batch / process_indexed_batch)
• add benchmark/connection/tensor_cache_locality.cpp and wire it into benchmark/connection/Makefile
• include no-data dispatch microbenchmarks for per-actor SML vs pooled SML vs direct baseline

Why

This isolates and addresses the 10k actor cache-locality issue by allowing a single SML router over contiguous storage and batched dispatch.

Local benchmark snapshot (median, CPU)

• direct actor array random: ~14.7us
• per-actor SML random: ~120.2us
• pooled SML random: ~40.7us
• pooled SML batch random: ~24.2us

Compared to per-actor SML, pooled batch dispatch cuts random no-data dispatch latency by roughly ~80% in this benchmark.

Validation

• built and ran: benchmark/connection/tensor_cache_locality.cpp with TEST_GBENCH
• command:
  • --benchmark_filter='BM_dispatch_(direct_actor_array_random|sml_actor_array_random|sml_pool_random|sml_pool_batch_random)'
  • --benchmark_min_time=0.2s --benchmark_repetitions=5 --benchmark_report_aggregates_only=true

[Copilot]

Pull request overview

This PR adds a new utility sm_pool to the Boost.SML library that enables pooled state machine storage with indexed event dispatch. The main goal is to address cache locality issues when managing many (10k+) state machine instances by allowing a single SML router over contiguous storage with batched dispatch capabilities.

Changes:

• Introduces boost::sml::utility::sm_pool class template for managing pooled state machines
• Adds comprehensive benchmarks comparing cache locality performance of different dispatch patterns
• Integrates the new benchmark into the build system

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
include/boost/sml/utility/sm_pool.hpp	New utility header providing sm_pool class template with indexed event dispatch and batch processing APIs
benchmark/connection/tensor_cache_locality.cpp	Comprehensive benchmark suite comparing per-actor SML, pooled SML, and direct dispatch patterns for cache locality analysis
benchmark/connection/Makefile	Adds tensor_cache_locality.cpp to the benchmark build targets

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The namespace closing order is incorrect. The BOOST_SML_NAMESPACE_END macro should come before the utility namespace closing brace, consistent with dispatch_table.hpp. Currently, the utility namespace is inside boost::ext::sml::v1_1_13, but it should be outside. This causes the utility to be placed in boost::ext::sml::v1_1_13::utility instead of boost::ext::sml::utility, breaking namespace consistency.

[Copilot]

The explicit keyword on a default constructor is unnecessary and unconventional. The explicit keyword is used to prevent implicit conversions for constructors with parameters. For a default constructor with no parameters, it has no practical effect and should be removed for clarity and consistency with other pool constructors in this file.

Suggested change

	explicit flat_tensor_pool() : slots(kTensorCount) {}
	flat_tensor_pool() : slots(kTensorCount) {}