memcopy-bench

A C/C++ research framework for evaluating data movement and frame-processing techniques in Telecom L2 scenarios. The initial focus is PDCP and GTP-U frame processing on x86, while reusable library code avoids unnecessary architecture-specific assumptions.

Layout

• lib/ — reusable frame, memory, transport, and scenario implementations. Libraries do not depend on benchmarking or reporting infrastructure. lib/test/ is reserved for shared test-only support; component unit tests remain colocated with their implementations.
• benchmarks/ — Google Benchmark executables. memory/ holds operation microbenchmarks; scenarios/ will hold complete workflow benchmarks.
• tests/integration/ — reserved for cross-component integration tests when they are needed. Unit tests are colocated with the library code they verify.
• tools/ — experiment automation and analysis utilities.
• results/ — raw benchmark output and experiment metadata (kept separate from analysis).
• reports/ — research findings and comparative analysis.

Prerequisites

Install Bazelisk. It selects the Bazel version pinned in .bazelversion. Commands are run from the repository root. Optimized mode is enabled by default in .bazelrc.

C++ editor tooling

Generate Bazel-aware compile commands for clangd and clang-tidy after changing C++ sources, targets, dependencies, or compile settings:

bazel run @hedron_compile_commands//:refresh_all

The generated root-level compile_commands.json is intentionally ignored by Git. Restart the editor language server after refreshing it if it does not reload the database automatically.

Build, lint, and test

bazel build //...
bazel test //...

Run the same C++ lint check used by CI (requires clang-tidy on PATH):

sh tools/run-clang-tidy.sh

Run tests with the supported sanitizers:

bazel test --config=asan //... --test_output=errors
bazel test --config=ubsan //... --test_output=errors

For example, the initial reusable algorithm and its colocated test are:

bazel build //lib/memory:memcpy
bazel test //lib/memory:memcpy_test

Run a benchmark

bazel run //benchmarks/memory:memcpy_benchmark

# Google Benchmark options follow `--`.
bazel run //benchmarks/memory:memcpy_benchmark -- \
  --benchmark_filter='BM_Memcpy/4096' \
  --benchmark_format=json \
  --benchmark_out=results/memcpy-4096.json

Store the matching experiment metadata (scenario, algorithm, input size, compiler configuration, CPU information, and source revision) alongside raw output in results/. Use reports/ for the derived analysis.

Development workflow

1. Add an algorithm or explicit frame-processing scenario under lib/.
2. Add colocated functional tests and run bazel test.
3. Add or extend the corresponding executable under benchmarks/.
4. Run controlled experiments and record raw measurements plus metadata in results/.
5. Analyze results in reports/.

A scenario describes what happens to a frame; algorithms describe how an individual operation is performed. Keep those concerns separate so algorithms can be compared across independent research cases.