cpp-performance-lab

A C++ microbenchmark repository for cache behavior, memory access, synchronization, communication paths, language/runtime overhead, allocator tradeoffs, container lookup, and syscall or network boundary cost.

Purpose

• Build intuition for cache hierarchy and memory access patterns
• Compare concurrency, communication, language, container, and allocator tradeoffs with reproducible microbenchmarks
• Measure syscall, IPC, and local transport overhead with small focused benchmarks
• Produce evidence-based performance notes from stable runs

1) Prerequisites

• CMake >= 3.20
• A C++20 compiler (clang++ or g++)
• Git + internet access (for fetching google/benchmark)

2) Build

cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build -j

3) Run all benchmarks (recommended)

scripts/run_all.sh

4) Run a specific benchmark

Standard pattern:

./build/benchmark/<binary_name> --benchmark_min_time=0.3s

Examples:

./build/benchmark/bm_stride_access --benchmark_min_time=0.3s
./build/benchmark/bm_cache_levels --benchmark_min_time=0.3s

Queue tuned run:

./build/benchmark/bm_queue \
  --benchmark_filter='BM_Queue(MutexTransfer/batch:64/backoff:0|SpscRingTransfer/batch:8/backoff:0)$' \
  --benchmark_min_time=1s \
  --benchmark_repetitions=10 \
  --benchmark_report_aggregates_only=true

5) Benchmarks

• bm_stride_access: locality loss from larger access stride
• bm_pointer_chasing: sequential access vs irregular pointer traversal
• bm_false_sharing: adjacent counters vs cache-line-padded counters
• bm_aos_vs_soa: layout sensitivity for dense vs sparse field usage
• bm_mutex_vs_atomic: contention scaling for shared counter updates
• bm_cache_levels: throughput drop as working set crosses cache levels
• bm_ilp: dependent vs independent instruction streams
• bm_branch_prediction: predictable, alternating, random, and branchless control flow
• bm_inlining_effects: forced inline, forced noinline, and function-pointer call shape
• bm_cache_associativity: friendly stride vs conflict-prone stride
• bm_queue: mutex queue vs tuned SPSC ring transfer
• bm_mpsc_mpmc_queue: mutex queue vs bounded lock-free queue under MPSC and MPMC load
• bm_cv_vs_spin: condition variable, yield, and spinning handoff cost
• bm_lock_variants: mutex, spinlock, and ticket-lock contention scaling
• bm_queue_message_size: queue throughput across multiple payload sizes
• bm_memory_pool: new/delete vs locked pool vs thread-local pool
• bm_tlb_pressure: contiguous, page-stride, and randomized page walks
• bm_cross_thread_free: producer allocation with consumer-side free across general-purpose, pool, and PMR paths
• bm_allocator_variants: new/delete, malloc/free, pmr, and arena-style allocation
• bm_allocator_mixed_size: mixed-size allocation across general-purpose and PMR pool paths
• bm_vector_deque_list: sequence-container scan cost across vector, deque, and list
• bm_mmap_vs_read: sequential read, random pread, and mapped-file scan
• bm_clock_overhead: chrono, clock_gettime, and gettimeofday call cost
• bm_mmap_cow: private first-touch/rewrite and shared mapped-write behavior
• bm_page_fault_mlock: first-touch, prefaulted, and mlock-backed page access
• bm_memory_order: throughput and correctness litmus tests across relaxed, release/acquire, and seq_cst
• bm_thread_affinity: default vs shared/split placement-hint thread handoff with verification counters
• bm_pipe_vs_shm: pipe syscall handoff vs shared-memory mailbox handoff
• bm_socketpair_vs_pipe: Unix stream socketpair vs pipe message handoff
• bm_virtual_vs_template_dispatch: template, virtual, and function-pointer dispatch
• bm_std_function_vs_lambda: lambda, functor, function pointer, and std::function
• bm_exception_vs_error_code: exception path vs optional-style error signaling
• bm_variant_vs_virtual: std::variant visitation vs virtual hierarchy dispatch
• bm_dynamic_cast_vs_tag: RTTI-based dispatch vs enum-tag dispatch
• bm_aliasing_effects: potential aliasing vs restrict-style no-alias access
• bm_container_lookup: map, unordered_map, and sorted-vector lookup
• bm_socket_loopback: local TCP vs Unix stream loopback message transfer

6) Outputs

• results-summary.md: current run summary and conclusions

Generate figures from benchmark runs:

python3 scripts/generate_plots.py

7) Figures

Cache Levels

Cache Associativity

Queue and Memory Pool

8) Source Layout

• benchmark/cache/: cache, locality, and working-set behavior
• benchmark/layout/: data layout experiments
• benchmark/concurrency/: synchronization, queues, and thread placement
• benchmark/memory/: allocator and pooling behavior
• benchmark/cpu/: instruction-throughput experiments
• benchmark/containers/: container and lookup tradeoff benchmarks
• benchmark/language/: dispatch and callable abstraction benchmarks
• benchmark/syscalls/: file and syscall boundary measurements
• benchmark/ipc/: communication-path benchmarks
• benchmark/network/: local socket and transport-path benchmarks