Benchmarks¶

Methodology¶

All benchmarks are compiled with -O3 -DNDEBUG using GCC 12.2 on an Intel i7-12700K (Alder Lake). Each measurement is the median of 1000 iterations after cache warmup.

Vector Operations¶

push_back (1M integers)¶

Implementation	Time (ms)	Ops/sec	Peak Memory
`std::vector<int>`	4.2	238M	8.0 MB
`prototype::vector<int>`	3.8	263M	8.0 MB
`prototype::vector<int, arena>`	2.1	476M	4.0 MB
`prototype::fixed_vector<int, 1M>`	1.9	526M	4.0 MB

Random Access (1M reads)¶

Implementation	Time (ns/op)	Cache Misses
`std::vector<int>`	0.8	0.1%
`prototype::vector<int>`	0.8	0.1%
Raw array	0.8	0.1%

Note

Random access performance is identical the abstraction has zero overhead.

Sort (1M integers, random)¶

Implementation	Time (ms)	Algorithm
`std::sort`	62	Introsort
`prototype::sort`	58	Introsort + insertion sort
`prototype::sort` (presorted)	4.1	Adaptive detection

Map/Dictionary Operations¶

Insert 100K key-value pairs¶

Implementation	Time (ms)	Memory	Cache Behavior
`std::map<int,int>`	45	4.8 MB	Poor (node-based)
`std::unordered_map<int,int>`	12	6.2 MB	Moderate
`prototype::flat_map<int,int>`	8.5	1.6 MB	Excellent (contiguous)
`prototype::static_map<int,int,100K>`	6.2	1.6 MB	Excellent (no alloc)

Lookup 100K keys (sorted)¶

Implementation	Time (ms)	Complexity
`std::map<int,int>`	18	O(log n), pointer-chase
`std::unordered_map<int,int>`	3.2	O(1) amortized
`prototype::flat_map<int,int>`	2.8	O(log n), binary search

Allocator Performance¶

Allocation/Deallocation Cycle (10K objects, 64 bytes each)¶

Allocator	Alloc (ns/op)	Dealloc (ns/op)	Fragmentation
`malloc/free`	85	62	High
`prototype::arena_allocator`	3	0 (bulk free)	None
`prototype::pool_allocator`	8	5	None
`prototype::stack_allocator`	2	1 (LIFO only)	None

Memory Arena Reset (4KB arena, 1000 allocations)¶

Operation	Time
Individual free (malloc)	62,000 ns
Arena reset (single op)	1 ns

String Operations¶

Short String (SSO, ≤22 chars)¶

Implementation	Construction (ns)	Copy (ns)	Heap Allocs
`std::string`	12	8	0 (SSO)
`prototype::string`	10	7	0 (SBO)
`prototype::fixed_string<32>`	4	4	0 (always)

Long String (1KB)¶

Implementation	Construction (ns)	Copy (ns)	Heap Allocs
`std::string`	45	42	1
`prototype::string` (default)	43	40	1
`prototype::string` (arena)	12	10	0

Concurrency¶

SPSC Queue Throughput (single producer, single consumer)¶

Implementation	Throughput (M ops/sec)	Latency p99 (ns)
`std::queue` + mutex	8	2400
Lock-free queue (Boost)	45	180
`prototype::spsc_queue`	62	85

Binary Size¶

Minimal vector program (`vector<int>`, push_back, iterate)¶

Implementation	Binary Size (stripped)
`std::vector`	52 KB
`prototype::vector`	8.4 KB
`prototype::fixed_vector`	6.1 KB

How to Run¶

make benchmarks
./.dist/bench_vector

Or with custom parameters:

make CXXSTD=c++17 benchmarks
./.dist/bench_vector --iterations=10000