Runtime Benchmarks

This page describes the results of benchmarking the performance of code generated by bon's builder macros.

Builder macros generate code that is easily optimizable by the compiler. This has been tested by the benchmarks below. The benchmarks compare regular positional function call syntax and builder syntax for functions annotated with #[builder].

In many cases rustc generates the same assembly code for the builder syntax as it would for a regular function call. Even when the generated assembly differs, the performance differences are negligible.

TIP

Don't take these microbenchmarks for granted. Do your own performance measurements in your application in real conditions. Feel free to open an issue if you find performance problems in bon.

Wallclock Statistics

Benchmark	Description	Assembly output	Run time
args_3	3 args of primitive types	Equal	regular: 6.2751ns builder: 6.3021ns
args_5	5 args of primitive types	Equal	regular: 7.8298ns builder: 7.8321ns
args_10	10 args of primitive types	Ordering diff	regular: 17.322ns builder: 17.178ns
args_10_structs	10 args of primitive types and structs	Instructions diff	regular: 2.7477ns builder: 2.7311ns
args_10_alloc	10 args of primitive and heap-allocated types	Instructions diff	regular: 91.666ns builder: 84.818ns (*)
args_20	20 args of primitive types	Equal	regular: 36.467ns builder: 36.786ns

(*)

Interestingly, in this case builder version performed even better. If you don't believe this, you can run these benchmarks for yourself. Maybe some ASM expert could explain this 😳?

High-Precision Statistics

Benchmark	Instructions count	L1 accesses	L2 accesses	RAM accesses
args_3	regular: 107 builder: 107	regular: 134 builder: 134	regular: 1 builder: 1	regular: 8 builder: 8
args_5	regular: 125 builder: 125	regular: 164 builder: 164	regular: 1 builder: 1	regular: 7 builder: 7
args_10	regular: 283 builder: 283	regular: 382 builder: 383	regular: 4 builder: 2	regular: 18 builder: 19
args_10_structs	regular: 22 builder: 22	regular: 30 builder: 31	regular: 2 builder: 1	regular: 5 builder: 5
args_10_alloc	regular: 2038 builder: 2037	regular: 2839 builder: 2837	regular: 1 builder: 1	regular: 33 builder: 34
args_20	regular: 557 builder: 557	regular: 775 builder: 775	regular: 1 builder: 1	regular: 32 builder: 32

Conditions

The code was compiled with opt-level = 3 and debug = 0.

Hardware

The benchmarks were run on a dedicated root server AX51-NVMe on Hetzner.

• OS: Ubuntu 22.04.4 (Linux 5.15.0-76-generic)
• CPU: AMD Ryzen 7 3700X 8-Core Processor (x86_64)
• RAM: 62.8 GiB

References

The source code of the benchmarks is available here.