static-vector-benchmark

This repository contains the source code for a micro benchmark of static vector operations.

Building

Use the enclosed Makefile. A system installation of Google benchmark is assumed to be available.

Running

Run the benchmark executable. The output should look something like this:

2022-12-13T21:46:14-05:00
Running ./benchmark
Run on (24 X 2660.09 MHz CPU s)
CPU Caches:
  L1 Data 32 KiB (x12)
  L1 Instruction 32 KiB (x12)
  L2 Unified 256 KiB (x12)
  L3 Unified 12288 KiB (x2)
Load Average: 1.16, 1.17, 1.12
----------------------------------------------------------------------------
Benchmark                  Time             CPU   Iterations UserCounters...
----------------------------------------------------------------------------
BM_vector/1000000     651827 ns       651795 ns          996 items_per_second=1.53423G/s

The benchmarks

These benchmarks measure the speed of proposed std::static_vector push_back variants:

• push_back_check If space is not available, throw std::bad_alloc, otherwise return a reference to the newly created element.
• try_push_back If space is not available, return 0, otherwise return a pointer to the newly created element.
• push_back_unsafe If space is not available the behavior is undefined, otherwise return a reference to the newly created element.

The "invisible" variants of the test remove the compiler's ability to determine that the vector does not change size in the inner loop. The intent is to eliminate the compiler's ability to push_back multiple values at the same time and better measure speeds.

Scripts

There are several scripts in the scripts directory that can be used to run the benchmark multiple times, collate the output into a more useful form, and create graphs.

MacOS Notes

Install dependencies

brew install google-benchmark
brew install pkg-config
brew install gnuplot
brew install jq

To build using the stock clang compiler, build like this:

make CXX="clang++ --std=c++20"