Add benchmarks to evaluate overhead

[brawner]

Based on some feedback from #10 I wanted to measure overhead of different choices with this infrastructure. This adds malloc/realloc benchmarks to measure the overhead of tracking them. It also enables linters for this package.

Signed-off-by: Stephen Brawner brawner@gmail.com

[brawner]

Building with benchmarks, --packages-select performance_test_fixture

• Linux
• Linux-aarch64
• macOS
• Windows

[cottsay]

A couple of comments. Also, please re-run CI with AMENT_RUN_PERFORMANCE_TESTS=1 to see what the results are like.

[brawner]

Addressed the msbuild warnings, and running this time with the benchmarks actually on (hopefully).

• Linux
• Linux-aarch64
• macOS
• Windows

[brawner]

Benchmark results can be seen here:
https://ci.ros2.org/job/ci_linux/13029/testReport/(root)/projectroot/benchmark_malloc_realloc/

[cottsay]

Looks good once BENCHMARK_MAIN has been dropped.

[cottsay]

This is great. I'd like to note, however, that the LD_PRELOAD is still in effect even though the memory tools aren't active, so there is still additional overhead associated with memory operations. I'm showing 270/70/12 ns for testing with memory tools, with memory tools disabled, and without preloading (respectively).

You'd have to use a separate test and add it with ament_add_google_benchmark, but you could test that here if you'd like. It's up to you.

[brawner]

Added a new test for use without memory tools. I also added logic so that it would be skipped if the other benchmark is skipped.

• Linux
• Linux-aarch64
• macOS
• Windows

[brawner]

I think I fixed the warnings above

• Linux
• Linux-aarch64
• macOS
• Windows

[brawner]

Without memory tools. I'm impressed by the memory latency of this build on cloud VMs.

---------------------------------------------------------------------------------------------------------------------------
Benchmark                                                                                 Time             CPU   Iterations
---------------------------------------------------------------------------------------------------------------------------
BenchmarkFixture/benchmark_on_malloc/Malloc Size:1                                     9.53 ns         9.53 ns     73326267
BenchmarkFixture/benchmark_on_malloc/Malloc Size:16                                    10.2 ns         10.2 ns     69127812
BenchmarkFixture/benchmark_on_malloc/Malloc Size:256                                   10.2 ns         10.2 ns     69155989
BenchmarkFixture/benchmark_on_malloc/Malloc Size:4096                                  23.2 ns         23.2 ns     30191050
BenchmarkFixture/benchmark_on_malloc/Malloc Size:65536                                 27.4 ns         27.4 ns     25594158
BenchmarkFixture/benchmark_on_malloc/Malloc Size:1048576                               25.3 ns         25.3 ns     27656313
BenchmarkFixture/benchmark_on_malloc/Malloc Size:16777216                              25.3 ns         25.3 ns     27711868
BenchmarkFixture/benchmark_on_malloc/Malloc Size:268435456                             4804 ns         4804 ns       145695

[brawner]

Windows is picky about the shift conversion, rebuilding after latest commit

• 

[cottsay]

A couple of thoughts on the new stuff.

[cottsay]

Should probably get one last round of CI, but this looks good to me. Thanks for iterating.

[brawner]

Hopefully the last round of CI

• Linux
• Linux-aarch64
• macOS
• Windows

[brawner]

Warnings on aarch and windows are just the expected ones for benchmarks.