wrenfold-benchmarks

See a summary of results here.

This repository contains runtime benchmarks for wrenfold-generated functions. We compare to SymForce and some handwritten implementations. When considering these benchmarks, it is important to remember that all of the following factors can materially impact the results:

• Choice of compiler and compiler version.
• Selected CPU features (ie. AVX, SSE). For example, enabling -march=native can improve or pessimize a given implementation.
• CPU architecture (ARM vs Intel).

As with all performance assessment, your mileage may vary and there is no perfect substitute for profiling in the context of your deployment environment.

Notes

• The original python symbolic implementations can be found in the scripts directory.
• SymForce outputs were last updated using version 0.9.0.
• For SymForce generated methods, we compare to chain-rule versions of the generated functions. See sections B.1 and B.2 of the paper.

Building

Make sure you have cloned the requisite submodules:

git submodule update --init --recursive

To configure and run the benchmark suite:

mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release -Wno-deprecated
cmake --build .
./run_benchmarks

Updating the generated functions

Functions can be updated by running:

python scripts/gen_all.py

Both wrenfold and symforce must be installed in your python environment.