Faust benchmark

[jpcima]

Below are some measures of output throughput with faust bench.cpp. (GCC)

• fast-math is able to get a performance increase most of times
• vec is almost always a massive decrease

Edit: the data is a measure of throughput, which means higher = better.

[jpcima]

And this is when using Clang as compiler.

Here it's some of the same conclusions except that vec is useful in some cases.

[sletz]

You should try the faustbench-llvm tool described here: https://github.com/grame-cncm/faust/tree/master-dev/tools/benchmark. See for instance this log: https://gist.github.com/sletz/1f2f07bde42dda5721c46a3ce4e9c578

[jpcima]

@sletz, very useful thanks I've had a few obstacles building it however, so I took a few notes.

---

• has macOS-only build flags

  "-dead_strip" in tools/benchmark/Makefile

• fails build on a case-sensitive FS

    /opt/faust/include/faust/dsp/dsp-bench.h:42:10: erreur fatale: faust/GUI/MapUI.h : Aucun fichier ou dossier de ce type
       42 | #include "faust/GUI/MapUI.h"

replace faust/GUI → faust/gui to resolve it

• builds against system headers instead of headers in-repository (fail with faust 2.20.2 on system)

[sletz]

Thanks, should be fixed in https://github.com/grame-cncm/faust/commit/ccedfc929e6d656b291bc29c8f306a5fdf995864.