Switched https://github.com/fmtlib/dtoa-benchmark to Google Benchmark for more reliable time measurement.

[vitaut@mastodon.social]

This is a benchmark of Theseus now: there is basically nothing left of the original Milo's benchmark.

[vitaut@mastodon.social]

I particularly like the macro-free method registration.

[pkhuong@discuss.systems]

@vitaut compile time is too weird, code that runs before main is much clearer!

[malwareminigun@infosec.exchange]

@vitaut GCC 15.2 on Ubuntu 26.04 on Docker on Windows on ARM

Snapdragon(R) X2 Elite Extreme - X2E94100 - Qualcomm Oryon(TM) CPU

[vitaut@mastodon.social]

@malwareminigun Nice, thanks for testing!

[malwareminigun@infosec.exchange]

@vitaut I tried to run it on Windows directly but too many *nix assumptions; I gave up after cache-contention.cc wanted <unistd.h>

[vitaut@mastodon.social]

@malwareminigun I'll disable it on Windows.

[malwareminigun@infosec.exchange]

@vitaut I mean, that's merely where I gave up, there were lots of other bits before that.

EDIT: But if you want to see our ostringstream be off at the freaking moon even on a log chart like this I'm sure it'll be funny

[vitaut@mastodon.social]

@malwareminigun PR is welcome

[malwareminigun@infosec.exchange]

@vitaut It looks like whatever picks the file name got confused, will be happy to send these somewhere if you want.

[malwareminigun@infosec.exchange]

@vitaut Here's the same hardware with Windows/MSVC directly:

[vitaut@mastodon.social]

@malwareminigun I just switched from CSV to Google Benchmark's native JSON

[malwareminigun@infosec.exchange]

@vitaut I'll get around to submitting https://github.com/BillyONeal/dtoa-benchmark/commit/be563ec2af8973840315cb644783db5144b8566b ... eventually. Want to see what it does on an x64 machine first but that machine is a bit busy ( https://infosec.exchange/@malwareminigun/116495687063912298 )