Conversation

Notices

1. Embed this notice
   adb (adb@infosec.exchange)'s status on Tuesday, 10-Jun-2025 09:13:16 JST adb

   in reply to

   • Tony Finch

   @fanf Still believe that baseline and Vaseline SHOULD rhyme.

   • feld likes this.
   • Embed this notice
     Tony Finch (fanf@mendeddrum.org)'s status on Tuesday, 10-Jun-2025 09:13:17 JST Tony Finch

     in reply to

     RIGHT

     i added a baseline to my benchmark which revealed my amd64 numbers were nonsense, so i revamped it and now the benchmark results are more plausible

     but the conclusion of the blog post is rather different!

     newly rewritten article now up at the same url https://dotat.at/@/2025-06-08-floats.html

   • Embed this notice
     Tony Finch (fanf@mendeddrum.org)'s status on Tuesday, 10-Jun-2025 09:13:18 JST Tony Finch

     in reply to

     aha!

     i think i have succeeded by passing arguments and return values via pointers and putting an MFENCE just before and just after the code under test

   • Embed this notice
     Tony Finch (fanf@mendeddrum.org)'s status on Tuesday, 10-Jun-2025 09:13:19 JST Tony Finch

     in reply to

     hmm no, `cpuid` is not a good replacement for `isb sy`

   • Embed this notice
     Tony Finch (fanf@mendeddrum.org)'s status on Tuesday, 10-Jun-2025 09:13:20 JST Tony Finch

     in reply to

     also bloody amd64 is running my baseline slower than my functions that do work

     how am i supposed to measure these things i ask you

   • Embed this notice
     Tony Finch (fanf@mendeddrum.org)'s status on Tuesday, 10-Jun-2025 09:13:21 JST Tony Finch

     in reply to

     oh good grief i have discovered that a better compiler significantly changes the results on arm64

   • Embed this notice
     Tony Finch (fanf@mendeddrum.org)'s status on Tuesday, 10-Jun-2025 09:13:22 JST Tony Finch

     on my blog!

     performance of random floats

     in which i examine the relative cost of bithacking floating point numbers vs more conventional conversions

     https://dotat.at/@/2025-06-08-floats.html