Conversation

Notices

1. Embed this notice
   John Regehr (regehr@mastodon.social)'s status on Wednesday, 02-Oct-2024 15:53:58 JST John Regehr

   I wasn't sure how much it would matter to vectorize scan loops such as [<] and [>>] in Brainfuck, but on my M1 Mac I'm seeing dbfi.b going about 10x faster due to just this one optimization

   http://www.hevanet.com/cristofd/08.html

   • Embed this notice
     John Regehr (regehr@mastodon.social)'s status on Wednesday, 02-Oct-2024 15:53:52 JST John Regehr

     in reply to

     • Owen Anderson

     @resistor one of the enterprising students in my class is attacking the problem that ~35% of LLVM unit tests trigger UB. 🤪 for real.

     Haelwenn /элвэн/ :triskell: likes this.
   • Embed this notice
     John Regehr (regehr@mastodon.social)'s status on Wednesday, 02-Oct-2024 15:53:54 JST John Regehr

     in reply to

     • Owen Anderson

     @resistor hoo boy. I want to say I'd use this in a performance comparison but LLVM 2.8 or whatev is pretty distant now

   • Embed this notice
     Owen Anderson (resistor@mastodon.online)'s status on Wednesday, 02-Oct-2024 15:53:54 JST Owen Anderson

     in reply to

     @regehr sounds like an extra credit opportunity for some enterprising undergrad 🤪

   • Embed this notice
     John Regehr (regehr@mastodon.social)'s status on Wednesday, 02-Oct-2024 15:53:55 JST John Regehr

     in reply to

     • Tommy Thorn

     @tommythorn ah, impossible to say without having worked on the other thing as well. what I can say about BF is that it's a bit a sweet spot for a compilers class because it has plenty of low-hanging optimizations, but they mostly require BF-specific knowledge, so for example LLVM alone would not be able to pick up most of these benefits.

   • Embed this notice
     John Regehr (regehr@mastodon.social)'s status on Wednesday, 02-Oct-2024 15:53:55 JST John Regehr

     in reply to

     • Tommy Thorn

     @tommythorn
     but so far in class we've only targeted native assembly. once we're done with this, we'll retarget everything to LLVM and pick up for free all of those amazing low-level optimizations that are too numerous for us to do by hand

   • Embed this notice
     Owen Anderson (resistor@mastodon.online)'s status on Wednesday, 02-Oct-2024 15:53:55 JST Owen Anderson

     in reply to

     @regehr I wrote a tracing JIT for Brainfuck using LLVM a really long time ago: https://github.com/resistor/BrainFTracing

   • Embed this notice
     Tommy Thorn (tommythorn@chaos.social)'s status on Wednesday, 02-Oct-2024 15:53:57 JST Tommy Thorn

     in reply to

     @regehr Serious question: how does the effort used to get to the current performance level for BF compare to what it would have taken for something less insane, say the LCC VM (which formed the bases for the Quake III VM).

   • Embed this notice
     John Regehr (regehr@mastodon.social)'s status on Wednesday, 02-Oct-2024 15:53:58 JST John Regehr

     in reply to

     it was not my intent for this to become an "optimizing Brainfuck" account. this will stop soon.