Conversation

Notices

1. Embed this notice
   pinskia (pinskia@hachyderm.io)'s status on Sunday, 10-Nov-2024 06:55:06 JST pinskia

   in reply to

   • moonchild
   • Marc B. Reynolds

   @mbr @moonchild Yes it was just a documentation thing. I had noticed it originally when I was looking through LLVM bug and came across https://github.com/llvm/llvm-project/issues/91674 .

   • Embed this notice
     Marc B. Reynolds (mbr@mastodon.gamedev.place)'s status on Sunday, 10-Nov-2024 06:55:12 JST Marc B. Reynolds

     in reply to

     • moonchild

     @moonchild Well it appears it was just a documentation thing only!

   • Embed this notice
     Marc B. Reynolds (mbr@mastodon.gamedev.place)'s status on Sunday, 10-Nov-2024 06:55:13 JST Marc B. Reynolds

     in reply to

     • moonchild

     @moonchild TBH: It's not clear to me. It may have but just not been documented?

   • Embed this notice
     moonchild (moonchild@social.applied-langua.ge)'s status on Sunday, 10-Nov-2024 06:55:14 JST moonchild

     in reply to

     • Marc B. Reynolds
     @mbr wait WHAT ik assoc barrier has been a thing for years did it not prevent fma contraction before😭
   • Embed this notice
     Marc B. Reynolds (mbr@mastodon.gamedev.place)'s status on Sunday, 10-Nov-2024 06:55:15 JST Marc B. Reynolds

     Nice. the next GCC:

     __builtin_assoc_barrier (a*b) + c

     will prevent generating an FMA.

     EDIT: This has been the case and it was just missing from the documentation.

     https://gcc.godbolt.org/z/Kje39YeY9