Next step I'm implementing matrix * matrix for large matrices with .NET 7 SIMD, starting to be 30-40% faster than Tensorflow, which is 😎 but I'm still 30-40% slower than MKL! 😅
Not that bad for a first try, but I will have to dig further if 1) I can optimize things further with some fancy AVX2 instructions, 2) If I can improve cache locality usage when going //

It’s been years in the making and we’re finally ready to show the world: don’t miss Yeoul’s EuroLLVM keynote, “-fbounds-safety: enforcing bounds safety for production C code”! https://discourse.llvm.org/t/the-2023-eurollvm-program/69514

Friday fun playing around with #fortnite #uefn running round a scan the Natural History Museum (inspired by tweet below). This platform is insanely good!

FUN FACT about Verse now that UEFN is out: the floating point is not quite IEEE-754 double, but it's close. Specifically the functional logic programming part of Verse (which is to say, its foundation) requires a few axioms that IEEE floats don't satisfy, so the float semantics are slightly tweaked.

Dependent Types mentioned in a GDC talk! Never thought I’d see the day 😀
[in the Q&A of the Verse talk]

#UnrealEngine references thread for Verse🧵