Alexandre Mutel

Hey, finally! I just published a new - and real - blog post "10x Performance with SIMD Vectorized Code in C#/.NET" https://xoofx.com/blog/2023/07/09/10x-performance-with-simd-in-csharp-dotnet/ 🎉

That was a quick write up, so my apologize for the poor phrasing, after 3 years without writing a blog post, I feel rusty. But it feels good to share again! 🤗 #dotnet

10x Performance with SIMD Vectorized Code in C#/.NET | xoofx

Jul 9, 2023 at 8:18pmWeb
Show threadnietras 👾Jul 9, 2023
@xoofx great post, this is basically Sep 👍is there a reason for doing permute *before* pack unsigned saturate?
Show threadAlexandre MutelJul 10, 2023

@nietras oh, good catch! 🙂 No reasons, I think I missed the fact that I could use _mm256_permute4x64_epi64 after instead of performing the swap before. It helps saving 3 permutes in the end, not bad! Thanks for the suggestion.

I have updated the blog post and added a link to Sep at the end of the article: That's actually a good example of real world usage of intrinsics for performance benefits! 😉

Show threadnietras 👾Jul 10, 2023
@xoofx thanks ☺️
Show threadnietras 👾Jul 11, 2023
@xoofx of course the generic versions can be improved too given there is ExtractMostSignificantBits (generic move mask) and not sure pack saturated is needed, can just Narrow. So should actually be able to make this fully generic. I think 😅
Show threadAlexandre MutelJul 11, 2023
@nietras yeah, definitely could be, I took the original code without digging further, anyway, I'm back to my holidays, I won't check that until next week! 😎🏖️
Show threadtjheuvelJul 10, 2023
@xoofx Now with Burst Intrinsics!
Show threadAlexandre MutelJul 10, 2023
@tjheuvel naan, the truth is that it wouldn't make a dime change. Burst and its intrinsics wouldn't be able to optimize more than this. 😉
We are working on bringing support for System.Runtime.Intrinsics.x86 and ARM in the future that should hopefully obsolete Burst intrinsics.
Show threadtjheuvelJul 20, 2023

@xoofx Curiousity got the better of me, but i cant get it to work at all.

For some reason SseSupported returns false, in an IL2CPP build using the performance test framework. Burst AOT settings has everything on too...

Show threadNeil HenningJul 11, 2023
@xoofx great post Alex! Great to see you writing again man. 🤗
Show threadNeil HenningJul 11, 2023
@xoofx meant to say, it looks like your RSS feed is empty? I use NetNewsWire to subscribe to things, was surprised when I added your site there was no posts!
Show threadAlexandre MutelJul 11, 2023
@neilhenning oh, I will have to double check, it should generate something but I never really tried with a real RSS reader!
Show threadAlexandre MutelAug 24, 2023
@neilhenning Finally took the time to have a look: I had to fix my static web site generator that I probably broke in some past versions. Discovered also that my 404 redirect page was no longer generated correctly. Thanks for letting me know, hopefully should be good now! 🤞
Show threadNeil HenningAug 24, 2023
@xoofx awesome!