mathnet-spatial
mathnet-spatial copied to clipboard
Consider to use Microsoft.Bcl.Simd internally
See:
- Microsoft.Bcl.Simd NuGet package
- RyuJIT: The next-generation JIT compiler for .NET
- SIMD Sample
- SIMD - Wikipedia
Maybe it makes sense to use their fixed-sized vectors, maybe not.
(low priority)
They are Win 8.1 right? If so are they worth it?
Yes, and it requires installing previews and enabling them by registry settings. I expect the SIMD vectors work on the current platform as well, but of course not vectorized and without leveraging the SIMD instruction sets - but I have not actually verified that assumption.
It may make a big difference in the future performance wise. Maybe we could do some experiments in a branch with some benchmarking. Only a significant speed-up would justify using them though, in my opinion. But as long as we do not make design changes that would make it very difficult to use them efficiently in the future we can also delay that until the library has stabilized.
Just to clarify, I'm happy with the current very direct struct + public readonly fields approach. I do not suggest to change them.
Ok, a branch sounds good, would be interesting to play around with it if nothing else.
Not relevant anymore. Microsoft.Bcl.Simd does not exist anymore, and the latest update was on 11.04.2014