Hello, I came across this, it looks just great, I was going to do exactly this myself for a physics video game project ive been working on forever, and this might save me a lot of time. Please forgive the issue report, I couldn't see any email address for you and i have a couple of questions if you have a minute.

But before i try to work with it, i'm wondering if you got this code working, I see no sample code or exe, maybe if you could upload your vvvv project or any tests? Also wondering if you did any benchmarking, is it just 3x faster or the promised 8-16x faster.

very promising, love what you did with the masks to avoid branching and something special with number iterations in case it's not converging exactly what i had planned.

I'm presuming you made this in managed c++ instead of C# because the Vector class in c# Numerics has reported issues with alignment mismatching making it slow to copy data back and forth to GPU, is it true? I am disappointed in 2020 its so difficult to write this code and that it can't just use + operators but you've done it the best way from my research. I have looked at BEPU physics engine for rigid bodies which makes a huge use of c# vector but has so much unmanaged code , illegible loops, and tricks for intrinsics it might has well be all c++.

, I'm shocked that JIT compilation doesn't give the compiler more opportunities for SIMD optimizing, since it can know that the hardware is and can then supply the best instructions at startup on the client's phone or pc.

My other plan was to pursue a compute shader.. but this will leave my gpu alone and maybe be easier to work with.

I am getting a warning based on the processor. "There was a mismatch between the processor architecture of the project being built "MSIL" and the processor architecture of the reference "SlimDX", "AMD64". This mismatch may cause runtime failure"

If you lost the VVVV code but you say that it works and is fast, i will go make some tests in C# through the interface.

thx