SIMD

Jan 18, 2013 at 8:05 PM
Edited Jan 18, 2013 at 8:05 PM

What are your guy's thoughts on implementing SIMD for the math in ANX.  I've only done some basic research at this point.  It seems feasible and should provide at least a slight performance boost, but likely a whole lot more.  If you guys think this is a good idea, I could submit a patch with my changes.

Feb 3, 2013 at 9:08 AM
From my experience, it's easier for both sides to just submit what you have. Just submit the patch and the team can decide if and what they want to integrate.

I'm looking forward to see a SIMD math module in ANX.
Coordinator
Feb 4, 2013 at 10:40 AM
I'm very interested in SIMD math for ANX. I'm doesn't feel confident that we will gain much performance using SIMD but I've never benchmarked it, so this is only a feeling ;-)

I'm really interested in this patch to evaluate what you've done.
Feb 5, 2013 at 6:19 AM
Unfortunately, there isn't much for me to submit. I've looked into a couple of different ways to provide the SIMD support. One of the most effective ways to do this is through Mono's Mono.SIMD namespace. This works great under the mono framework, but just uses emulated SIMD under the .Net 4 framework. On a few of my the tests using basic Vector2s, it only supplies a marginal gain in performance. Beyond that I don't currently have much more to report. I plan to continue my research and I'll keep you posted.