Add SSE2NEON for ARM (Nvidia Tegra, PI, etc) support

[israelshirk]

What

Adds support for ARM processors. In my case, this means a quadcopter self-navigating for SAR. There are a number of possible uses, but in general PI and Nvidia Tegra's seem to run the small robotics world at the moment.

How

Including SSE2NEON.h only when SSE is not present keeps things nicely segregated - SSE2NEON takes care of all the translations, and any required compatibility fixes would belong to the upstream. There are other options such as simde; however they didn't work on the very first try :)

Related issues

#57 and #31

Changes

I've tried to keep this as loosely-coupled as possible. Ping me if you'd like any changes, such as just including SSE2NEON.h in the utils folder instead of using the submodule.

Thanks! Israel

[JakobEngel]

thanks!
I don't have a way of testing it on ARM, however given the comments it seems to work well!
I changed the ifdef checks to
#if !defined(SSE3) && !defined(SSE2) && !defined(SSE1)
(I only have SSE2 on my workstation, i.e. SSE3 is not defined and compilation failed).

Best,
Jakob

[israelshirk]

Awesome, thanks! I'll run some tests tomorrow but I can't imagine anything would be broken - I was just using SSE3 based off of some intel docs.

[guohengkai]

@israelshirk Hi, it seems that in the newest NDK the codes fail to compile. The errors were:
AccumulatedTopHessian.s: Assembler messages:
AccumulatedTopHessian.s:28297: Error: r13 not allowed here -- `sub.w sp,r2,#80'
clang++: error: assembler command failed with exit code 1 (use -v to see invocation)

Do you have any idea?

[israelshirk]

@guohengkai Could you create a new issue for this? Thanks!

[guohengkai]

@israelshirk Done: #96