enable ARM NEON (128 bit) vector registers via compiler defined macros

[roiser]

enable ARM NEON

[valassi]

Hi @roiser I added several comments about this in #221.

Bottom line:

• it presently does not support a way to disable SIMD (ie the 'none' option)
• it addresses the sam eissues as 128-bit vsx on Power9, but in a different way
• for these reasons I would modify this to use the same approach as on Power9 (ie add -D__SSE42__ if you want 128bit neon vectors, and add nothing if you want to keep no simd)
• later on this should probably be changed to use a define like -Dmg_prefer_vector_width128 instead of -D__sse42, and use that on all of intel, power9, arm

[valassi]

Hi @roiser I am closing this one and replacing it by PR #425. Please have a look (but still in WIP)... thanks!

[valassi]

I have now merged #425 which includes and replaces this #421.

[roiser]

• later on this should probably be changed to use a define like -Dmg_prefer_vector_width128 instead of -D__sse42, and use that on all of intel, power9, arm

On this point, yes I think we should go away from Intel specific names and use generic ones. All very fine with me, thanks