neon: Fix errors related to floating point ops in NEON
Hello,
This MR fixes errors related to floating point operators in NEON (divf, sqrtf, addf, subf, etc.):
- "Out-of-shift" errors when compiling some ops
- divf and sqrtf only processing half of the input (two numbers out of four)
- Test suite not passing because of NEON-related imprecisions
This should fix #33 (closed) #20 (closed), and help with #2 (closed).
It has been tested on 32-bit and 64-bit.
Thank you!