SP int inline asm:
- allow input variables to be either registers or memory for Intel
x86/x64 (minor performance improvement)
- don't have memory in clobber list if output variables are registers
- remove empty clobber line in arm32/thumb2 code for old versions of
gcc
_sp_mont_red():
- simplify the code by not using extra variables
- don't add to j in for loop check.