Add initial support for half precision builtins

Hi,

I've tried to post this patch but it's too big for the ML (and I don't
know who to bug to get it released).

It adds basic fp16 builtins. The rest will probably need a fp32 wrapper
(especially the sw implementations of math functions), but that needs
fixed conversion routines.

This should be enough to make clpeak happy (BZ: 96897)

Jan

0001-Add-initial-support-for-half-precision-builtins.patch (40.7 KB)