On 22 November 2013 04:46, Mike Pall <mike-1311@xxxxxxxxxx> wrote: > Daurnimator wrote: > > This is what I ended up with: > https://gist.github.com/daurnimator/7582543 > > Hopefully I didn't overlook anything else... > > Wrong mantissa condition, use of uint32_t, endianess issues, > efficiency issues. > Thanks :) What is wrong with using unsigned here (just performance?) What was incorrect about my previous mantissa condition? local lohi_p = ffi.typeof("struct { int32_t "..(ffi.abi("le") and "lo, > hi" or "hi, lo").."; } *") > Nice trick for an endianess fix :) Will have to remember this. I've updated the gist with your code, thanks again for the supervision!