GSQRT, GSIN, GCOS, GTAN: if result size is float (as opposed to double),
truncate the result to that length by writing it into memory and
getting it back again (duh!), since that's what gcc does.