From: gah@ugcs.caltech.edu   
      
   On 11/23/23 7:28 AM, Steve Lionel wrote:   
   > On 11/23/2023 5:29 AM, Woozy Song wrote:   
   >> Where X and Y are both real, and W is double precision. I presume they   
   >> wanted extra precision for the arithmetic, but then it gets quantised   
   >> back to 32-bit float.   
   >> By the way, this code came from a large oil company with a team of   
   >> programmers.   
      
   > The multiplication by 1.0D0 of each term has no effect whatsoever. If   
   > those were removed, the multiplication of Y*W would convert Y to double,   
   > due to "mixed-mode arithmetic" rules, before doing the operation. Then   
   > when X is added, it too would be converted to double before the   
   > addition. Then the whole result would be rounded back to single   
   > precision for the assignment.   
      
   > If I were writing this, I'd use the standard conversion intrinsic   
   > functions to make it clear what is happening, as I hate seeing implicit   
   > conversion. For example:   
      
   > X = REAL(DBLE(X)+(DBLE(Y)*W))   
      
   In the Fortran 66 days, there was worry that constants would   
   be converted at run-time, if needed.   
      
   We were told to use 1.0 instead of 1 in REAL expressions.   
      
   As well as I know, compilers did compile time conversions long   
   before this recommendation went away.  Though often enough, I still   
   write 1.0, as it makes it more obvious that the expression is REAL.   
      
   As for conversions, Java requires a cast for narrowing conversions.   
      
   Widening conversions in the order:   
      
   byte --> {short, char} --> int --> long --> float --> double   
      
   don't require a cast, but the other way, narrowing, do.   
      
   Note that this is true, even though significant bits might be lost   
   in the int --> float and long --> double cases.   
      
   The unsigned 16 bit char, and signed 16 bit short, require a   
   cast in both directions.   
      
   As to the original question, often double precision is needed   
   in intermediate values, even when the final result is single precision.   
      
   But also, it is possible that the type declarations changed over the   
   years, but not the line in question.   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)