如何将REAL48 float转换为double

Question

I am connecting to a Pervasive SQL database which splits some data over two fields. DOUBLE fields are actually split into fieldName_1 and fieldName_2 where _1 is a 2 byte int and _2 is a 4 byte int.

I want to take these values and convert them using PHP into a usable value. I have some example code to do the conversion, but it is written in Delphi which I do not understand:

{ Reconstitutes a SmallInt and LongInt that form }
{ a Real into a double.                          }
Function EntConvertInts (Const Int2 : SmallInt;
                         Const Int4 : LongInt) : Double; StdCall;
Var
  TheRealArray : Array [1..6] Of Char;
  TheReal      : Real;
Begin
  Move (Int2, TheRealArray[1], 2);
  Move (Int4, TheRealArray[3], 4);
  Move (TheRealArray[1], TheReal, 6);

  Result := TheReal;
End;

Some data [fieldName_1,fieldName_2]

[132, 805306368] -> this should be 11

[132, 1073741824] -> this should be 12

I don't understand the logic enough to be able to port this into PHP. Any help would be most appreciated. Thanks

EDIT. This is the C code that they provided, showing sign/exponent:

double real_to_double (real r)
/* takes Pascal real, return C double */
{
    union doublearray da;
    unsigned x;

    x = r[0] & 0x00FF;  /* Real biased exponent in x */
    /* when exponent is 0, value is 0.0 */
    if (x == 0)
        da.d = 0.0;
    else {
        da.a[3] = ((x + 894) << 4) |  /* adjust exponent bias */
                  (r[2] & 0x8000) |  /* sign bit */
                  ((r[2] & 0x7800) >> 11);  /* begin significand */
        da.a[2] = (r[2] << 5) |  /* continue shifting significand */
                  (r[1] >> 11);
        da.a[1] = (r[1] << 5) |
                  (r[0] >> 11);
        da.a[0] = (r[0] & 0xFF00) << 5; /* mask real's exponent */
    }
    return da.d;
}

Answer 1

Adding this as another answer because I've finally figured this out. Here is PHP code which will convert the values. It has to be manually calculated because PHP does not know how to unpack a Real48 (non standard). Explanation in comments below.

function BiIntToReal48($f1, $f2){
  $x = str_pad(decbin($f1), 16, "0", STR_PAD_LEFT);
  $y = str_pad(decbin($f2), 32, "0", STR_PAD_LEFT);
  //full Real48 binary string
  $real48 = $y . $x;

  //Real48 format is V = (-1)^s * 1.f * 2^(exp-129)
  // rightmost eight bits are the exponent  (bits 40-->47)
  // subtract 129 to get the final value
  $exp = (bindec(substr($real48, -8)) - 129);

  //Sign bit is leftmost bit (bit[0])
  $sign =$real48[0];

  //Now work through the significand - bits are fractional binary 
  //(1/2s place, 1/4s place, 1/8ths place, etc)
  // bits 1-->39 
  // significand is always 1.fffffffff... etc so start with 1.0
  $sgf = "1.0";

  for ($i = 1; $i <= 39; $i++){
      if ($real48[$i] == 1){
        $sgf = $sgf + pow(2,-$i); 
      }       
  } 
  //final calculation
  $final = pow(-1, $sign) * $sgf * pow(2,$exp);
  return($final);
}
$field_1 = 132;
$field_2 = 805306368;      
$ConvVal = BiIntToReal48($field_1, $field_2);
// ^ gives $ConvVal = 11, qed