I'm dealing with a large number of data coming from an external device. They are big-endian 16 bits signed numbers, which must be converted to integer and to real to perform calculations.

Besides the obvious solutions of treating them as isolated bytes, multipying by 256 the highest, adding the lower, and oring wit $FFFF0000, or whatever sizeof(Integer) suggests if the result is bigger than 32767, is there a more efficient and elegant way to achieve the result?

I'm a bit lost on all available data types, and unable to tell apart the ones which are platform specific and the ones which are not, which assignments give the proper result and which do not.

Any suggestion would be greatly appreciated.

Giuliano Colla

Just use the BEtoN function. That will convert whatever big-endian number to your native endian.

I don't know what speed it has, but it sure is elegant :)

Giuliano Colla wrote

I'm dealing with a large number of data coming from an external device. They are big-endian 16 bits signed numbers, which must be converted to integer and to real to perform calculations. Besides the obvious solutions of treating them as isolated bytes, multipying by 256 the highest, adding the lower, and oring wit $FFFF0000, or whatever sizeof(Integer) suggests if the result is bigger than 32767, is there a more efficient and elegant way to achieve the result?

Call BEtoN() on them (Big Endian to Native byte order). Note that this routine is overloaded for all integer sizes > 1, and that FPC extends the size of any number < 32/64 bits (depending on the architecture) to 32/64 bits when performing almost any operation on it. It's therefore probably the safest to call it as BEtoN(smallint(value)) to ensure that the correct overload is selected.

Jonas

Il 16/04/2012 00:42, Jonas Maebe ha scritto:

I was sure that hidden somewhere there was a better way! Thanks a lot

Giuliano

There exists the BEtoN and LEtoN functions you can use. (NtoLE and NtoBE also)

var
  Data: SmallInt; // 16 bit signed type
...
Data := GetBENumberFromWherever;
Data := BEtoN(Data);

I'm not sure about real types. If they are 32 bits and sent in two pieces then you could just do

var
  Data: Single; // 4 bytes
  DataArray: array[0..1] of Word absolute Data;
...
// I'm not sure how the data is represented in memory for floats.
  DataArray[1] := BEtoN(Word(GetBENumberFromWherever));
  DataArray[0] := BEtoN(Word(GetBENumberFromWherever));

Anyway you get the idea. Probably you will have to experiment to achieve exactly what you want. Someone else probably has a better answer.

Regards,

Andrew

I asked this a few weeks ago in the context of doing endianness-changes in-place in a data structure, and Martin (I think) suggested overloading the := operator.

I did however find that one has to be extremely careful inside the custom operator, I ended up using Move() rather than an assignment to the result since it was tending to go recursive- some way of inhibiting all automatic type conversions would be useful.

(* This type represents a little-endian word, irrespective of the
platform. It  *)
(* is of known size (16 bits, tested by assertion) but is not directly
         *)
(* compatible with any other type.
         *)
//
type    WordLE= packed record b0, b1: byte; end;
(* Assign a little-endian word to an (unsigned) integer, with
appropriate       *)
(* conversion.
         *)
//
operator := (wle: WordLE): word;
begin
..
Type TAWSHeader=packed Record
        ThisSize: WordLE;
        PrevSize: WordLE;
        Flags:    Byte;
        Rsvd:     Byte;
      End;

Il 16/04/2012 01:44, Andrew Haines ha scritto:

There exists the BEtoN and LEtoN functions you can use. (NtoLE and NtoBE also)

Thanks to everybody.
Now I believe to have enough information to proceed. I'll keep this little piece of discussion as an example to show when someone asks me why I prefer open source software to proprietary one!

Giuliano