Hi,

I have a problem when trying to pass some variable parameters to a procedure writen using integrated assembler. I have reduced the problem to the following code:

program IncrementDemo;
{$R+}
{$asmmode intel}
var
   a: dword;

procedure increment(var x: dword); assembler;
asm
push ds { Is DS saved by default? }

lds ebx, x
mov eax, [ebx] { Line with the problem }
inc eax
mov [ebx], eax

      pop ds
   end;
begin
   write('Enter a number: ');
   readln(a);
   increment(a);
   writeln('Result is: ', a, '.')
end.

Line with the problem is marked above. If I tried using mov eax, [bx]
integrated environment would report that 16-bit references are not supported. With EBX instead of BX program reports run-time error 216.

I have tried using both DOS and Win32 targets. FPC 1.0.10, Win32.

Another Q: Is it necessary to save the segment registers' contents when entering an assembler procedure?

Thanks,
Ivan

FPC is 32 bit (flat memory modell) compiler so don't mess with segment registers!

So:

  procedure increment(var x: dword); assembler;
     asm
       mov ebx, x
       mov eax, [ebx]     { Line with the problem }
       inc eax
       mov [ebx], eax
     end;

The asm block can be this way:

procedure increment(var x: dword); assembler;
asm
   inc [eax]
end;

You don't need to worry about segment registers, but don't mess with them! I assume that the parameter x is passed on the eax register as Delphi does. Please correct me if I'm wrong...

You are wrong.

He is using 1.0.x, which doesn't have register parameters. That is 1.9.2 or 1.9.4+

Ok, then this can be this way:

procedure increment(var x: dword); assembler;
asm
mov eax, x
inc [eax]
end;

Is this correct? If the compiler uses registers, the first line is translated to a 'mov eax, eax', otherwise it is something like 'mov eax, [esp-4]'.

Regards,

Nelson

Afaik, it is correct, unless there is something with inc [eax] (e.g. not working on older ones)

On Fri, 6 Aug 2004 11:18:01 +0200 (CEST), Marco van de Voort wrote

Almost OK - you need to specify the operand size explicitely ("inc dword ptr [eax]" in Intel syntax, or "incl (%eax)" in AT&T syntax).

Tomas

On Thu, 05 Aug 2004 16:56:28 +0200, Florian Klaempfl wrote

In addition, please, note that ebx needs to be preserved in assembler procedures in later FPC versions (although 1.0.10 is OK because of a different default calling convention), so it would be better to either use indirect addressing as mentioned in another message in this thread, or use a different register (to avoid having to save ebx at the beginning and restore it at the end) in order to ensure compatibility of your code with later FPC versions.

Tomas