[fpc-pascal] fast integer multiplication

[Peter Vreman]

>>>Since you ask the question, I suppose you cannot simply
>>>suppress the "movl $0,%edx" line generation [*]. What about
>>>using a compiler directive, something like {$EXTENDEDMUL ON/OFF}
>>>for instance?
>>
>> I don't like this. This means we have to add directives for all kind of
>> special cpu features to improve some special corner cases.
>
> But it is a special case because FPC makes it special. As I wrote
> yesterday, when you write Word := Byte * Byte, what you get is
> the exact result (not the result mod 2^8), when you write
> Longword := Word * Word, what you get is the exact result (not the
> Result mod 2^16), but when you write QWord := Longword * Longword,
> there, what you get is not (always) the exact result but the result
> mod 2^32. This is FPC which makes special this case by setting the
> product high part to 0 (what it doesn't do with other cases).
>
> Being said that, I agree with you that adding compiler directives
> is not a good thing [*]. Here, it would be sufficient to suppress
> the generation of the line "movl $0,%edx". Since you do not do so,
> I suppose there would be some side effects I don't see at the
> moment, but it would be the simplest solution.
>
> [*] If QWord := Longword * Longword gave the exact result, it would
> allow me to suppress a lot of compiler directives :-)

The rules are the same as with int64:=longint*longint. Below is a test.
The result is -2 on both fpc and delphi.

var
  e : int64;
  i,j : longint;

begin
  i:=$7fffffff;
  j:=2;
  e:=i*j;
  writeln(e);
end.