Hi *,
is it possible inline also methods of objects (not regular procedures) ? I have very simple class method, which call I need inline. (looking at assembler code it is not inlined; I have looked also in documentation, but I do not found article about inline and methods)

Example:

function T2DArray.Data(x, y: integer): PElement; inline;
begin
   Result := @FArray[x+y*Fx];
end;

Thanks
-Laco.

documentation, but I do not found article about inline and methods)

Yes. Inline needs to be declared at the declaration however and not the implementation and any function or method that uses it (in the same unit) needs to come afterwards if you want it to be inlined (for other units that doesn't matter as long as you don't have circular dependencies, cause then things get messy). On addition to that "inline" is only a hint for the compiler, so there might be cases when the compiler decides not to inline.

Regards,
Sven

documentation, but I do not found article about inline and methods)

implementation and any function or method that uses it (in the same unit) needs to come afterwards if you want it to be inlined (for other units that doesn't matter as long as you don't have circular dependencies, cause then things get messy). On addition to that "inline" is only a hint for the compiler, so there might be cases when the compiler decides not to inline.

Oh and it obviously won't work with virtual methods.

Regards,
Sven

I thought it was, and believe that in the past I've been told that it was a useful optimisation aid hence in my code I find e.g.

(*
Setters marked "inline" are trivial except for an assertion to check the parameter range. Most of the associated code will be optimised away if assertions are ignored by the compiler.
*)

[Sorry, I've got this bad habit of commenting stuff which some people frown on. But I think that for something like this- i.e. /why/ something is being done- it's justified.]

I have "inline" on both places also on declaration

and any function or method that uses it (in the same unit) needs to come afterwards

Yes it comes afterwards

In my case it is not virtual method.

But I forgot to mention, that it is method of generic object:

   generic T2DArray<T> = object
    ...
     public
       constructor Init(oobMode: TOutOfBoundsMode); overload;
       constructor Init(x,y: integer); overload;
       ...
       function Data(x,y: integer): PElement; overload; inline;
   end;

So what can be a case, that method is not inlined, can it be track down or is there something which I can do to force inlining ? (FPC 2.6.4)
Thanks
-Laco.

Did you add {$INLINE ON} ?

If not, the compiler will ignore the ;inline; modifier.

Michael.

No, but I have enabled it using Project Options.
Now I added to source unit {$INLINE ON}, but it does not help. I have in assembler still CALL to method.
(In same source unit I have regular procedures and these are inlined as expected; My problem is relatod to object methods only)

Any other ideas, what leads compiler to not inline ? Thanks

-Laco.

LacaK wrote on Wed, 18 May 2016:

Any other ideas, what leads compiler to not inline ?

If you compile with -vd, the compiler will tell you in most cases why it's not inlining. Additionally, it's possible that it's not possible to inline generic routines in regular routines, because IIRC the code for specialisations is generated after the code for all regular routines has been generated (and hence it's probably never available yet while generating code for regular routines, unless perhaps if the specialisation is defined in another unit).

Jonas

it's not inlining. Additionally, it's possible that it's not possible to inline generic routines in regular routines, because IIRC the code for specialisations is generated after the code for all regular routines has been generated (and hence it's probably never available yet while generating code for regular routines, unless perhaps if the specialisation is defined in another unit).

Yes, that might indeed be the most probable cause. That's something that I still need to fix in trunk, cause I really want inlining to work with generic functions (with generic records and classes it should work if the specialization was done in the interface section of another unit).

Regards,
Sven

Hm, I have checked in Lazarus on Project Options / Verbosity appropriate -vd option, but in output window I do not see any more info ... How should look such info message ?

Additionally, it's possible that it's not possible to inline generic routines in regular routines,

It is not my case.
I use this generic in another generic and then it is specialized in interface section of same unit.

Later this specialized object is used in another class type, which comes after specialization.
So this object is not used in regular procedure, but in methods of another class ...

I did also attempt to put all this stuff (with generics and specializations) into separate unit and then I have used this unit ( in interface section ) in "main" unit
But still does not inline ...

-Laco.

In our previous episode, Sven Barth said:

Yes, that might indeed be the most probable cause. That's something that I still need to fix in trunk, cause I really want inlining to work with generic functions (with generic records and classes it should work if the specialization was done in the interface section of another unit).

(delphi doesn't inline generics either. At least it didn't at XE7 level, haven't tested XE10 yet)

Did you build your project clean?
Changing verbosity flags does not recompile units.

Mattias

"Marco van de Voort" wrote

In our previous episode, Sven Barth said: > Yes, that might indeed be the most probable cause. That's something

that I

> still need to fix in trunk, cause I really want inlining to work with > generic functions (with generic records and classes it should work if

the

> specialization was done in the interface section of another unit). (delphi doesn't inline generics either. At least it didn't at XE7 level, haven't tested XE10 yet)

Doesn't mean that we can't do that better... ;)

Regards,
Sven

"LacaK" wrote

> Additionally, it's possible that it's not possible to inline generic

routines in regular routines,

It is not my case. I use this generic in another generic and then it is specialized in

interface section of same unit.

Later this specialized object is used in another class type, which comes

after specialization.

So this object is not used in regular procedure, but in methods of

another class ...

I did also attempt to put all this stuff (with generics and

specializations) into separate unit and then I have used this unit ( in interface section ) in "main" unit

But still does not inline ...

You need to call the function in another unit than where the specialization is. Maybe provide a small (compileable) example that shows what you're doing so that we can check that ourselves. It could also be that 2.6.x had still problems there...

Regards,
Sven

> Hm, I have checked in Lazarus on Project Options / Verbosity appropriate > -vd option, but in output window I do not see any more info ... > How should look such info message ? Did you build your project clean?

Yes, I have used "Clean and build" option.
May be that this diagnostic messages are not shown in "Messages" window. (because when I compile from command line I see also other lines, which are not shown in Lazarus)
L.

Yes I do it so (in attached example it is in one unit, but in my real generic and specialization is in separate unit, that is used by another in interface section)

Maybe provide a small (compileable) example that shows what you're doing so that we can check that ourselves.

Attached

It could also be that 2.6.x had still problems there...

I tried also with FPC 3.0, but result is same AFAICS

Thanks
-Laco.

Right click on Messages window / Filter...

Mattias

Very nice!
(I was not aware of such option I have searched through menu Tools/Environment options ...)
Thanks
-Laco.

"LacaK" wrote

> > >> Additionally, it's possible that it's not possible to inline generic

routines in regular routines,

> > > > It is not my case. > > I use this generic in another generic and then it is specialized in

interface section of same unit.

> > > > Later this specialized object is used in another class type, which

comes after specialization.

> > So this object is not used in regular procedure, but in methods of

another class ...

> > > > I did also attempt to put all this stuff (with generics and

specializations) into separate unit and then I have used this unit ( in interface section ) in "main" unit

> > But still does not inline ... > > You need to call the function in another unit than where the

specialization is.

> Yes I do it so (in attached example it is in one unit, but in my real

generic and specialization is in separate unit, that is used by another in interface section)

> Maybe provide a small (compileable) example that shows what you're doing

so that we can check that ourselves.

> Attached

Moving GT and T1 to a separate unit results in inlined code, as expected (tested with 2.6.4, 3.0.0 and 3.1.1), also if T2 is in another separate unit.

Regards,
Sven

Dňa 20.5.2016 o 14:42 Sven Barth napísal(a):

Yes I can confirm, that this works .
So if I have specialized object type (with inline methods), which is used by another type in same unit inline methods are not inlined ? (assume, that specialized type is defined before type which uses it in interface section and is implemented also before in implementation section) Can this be fixed or is this some kind of design limitation, which can not be bypassed (nor in future) ?
Thanks
-Laco.

generic routines in regular routines,

> >> > > >> > It is not my case. > >> > I use this generic in another generic and then it is specialized in

interface section of same unit.

> >> > > >> > Later this specialized object is used in another class type, which

comes after specialization.

> >> > So this object is not used in regular procedure, but in methods of

another class ...

> >> > > >> > I did also attempt to put all this stuff (with generics and

specializations) into separate unit and then I have used this unit ( in interface section ) in "main" unit

> >> > But still does not inline ... > >> > >> You need to call the function in another unit than where the

specialization is.

> >> > > Yes I do it so (in attached example it is in one unit, but in my real

generic and specialization is in separate unit, that is used by another in interface section)

> > > > > >> Maybe provide a small (compileable) example that shows what you're

doing so that we can check that ourselves.

> >> > > Attached > > Moving GT and T1 to a separate unit results in inlined code, as expected

(tested with 2.6.4, 3.0.0 and 3.1.1), also if T2 is in another separate unit.

> Yes I can confirm, that this works . So if I have specialized object type (with inline methods), which is used

by another type in same unit inline methods are not inlined ?

(assume, that specialized type is defined before type which uses it in

interface section and is implemented also before in implementation section)

Your assumption is correct.

Can this be fixed or is this some kind of design limitation, which can

not be bypassed (nor in future) ?

This *might* get fixed in the future. After all by definition inlining is completely up to the compiler, so even if you add "inline" to some method or function there is no guarantee that it will be inlined.

Regards,
Sven