Performance of immediate objects

Performance of immediate objects

[Chris King]

Is there a substantial difference to the way in which

class foo (i:int) = object
 val v = i
 method bar = v
end

let mk_foo i = new foo i

and

let mk_foo (i:int) = object
 val v = i
 method bar = v
end

are compiled?  I've run a couple tests with the above and immediate
objects seem to be about 15% slower than classes.  Is this because a
new method table is created for the immediate object every time or is
that the case for classes also?

- Chris King

Re: [Caml-list] Performance of immediate objects

[Jacques Garrigue]

From: "Chris King" <colanderman@gmail.com>

> Is there a substantial difference to the way in which
> 
> class foo (i:int) = object
>  val v = i
>  method bar = v
> end
> 
> let mk_foo i = new foo i
> 
> and
> 
> let mk_foo (i:int) = object
>  val v = i
>  method bar = v
> end
> 
> are compiled?  I've run a couple tests with the above and immediate
> objects seem to be about 15% slower than classes.  Is this because a
> new method table is created for the immediate object every time or is
> that the case for classes also?

Certainly not. It would not be a 15% slowdown, but rather 10000%.
Initialization occurs only once, but one needs to check whether it was
done every time an object is created.
However, the 15% slowndown is for bytecode. For native code this is a
15% speedup. Guess which one we try to optimize.

Jacques Garrigue

Re: [Caml-list] Performance of immediate objects

[Chris King]

On 7/11/06, Jacques Garrigue <garrigue@math.nagoya-u.ac.jp> wrote:
> Initialization occurs only once, but one needs to check whether it was
> done every time an object is created.

Thanks, now the assembly output makes more sense to me.  I saw the
initialization code in the mk_foo function but didn't realize that it
was executed only once.

- Chris King