Re: Should Inline be private in the private part of a package spec?

["Vasiliy Molostov" <molostoff@gmail.com>]

Simon Wright <simon@pushface.org> писал(а) в своём письме Tue, 07 Aug 2012  
21:28:25 +0400:

> "Vasiliy Molostov" <molostoff@gmail.com> writes:
>
>> Simon Wright <simon@pushface.org> писал(а) в своём письме Tue, 07 Aug
>> 2012 14:01:27 +0400:
>
..
>
> which you would think would be a likely candidate for inlining.

Perhaps. Since it is an environment dependent, e.g. how internal data flow  
in a resulting image is organized by a compiler (e.g. stack and register  
alchemy) on a given hardware.

By using a word "pragma inline" we suppose, I think, "un-roll" expensive  
call with many parameters, referencing them directly to locals.

This can increase perfomance in loops where such a call is used, and is  
called many times. So, the more parameters we passed the more expensive  
becomes that call. As parameter count goes to be less as the expensiveness  
goes to be less, too.

I am not sure that the expense becomes zero with a parameter count equal  
to one, since it is a compiler (implementation) and architecture dependent  
thing, but indeed and doubtless, inlining in general can improve  
perfomance with this  call-in-a-loop stuff.

It is my opinion only.

> I was agreeing that inlining may not improve speed.

Yes, may not.

-- 
Написано в почтовом клиенте браузера Opera: http://www.opera.com/mail/