Re: [GD-General] Scripting Systems

[Eero P. <epa...@ko...>]

Alen Ladavac wrote:

> From: "Eero Pajarre" <epa...@ko...>
> 
>>For me GC has not been the main problem speed wise, and I do trigger
>>GC each frame, so that I can minimize "real time" problems with it.
> 
> 
> Interesting idea. It really must smooth the performance. Though I would
> expect it to be slooow. It certainly does trash the cache big time, doesn't
> it?
> 

It might be just me not noticing it, but as I said it has not
been a problem yet. I certainly have been "garbage aware" in my
code, so I don't produce that much of it. Also my data structures
at lua side are not that huge. The main objects are on C++ side
and just referred from the Lua side, although I typically
still have also a Lua table for each main object.

> 
>>Profilers show the Lua name lookup as a spike in the curves, this might
> 
> That's same thing that I heard before, and one of main reasons why we didn't
> choose Lua for in-game scripting. I just couldn't find anyone who could
> reaffirm that.
> 
> 
>>Local variables them self AFAIK do not need the name
>>lookup.
> 
> That is good. However, any member variable/function access would do a name
> lookup, right?
> 

Yes, although the access is based on hashed arrays etc, and is
quick.

> 
> Out of curiosity, does the interpreter have a function table for
> instructions? If yes, then I don't see how so much misprediction could
> happen. :/
> 

The main loop is actually switch based, but I think that even
with function table the main switching point would predict
rather badly. Also there is some unpacking of parameters in
the virtual machine, which also doesn't branch predict too well.


		Eero