inlining and optimations when instantiating templates

[kedaman]

will

if (0)
  do stuff a;
else
  do stuff b;

compile to

  do stuff b;

?

also, as this is in a template, will the instance have to support functions specified in "do stuff b", in case 0 is evaluated above, as a result of the template parameters, by calling an inline function on them, say:

if (GetVal())
  do stuff a;
else
  do stuff b;

for:

template <class x, class y>
class Stick{
public:
  inline bool GetVal(){return a.GetVal() || b.GetVal();}
  .
  .
  .
private:
  x a;
  y b;
}

Stick template is used recursively, upto about 8 times, but will that be a problem for inlining?

[HarryW]

Any instances of templates 'things' (classes, functions, structs, whatever you can make a template of) are translated early on in compilation into a set of equivalent 'things', one for each different use of the template you have specified. So if you want to consider optimisations, you should remember that your template class will be split up into several normal classes, one for each different combination of the classes x and y you use.

Anything that is done recursively will probably not get inlined, whether you specify it as inline or not. Some compilers are very fussy about what they will and will not inline, particularly Microsoft's VC++ compiler, which likes to think it knows better than you. Just like all their other products....

There are some preprocessor directives you can use to help enforce your will on different compilers, usually using the #pragma directive which is compiler-specific.

AFAIK anything that always evaluates to false will just get treated as a constant value of false if/when the compiler optimises it, but they're strange things, compilers, so I couldn't say for sure.

[kedaman]

I was thinking more like template recursion:
Stick<a,Stick<b,Stick<c> > >
I saw an article on flipcode.com which mentioned three pragma's you could add to prevent VC++ to get fussy about inlining, but would this form of recursion prevent it?

The thing is that i can't put precompiler #ifdefs to determine what the instances should contain, but they are templates right, and they don't instantiate what isn't unneeded? The problem I have is that for a class to be a valid template parameter, it has to have the methods used on it in the template, but i have methods both in "do stuff a" and "do stuff b" that should not be together in the same instance. Will the compiler be smart enough to remove the code of either a or b and not complain about the class parameter not fitting the way it is used?

[parksie]

Methods that aren't required by a specific instantiation are not provided with code.

[HarryW]

Uhm.... Maybe they could inherit from a base class that had trivial definitions of the unnecessary functions (just return 0 or something), and the subclasses could override them with more useful local definitions of the functions that would be needed.

Stick<a,Stick<b,Stick<c> > > isn't really what I would call recursion, but I see what you mean. I think maybe the best way to find out if your compiler is optimising it the way you would like would be to write a simple example (I mean really simple) and look at the assembly code that it produces. You'd have to look at a non-Windows app though to avoid all the console code.

There was a good article on templates and optimisations at Flipcode a while ago, aimed at efficient matrix calculations. That might be the one you were thinking of earlier, but if not check it out.

[kedaman]

Uhm.... Maybe they could inherit from a base class that had trivial definitions of the unnecessary functions (just return 0 or something), and the subclasses could override them with more useful local definitions of the functions that would be needed.

virtual functions would ruin it all, or would it? I'm almost certain the methods won't inline if they are virtual.

The page i was talking about was about vector math calculations, so i might have a look at the matrix one as well.

I'll try testing some really simple templates, how do I get the asembly listings in VC++6?

[HarryW]

Well you can look at the source using any kind of disassembler. A lot of hex editors have built-in disassemblers. I expect you can view the assembly code in the VC++6 IDE if you try it.

The only problem with that is that you might get lots of unwanted Windows code messing things up, making it unclear.

Do the functions really need to be virtual? Been a while since I used C++ now, I'm more into Python at the moment.

[kedaman]

Do you have a hex editor with built in disasembler since that would be nice?
if i want to override a function from a base class, i need the function to be virtual, if it's suppose to be called. I'm not quite sure but some compilers will inline and some not, if the base object pointer is known to be of the specific derived class under design time. In any case, I don't think that would help much, the functions i use for the different cases are not suppose to be in the same parameter class.

of interest, what can you do with Python? I have a mapeditor for quake, halflife and some other similar games called quark (quake army knife) that runs on Python.

[HarryW]

Python is an object-oriented scripting language. I use it with Zope (www.zope.org , it's a dynamic webpage delivery type thing, open source and with lots of extendability) to make Zope Products at the moment. It's pretty cool. You can compile it to Java bytecode with the right compiler too

Some games are scripted in Python, such as Severance. If you know Python you can add your own mods to the game using calls to the game engine, which is in a load of dlls I think.

I don't have a hex editor that does it here, but a lot of supposed 'hacking tools' (or 'toolz' I guess) are actually just hex editors with a disassembler attached. SO you might want to search for them on www.astalavista.com or somewhere like that.

[kedaman]

ok, thanks for the suggestions Harry I'll have a look at Python sooner or later. If you develope a game using Python, does it mean everything runs on script? I mean, can I have the core implemented in C++ and then use Python so that the user can make mods to it?