The right way to code.

My application now becomes an oversized giant. There's a small lag and performance drop as the code keeps getting bigger. I suspected that it's all because of the unorganized and poor OOP implementation in my part. Here's what I did (and didn't) do. Oh and FYI, I'm making a simple game using SDL:

1. I load all the bitmaps into SDL_surfaces in the memory (200 megs of images BTW). And from those surfaces, I created objects like background, sprites, etc.

2. Didn't do any design pattern whatsoever. I created classes when I think I need one.

3. No inlined, or static function / variable

4. Few assert and exception handler

5. Most of the variables are referenced by pointer (eg. surfaces, sprites, etc)

6. My animation routine is like this:
1. Blit the whole background
2. Blit objects
3. Flip the surface
4. Loop to 1

The code itself is one big mess. My questions are:

1. Is the performance boost from an inlined function significant to the whole application's performance?

2. If I want to create objects / variables that can be accessed by multiple objects and if the accessed objects / variables were changed, the change will be carried out to all the accessing objects, is it better to use static or a reference by pointer ? What's the benefit of static over a reference by pointer (or vice versa)?

3. Do unused functions and variables in a class have any impact to the performance?

4. Does any of you have any advice to help me make a better code?

Thanks alot in advance.

Search C++ inline functions you will find a lot of stuff... most of it will say profile your code with and without to tell if inlines are helping.

Being static and being referenced by a pointer are two different things, express this in code example please.

Yes, because space has to be alocated for all parts of an object.

Yeah, rewrite it from scratch . Seriously though, first find parts that are obviously lagging down your program, and rewrite those, then when you can't tell as easily look into profiling your code to see where the bottlenecks are and work on fixing those up. This is where design patterns are something you need. That way you make interfaces to areas of your code, and can change the implementation without worrying about the affects on the rest of your code.

You will iterate through this "find the slow stuff or stuff that doesn't run how you want to to"-> "rewrite it." thing untill the program as a whole satisfies you. Then you will release it (probably) and users will do the first part (aka bugs) and leave the second part up to you.

It is extremely hard and generally a bad idea to guess at performance bottlenecks in code you see. It's impossible to do it with code you only get to describe. As Wraithan said, profile the code.

But if I had to guess, I'd say the performance penalties come from the 200MB of pictures you have loaded (how much RAM do you have?), especially as you can't keep them all in graphics memory all the time and it will probably be used poorly.

More bad performance might come from a poor choice of algorithms on you part. Not knowing what kind of game you're making, I can't even begin to guess what algorithms you might have, though.

Originally Posted by Wraithan

Search C++ inline functions you will find a lot of stuff... most of it will say profile your code with and without to tell if inlines are helping.

Being static and being referenced by a pointer are two different things, express this in code example please.

Yes, because space has to be alocated for all parts of an object.

Yeah, rewrite it from scratch . Seriously though, first find parts that are obviously lagging down your program, and rewrite those, then when you can't tell as easily look into profiling your code to see where the bottlenecks are and work on fixing those up. This is where design patterns are something you need. That way you make interfaces to areas of your code, and can change the implementation without worrying about the affects on the rest of your code.

You will iterate through this "find the slow stuff or stuff that doesn't run how you want to to"-> "rewrite it." thing untill the program as a whole satisfies you. Then you will release it (probably) and users will do the first part (aka bugs) and leave the second part up to you.

I've searched the web about inlining, and it's pretty much like you said.

About static and a reference by pointer example:
Pointer:
Say, if I have two objects instances from a this class A:
clas A
{
public:
B *pointerOfB;
void createB() {pointerOfB=new(B)}
operator=(A*) { pointerOfB=A->pointerOfB }
}

named A1 and A2. In A1, I new-ed B by calling createB(). And in A2, I called the copy constructor A2=A1. CMIIW, the pointer reference of B in A1 and A2 will be the same. Would it be better if I use the pointer reference or just make the B variable static?

Originally Posted by CornedBee

It is extremely hard and generally a bad idea to guess at performance bottlenecks in code you see. It's impossible to do it with code you only get to describe. As Wraithan said, profile the code.

But if I had to guess, I'd say the performance penalties come from the 200MB of pictures you have loaded (how much RAM do you have?), especially as you can't keep them all in graphics memory all the time and it will probably be used poorly.

More bad performance might come from a poor choice of algorithms on you part. Not knowing what kind of game you're making, I can't even begin to guess what algorithms you might have, though.

Yeah, that's what I thought too. But I think it's better to load all the bitmaps than only load 'em when needed from the file. I mean there's the file reading cost and all. I think it's will be slower if I keep reading bitmap files, creating and freeing surfaces all the time. BTW, the reason it's so big is because it's in 32 bpp and in a resolution of 1024X768 (the background). And my RAM is 512 megs.

Well, the only algorithm I use is the blitting algorithm because it's not the sophisticated type of game. Only a board game.

Oh yeah another question please, I know that this question seemed rather elementary, but how do I profile my code?

Thanks.

> Oh yeah another question please, I know that this question seemed rather elementary, but how do I profile my code?

If someone tells you to profile your code, it usually means that person wants you to install and run a tool called a profiler, such as this one or this one. In addition, some IDEs come with profilers already built in. People have their favorites I suppose, but there are many to choose from, and you'll want to get one that works with your compiler.

Originally Posted by citizen

> Oh yeah another question please, I know that this question seemed rather elementary, but how do I profile my code?

If someone tells you to profile your code, it usually means that person wants you to install and run a tool called a profiler, such as this one or this one. In addition, some IDEs come with profilers already built in. People have their favorites I suppose, but there are many to choose from, and you'll want to get one that works with your compiler.

OIC, thanks. BTW, is there any free and integratable to KDevelop C++ profiler out there? Dunno if I have any money to buy the commercial one.

KCachegrind. In fact it is right there under "Debug". (NB. I have no experience with it.)

Or just compile with -pg and use gprof(1).

Ah, thanks a lot.