irrArray extension

[Dareltibus]

Feature request(is that the right place? ):
a new method in "irrArray.h" for the array class:

Code: Select all

//! erase the array from "index" to "end" if option=false
/**erase the array from "index to "start" if option=true,*/
void erase(u32 index, bool option ) //no default value for "option"
{ 
       //check if was entered an invalid value
	if(index >= used)
		return;

   u32 i=0;
	if(option)
	{
		for (; i<index; ++i)
			allocator.destruct(&data[i]);
		

		for (i=0; i<(used-index); ++i)
			allocator.construct(&data[i], data[i+index]);

		
		for (; i<used; ++i)
			allocator.destruct(&data[i]);


		used-=index;
	}
	else
	{
		for (i=index; i<used; ++i)
			allocator.destruct(&data[i]);
		used=index;
	}
}

it will be included in the engine or it is not a general case?

This method is only optimized for deleting the array from "start to index" or from "index to end" (the latest one is the fastest).
I'm actually using it in my game. If this is not suitable for the engine itself I will put that code in code snippets.
(i'm testing it, but if someone can tell me that it is correct I'm happy twice)

[hybrid]

No, that was not the correct place. The 1.7 beta phase is gone and closed. There would be no reason to revive such an old thread. Better make a new one, as I did for you now.
However, what's the good thing of this over the existing erase(index, number) method? That one would work exactly as yours, you just need to give the correct number of elements from index to end or from start to index.

[Dareltibus]

Ok thanks.

I know .It should be a little faster than the old one (but only in two cases)

and i have no need to check for the last element of the array before calling it, so it is also simpler in use if you need to erase exactly from the start or from the end.

[hybrid]

Well, actually the same happens here as there. The second loop in the existing method is also only invoked where necessary. So I'm not sure where your method is faster, once you did the size-index calculation.

[Dareltibus]

u are right

erase to end:

Code: Select all

u32 time0,time1,time2;
	array<u8> a,b;
	for(u32 i=0; i<40000000; i++)
	{
		a.push_back(3);
		b.push_back(3);
	}
	;
	time0=device->getTimer()->getRealTime();
	time0=device->getTimer()->getRealTime();
	a.erase(5000000,3500000);
	time1=device->getTimer()->getRealTime();
	;
	b.erase(5000000, false);
	time2=device->getTimer()->getRealTime();

	time2-=time1; //value = 1383
	time1-=time0; //value = 1385
        bool test = time2<time1 // test = true (not a real gain)

erase to start

Code: Select all

u32 time0,time1,time2;
	array<u8> a,b;
	for(u32 i=0; i<40000000; i++)
	{
		a.push_back(3);
		b.push_back(3);
	}
	;
	time0=device->getTimer()->getRealTime();
	time0=device->getTimer()->getRealTime();
	a.erase(0,3500000);
	time1=device->getTimer()->getRealTime();
	;
	b.erase(35000000, true);
	time2=device->getTimer()->getRealTime();

	time2-=time1; //value = 3146 
	time1-=time0; //value = 2185
        bool test = time2<time1 // test = false result

all values are a average on 10 tests.

ok sorry close it O_O.
I tested it only with little arrays (1000000 size). with bigger arrays my code is slower . original code of irrlicht is 50% faster