For loop takes time

[Shakti]

I have a very weird (at least how I see it) problem. In a game I am making I perform some memory copying using a for loop and when I loop through it 512*512 times (262144 times) it is very very very slow, but when I loop through it 600*650 times (390000 times) the loop goes around 4 times as fast. Anybody have any idea what is going on here? I am performing the same operations in both cases and the timing is done the same way in both cases.

[Laserve]

I'd love to see some code ^^

[Shakti]

Here is the loop itself.

for (DWORD screencounter=0;screencounter<screensize;screencounter++)
{
	bgBuffer[screenoffset]=imageBuf[spritepixeloffset];

	screenoffset+=screenX;
	spritepixeloffset+=sizeX;

	ycounter++;
	spriteycounter++;

	if (spriteycounter >= sizeY) 
	{
		spritepixeloffset=spritexcounter;
		spriteycounter=0;
	}

	if (ycounter>=screenY)
	{
		ycounter=0;
		xcounter++;
		spritexcounter++;
		spriteycounter=spritepixely;
		if (spritexcounter>=sizeX) spritexcounter-=sizeX;
		
                spritepixeloffset=spriteybottom+spritexcounter;
            
		screenoffset=xcounter;
	}
}

imageBuf is a buffer of DWORDs with 256 elements. bgBuffer is a buffer with the size mentioned in my first post.

[Salem]

Check the alignment of the destination buffer in both cases (just print its address in hex format).

If the address doesn't end in 0, 4, 8 or C, then it's unaligned memory, which can be a lot more expensive to access than aligned memory.

0x2468 is an aligned address
0x1357 isn't

[Shakti]

The memory is aligned.

[Salem]

What happens if you do the 600*650 test before the 512*512 test?
Do you re-use the same (large) buffer for both tests?

[Shakti]

The size of the buffer is at the moment decided compiletime but allocated dynamicly, this is then updated every frame when a camera moves.

[Zach L.]

I take it that it is repeatable? (And hence, not the system allocating cpu time for something else instead of your program on that given test.)

[Shakti]

It happends every time I run it.

[Zach L.]

I figured... Just something to check

Not sure it'll help too much, but can we see some more relevant code?
And, what system, compiler, etc is this occuring on?

[Shakti]

ok here is the whole function:

DWORD screenY = 600;
DWORD screenX = 650;

DWORD addr1=0,addr2=0;

if(bgBuffer == NULL)
{
	try
	{
		bgBuffer = new DWORD[screenY*screenX];
	}
	catch(std::bad_alloc error)
	{
		MessageBox(0, error.what(), 0, 0);
		PostQuitMessage(0);
	}
}

int camX = (int)cam->GetBgX();
int camY = (int)cam->GetBgY();

DWORD screenoffset=0;
int imgwidth=sizeX;
int spritepixelx=((camX) % sizeX);
int spritepixely=((camY) % sizeY);
if(spritepixelx < 0 || spritepixely < 0)
	return;

WORD spritepixeloffset=(spritepixely*sizeX)+spritepixelx; //This should be spritepixelx
WORD ycounter=0;
WORD xcounter=0;

DWORD screensize=screenY*screenX;    
    
WORD spritexcounter=spritepixelx;
WORD spriteycounter=spritepixely;
spriteycounter = spriteycounter;

WORD spriteybottom = spritepixely*sizeX;

for(DWORD screencounter=0;screencounter<screensize;screencounter++)
{
	bgBuffer[screenoffset]=imageBuf[spritepixeloffset];

	screenoffset+=screenX;
	spritepixeloffset+=sizeX;

	ycounter++;
	spriteycounter++;

	if (spriteycounter >= sizeY) 
	{
		spritepixeloffset=spritexcounter;
		spriteycounter=0;
	}

	if (ycounter>=screenY)
	{
		ycounter=0;
		xcounter++;
		spritexcounter++;
		spriteycounter=spritepixely;
		if (spritexcounter>=sizeX) spritexcounter-=sizeX;
		spritepixeloffset=spriteybottom+spritexcounter;
            
		screenoffset=xcounter;
	}
}

if(!glIsEnabled(GL_TEXTURE_2D))
	glEnable(GL_TEXTURE_2D);
if(bgTexture)
	glDeleteTextures(1,bgTexture);

glGenTextures(1, bgTexture);
glBindTexture(GL_TEXTURE_2D, bgTexture[0]);
glTexParameteri(GL_TEXTURE_2D,GL_TEXTURE_MIN_FILTER,GL_LINEAR);
glTexParameteri(GL_TEXTURE_2D,GL_TEXTURE_MAG_FILTER,GL_LINEAR);


glTexImage2D(GL_TEXTURE_2D, 0, 4, screenX, screenY, 0, 
			GL_RGBA, GL_UNSIGNED_BYTE, (DWORD*)bgBuffer);

glDisable(GL_TEXTURE_2D);

This happends on a winXP home system, AMD Athlon 2200+, and 512 ram with VS.NET 2002.

[DerelictDream]

for(DWORDscreencounter=0;screencounter<screensize;screencounter++)

I know you had it correct in the top post, just noting the typo here...

[Shakti]

Oh yah, it became like that when I pasted it. Should be fixed now.