optimizing the scaling code

[Michael Thaler]

Hello,

I think the main speed problem in my current scaling code is that I read 
single pixels using

QUANTUM *tmp = new QUANTUM[ pixelSize * sizeof(QUANTUM) ];
m_dev -> readBytes(tmp, x, y, 1, 1);

I am currently rewriting the code to read whole lines and I wonder what is the 
best way to do this. Casper's transform visitor does this:

KisHLineIteratorPixel srcIt = src->createHLineIterator(left, y, w, true);
int i = 0;
while(!srcIt.isDone())
{
        Q_UINT8 *data;
        data = srcIt.rawData();
        memcpy(&tmpLine[i*pixelSize], data, pixelSize);
        ++srcIt;
        i++;
}

which creates a temporary array with the pixel data of a line.

Another method would be to just use

QUANTUM *tmp = new QUANTUM[ width * pixelSize * sizeof(QUANTUM) ];
m_dev -> readBytes(tmp, 0, y, width, 1);

What is faster? Or is there any other method to read a line as fast as 
possible?