too long time without a Fastest something ! so I decided to share another nice code from 3DzzD 
the following function enable to filter 4 pixels using two interpolation factors to perform fast/perpixel “bicubic bilinear/cubic filtering” of an image
the filtering is based on area covered by pixels based on their interpolation factors, and the result is something that look very similar to bicubic filtering :
- make pixels appear rounded when zoomed a lot and antialiased when zoomed out
- you can see such effect in most 3DzzD demo if you get very close of a texture : this enable high quality when texture ar zoomed out (under pixel presicion up to 1/256 with in conjonction with mipmapping enable some kind of 2*2 anisiotropic filtering) and also zoomed in that make pixels blend nice and appear rounded.
NB: for readability it is adapted from the original version wich is a little different
/**
* Perform a bilinear interpolation of 4 RGB/vector each packed in an int as three unsigned bytes 0xXXYYZZ or 0xRRGGBB.
*
* can be either RGB color for image bicubic filtering or vector XYZ (only positive)
*
* @param c1 value/color 1 (upper left value/RGB pixel)
* @param c2 value/color 2 (upper right value/RGB pixel)
* @param c3 value/color 3 (lower left value/RGB pixel)
* @param c4 value/color 4 (lower right value/RGB pixel)
* @param bX x interpolation factor (range 0-256)
* @param bY y interpolation factor (range 0-256)
*
* @return interpolated value(packed RGB pixel) of c1,c2,c3,c4 for given factors bX & bY as three packed unsigned bytes
*
* @author Bruno Augier http://dzzd.net/
*/
public int interpolate(int c1,int c2,int c3,int c4,int bX, int bY)
{
int f24=(bX*bY)>>8;
int f23=bX-f24;
int f14=bY-f24;
int f13=((256-bX)*(256-bY))>>8; // this one can be computed faster
return ((((c1&0xFF00FF)*f13+(c2&0xFF00FF)*f23+(c3&0xFF00FF)*f14+(c4&0xFF00FF)*f24)&0xFF00FF00)|
(((c1&0x00FF00)*f13+(c2&0x00FF00)*f23+(c3&0x00FF00)*f14+(c4&0x00FF00)*f24)&0x00FF0000))>>>8;
}
EDIT: correction missnamed : bilinear/cubic rather than bicubic…