Assembler Optimization on PC

I need an optimized assembler program to take a 640x480 8-bit image and reduce it to 160x120 resolution as follows:

For each 4x4 collection of pixels in the source image, calculate the average of all the non-zero pixels and store that average in the destination 160x120 image. If a pixel value in the source image is zero it must be ignored and not included in the destination image. If all source image pixels in a block are zero then the destination image should be set to zero.

The C code for this would look something like:

ImageAverageShrinkNoZeros(unsigned char *pSource, unsigned char *pDest)
unsigned char *pRow1, *pRow2, *pRow3, *pRow4;
int nTotal;
int nPixelCnt;
int nRow,nCol,I;

pRow1 = pSource;

for (nRow=0; nRow<480; nRow+=4)
{
pRow1 = pSource + 640 * nRow;
pRow2 = pRow1 + 640;
pRow3 = pRow2 + 640;
pRow4 = pRow3 + 640;

for(nCol=0; nCol<640; nCol+=4)
{
nTotal = 0;
nPixelCnt = 0;

for(i=0;i<4;i++)
{
if(*pRow1 > 0) { nTotal += *pRow1++; nPixelCnt++; } else pRow1++;
if(*pRow2 > 0) { nTotal += *pRow2++; nPixelCnt++; } else pRow2++;
if(*pRow3 > 0) { nTotal += *pRow3++; nPixelCnt++; } else pRow3++;
if(*pRow4 > 0) { nTotal += *pRow4++; nPixelCnt++; } else pRow4++;
}
if(nPixelCnt > 0)
*pDest++ = (unsigned char) (nTotal / nPixelCnt);
else
*pDest++ = 0;
}
}

I need this to run very quickly on a PC in windows.