# PP assignment 6
###### tags: `PP Assignment`

### Q1
**Explain your implementation. How do you optimize the performance of convolution?**
* In ```hostFE.c```, I set work-group size to be 25 x 25. This ensure every pixel is being processed by a work item and it's optimized for the device.
* The kernel function lies in ```kernel.cl```. The kernel function start off by getting the item ID of the current work-item. Finally the pixel is calculated using zero-padding convlution based on the function in ```serialConv.c```

__kernel void convolution(){
    int i = get_global_id(0);
    int j = get_global_id(1);

    conv = 0;

    for (k = -halffilterSize; k <= halffilterSize; k++){
        for (l = -halffilterSize; l <= halffilterSize; l++){
            if (j + k >= 0 && j + k < imageHeight && 
                i + l >= 0 && i + l < imageWidth){
                    conv += inputImage[(j + k) * imageWidth + i + l]
                    * filter[(k + halffilterSize) * 
                    filterWidth + l + halffilterSize];
            }
        }
    }
    output[j * imageWidth + i] = conv;
}