Converting a c# function into openCL which is basically c

user_17872531 0
Viewed 36

I am trying to implement a blur-function I wrote in c#. I want it to run in an openCL Kernel

The function is as follows:

     private static int cSharpBlur(double[,] blur, int width, int height, int[,] imageToInt, int[,]   outputImage, int i, int j)
    {
        int maskSize = 1;
        double sum = 0.0f;
        // Collect neighbor values and multiply with gaussian
        
        for (int a = -maskSize; a < maskSize + 1; a++)
        {
            for (int b = -maskSize; b < maskSize + 1; b++)
            {
                sum += blur[a+1, b+1] * imageToInt[Clamp(i+a,0,width-1), Clamp(j+b,0,height-1)];
                
            }
        }
        byte[] values = BitConverter.GetBytes(imageToInt[i,j]);
        int alpha = values[3];
        int alphasum = alpha * (int)sum;
        values[3] = (byte)alphasum;
        int newValue = BitConverter.ToInt32(values,0);
        return newValue;
    }

Now I obviously don't have .GetBytes and BitConverter.ToInt32 in openCL. Neither do I have a 2 dimensional array.

I solved this via __kernel void gaussianBlur(__global int* imageToInt, int width, int height, __global double* blurBuffer, int blurBufferSize, __global int* outputBuffer){ int col = get_global_id(0); int row = get_global_id(1);

double sum = 0.0f;
for (int a = -1; a < 2; a++)
{
    for (int b = -1; b < 2; b++)
    {
        sum += blurBuffer[a+b+2] * imageToInt[col+width*row];
    }
}
outputBuffer[col + width * row] = sum;

What's missing is the entire getBytes and ToInt stuff.

How can I do that in openCL?

Thanks in advance and have a great weekend!

opencl c c#
1 Answers

For getBytes, use as_uchar4:

uchar4 values = as_uchar4(imageToInt[col+width*row]);

and for ToInt32, use as_int:

int newValue = as_int(values);

So your kernel should look something like this:

__kernel void gaussianBlur(__global int* imageToInt, int width, int height, __global double* blurBuffer, int blurBufferSize, __global int* outputBuffer) {
    int col = get_global_id(0); int row = get_global_id(1);
    int maskSize = 1;
    double sum = 0.0f;
    for(int a = -maskSize; a < maskSize + 1; a++) { // Collect neighbor values and multiply with gaussian
        for(int b = -maskSize; b < maskSize + 1; b++) {
            sum += blurBuffer[a+1+(b+1)*2*maskSize] * imageToInt[col+width*row];
        }
    }
    uchar4 values = as_uchar4(imageToInt[col+width*row]);
    int alpha = (int)values.s3;
    int alphasum = alpha * (int)sum;
    values.s3 = (uchar)alphasum;
    int newValue = as_int(values);
    outputBuffer[col+width*row] = newValue;
}

These as_... functions come built-in with OpenCL C, and enable you to reinterpret the bits that make up a number, as long as the total number of bits remains the same. In your case, an int is made up o 4 bytes, just like the uchar4 vector data type. With uchar4, you can address the individual bytes with .s1, .s1, .s2, .s3, or with .x, .y, .z, .w.

The as_... functions can also be used to get the individual bits of a float number for example:

float x = 1.0f;
int bits = as_int(x);

In plain C, you can also manually write such a function:

uint as_int(const float x) {
    return *(int*)&x;
}

OpenCL C comes packed with math functionality, more than most other languages. All of the built-in functions are listed in this super helpful reference card.

user_91789920
Related