static void Main(string[] args)
{
//Generate the arrays for the demo
const int arrayLength = 20;
int[] a = new int[arrayLength];
int[] b = new int[arrayLength];
int[] c = new int[arrayLength]; //this will be the array storing the sum result
for (int i = 0; i < arrayLength; i++)
{
a[i] = i;
b[i] = i;
}
//initialize the OpenCL object
OpenCL.OpenCL cl = new OpenCL.OpenCL();
cl.Accelerator = AcceleratorDevice.GPU;
//Load the kernel from the file into a string
string kernel = File.ReadAllText("kernel.cl");
//Specify the kernel code (in our case in the variable kernel) and which function from the kernel file we intend to call
cl.SetKernel(kernel, "addArray");
//Specify the parameters in the same order as in the function definition
cl.SetParameter(a, b, c);
/*
* Specify the number of worker threads,
* in our case the same as the number of elements since every threads processes only one element,
* and launch the code on the GPU
*/
cl.Execute(arrayLength);
Console.WriteLine("Done...");
for (int i = 0; i < arrayLength; i++)
{
Console.WriteLine(c[i].ToString()); //print the result on screen
}
Console.ReadKey();
}
static void Main(string[] args)
{
OpenCL.OpenCL cl = new OpenCL.OpenCL();
cl.Accelerator = AcceleratorDevice.GPU;
string kernel = File.ReadAllText(@"../../../kernel.cl");
const int aX = 2000, aY = 2000, bX = 2000, bY = 2000; //x=number of lines, y = number of columns
double[] a = new double[aX * aY];
double[] b = new double[bX * bY];
double[] c = new double[aX * bY]; // resulting matrix, with aX*bY dimensions
int[] dimensions = new int[3] {
aX, aY, bY
};
Random rand = new Random();
for (int i = 0; i < aX; i++)
{
for (int j = 0; j < aY; j++)
{
a[i * aY + j] = rand.NextDouble();
}
}
for (int i = 0; i < bX; i++)
{
for (int j = 0; j < bY; j++)
{
b[i * bY + j] = rand.NextDouble();
}
}
cl.SetKernel(kernel, "MatrixMulti");
//Pass the dimensions and matrix pointers:
cl.SetParameter(dimensions, a, b, c);
//Each cell of the result will be computed at the same time
Stopwatch stopwatch = new Stopwatch();
stopwatch.Start();
cl.Execute(aX * bY);
stopwatch.Stop();
Console.WriteLine("GPU parallel multiplication done in " + stopwatch.ElapsedMilliseconds / 1000.00 + " seconds");
for (int i = 0; i < aX; i++)
{
for (int j = 0; j < bY; j++)
{
//Console.Write(c[i*bY + j] + " ");
}
//Console.WriteLine();
}
//Run linearily on CPU:
stopwatch.Reset();
stopwatch.Start();
for (int id = 0; id < aX * bY; id++)
{
int NLin_1 = dimensions[0]; //number of lines of first matrix
int NCol_1 = dimensions[1]; //number of columns of first matrix
int NCol_2 = dimensions[2]; //number of columns of second matrix
int L = id / NCol_2; //get the position in the final matrix from the id
int C = id - L * NCol_2;
double element = 0;
for (int i = 0; i < NCol_1; i++)
{
element = element + a[L * NCol_1 + i] * b[C + NCol_2 * i];
}
c[id] = element;
}
stopwatch.Stop();
Console.WriteLine("CPU linear multiplication done in " + stopwatch.ElapsedMilliseconds / 1000.00 + " seconds");
stopwatch.Reset();
stopwatch.Start();
Parallel.For(0, aX * bY, id =>
{
int NLin_1 = dimensions[0]; //number of lines of first matrix
int NCol_1 = dimensions[1]; //number of columns of first matrix
int NCol_2 = dimensions[2]; //number of columns of second matrix
int L = id / NCol_2; //get the position in the final matrix from the id
int C = id - L * NCol_2;
double element = 0;
for (int i = 0; i < NCol_1; i++)
{
element = element + a[L * NCol_1 + i] * b[C + NCol_2 * i];
}
c[id] = element;
}
);
stopwatch.Stop();
Console.WriteLine("CPU parallel multiplication done in " + stopwatch.ElapsedMilliseconds / 1000.00 + " seconds");
Console.ReadKey();
}
Task Enqueue(int from, int to, OpenCL acc)
{
return(Task.Run(() => acc.Execute(from, to - from, -1)));
}
