public static float[] RunMatrixMulShared(float[][] a, float[][] b, int N, ref Stopwatch sw)
{
//Create context and accelerator
var gpu = new CudaAccelerator(new Context());
//Create typed launcher
var matrixMulKernelShared = gpu.LoadStreamKernel <
ArrayView <float>,
ArrayView <float>,
ArrayView <float>,
int>(MatrixMulShared);
//Allocate memory
var buffSize = N * N;
MemoryBuffer <float> d_a = gpu.Allocate <float>(buffSize);
MemoryBuffer <float> d_b = gpu.Allocate <float>(buffSize);
MemoryBuffer <float> d_c = gpu.Allocate <float>(buffSize);
d_a.CopyFrom(FlatternArr(a), 0, Index1.Zero, buffSize);
d_b.CopyFrom(FlatternArr(b), 0, Index1.Zero, buffSize);
//Groups per grid dimension
int GrPerDim = (int)Math.Ceiling((float)N / groupSize);
KernelConfig dimension = (
new Index2(GrPerDim, GrPerDim), // Number of groups
new Index2(groupSize, groupSize)); // Group size (thread count in group)
sw.Restart();
matrixMulKernelShared(dimension, d_a.View, d_b.View, d_c.View, N);
// Wait for the kernel to finish...
gpu.Synchronize();
sw.Stop();
var c = d_c.GetAsArray();
return(c);
}