public override float GetElementAsFloat(long index)
{
CUdeviceptr ptr = DevicePtrAtElement(index);
if (ElementType == DType.Float32)
{
float[] result = new float[1]; context.CopyToHost(result, ptr); return(result[0]);
}
else if (ElementType == DType.Float64)
{
double[] result = new double[1]; context.CopyToHost(result, ptr); return((float)result[0]);
}
else if (ElementType == DType.Int32)
{
int[] result = new int[1]; context.CopyToHost(result, ptr); return(result[0]);
}
else if (ElementType == DType.UInt8)
{
byte[] result = new byte[1]; context.CopyToHost(result, ptr); return(result[0]);
}
else
{
throw new NotSupportedException("Element type " + ElementType + " not supported");
}
}
static double[] SumMatrixManagedCuda(double[][,] matrix)
{
int Z = matrix.Length;
int Y = matrix[0].GetLength(0);
int X = matrix[0].GetLength(1);
var result = new double[Y * X];
var lm = ToLinearArray(matrix);
int N = lm.Length;
matrixSumCude.SetComputeSize((uint)X, (uint)Y);
//matrixSumCude.BlockDimensions = 128;
//matrixSumCude.GridDimensions = (N + 127) / 128;
var da = cntxt.AllocateMemory(N * sizeof(double));
var db = cntxt.AllocateMemory(result.Length * sizeof(double));
cntxt.CopyToDevice(da, lm);
cntxt.CopyToDevice(db, result);
//CudaDeviceVariable<int> dA = a;
//CudaDeviceVariable<int> dB = b;
//CudaDeviceVariable<int> dC = new CudaDeviceVariable<int>(N);
// Invoke kernel
//kernel.Run(dA.DevicePointer, dC.DevicePointer, dimX, dimY, dimZ);
matrixSumCude.Run(db, da, X, Y, Z);
cntxt.CopyToHost <double>(result, db);
return(result);
}
unsafe int Sum(CUdeviceptr array_ptr, int size)
{
if (size == 0)
{
return(0);
}
if (reductions_tmp == null || reductions_tmp.Length < size)
{
reductions_tmp = new int[size];
}
int nresult_alt = 0;
fixed(int *ptr = reductions_tmp)
{
IntPtr intptr = (IntPtr)ptr;
ctx.CopyToHost(intptr, array_ptr, (uint)(size * sizeof(int)));
for (int i = 0; i < size; i++)
{
nresult_alt += ptr[i];
}
return(nresult_alt);
}
}
public override int[] GetElementsAsInt(long index, int length)
{
CUdeviceptr ptr = DevicePtrAtElement(index);
if (ElementType == DType.Int32)
{
int[] result = new int[length]; context.CopyToHost(result, ptr); return(result);
}
else
{
throw new NotSupportedException("Element type " + ElementType + " not supported");
}
}
public uint[] Run()
{
var ptx = @"C:\Src\_Tree\SmallPrograms\Buddhabrot\Buddhabrot.Cuda70\x64\Release\Buddhabrot.ptx";
var context = new CudaContext();
var module = new CudaModuleHelper(context, ptx);
var init = module.GetKernel("Init");
var setSettings = module.GetKernel("SetSettings");
var runBuddha = module.GetKernel("RunBuddha");
var nBlocks = 4196;
var nThreads = 256;
var dSettings = context.AllocateMemoryFor(settings);
context.CopyToDevice(dSettings, settings);
var array = new uint[settings.Width * settings.Height];
var dState = context.AllocateMemory(nThreads * nBlocks * SizeOfCurandState);
var dArray = context.AllocateMemoryFor(array);
context.CopyToDevice(dArray, array);
init.Launch(nBlocks, nThreads, dState);
setSettings.Launch(1, 1, dSettings);
Console.WriteLine("Starting...");
var sw = Stopwatch.StartNew();
long i = 0;
while (!IsStopping)
{
runBuddha.Launch(nBlocks, nThreads, dArray, dState);
double count = (++i * nBlocks * nThreads);
if (i % 5 == 0)
{
Console.WriteLine("Generated {0:0.0} Million samples in {1:0.000} sec", count / 1000000.0, sw.ElapsedMilliseconds / 1000.0);
}
if (maxSamples.HasValue && count >= maxSamples)
break;
}
context.CopyToHost(array, dArray);
return array;
}
public void CopyToHost(float[] target)
{
_context.CopyToHost <float>(target, _ptr.DevicePointer);
}
public void CopyToHost(float[] target)
{
_context.CopyToHost <float>(target, DeviceVariable.DevicePointer);
}