/// <summary>Computes the Discrete Fourier Transform of a double2 vector x whose length is a power of 16.
/// x = { Re[x0] Im[x0] Re[x1] Im[x1] ... Re[xn] Im[xn] }, n = power of 16 (Length = 2*pow(16,n))</summary>
public static CLCalc.Program.Variable FFT16(ref CLCalc.Program.Variable CLx)
{
if (CLy == null || CLy.OriginalVarLength != CLx.OriginalVarLength) CLy = new CLCalc.Program.Variable(new float[CLx.OriginalVarLength]);
if (CLCalc.CLAcceleration != CLCalc.CLAccelerationType.UsingCL) return null;
int nn = (int)Math.Log(CLx.OriginalVarLength >> 1, 16);
nn = 1 << ((nn << 2) + 1);
if (nn != CLx.OriginalVarLength) throw new Exception("Number of elements should be a power of 16 ( vector length should be 2*pow(16,n) )");
if (kernelfft_radix16 == null)
{
InitKernels();
}
CLCalc.Program.Variable[] args = new CLCalc.Program.Variable[] { CLx, CLy, CLp };
CLCalc.Program.Variable[] args2 = new CLCalc.Program.Variable[] { CLy, CLx, CLp };
bool usar2 = true;
int[] p = new int[] { 1 };
CLp.WriteToDevice(p);
int n = CLx.OriginalVarLength >> 5;
while (p[0] <= n)
{
usar2 = !usar2;
if (usar2)
kernelfft_radix16.Execute(args2, n);
else
kernelfft_radix16.Execute(args, n);
p[0] = p[0] << 4;
CLp.WriteToDevice(p);
}
if (usar2)
{
CLCalc.Program.Variable temp = CLx;
CLx = CLy; CLy = temp;
}
return CLy;
}
/// <summary>Computes the inverse Discrete Fourier Transform of a float2 vector x whose length is a power of 4.
/// x = { Re[x0] Im[x0] Re[x1] Im[x1] ... Re[xn] Im[xn] }, n = power of 4 (Length = 2*pow(4,n))</summary>
public static CLCalc.Program.Variable iFFT4(CLCalc.Program.Variable CLx)
{
if (CLScale == null) CLScale = new CLCalc.Program.Variable(new float[1]);
//Trick: DFT-1 (x) = DFT(x*)*/N;
//Conjugate
float[] vx = new float[CLx.OriginalVarLength];
CLx.ReadFromDeviceTo(vx);
float[] scale = new float[] { 1 };
CLScale.WriteToDevice(scale);
kernelConjugate.Execute(new CLCalc.Program.Variable[] { CLx, CLScale }, CLx.OriginalVarLength >> 1);
CLx.ReadFromDeviceTo(vx);
CLy = FFT4(ref CLx);
scale[0] = 1 / (float)(CLx.OriginalVarLength >> 1);
CLScale.WriteToDevice(scale);
kernelConjugate.Execute(new CLCalc.Program.Variable[] { CLy, CLScale }, CLy.OriginalVarLength >> 1);
return CLy;
}
/// <summary>Calculates LU decomposition of M matrix</summary>
/// <param name="M">Matrix to decompose</param>
/// <param name="n">Matrix dimension</param>
/// <param name="varindx">Swap index</param>
private CLCalc.Program.Variable LUDecomp(double[,] M, int n, out CLCalc.Program.Variable varindx)
{
//arguments and work_dim
CLCalc.Program.Variable[] args;
int[] max;
//Matrix to vector
double[] vecM = MatrixToVector(M, ref n, ref n);
CLCalc.Program.Variable varM = new Program.Variable(vecM);
//Scaling transformation
double[] vv = new double[n];
CLCalc.Program.Variable varvv = new Program.Variable(vv);
max = new int[1] { n };
args = new CLCalc.Program.Variable[] { varM, varvv };
doubleLUScale.Execute(args, max);
//In order LU factorization (Crout)
int[] J = new int[1] { 0 };
CLCalc.Program.Variable varJ = new Program.Variable(J);
int[] N = new int[1] { n };
CLCalc.Program.Variable varN = new Program.Variable(N);
int[] indx = new int[n];
varindx = new Program.Variable(indx);
args = new Program.Variable[] { varM, varJ, varN, varindx, varvv };
for (J[0] = 0; J[0] < n; J[0]++)
{
varJ.WriteToDevice(J);
max[0] = J[0];
doubleLUCalcBetas.Execute(args, max);
max[0] = n - J[0];
doubleLUCalcAlphas.Execute(args, max);
max[0] = 1;
doubleLUCalcPivo.Execute(args, max);
max[0] = n;
doubleLUTrocaCols.Execute(args, max);
if (J[0] != n - 1)
{
max[0] = n - J[0] - 1;
doubleLUDivByPivot.Execute(args, max);
}
}
return varM;
}
/// <summary>Sets kernel arguments</summary>
/// <param name="Variables">Variables to be set as arguments</param>
private void SetArguments(CLCalc.Program.MemoryObject[] Variables)
{
//if (Variables.Length != nArgs) throw new Exception("Wrong number of arguments");
if (Vars != Variables)
{
Vars = Variables;
for (int i = 0; i < Variables.Length; i++)
{
Variables[i].SetAsArgument(i, kernel);
}
}
}
private CLCalc.Program.Variable LUBackSubstitute(CLCalc.Program.Variable MLUDecomp, double[] b, int n, CLCalc.Program.Variable varindx)
{
CLCalc.Program.Variable varx = new Program.Variable(b);
CLCalc.Program.Variable varN = new Program.Variable(new int[1] { n });
int[] J = new int[1];
CLCalc.Program.Variable varJ = new Program.Variable(J);
CLCalc.Program.Variable[] args = new Program.Variable[] { MLUDecomp, varx, varindx, varN, varJ };
int[] max = new int[1];
//ajeita o vetor com respeito as trocas de linha
max[0] = 1;
doubleLUUnscramble.Execute(args, max);
//Forward subst
for (int i = n - 1; i >= 1; i--)
{
max[0] = i;
doubleLUForwardSubs.Execute(args, max);
}
//Backward subst
for (int j = n - 1; j >= 1; j--)
{
max[0] = 1;
J[0] = j;
varJ.WriteToDevice(J);
doubleLUDivide.Execute(args, max);
max[0] = j;
doubleLUBackSubs.Execute(args, max);
}
//Primeiro elemento
max[0] = 1;
J[0] = 0;
varJ.WriteToDevice(J);
doubleLUDivide.Execute(args, max);
return varx;
}
/// <summary>Execute this kernel</summary>
/// <param name="Arguments">Arguments of the kernel function</param>
/// <param name="GlobalWorkSize">Array of maximum index arrays. Total work-items = product(max[i],i+0..n-1), n=max.Length</param>
/// <param name="LocalWorkSize">Local work sizes</param>
/// <param name="WorkOffSet">Global offset</param>
public void Execute(CLCalc.Program.MemoryObject[] Arguments, int[] GlobalWorkSize, int[] LocalWorkSize, int[] WorkOffSet)
{
Execute(CommQueues[DefaultCQ], Arguments, GlobalWorkSize, LocalWorkSize, WorkOffSet, null);
}
/// <summary>Execute this kernel</summary>
/// <param name="GlobalWorkSize">Array of maximum index arrays. Total work-items = product(max[i],i+0..n-1), n=max.Length</param>
/// <param name="LocalWorkSize">Local work sizes</param>
/// <param name="Arguments">Arguments of the kernel function</param>
/// <param name="events">Events list</param>
public void Execute(CLCalc.Program.MemoryObject[] Arguments, int[] GlobalWorkSize, int[] LocalWorkSize, ICollection<ComputeEventBase> events)
{
//CLEvent Event=new CLEvent();
Execute(CommQueues[DefaultCQ], Arguments, GlobalWorkSize, LocalWorkSize, events);
//OpenCLDriver.clReleaseEvent(Event);
}
/// <summary>Execute this kernel</summary>
/// <param name="CQ">Command queue to use</param>
/// <param name="Arguments">Arguments of the kernel function</param>
/// <param name="GlobalWorkSize">Array of maximum index arrays. Total work-items = product(max[i],i+0..n-1), n=max.Length</param>
/// <param name="events">Event of this command</param>
public void Execute(ComputeCommandQueue CQ, CLCalc.Program.MemoryObject[] Arguments, int[] GlobalWorkSize, ICollection<ComputeEventBase> events)
{
SetArguments(Arguments);
long[] globWSize=new long[GlobalWorkSize.Length];
for (int i=0;i<globWSize.Length;i++) globWSize[i]=GlobalWorkSize[i];
CQ.Execute(kernel, null, globWSize, null, events);
}
/// <summary>Execute this kernel</summary>
/// <param name="CQ">Command queue to use</param>
/// <param name="Arguments">Arguments of the kernel function</param>
/// <param name="GlobalWorkSize">Array of maximum index arrays. Total work-items = product(max[i],i+0..n-1), n=max.Length</param>
/// <param name="LocalWorkSize">Local work sizes</param>
/// <param name="events">Event of this command</param>
public void Execute(ComputeCommandQueue CQ, CLCalc.Program.MemoryObject[] Arguments, int[] GlobalWorkSize, int[] LocalWorkSize, ICollection<ComputeEventBase> events)
{
SetArguments(Arguments);
if (LocalWorkSize != null && GlobalWorkSize.Length != LocalWorkSize.Length) throw new Exception("Global and local work size must have same dimension");
long[] globWSize = new long[GlobalWorkSize.Length];
for (int i = 0; i < globWSize.Length; i++) globWSize[i] = GlobalWorkSize[i];
long[] locWSize = null;
if (LocalWorkSize != null)
{
locWSize = new long[LocalWorkSize.Length];
for (int i = 0; i < locWSize.Length; i++) locWSize[i] = LocalWorkSize[i];
}
CQ.Execute(kernel, null, globWSize, locWSize, events);
}
/// <summary>Classifies multiple samples stored in OpenCL memory</summary>
/// <param name="Samples">Samples data to classify</param>
/// <param name="svm">SVM to use as classifier</param>
public static float[] MultiClassify(SVM svm, CLCalc.Program.Image2D Samples)
{
float[] resp = new float[Samples.Height];
//svm.WriteToDevice();
if ((Samples.Width << 2) != svm.HostVLen[0]) throw new Exception("Invalid Samples width, should be the same length of training features");
if (svm.CLKernelValuesMultiClassify == null || svm.CLKernelValuesMultiClassify.OriginalVarLength != svm.alphaList.Count * Samples.Height)
{
svm.CLKernelValuesMultiClassify = new CLCalc.Program.Variable(new float[svm.alphaList.Count * Samples.Height]);
}
if (svm.CLAlphas == null || svm.CLAlphas.OriginalVarLength != svm.alphaList.Count)
{
svm.CLAlphas = new CLCalc.Program.Variable(svm.alphaList.ToArray());
float[] ys = new float[svm.TrainingSet.trainingArray.Count];
for (int i = 0; i < ys.Length; i++) ys[i] = svm.TrainingSet.trainingArray[i].y;
svm.CLys = new CLCalc.Program.Variable(ys);
}
if (svm.CLb==null)
{
svm.CLb = new CLCalc.Program.Variable(new float[] { svm.b });
svm.CLQtdSupVecs = new CLCalc.Program.Variable(new int[] { svm.alphaList.Count });
CLMultiClassifSums = new CLCalc.Program.Variable(new float[Samples.Height]);
}
if (CLMultiClassifSums.OriginalVarLength != Samples.Height)
{
CLMultiClassifSums = new CLCalc.Program.Variable(new float[Samples.Height]);
}
//svm.CLAlphas.WriteToDevice(svm.alphaList.ToArray());
//svm.CLys.WriteToDevice(ys);
//svm.CLb.WriteToDevice(new float[] { svm.b });
//svm.CLQtdSupVecs.WriteToDevice(new int[] { svm.alphaList.Count });
CLCalc.Program.MemoryObject[] args = new CLCalc.Program.MemoryObject[] { svm.CLTrainingFeatures, svm.CLQtdSupVecs, svm.CLXVecLen, Samples, svm.CLKernelValuesMultiClassify, svm.CLLambda };
kernelComputeMultiKernelRBF.Execute(args, new int[] { svm.alphaList.Count, Samples.Height });
CLCalc.Program.Sync();
args = new CLCalc.Program.MemoryObject[] { svm.CLAlphas, svm.CLQtdSupVecs, svm.CLXVecLen, svm.CLys, svm.CLKernelValuesMultiClassify, svm.CLb, CLMultiClassifSums };
kernelSumKernels.Execute(args, Samples.Height);
CLMultiClassifSums.ReadFromDeviceTo(resp);
return resp;
}
public Kernels(CLCalc.Program.Image2D CLGLRenderImage)
{
}
private void MassaMola(CLCalc.Program.Variable x, CLCalc.Program.Variable y, CLCalc.Program.Variable dydx)
{
CLCalc.Program.Variable[] args = new CLCalc.Program.Variable[3] { x, y, dydx };
KernelDerivs.Execute(args, nMassas);
}
public Kernels(CLCalc.Program.Image2D CLGLRenderImage)
{
this.CLimg = CLGLRenderImage;
dimensions = new int[CLimg.Width * CLimg.Height];
for (int y = 0; y < CLimg.Height; y++)
{
CLY.VarSize += 100;
}
for (int x = 0; x < CLimg.Width; x++)
{
CLX.VarSize += dimensions[CLimg.Height - 1 + CLimg.Height * x];
CLZ.VarSize += dimensions[CLimg.Height * x] = 240;
}
CLDimensions = new CLCalc.Program.Variable(dimensions);
}
private static void DoOCL(CLCalc.Program.Kernel VectorSum, CLCalc.Program.Variable[] args, int[] workers)
{
VectorSum.Execute(args, workers);
}
/// <summary>Creates a new isosurface calculator. You may pass variables created from a OpenGL context to the CL variables if you are using interop or NULL
/// if not using OpenCL/GL interop.</summary>
/// <param name="FuncValues">Values of the evaluated 3D function f(x,y,z). FuncValues=float[maxX,maxY,maxZ]</param>
/// <param name="CLEdgeCoords">OpenCL variable (float) to hold edge coordinates. Dimension has to be 9 * maxX * maxY * maxZ</param>
/// <param name="CLEdgeNormals">OpenCL variable (float) to hold edge normals. Dimension has to be 9 * maxX * maxY * maxZ</param>
/// <param name="CLElementArrayIndex">OpenCL variable (int) to hold element array index. Dimension has to be 5 * 3 * (maxX - 1) * (maxY - 1) * (maxZ - 1)</param>
private void InitMarchingCubes(float[, ,] FuncValues, CLCalc.Program.Variable CLEdgeCoords, CLCalc.Program.Variable CLEdgeNormals, CLCalc.Program.Variable CLElementArrayIndex)
{
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.Unknown) CLCalc.InitCL();
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.UsingCL)
{
//Reads maximum lengths
int maxX = FuncValues.GetLength(0);
int maxY = FuncValues.GetLength(1);
int maxZ = FuncValues.GetLength(2);
max = new int[] { maxX, maxY, maxZ };
#region Creating variables
//Isolevel
isoLevel = new float[1] { 1.32746E-5f };
varIsoLevel = new CLCalc.Program.Variable(isoLevel);
//Step size and x0,y0,z0
varStep = new CLCalc.Program.Variable(step);
varInitVals = new CLCalc.Program.Variable(initVals);
//Create and copy function values
funcVals = new float[maxX * maxY * maxZ];
CLFuncVals = new CLCalc.Program.Variable(funcVals);
SetFuncVals(FuncValues);
//Edge coordinates - 3 coords * 3 possible directions * number of points
edgeCoords = new float[9 * maxX * maxY * maxZ];
if (CLEdgeCoords != null)
{
varEdgeCoords = CLEdgeCoords;
varEdgeCoords.WriteToDevice(edgeCoords);
}
else varEdgeCoords = new CLCalc.Program.Variable(edgeCoords);
//4 preliminary normals per edge - has to be averaged afterwards
edgePrelimNormals = new float[36 * maxX * maxY * maxZ];
varEdgePrelimNormals = new CLCalc.Program.Variable(edgePrelimNormals);
//Edge normals
edgeNormals = new float[9 * maxX * maxY * maxZ];
if (CLEdgeNormals != null)
{
varEdgeNormals = CLEdgeNormals;
varEdgeNormals.WriteToDevice(edgeNormals);
}
else varEdgeNormals = new CLCalc.Program.Variable(edgeNormals);
//Number of cubes: (maxX-1)*(maxY-1)*(maxZ-1)
//Marching cube algorithm: each cube can have 5 triangles drawn, 3 vertexes per triangle
//q-th vertex of p-th triangle of the ijk-th cube: [(5*(i+(maxX-1)*j+k*(maxX-1)*(maxY-1))+p)*3+q]
elementIndex = new int[5 * 3 * (maxX - 1) * (maxY - 1) * (maxZ - 1)];
if (CLElementArrayIndex != null)
{
varElemIndex = CLElementArrayIndex;
varElemIndex.WriteToDevice(elementIndex);
}
else varElemIndex = new CLCalc.Program.Variable(elementIndex);
//Edge remapping to build output
edges = new int[edgeCoords.Length / 3];
for (int i = 0; i < edges.Length; i++) edges[i] = -1;
#endregion
#region Compile code and create kernels
CLMarchingCubesSrc cmsrc = new CLMarchingCubesSrc();
CLCalc.Program.Compile(new string[] { cmsrc.definitions, cmsrc.src });
kernelInterpPts = new CLCalc.Program.Kernel("interpPts");
kernelPolygonize = new CLCalc.Program.Kernel("Polygonize");
kernelSmoothNormals = new CLCalc.Program.Kernel("SmoothNormals");
kernelPolygonizeNoNormals = new CLCalc.Program.Kernel("PolygonizeNoNormals");
#endregion
}
else throw new Exception("OpenCL not available");
}
/// <summary>Execute this kernel using work_dim = 1</summary>
/// <param name="GlobalWorkSize">Global work size in one-dimension. global_work_size = new int[1] {GlobalWorkSize}</param>
/// <param name="Arguments">Arguments of the kernel function</param>
public void Execute(CLCalc.Program.MemoryObject[] Arguments, int GlobalWorkSize)
{
//CLEvent Event=new CLEvent();
Execute(CommQueues[DefaultCQ], Arguments, new int[] { GlobalWorkSize }, null, null);
//OpenCLDriver.clReleaseEvent(Event);
}
/// <summary>Creates a new isosurface calculator. You may pass variables created from a OpenGL context to the CL variables if you are using interop or NULL
/// if not using OpenCL/GL interop.</summary>
/// <param name="FuncValues">Values of the evaluated 3D function f(x,y,z). FuncValues=float[maxX,maxY,maxZ]</param>
/// <param name="CLEdgeCoords">OpenCL variable (float) to hold edge coordinates. Dimension has to be 9 * maxX * maxY * maxZ</param>
/// <param name="CLEdgeNormals">OpenCL variable (float) to hold edge normals. Dimension has to be 9 * maxX * maxY * maxZ</param>
/// <param name="CLElementArrayIndex">OpenCL variable (int) to hold element array index. Dimension has to be 5 * 3 * (maxX - 1) * (maxY - 1) * (maxZ - 1)</param>
public MarchingCubes(float[, ,] FuncValues, CLCalc.Program.Variable CLEdgeCoords, CLCalc.Program.Variable CLEdgeNormals, CLCalc.Program.Variable CLElementArrayIndex)
{
InitMarchingCubes(FuncValues, CLEdgeCoords, CLEdgeNormals, CLElementArrayIndex);
}
/// <summary>Execute this kernel</summary>
/// <param name="CQ">Command queue to use</param>
/// <param name="Arguments">Arguments of the kernel function</param>
/// <param name="GlobalWorkSize">Array of maximum index arrays. Total work-items = product(max[i],i+0..n-1), n=max.Length</param>
/// <param name="LocalWorkSize">Local work sizes</param>
/// <param name="events">Event of this command</param>
public void Execute(ComputeCommandQueue CQ, CLCalc.Program.MemoryObject[] Arguments, int[] GlobalWorkSize, int[] LocalWorkSize, ICollection<ComputeEventBase> events)
{
Execute(CQ, Arguments, GlobalWorkSize, LocalWorkSize, null, events);
}