static void Main()
{
//Initializes OpenCL Platforms and Devices and sets everything up
CLCalc.InitCL();
//Create vectors with 2000 numbers
float[] v1 = new float[n], v2 = new float[n];
var vResult = new float[n];
//Creates population for v1 and v2
for (int i = 0; i < n; i++)
{
v1[i] = (float)i / 10;
v2[i] = -(float)i / 9;
}
//var prog = new ComputeProgram(CLCalc.Program.Context, "");
//Compiles the source codes. The source is a string array because the user may want
//to split the source into many strings.
CLCalc.Program.Compile(new string[] { vecSum });
//Gets host access to the OpenCL floatVectorSum kernel
CLCalc.Program.Kernel VectorSum = new CLCalc.Program.Kernel("floatVectorSum");
//Creates vectors v1 and v2 in the device memory
CLCalc.Program.Variable varV1 = new CLCalc.Program.Variable(v1);
CLCalc.Program.Variable varV2 = new CLCalc.Program.Variable(v2);
//Arguments of VectorSum kernel
CLCalc.Program.Variable[] args = new CLCalc.Program.Variable[] { varV1, varV2 };
//How many workers will there be? We need "n", one for each element
int[] workers = new int[1] { n };
sw.Start();
//Execute the kernel
for (int i = 0; i < count; i++) DoOCL(VectorSum, args, workers);
sw.Stop();
//Read device memory varV1 to host memory vResult
varV1.ReadFromDeviceTo(vResult);
Console.WriteLine("OpenCL: {0}", sw.ElapsedTicks);
sw.Restart();
for (int i = 0; i < count; i++) DoCPU(v1, v2, vResult);
sw.Stop();
Console.WriteLine("CPU: {0}", sw.ElapsedTicks);
PressAny();
}
/// <summary>Constructor. Initializes OpenCL</summary>
public CLMatrixMult()
{
if (CLCalc.GLAcceleration == CLCalc.GLAccelerationType.Unknown)
CLCalc.InitCL();
//Using OpenCL, declare variables
if (CLCalc.GLAcceleration == CLCalc.GLAccelerationType.UsingGL)
{
floatMatrixMultSource src = new floatMatrixMultSource();
CLCalc.Program.Compile(new string[] { src.matrixMultLocals, src.matrixMultNoLocals });
//multiplication
floatMatrixMultNoLocals = new CLCalc.Program.Kernel("floatMatrixMult");
floatMatrixMultLocals = new CLCalc.Program.Kernel("floatMatrixMultLocals");
}
}
/// <summary>Initializes class</summary>
private static void Init(int FilterSize)
{
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.Unknown) CLCalc.InitCL();
//Compiles source code
CLCalc.Program.Compile((new CLFilterSrc()).src);
//Creates kernel
kernelApplyFilter = new CLCalc.Program.Kernel("ApplyFilter");
kernelApplyFilterWorkDim2 = new CLCalc.Program.Kernel("ImgFilter");
//Creates filter
varFilter = new CLCalc.Program.Variable(new float[3 * FilterSize * FilterSize]);
//Width
varWidth = new CLCalc.Program.Variable(new int[1]);
Initialized = true;
}
/// <summary>Static Constructor. Builds kernels</summary>
static SparseLinalg()
{
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.Unknown)
{
try { CLCalc.InitCL(); }
catch
{
}
}
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.UsingCL)
{
//Kernel
CLLinalgSrc src = new CLLinalgSrc();
CLCalc.Program.Compile(new string[] { src.srcDotProd, src.srcMatVecMult, src.srcLinConjGrad });
kernelDotProduct = new CLCalc.Program.Kernel("dotProd");
kernelSum = new CLCalc.Program.Kernel("sumDotProd");
kernelGetDotSum = new CLCalc.Program.Kernel("GetResp");
kernelSparseMatrixVecMult = new CLCalc.Program.Kernel("SparseMatrixVecMult");
//Linear solving
kernelInitRP = new CLCalc.Program.Kernel("InitRP");
kernelMultiplyAdd = new CLCalc.Program.Kernel("MultiplyAdd");
kernelCopyToTemp = new CLCalc.Program.Kernel("CopyToTemp");
}
}
public OpenCLTest()
{
CLCalc.InitCL();
var buildlogs = new List<string>();
CLCalc.Program.Compile(new[] { run }, out buildlogs);
Random r = new Random();
float[] sample = Enumerable.Range(0, 768).Select(x => (float)r.NextDouble() * 2 - 1).ToArray();
float[] hlW = Enumerable.Range(0, 10000 * 769).Select(x => (float)r.NextDouble() * 2 - 1).ToArray();
float[] olW = Enumerable.Range(0, 10001 * 10).Select(x => (float)r.NextDouble() * 2 - 1).ToArray();
float[] hlDest = new float[10000];
float[] olDest = new float[10];
//float[] sample = new[] { .3f, .7f };
//float[] hlW = new[] { .4f, .4f, .1f, .5f, .5f, .7f };
//float[] olW = new[] { .4f, .6f, .8f };
//float[] hlDest = new[] { 0f, 0f };
//float[] olDest = new[] { 0f };
var sw = Stopwatch.StartNew();
var sw2 = Stopwatch.StartNew();
var gSample = sample.ToGraphics();
var ghlW = hlW.ToGraphics();
var golW = olW.ToGraphics();
var ghlDest = hlDest.ToGraphics();
var golDest = olDest.ToGraphics();
sw2.Stop();
var kernel = new CLCalc.Program.Kernel("runNN");
var args = new[] {gSample, new CLCalc.Program.Variable(new[] { sample.Length}), ghlDest,
new CLCalc.Program.Variable(new[] { hlDest.Length}), ghlW};
kernel.Execute(args, hlDest.Length);
args = new CLCalc.Program.Variable[5];
args[0] = ghlDest;
args[1] = new CLCalc.Program.Variable(new[] { hlDest.Length });
args[2] = golDest;
args[3] = new CLCalc.Program.Variable(new[] { olDest.Length });
args[4] = golW;
kernel.Execute(args, olDest.Length);
sw2.Start();
ghlDest.ReadFromDeviceTo(hlDest);
golDest.ReadFromDeviceTo(olDest);
sw2.Stop();
sw.Stop();
Console.WriteLine("hlDest\n" + string.Join("\n", hlDest));
Console.WriteLine("olDest\n" + string.Join("\n", olDest));
Console.WriteLine("Total {0}ms", sw.ElapsedMilliseconds);
Console.WriteLine("Memory {0}ms", sw2.ElapsedMilliseconds);
}
/// <summary>Constructor. Builds OpenCL program.</summary>
public doubleLinearAlgebra()
{
if (CLCalc.CLDevices == null)
{
try
{
CLCalc.InitCL();
}
catch
{
throw new Exception("Could not initialize OpenCL");
}
}
try
{
LinalgSource Source = new LinalgSource();
string[] s = new string[] { Source.dblInclude, Source.vecSum, Source.matrixMult,
Source.GaussSeidel, Source.LUDecomp };
CLCalc.Program.Compile(s);
//sum
doubleVecSum = new Program.Kernel("doubleVectorSum");
//multiplication
doubleMatrixMult = new Program.Kernel("doubleMatrixMult");
//Linear system (Gauss Seidel)
doubleGaussSeidel = new Program.Kernel("doubleGaussSeidel");
doubleGaussSeidelError = new Program.Kernel("doubleGaussSeidelError");
doubleCalcMtM = new Program.Kernel("doubleCalcMtM");
doubleCalcMtb = new Program.Kernel("doubleCalcMtb");
//Linear system (LU factorization)
doubleLUScale = new Program.Kernel("doubleLUScale");
doubleLUCalcBetas = new Program.Kernel("doubleLUCalcBetas");
doubleLUCalcAlphas = new Program.Kernel("doubleLUCalcAlphas");
doubleLUCalcPivo = new Program.Kernel("doubleLUCalcPivo");
doubleLUTrocaCols = new Program.Kernel("doubleLUTrocaCols");
doubleLUDivByPivot = new Program.Kernel("doubleLUDivByPivot");
doubleLUForwardSubs = new Program.Kernel("doubleLUForwardSubs");
doubleLUBackSubs = new Program.Kernel("doubleLUBackSubs");
doubleLUDivide = new Program.Kernel("doubleLUDivide");
doubleLUUnscramble = new Program.Kernel("doubleLUUnscramble");
doubleSolveError = new Program.Kernel("doubleSolveError");
doubleLUSubErr = new Program.Kernel("doubleLUSubErr");
}
catch
{
throw new Exception("Could not compile program");
}
}
/// <summary>Constructor.</summary>
/// <param name="nMasses">Number of masses in the system</param>
/// <param name="nConnections">Number of connections</param>
/// <param name="Masses">Mass of each vertex</param>
/// <param name="InitialStateSpace">Position and velocity of vertexes
/// [2*3*i] - posx, [2*(3*i+1)] - posy, [2*(3*i+2)] - posz,
/// [1+2*3*i] - velx, [1+2*(3*i+1)] - vely, [1+2*(3*i+2)] - velz</param>
/// <param name="Origins">Origin vertex of connections. Spring connects Origin[i] to Dests[i]</param>
/// <param name="Dests">Destination vertex of connections. Spring connects Origin[i] to Dests[i]</param>
/// <param name="SpringKs">Spring constant for each connection</param>
/// <param name="GroundKs">Spring constant for each mass, connecting to ground (nMass)</param>
/// <param name="Damp">Structural damping (relative-speed dependant) (nConnections)</param>
/// <param name="GroundDamp">Absolute damping proportional to speed relative to Earth (nMass)</param>
public floatDEM(int nMasses, int nConnections,
float[] Masses, float[] InitialStateSpace,
int[] Origins, int[] Dests, float[] SpringKs, float[] Damp, float[] GroundKs, float[] GroundDamp)
{
#region Consistency check
if (Masses.Length != nMasses)
throw new Exception("Invalid Masses length (!=nMasses)");
if (InitialStateSpace.Length != 6 * nMasses)
throw new Exception("Invalid positions length (!=6*nMasses - x, y, z)");
if (Origins.Length != nConnections)
throw new Exception("Invalid Origins length (!=nConnections)");
if (Dests.Length != nConnections)
throw new Exception("Invalid Dests length (!=nConnections)");
if (SpringKs.Length != nConnections)
throw new Exception("Invalid SpringKs length (!=nConnections)");
if (GroundKs.Length != nMasses)
throw new Exception("Invalid GroundKs length (!=nMasses)");
if (Damp.Length != nConnections)
throw new Exception("Invalid Damp length (!=nConnections)");
if (GroundDamp.Length != nMasses)
throw new Exception("Invalid GroundDamp length (!=nMasses)");
#endregion
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.Unknown)
{
CLCalc.InitCL();
}
#region Variables reading
//Sizes
nConn = new int[1] { nConnections };
nM = new int[1] { nMasses };
//Inputs
m = new Program.Variable(Masses);
float[] InitialPositions = new float[3 * nMasses];
for (int i = 0; i < 3 * nMasses; i++) InitialPositions[i] = InitialStateSpace[2 * i];
posOrig = new Program.Variable(InitialPositions);
origs = new Program.Variable(Origins);
dests = new Program.Variable(Dests);
k = new Program.Variable(SpringKs);
kGround = new Program.Variable(GroundKs);
c = new Program.Variable(Damp);
cGround = new Program.Variable(GroundDamp);
//Outputs
L0 = new Program.Variable(new float[nConnections]);
forces = new Program.Variable(new float[3 * nMasses]);
connForces = new Program.Variable(new float[3 * nConnections]);
nConnec = new Program.Variable(new int[1] { nConnections });
int[] nodesConnects = new int[30 * nMasses];
for (int i = 0; i < nodesConnects.Length; i++) nodesConnects[i] = -1;
nodesConnections = new Program.Variable(nodesConnects);
#endregion
#region Kernels initialization
DEMSource Source = new DEMSource();
string[] s = new string[] { Source.floatcalcL0, Source.floatresetForces, Source.floatcalcForces, Source.floatderivs, Source.floatcalcGroundForces, Source.floatcalcNodesConnections };
CLCalc.Program.Compile(s);
KernelcalcL0 = new Program.Kernel("floatcalcL0");
argscalcL0 = new Program.Variable[] { posOrig, origs, dests, L0 };
KernelcalcNodesConnections = new Program.Kernel("floatcalcNodesConnections");
argscalcNodesConnections = new Program.Variable[] { nodesConnections, nConnec, origs, dests };
KernelresetForces = new Program.Kernel("floatresetForces");
argsresetForces = new Program.Variable[] { forces };
KernelcalcForces = new Program.Kernel("floatcalcForces");
KernelcalcGroundForces = new Program.Kernel("floatcalcGroundForces");
Kernelderivs = new Program.Kernel("floatderivs");
#endregion
// Initial lengths calculation
KernelcalcL0.Execute(argscalcL0, nConn);
//Connections calculation
KernelcalcNodesConnections.Execute(argscalcNodesConnections, nM);
//nodesConnections.ReadFromDeviceTo(nodesConnects);
}
public float[] Add(float[] B, int startx, int starty, int startz, int Bx, int By)
{
float[] start = new float[3] { startx, starty, startz };
float[] size = new float[6] { X, Y, Z, B.GetLength(0), Bx, By };
float[] csResult = new float[X * Y * Z];
/*In order to avoid OOM, will start at the startz value, and pass
in one 2D matrix at a time.
* */
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.Unknown)
CLCalc.InitCL();
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.UsingCL)
{
CLSource source = new CLSource();
CLCalc.Program.Compile(new string[] { source.AddSubset_3string });
Matrix = new CLCalc.Program.Kernel("AddSubset3");
}
CLCalc.Program.Variable dev_A = new CLCalc.Program.Variable(M);
CLCalc.Program.Variable dev_B = new CLCalc.Program.Variable(B);
CLCalc.Program.Variable dev_start = new CLCalc.Program.Variable(start);
CLCalc.Program.Variable dev_size = new CLCalc.Program.Variable(size);
CLCalc.Program.Variable dev_Result = new CLCalc.Program.Variable(csResult);
CLCalc.Program.Variable[] args = new CLCalc.Program.Variable[5] { dev_A, dev_B, dev_start, dev_size, dev_Result };
Matrix.Execute(args, new int[] { X * Y * Z });
dev_Result.ReadFromDeviceTo(csResult);
M = csResult;
return csResult;
}
private static void InitKernels()
{
string s = new CLFFTSrc().s;
CLCalc.InitCL();
try
{
CLCalc.Program.Compile(s);
}
catch
{
}
kernelfft_radix16 = new CLCalc.Program.Kernel("fft_radix16");
kernelfft_radix4 = new CLCalc.Program.Kernel("fft_radix4");
kernelConjugate = new CLCalc.Program.Kernel("Conjugate");
CLp = new CLCalc.Program.Variable(new int[1]);
}
private void frmCLInfo_Load(object sender, EventArgs e)
{
CLCalc.InitCL(ComputeDeviceTypes.All);
if (CLCalc.CLAcceleration != CLCalc.CLAccelerationType.UsingCL)
{
cmbPlat.Items.Add("OpenCL ERROR");
if (cmbPlat.Items.Count > 0) cmbPlat.SelectedIndex = 0;
}
else
{
foreach(ComputePlatform p in CLCalc.CLPlatforms)
{
cmbPlat.Items.Add(p.Name + " " + p.Profile + " " + p.Vendor + " " + p.Version);
}
if (cmbPlat.Items.Count > 0) cmbPlat.SelectedIndex = 0;
int i=0;
foreach (ComputeDevice d in CLCalc.CLDevices)
{
//if (d.CLDeviceAvailable)
//{
cmbDevices.Items.Add(d.Name + " " + d.Type + " " + d.Vendor + " " + d.Version);
cmbCurDevice.Items.Add(d.Name + " " + d.Type + " " + d.Vendor + " " + d.Version);
//}
//else
//{
// cmbDevices.Items.Add("NOT AVAILABLE: " + d.CLDeviceName + " " + d.CLDeviceType + " " + d.CLDeviceVendor + " " + d.CLDeviceVersion);
// cmbCurDevice.Items.Add("NOT AVAILABLE: " + d.CLDeviceName + " " + d.CLDeviceType + " " + d.CLDeviceVendor + " " + d.CLDeviceVersion);
//}
i++;
}
if (cmbDevices.Items.Count > 0)
{
cmbDevices.SelectedIndex = 0;
cmbCurDevice.SelectedIndex = CLCalc.Program.DefaultCQ;
}
}
ReadImportantRegistryEntries();
//int[] n = new int[3] {1,1,1};
//int[] nn = new int[3];
//CLCalc.Program.Variable v = new CLCalc.Program.Variable(n);
//v.WriteToDevice(n);
//v.ReadFromDeviceTo(nn);
string s = @" kernel void teste() {}";
CLCalc.Program.Compile(s);
try
{
CLCalc.Program.Kernel k = new CLCalc.Program.Kernel("teste");
}
catch
{
MessageBox.Show("");
}
}
/// <summary>Compiles code and initializes kernel for this svm stance</summary>
private void CLSVMInit()
{
if (CLResource == null) CLResource = new int[0];
lock (CLResource)
{
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.Unknown) CLCalc.InitCL();
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.UsingCL)
{
if (kernelComputeKernelRBF == null)
{
CLSVMSrc s = new CLSVMSrc();
CLCalc.Program.Compile(new string[] { s.srcKernels, s.srcFindMaxMinErr, s.srcMultClass });
//Kernel computation
kernelComputeKernelRBF = new CLCalc.Program.Kernel("ComputeKernelRBF");
kernelMaxErr = new CLCalc.Program.Kernel("maxErr");
kernelComputeMax = new CLCalc.Program.Kernel("computeMax");
kernelMinErr = new CLCalc.Program.Kernel("minErr");
kernelComputeMin = new CLCalc.Program.Kernel("computeMin");
kernelGetResp = new CLCalc.Program.Kernel("getResp");
//Update error
kernelUpdateErr = new CLCalc.Program.Kernel("UpdateErr");
//Multiple classification
kernelComputeMultiKernelRBF = new CLCalc.Program.Kernel("ComputeMultiKernelRBF");
kernelSumKernels=new CLCalc.Program.Kernel("SumKernels");
}
//Memory obbjects
//Find max/min
CLErrLen = new CLCalc.Program.Variable(new int[1]);
HostResp = new int[1];
CLResp = new CLCalc.Program.Variable(HostResp);
CLMaxMinErrs = new CLCalc.Program.Variable(new float[MAXMINWORKSIZE]);
CLMaxMinInds = new CLCalc.Program.Variable(new int[MAXMINWORKSIZE]);
//Update error
CLUpdtErrParams = new CLCalc.Program.Variable(new float[3]);
}
}
}
private void button4_Click(object sender, EventArgs e)
{
nMassas = new int[1] { 10 };
//Integrador de EDO
float x0 = 0;
float[] y0 = new float[2 * nMassas[0]];
//y0[0]=posicao, y0[1] = velocidade
for (int i = 0; i < 2 * nMassas[0]; i += 2)
{
y0[i] = 1;
}
CLCalc.CLPrograms.floatODE46 ode46 = new CLCalc.CLPrograms.floatODE46(x0, 0.005f, y0, MassaMola);
//CLCalc.Program.DefaultCQ = 0;
//Retorno de derivadas
string[] s = new string[] {CLCalc.EnableDblSupport, floatDerivsMMola };
CLCalc.Program.Compile(s);
KernelDerivs = new CLCalc.Program.Kernel("floatDerivs");
System.Diagnostics.Stopwatch sw = new System.Diagnostics.Stopwatch();
sw.Start();
ode46.Integrate(20);
sw.Stop();
this.Text = sw.ElapsedMilliseconds.ToString();
Application.DoEvents();
float[] y = ode46.State;
ode46.Integrate(25);
y = ode46.State;
float[] yerr = ode46.AbsError;
float x = ode46.IndepVar;
}
private void button3_Click(object sender, EventArgs e)
{
float[] x = new float[] { 1, 2, 3, 0.123f };
float[] y = new float[] { 1, 2, 1, 1 };
string s = @"
kernel void
sum (global float4 * x, global float4 * y)
{
x[0] = x[0] + y[0];
}
";
CLCalc.Program.Compile(new string[] { s });
CLCalc.Program.Kernel sum = new CLCalc.Program.Kernel("sum");
CLCalc.Program.Variable varx=new CLCalc.Program.Variable(x);
CLCalc.Program.Variable vary=new CLCalc.Program.Variable(y);
CLCalc.Program.Variable[] args = { varx, vary };
int[] max = new int[] { 1 };
sum.Execute(args, max);
varx.ReadFromDeviceTo(x);
//float[] t = new float[1] { 0 };
//float[] pos = new float[] { 1, 2, 3, 4, 5, 6 };
//float[] vel = new float[] { 1, 0, 0, 0, 0, 0 };
//float[] forces = new float[] { 0, 1, 0, 0, 0, 0 };
//float[] masses = new float[] { 1, 1 };
//float[] colSizes = new float[] { 0.1f, 0.1f };
//CLCalc.InitCL();
//CLCalc.CLPrograms.floatBodyPhysics phys = new CLCalc.CLPrograms.floatBodyPhysics(10);
//CLCalc.CLPrograms.floatBodyPhysics phys2 = new CLCalc.CLPrograms.floatBodyPhysics(20);
//CLCalc.FinishCL();
}
private void button1_Click(object sender, EventArgs e)
{
CLCalc.InitCL();
double[] a = new double[] { 2, 147483647, 2, 7 };
double[] b = new double[] { 1, 2, 7, 4 };
double[] c = new double[4];
CLCalc.Program.Variable v1 = new CLCalc.Program.Variable(a);
CLCalc.Program.Variable v2 = new CLCalc.Program.Variable(b);
CLCalc.Program.Variable v3 = new CLCalc.Program.Variable(c);
CLCalc.CLPrograms.VectorSum VecSum = new CLCalc.CLPrograms.VectorSum();
CLCalc.CLPrograms.MinDifs Mdifs = new CLCalc.CLPrograms.MinDifs();
//string[] s = new string[] { VecSum.intVectorSum, VecSum.floatVectorSum };
string[] s = new string[] { VecSum.doubleVectorSum };
CLCalc.Program.Compile(s);
CLCalc.Program.Kernel k = new CLCalc.Program.Kernel("doubleVectorSum");
//CLCalc.Program.Kernel k2 = new CLCalc.Program.Kernel("intVectorSum");
//CLCalc.Program.Kernel k = new CLCalc.Program.Kernel("floatMinDifs");
CLCalc.Program.Variable[] vv = new CLCalc.Program.Variable[3] { v1, v2, v3 };
int[] max=new int[1] {a.Length};
k.Execute(vv, max);
CLCalc.Program.Sync();
v3.ReadFromDeviceTo(c);
CLCalc.FinishCL();
}
static Kernels()
{
try
{
CLCalc.Program.Compile(src);
CLCalc.Program.MemoryObject[] Args = new CLCalc.Program.MemoryObject[100]; ;
int globalWorkSize = 4;
// compile the kernels
KernelStart = new CLCalc.Program.Kernel("KernelStart");
coalesced = new CLCalc.Program.Kernel("coalesced");
// run kernel start
KernelStart.Execute(Args, globalWorkSize);
}
catch (NullReferenceException nre)
{
System.Console.WriteLine("" + nre);
}
// System.Diagnostics.Debug.WriteLine("Hello");
}
/// <summary>Constructor.</summary>
/// <param name="InitialState">Initial state of system</param>
/// <param name="StepSize">Desired step per integration pass</param>
/// <param name="InitialIndepVarValue">Initial independent variable value</param>
/// <param name="DerivativeCalculator">Function to calculate derivatives vector</param>
public doubleODE46(double InitialIndepVarValue, double StepSize, double[] InitialState, DerivCalcDeleg DerivativeCalculator)
{
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.Unknown)
{
CLCalc.InitCL();
}
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.NotUsingCL)
throw new Exception("OpenCL not available");
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.UsingCL)
{
ODE46Source Source = new ODE46Source();
string[] s = new string[] { @"
#pragma OPENCL EXTENSION cl_khr_fp64 : enable
", Source.doubleStep2, Source.doubleStep3, Source.doubleStep4, Source.doubleStep5, Source.doubleStep6, Source.doubleFinalizeCalc };
CLCalc.Program.Compile(s);
//Calculador de derivada
Derivs = DerivativeCalculator;
//Scalars
double[] xx = new double[1] { InitialIndepVarValue };
x = new CLCalc.Program.Variable(xx);
xsav = new CLCalc.Program.Variable(xx);
//Sets initial values to Device and local variables
hdid = new CLCalc.Program.Variable(xx);
currentX = InitialIndepVarValue;
SetStep(StepSize);
//Vectors
yy = new double[InitialState.Length];
for (int i = 0; i < InitialState.Length; i++) yy[i] = InitialState[i];
ysav = new CLCalc.Program.Variable(yy);
k1 = new CLCalc.Program.Variable(InitialState);
k2 = new CLCalc.Program.Variable(InitialState);
k3 = new CLCalc.Program.Variable(InitialState);
k4 = new CLCalc.Program.Variable(InitialState);
k5 = new CLCalc.Program.Variable(InitialState);
k6 = new CLCalc.Program.Variable(InitialState);
absError = new CLCalc.Program.Variable(new double[InitialState.Length]);
y = new CLCalc.Program.Variable(yy);
//Kernels
KernelFinalizeCalc = new CLCalc.Program.Kernel("doubleFinalizeCalc");
KernelUpdateX = new CLCalc.Program.Kernel("doubleUpdateX");
KernelRK46YStep2 = new CLCalc.Program.Kernel("doubleYStep2");
KernelRK46XStep2 = new CLCalc.Program.Kernel("doubleXStep2");
KernelRK46YStep3 = new CLCalc.Program.Kernel("doubleYStep3");
KernelRK46XStep3 = new CLCalc.Program.Kernel("doubleXStep3");
KernelRK46YStep4 = new CLCalc.Program.Kernel("doubleYStep4");
KernelRK46XStep4 = new CLCalc.Program.Kernel("doubleXStep4");
KernelRK46YStep5 = new CLCalc.Program.Kernel("doubleYStep5");
KernelRK46XStep5 = new CLCalc.Program.Kernel("doubleXStep5");
KernelRK46YStep6 = new CLCalc.Program.Kernel("doubleYStep6");
KernelRK46XStep6 = new CLCalc.Program.Kernel("doubleXStep6");
//Kernel arguments
ArgsFinalize = new CLCalc.Program.Variable[] { x, hdid, y, ysav, absError, k1, k2, k3, k4, k5, k6 };
ArgsRK46Y = new CLCalc.Program.Variable[] { x, hdid, y, ysav, k1, k2, k3, k4, k5, k6 };
ArgsRK46X = new CLCalc.Program.Variable[] { x, hdid, xsav };
NStates = new int[1] { InitialState.Length };
NScalar = new int[1] { 1 };
//Data retrieving
yerr = new double[NStates[0]];
xRet = new double[NScalar[0]];
}
}
/// <summary>Initializes physics program. Components indexes: [i] - x, [i+1] - y, [i+2] - z</summary>
/// <param name="nParticles">Number of particles</param>
public floatBodyPhysics(int nParticles)
{
string[] s = new string[] { CollisionAppliers, ForceAppliers, ConstAccelMotionEDOSolver };
Program.Compile(s);
//Kernels
MotionStep = new Program.Kernel("constAccelStep");
Kernel_ApplyGravity = new Program.Kernel("ApplyGravity");
Kernel_FloorCollision = new Program.Kernel("FloorCollision");
Kernel_SelfCollision = new Program.Kernel("SelfCollision");
Kernel_WallCollision = new Program.Kernel("WallCollision");
Kernel_ResetForces = new Program.Kernel("ResetForces");
Kernel_ResetCloseNeighbors = new Program.Kernel("ResetCloseNeighbors");
float[] t = new float[1] { 0 };
float[] gg = new float[3] { 0, 0, 0 };
step = new float[1] { 0 };
//Tamanho de alocacao de velocidades e posicoes
float[] aloc = new float[nParticles * 3];
//Tamanho de alocacao de caracteristicas das particulas
float[] alocPart = new float[nParticles];
//3*Nparticulas
CL_pos = new CLCalc.Program.Variable(aloc);
CL_vel = new CLCalc.Program.Variable(aloc);
CL_forces = new CLCalc.Program.Variable(aloc);
//Nparticulas
closeNeighbors = new int[nParticles];
CL_closeNeighbors = new CLCalc.Program.Variable(closeNeighbors);
for (int i = 0; i < nParticles; i++) alocPart[i] = 1f; //inicializa massas como 1 e tamanhos de colisao como 1
CL_masses = new CLCalc.Program.Variable(alocPart);
CL_collisionSizes = new CLCalc.Program.Variable(alocPart);
//escalares
CL_t = new CLCalc.Program.Variable(t);
CL_step = new CLCalc.Program.Variable(step);
//gravidade
CL_g = new CLCalc.Program.Variable(gg);
//Argumentos de funcoes
stepArgs = new CLCalc.Program.Variable[] { CL_t, CL_step, CL_forces, CL_masses, CL_pos, CL_vel };
applyGravArgs = new CLCalc.Program.Variable[] { CL_forces, CL_masses, CL_g };
floorCollisionArgs = new CLCalc.Program.Variable[] { CL_vel, CL_pos, CL_collisionSizes };
wallCollisionArgs = floorCollisionArgs;
selfCollisionArgs = new CLCalc.Program.Variable[] { CL_vel, CL_pos, CL_masses, CL_forces, CL_closeNeighbors, CL_collisionSizes };
resetForcesArgs = new Program.Variable[] { CL_forces };
resetCloseNeighborsArgs = new Program.Variable[] { CL_closeNeighbors };
nArgs = new int[1] { nParticles * 3 };
nPartics = new int[1] { nParticles };
nPartics2 = new int[2] { nParticles, nParticles };
}
/// <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>Initializes CL kernels</summary>
public static void Init()
{
if (kernelCholeskyDiagBlock == null)
{
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.Unknown) CLCalc.InitCL();
if (CLCalc.CLAcceleration == CLCalc.CLAccelerationType.UsingCL)
{
if (kernelCholeskyDiagBlock == null)
{
SUBMATRIXSIZE = (int)Math.Sqrt((double)CLCalc.Program.CommQueues[CLCalc.Program.DefaultCQ].Device.MaxWorkGroupSize);
SUBMATRIXSIZE = Math.Min(16, SUBMATRIXSIZE);
string strSubSize = SUBMATRIXSIZE.ToString();
string strTotSize = (SUBMATRIXSIZE * (SUBMATRIXSIZE + 1) / 2).ToString();
LinalgSrc src = new LinalgSrc();
string srcBlockChol = src.srcBlockCholesky.Replace("CONSTSUBMATRIXSIZE", strSubSize).Replace("CONSTGLOBALSIZE", strTotSize);
CLCalc.Program.Compile(new string[] { srcBlockChol, src.srcBkSubs, src.srcOperations, src.srcVecSum,
src.srcpNorm, src.strFeasibFunc, src.srcLogistReg });
kernelCholeskyDiagBlock = new CLCalc.Program.Kernel("CholeskyDiagBlock");
kernelCholeskyComputePanel = new CLCalc.Program.Kernel("CholeskyComputePanel");
kernelCholeskyForwardProp = new CLCalc.Program.Kernel("CholeskyForwardProp");
kernelFwdUpperBackSubs = new CLCalc.Program.Kernel("FwdUpperBackSubs");
kernelBkLowerBackSubs = new CLCalc.Program.Kernel("BkLowerBackSubs");
kernelFwdPropag = new CLCalc.Program.Kernel("FwdPropag");
kernelFwdPropag2 = new CLCalc.Program.Kernel("FwdPropag2");
kernelBackPropag = new CLCalc.Program.Kernel("BackPropag");
kernelBackPropag2 = new CLCalc.Program.Kernel("BackPropag2");
kernelInPlaceSubtract = new CLCalc.Program.Kernel("InPlaceSubtract");
kernelElemWiseAbs = new CLCalc.Program.Kernel("ElemWiseAbs");
kernelInnerProd = new CLCalc.Program.Kernel("InnerProd");
//Linear algebra
kernelSymMatrVecMultiply = new CLCalc.Program.Kernel("SymMatrVecMultiply");
kernelSymMatrMatrMultiply = new CLCalc.Program.Kernel("SymMatrMatrMultiply");
kernelComputeAtWA = new CLCalc.Program.Kernel("ComputeAtWA");
kernelComputeAinvHAt = new CLCalc.Program.Kernel("ComputeAinvHAt");
kernelRegularMatrTranspMatrProd = new CLCalc.Program.Kernel("RegularMatrTranspMatrProd");
kernelCopyBuffer = new CLCalc.Program.Kernel("CopyBuffer");
kernelLinearComb = new CLCalc.Program.Kernel("LinearComb");
kernelMatrVecProd = new CLCalc.Program.Kernel("MatrVecProd");
kernelTranspMatrVecProdW = new CLCalc.Program.Kernel("TranspMatrVecProdW");
kernelMatrVecProdSumVec = new CLCalc.Program.Kernel("MatrVecProdSumVec");
kernelDiagVecProd = new CLCalc.Program.Kernel("DiagVecProd");
kernelDiagTranspMatProd = new CLCalc.Program.Kernel("DiagTranspMatProd");
kernelElemWiseProd = new CLCalc.Program.Kernel("ElemWiseProd");
kernelElemWiseInv = new CLCalc.Program.Kernel("ElemWiseInv");
kernelElemWiseInv2 = new CLCalc.Program.Kernel("ElemWiseInv2");
kernelClear = new CLCalc.Program.Kernel("ClearResps");
kernelPreSum = new CLCalc.Program.Kernel("PreSum");
kernelCoalLocalSum = new CLCalc.Program.Kernel("CoalLocalSum");
kernelHasPositiveEntry = new CLCalc.Program.Kernel("HasPositiveEntry");
//pNorm minimization
floatOptimization.CurveFitting.kernelpNorm = new CLCalc.Program.Kernel("pNorm");
floatOptimization.CurveFitting.kerneldpNorm = new CLCalc.Program.Kernel("dpNorm");
//Logistic regression
floatOptimization.LogisticRegression.kernelComputeLogistRegParams = new CLCalc.Program.Kernel("ComputeLogistRegParams");
floatOptimization.LogisticRegression.kernelpNorm = floatOptimization.CurveFitting.kernelpNorm;
floatOptimization.LogisticRegression.kerneldpNorm = floatOptimization.CurveFitting.kerneldpNorm;
//Feasibility
floatOptimization.QuadraticProgramming.kernelgetLast = new CLCalc.Program.Kernel("getLast");
}
}
}
}
