////////////////////////////////////////////////////////////////////////////////////////////////////
/// <summary> Creates the kernel. </summary>
///
/// <seealso cref="M:KelpNet.Common.Functions.IParallelizable.CreateKernel()"/>
////////////////////////////////////////////////////////////////////////////////////////////////////
public void CreateKernel()
{
if (GpuEnable)
{
string kernelSource = ActivateFunctionString + ActivateKernelString;
ComputeProgram program = Weaver.CreateProgram(kernelSource);
ForwardKernel = program?.CreateKernel(ForwardKernelName);
BackwardKernel = program?.CreateKernel(BackwardKernelName);
}
}
public BlockedMatMulThread(ComputePlatform platform, ComputeDevice device, int inN, float[] inA, float[] inB)
{
ComputeDevice[] devices = { device };
context = new ComputeContext(devices, new ComputeContextPropertyList(platform), null, IntPtr.Zero);
program = new ComputeProgram(context, clProgramSource);
try
{
program.Build(null, null, null, IntPtr.Zero);
}
catch (BuildProgramFailureComputeException e)
{
string buildLog = program.GetBuildLog(device);
Console.WriteLine(buildLog);
throw;
}
kernel = program.CreateKernel("mmul");
queue = new ComputeCommandQueue(context, device, ComputeCommandQueueFlags.None);
n = inN;
a = new ComputeBuffer <float>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, inA);
b = new ComputeBuffer <float>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, inB);
int resultCount = (int)(device.MaxMemoryAllocationSize / n / n / sizeof(float));
results = new ComputeBuffer <float> [resultCount];
for (int i = 0; i < results.Length; ++i)
{
results[i] = new ComputeBuffer <float>(context, ComputeMemoryFlags.WriteOnly, n * n);
}
HaltRequested = false;
operationCounter = 0;
completionCounter = 0;
startTime = 0;
}
ComputeKernel CreateKernel(string Method, object[] args)
{
if (args == null)
{
throw new ArgumentException("You have to pass an argument to a kernel");
}
ComputeKernel kernel;
if (LastMethod == Method && LastKernel != null) //Kernel caching, do not compile twice
{
kernel = LastKernel;
}
else
{
kernel = program.CreateKernel(Method);
LastKernel = kernel;
}
LastMethod = Method;
for (int i = 0; i < args.Length; i++)
{
Setargument(kernel, i, args[i]);
}
return(kernel);
}
/// <summary>
/// 栅格化
/// </summary>
/// <param name="pointlist"></param>
/// <param name="W"></param>
/// <param name="H"></param>
/// <param name="T"></param>
/// <param name="program"></param>
/// <returns></returns>
public int[] Insert(int[] test1, int[] test2, ComputeProgram program)
{
int c1 = test1.GetLength(0);
int c2 = test2.GetLength(0);
ComputeBuffer <int> buffer1 = new ComputeBuffer <int>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, test1);
ComputeBuffer <int> buffer2 = new ComputeBuffer <int>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, test2);
ComputeBuffer <int> result = new ComputeBuffer <int>(context, ComputeMemoryFlags.ReadWrite, c1);
ComputeKernel kernel = program.CreateKernel("Insert");
kernel.SetMemoryArgument(0, buffer1);
kernel.SetMemoryArgument(1, buffer2);
kernel.SetMemoryArgument(2, result);
commands.Execute(kernel, null, new long[] { c1 }, null, events);
int[] resultnum = new int[c1];
GCHandle arrCHandle = GCHandle.Alloc(resultnum, GCHandleType.Pinned);
commands.Read(result, true, 0, c1, arrCHandle.AddrOfPinnedObject(), events);
arrCHandle.Free();
kernel.Dispose();
buffer1.Dispose();
buffer2.Dispose();
result.Dispose();
return(resultnum);
}
public TFP Hello(float num)
{
ComputeBuffer <TFP> result = new ComputeBuffer <TFP>(context, ComputeMemoryFlags.WriteOnly, 1);
string source = Encoding.ASCII.GetString(FZYK.Nest.Properties.Resources.nest);
if (fpType == FPType.FP64AMD)
{
source = "#define AMDFP64\n" + source;
}
else if (fpType == FPType.FP64)
{
source = "#define FP64\n" + source;
}
ComputeProgram program = new ComputeProgram(context, source);
try
{
program.Build(null, null, null, IntPtr.Zero);
}
catch (Exception)
{
var log = program.GetBuildLog(context.Devices[0]);
Debugger.Break();
}
ComputeKernel kernel = program.CreateKernel("hello");
TFP[] myresult = RunKernalTest(num, result, kernel);
return(myresult[0]);
}
public override void Init(WaveFormat AFormat)
{
base.Init(AFormat);
try
{
FDevice = App.Settings.Preferences.OpenCL.GetComputeDevice();
FContext = new ComputeContext(new ComputeDevice[] { FDevice }, new ComputeContextPropertyList(FDevice.Platform), null, IntPtr.Zero);
program = new ComputeProgram(FContext, FProgramSource);
program.Build(new[] { FDevice }, null, null, IntPtr.Zero);
kernel = program.CreateKernel("Wave");
commands = new ComputeCommandQueue(FContext, FContext.Devices[0], ComputeCommandQueueFlags.None);
}
catch (BuildProgramFailureComputeException bex)
{
Debug.WriteLine(bex.Message);
Debug.WriteLine(program.GetBuildLog(FDevice));
throw;
}
catch (Exception ex)
{
Debug.WriteLine(ex.Message);
throw;
}
}
public string vectorSum()
{
string vecSum = @"
__kernel void vectorSum(__global float *v1, __global float *v2, __global float *v3) {
int i = get_global_id(0);
v3[i] = v1[i] + v2[i];
}
";
int size = 100000;
float[] v1_ = new float[size];
float[] v2_ = new float[size];
float[] v3_ = new float[size];
for (var i = 0; i < size; i++)
{
v1_[i] = (float)i;
v2_[i] = (float).5f;
}
var platform_ = ComputePlatform.Platforms[0];
ComputeContextPropertyList properties = new ComputeContextPropertyList(platform_);
ComputeContext ctx = new ComputeContext(ComputeDeviceTypes.Gpu, properties, null, IntPtr.Zero);
ComputeCommandQueue commands = new ComputeCommandQueue(ctx, ctx.Devices[0], ComputeCommandQueueFlags.None);
ComputeProgram program = new ComputeProgram(ctx, vecSum);
try
{
program.Build(null, null, null, IntPtr.Zero);
Console.WriteLine("program build completed");
}
catch
{
string log = program.GetBuildLog(ctx.Devices[0]);
}
ComputeBuffer <float> v1, v2, v3;
v1 = new ComputeBuffer <float>(ctx, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, v1_);
v2 = new ComputeBuffer <float>(ctx, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, v2_);
v3 = new ComputeBuffer <float>(ctx, ComputeMemoryFlags.WriteOnly | ComputeMemoryFlags.CopyHostPointer, v3_);
long[] worker = { size };
commands.WriteToBuffer(v1_, v1, false, null);
commands.WriteToBuffer(v2_, v2, false, null);
ComputeKernel sumKernal = program.CreateKernel("vectorSum");
Console.WriteLine("kernal created");
sumKernal.SetMemoryArgument(0, v1);
sumKernal.SetMemoryArgument(1, v2);
sumKernal.SetMemoryArgument(2, v3);
commands.Execute(sumKernal, null, worker, null, null);
Console.WriteLine("Executed");
commands.ReadFromBuffer <float>(v3, ref v3_, false, null);
StringBuilder sb = new StringBuilder();
for (int i = 0; i < size; i++)
{
sb.AppendFormat("{0} + {1} = {2}<br>", v1_[i].ToString(), v2_[i].ToString(), v3_[i].ToString());
}
var sum_expression_result = sb.ToString();
return(sum_expression_result);
}
private void Initialize()
{
// get the intel integrated GPU
_integratedIntelGPUPlatform = ComputePlatform.Platforms.Where(n => n.Name.Contains("Intel")).First();
// create the compute context.
_context = new ComputeContext(
ComputeDeviceTypes.Gpu, // use the gpu
new ComputeContextPropertyList(_integratedIntelGPUPlatform), // use the intel openCL platform
null,
IntPtr.Zero);
// the command queue is the, well, queue of commands sent to the "device" (GPU)
_commandQueue = new ComputeCommandQueue(
_context, // the compute context
_context.Devices[0], // first device matching the context specifications
ComputeCommandQueueFlags.None); // no special flags
string kernelSource = null;
using (StreamReader sr = new StreamReader("kernel.cl"))
{
kernelSource = sr.ReadToEnd();
}
// create the "program"
_program = new ComputeProgram(_context, new string[] { kernelSource });
// compile.
_program.Build(null, null, null, IntPtr.Zero);
_kernel = _program.CreateKernel("ComputeMatrix");
}
public void SetDevice(int deviceIndx)
{
if ((deviceIndx < 0) || (deviceIndx >= oclDevices.Count))
{
throw new IndexOutOfRangeException("Invalid OpenCL device index.");
}
if (oclContext != null)
{
oclContext.Dispose();
oclContext = null;
}
if (oclCommandQueue != null)
{
oclCommandQueue.Dispose();
oclCommandQueue = null;
}
if (oclKernel != null)
{
oclKernel.Dispose();
oclKernel = null;
}
ComputeProgram oclProgram = null;
try
{
oclContext = new ComputeContext(new ComputeDevice[] { oclDevices[deviceIndx] },
new ComputeContextPropertyList(oclDevices[deviceIndx].Platform), null, IntPtr.Zero);
oclCommandQueue = new ComputeCommandQueue(oclContext, oclDevices[deviceIndx],
ComputeCommandQueueFlags.None);
oclProgram = new ComputeProgram(oclContext,
Encoding.Default.GetString(Properties.Resources.Test));
oclProgram.Build(new ComputeDevice[] { oclDevices[deviceIndx] }, "", null, IntPtr.Zero);
oclKernel = oclProgram.CreateKernel("Test");
}
catch (BuildProgramFailureComputeException ex)
{
string buildLog = oclProgram.GetBuildLog(oclDevices[deviceIndx]);
throw new Exception(buildLog, ex);
}
catch (Exception)
{
throw;
}
finally
{
if (oclProgram != null)
{
oclProgram.Dispose();
oclProgram = null;
}
}
}
// initialiseer OpenCL
static void InitCL()
{
// kies platform 0 (op sommige machines moet dit 1 of 2 zijn)
var platform = ComputePlatform.Platforms[CLPlatform];
Console.Write( "initializing OpenCL... " + platform.Name + " (" + platform.Profile + ").\n" );
Console.Write( platform.Devices.First().Name + " (" + platform.Devices.First().Type + ")\n");
Console.Write( (platform.Devices.First().GlobalMemorySize / 1024 / 1024) );
Console.WriteLine( " MiB global memory / " + (platform.Devices.First().LocalMemorySize / 1024) + " KiB local memory");
// maak een compute context
context = new ComputeContext( ComputeDeviceTypes.Gpu, new ComputeContextPropertyList( platform ), null, IntPtr.Zero );
// laad opencl programma
var streamReader = new StreamReader( "../../program.cl" );
string clSource = streamReader.ReadToEnd();
streamReader.Close();
// compileer opencl source code
program = new ComputeProgram( context, clSource );
try
{
program.Build( null, null, null, IntPtr.Zero );
}
catch
{
// fout in OpenCL code; check console window voor details.
Console.Write( "error in kernel code:\n" );
Console.Write( program.GetBuildLog( context.Devices[0] ) + "\n" );
}
// maak een commandorij
queue = new ComputeCommandQueue( context, context.Devices[0], 0 );
// lokaliseer de gewenste kernel in het programma
kernel = program.CreateKernel( "device_function" );
// alloceer data in RAM
}
/// <summary>
/// 栅格化范围数组
/// </summary>
/// <param name="pm"></param>
/// <param name="T"></param>
/// <param name="program"></param>
/// <returns></returns>
public int[,] GetGridArray(PlateModel pm, float T, ComputeProgram program)
{
List <PointF> pointlist = pm.OutModel.ExpandPoint;
float W = pm.Rect.Width;
float H = pm.Rect.Height;
int WI = Convert.ToInt32(Math.Ceiling(W / T));
int HI = Convert.ToInt32(Math.Ceiling(H / T));
int pc = pointlist.Count;
MyPoint[] plist = new MyPoint[pc];
for (int i = 0; i < pointlist.Count; i++)
{
plist[i].X = pointlist[i].X;
plist[i].Y = pointlist[i].Y;
}
ComputeBuffer <MyPoint> pbuffer = new ComputeBuffer <MyPoint>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, plist);
int[,] myresult = new int[HI, WI];
ComputeBuffer <int> outdata = new ComputeBuffer <int>(context, ComputeMemoryFlags.WriteOnly, WI * HI);
ComputeKernel kernelArray = program.CreateKernel("InitArray");
kernelArray.SetMemoryArgument(0, outdata);
commands.Execute(kernelArray, null, new long[] { WI *HI }, null, events);
ComputeKernel kernel = program.CreateKernel("GetGridValue");
kernel.SetMemoryArgument(0, pbuffer);
kernel.SetValueArgument(1, pc);
kernel.SetValueArgument(2, W);
kernel.SetValueArgument(3, H);
kernel.SetValueArgument(4, T);
kernel.SetValueArgument(5, WI);
kernel.SetValueArgument(6, HI);
kernel.SetMemoryArgument(7, outdata);
commands.Execute(kernel, null, new long[] { WI + HI }, null, events);
GCHandle arrCHandle = GCHandle.Alloc(myresult, GCHandleType.Pinned);
commands.Read(outdata, true, 0, WI * HI, arrCHandle.AddrOfPinnedObject(), events);
arrCHandle.Free();
kernel.Dispose();
pbuffer.Dispose();
outdata.Dispose();
return(myresult);
}
public FindPointOpenCl()
{
cpl = new ComputeContextPropertyList(ComputePlatform.Platforms[0]);
context = new ComputeContext(ComputeDeviceTypes.Gpu, cpl, null, IntPtr.Zero);
program = new ComputeProgram(context, new string[] { FindPointCalculationGpu });
program.Build(null, null, null, IntPtr.Zero);
kernel = program.CreateKernel("FindPointCalculationGpu");
}
public bool SetParallel(bool enable)
{
KernelSource = OpenCL.GetKernelSource(Resources.Dropout);
this.IsParallel = enable & OpenCL.Enable;
if (IsParallel)
{
ComputeProgram program = OpenCL.CreateProgram <T>(this.KernelSource);
ForwardKernel = program.CreateKernel("DropoutForward");
BackwardKernel = program.CreateKernel("DropoutBackward");
}
this.InitFunc(new StreamingContext());
return(IsParallel);
}
////////////////////////////////////////////////////////////////////////////////////////////////////
/// <summary> Creates the kernel. </summary>
///
/// <seealso cref="M:KelpNet.Common.Functions.IParallelizable.CreateKernel()"/>
////////////////////////////////////////////////////////////////////////////////////////////////////
public void CreateKernel()
{
if (GpuEnable)
{
string kernelSource = KernelString;
if (Activator != null)
{
kernelSource = Activator.ActivateFunctionString + KernelString;
}
ComputeProgram program = Weaver.CreateProgram(kernelSource);
ForwardKernel = program?.CreateKernel(ForwardKernelName);
BackwardgWKernel = program?.CreateKernel(BackwardgWKernelName);
BackwardgXKernel = program?.CreateKernel(BackwardgXKernelName);
}
}
public void CreateKernel()
{
if (this.GpuEnable)
{
string kernelSource = this.KernelString;
if (this.Activator != null)
{
kernelSource = this.Activator.ActivateFunctionString + this.KernelString;
}
ComputeProgram program = Weaver.CreateProgram(kernelSource);
this.ForwardKernel = program.CreateKernel(this.ForwardKernelName);
this.BackwardgWKernel = program.CreateKernel(this.BackwardgWKernelName);
this.BackwardgXKernel = program.CreateKernel(this.BackwardgXKernelName);
}
}
public static Tuple <List <List <int> >, TimeSpan> MultiplyParallel(List <List <int> > matrixOne, List <List <int> > matrixTwo)
{
if (!isRegularMatrix(matrixOne) || !isRegularMatrix(matrixTwo))
{
throw new ArgumentException("Non regular matrix detected. Rows size mismatch detected.");
}
if (matrixOne[0].Count != matrixTwo.Count)
{
throw new ArgumentException("Matrixes is not compatible. Columns count of first matrix is not equal to rows count of second matrix.");
}
List <List <int> > result = new List <List <int> >();
ComputePlatform platform = GetGPU();
if (platform is null)
{
throw new PlatformNotSupportedException("Platform doesn't have a dedicated GPU. Run is impossible.");
}
ComputeContext context = new ComputeContext(ComputeDeviceTypes.Gpu, new ComputeContextPropertyList(platform), null, IntPtr.Zero);
ComputeCommandQueue queue = new ComputeCommandQueue(context, context.Devices[0], ComputeCommandQueueFlags.None);
ComputeProgram program = new ComputeProgram(context, CalculateKernel);
program.Build(null, null, null, IntPtr.Zero);
ComputeKernel kernel = program.CreateKernel("Multiply");
List <ComputeBuffer <int> > rowsMatrixOne = matrixOne.TransformMatrixToComputerBuffersOfRows(context);
List <ComputeBuffer <int> > columnsMatrixTwo = matrixTwo.TransformMatrixToComputerBuffersOfColumns(context);
List <ComputeBuffer <int> > resultRowsMatrix = TwoDToOneDResult(matrixOne.Count, matrixTwo[0].Count, context);
Stopwatch stopwatch = new Stopwatch();
stopwatch.Start();
for (int i = 0; i < resultRowsMatrix.Count; ++i)
{
for (int j = 0; j < resultRowsMatrix[i].Count; ++j)
{
kernel.SetMemoryArgument(0, rowsMatrixOne[i]);
kernel.SetMemoryArgument(1, columnsMatrixTwo[j]);
kernel.SetMemoryArgument(2, resultRowsMatrix[i]);
kernel.SetValueArgument(3, matrixTwo.Count);
kernel.SetValueArgument(4, j);
queue.ExecuteTask(kernel, null);
}
}
queue.Finish();
stopwatch.Stop();
for (int i = 0; i < resultRowsMatrix.Count; ++i)
{
int[] res = new int[resultRowsMatrix[i].Count];
GCHandle gCHandle = GCHandle.Alloc(res, GCHandleType.Pinned);
queue.Read <int>(resultRowsMatrix[i], true, 0, res.Length, gCHandle.AddrOfPinnedObject(), null);
result.Add(new List <int>(res));
}
return(new Tuple <List <List <int> >, TimeSpan>(result, stopwatch.Elapsed));
}
public void CreateKernel()
{
if (this.GpuEnable)
{
string kernelSource = this.ActivateFunctionString;
foreach (var parameter in this.ActivationParameters)
{
kernelSource = this.ActivateFunctionString.Replace(parameter.Key, parameter.Value);
}
kernelSource += this.ActivateKernelString;
ComputeProgram program = Weaver.CreateProgram(kernelSource);
this.ForwardKernel = program.CreateKernel(this.ForwardKernelName);
this.BackwardKernel = program.CreateKernel(this.BackwardKernelName);
}
}
public async Task Can_multiply_using_pinned_host_pointer()
{
// This doesn't seem to work on NVidia platforms
var platform = ComputePlatform.Platforms.First();
var device = platform.Devices.First();
var context = new ComputeContext(new[] { device }, new ComputeContextPropertyList(platform), null,
IntPtr.Zero);
using var program = new ComputeProgram(context, Sum_kernel);
program.Build(new[] { device }, "-cl-std=CL1.2", (handle, ptr) => OnProgramBuilt(program, device),
IntPtr.Zero);
using var queue = new ComputeCommandQueue(context, device, ComputeCommandQueueFlags.None);
using var kernel = program.CreateKernel("sum");
var source = new byte[] { 1, 2, 3, 4, 5 };
var sourceHandle = GCHandle.Alloc(source, GCHandleType.Pinned);
var sourcePtr = sourceHandle.AddrOfPinnedObject();
using var sourceBuffer = new ComputeBuffer <byte>(
context,
ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.UseHostPointer,
source.LongLength,
sourcePtr);
var target = new byte[5];
var targetHandle = GCHandle.Alloc(target, GCHandleType.Pinned);
var targetPtr = targetHandle.AddrOfPinnedObject();
using var targetBuffer = new ComputeBuffer <byte>(
context,
ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.UseHostPointer,
target.LongLength,
targetPtr);
try
{
kernel.SetMemoryArgument(0, sourceBuffer);
kernel.SetMemoryArgument(1, targetBuffer);
var events = new List <ComputeEventBase>();
queue.Execute(kernel, null, new long[] { source.Length }, null, events);
//await events.WaitForEvents();
for (var i = 0; i < target.Length; i++)
{
Console.WriteLine($"{source[i]} * 2 = {target[i]}");
}
}
finally
{
sourceHandle.Free();
targetHandle.Free();
}
}
public async Task Can_use_struct()
{
var platform = ComputePlatform.Platforms.First();
var device = platform.Devices.First();
var context = new ComputeContext(new[] { device }, new ComputeContextPropertyList(platform), null, IntPtr.Zero);
using var program = new ComputeProgram(context, Struct_kernel);
try
{
program.Build(new[] { device }, "-cl-std=CL1.2", null, IntPtr.Zero);
}
catch (Exception ex)
{
OnProgramBuilt(program, device);
return;
}
using var queue = new ComputeCommandQueue(context, device, ComputeCommandQueueFlags.None);
using var kernel = program.CreateKernel("fill");
var target = new Target[10];
var gcHandle = GCHandle.Alloc(target, GCHandleType.Pinned);
var ptr = gcHandle.AddrOfPinnedObject();
var buffer = new ComputeBuffer <Target>(
context,
ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.UseHostPointer,
10,
ptr);
kernel.SetMemoryArgument(0, buffer);
var events = new List <ComputeEventBase>();
queue.Execute(kernel, null, new long[] { 10 }, null, events);
//await events.WaitForEvents();
unsafe
{
target[0].a.Should().Be(0);
target[0].b.Should().BeApproximately(0.5f, 0.01f);
target[0].c.Should().Be(1);
target[0].d[0].Should().Be(1);
target[0].d[1].Should().Be(2);
target[1].a.Should().Be(2);
target[1].b.Should().BeApproximately(1.5f, 0.01f);
target[1].c.Should().Be(2);
target[1].d[0].Should().Be(1);
target[1].d[1].Should().Be(2);
target[9].a.Should().Be(18);
target[9].b.Should().BeApproximately(9.5f, 0.01f);
target[9].c.Should().Be(10);
target[9].d[0].Should().Be(1);
target[9].d[1].Should().Be(2);
}
}
public static void CLBuilder(ComputeContext context)
{
UIConsole.Log("Loading kernel.cl");
var clProgramSource = File.ReadAllText("kernel.cl");
UIConsole.Log("Compiling kernel");
program = new ComputeProgram(context, clProgramSource);
try {
program.Build(null, null, null, IntPtr.Zero);
} catch (Exception e) {
UIConsole.Error("Build Log: \n" + program.GetBuildLog(context.Devices[0]));
throw e;
}
reprojectKernel = program.CreateKernel("reproject");
apply2DLUTKernel = program.CreateKernel("apply2DLUT");
applyCurveKernel = program.CreateKernel("applyCurve");
UIConsole.Log("Building Curve LUT");
curveLut = new byte[256];
for (int i = 0; i < 256; i++)
{
float v = 255 * OpenSatelliteProject.Presets.NEW_VIS_FALSE_CURVE[i];
curveLut[i] = (byte)(((int)Math.Floor(v)) & 0xFF);
}
curveLutBuffer = new ComputeBuffer <byte>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, curveLut);
UIConsole.Log("Loading LUT2D");
byte[] buffer = ReadFileFromOSPAssembly("falsecolor.png");
Bitmap lutBmp;
using (MemoryStream stream = new MemoryStream(buffer)) {
lutBmp = new Bitmap(stream);
}
lut2D = new uint[256 * 256];
for (int i = 0; i < 256; i++)
{
for (int j = 0; j < 256; j++)
{
lut2D[(i * 256) + j] = (uint)(lutBmp.GetPixel(j, i).ToArgb() & 0xFFFFFFFF);
}
}
lut2DBuffer = new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, lut2D);
}
public GPU()
{
context = new ComputeContext(ComputeDeviceTypes.Gpu, new ComputeContextPropertyList(ComputePlatform.Platforms[0]), null, IntPtr.Zero);
queue = new ComputeCommandQueue(context, context.Devices[0], ComputeCommandQueueFlags.None);
var program = new ComputeProgram(context, ArrayCopy);
program.Build(null, null, null, IntPtr.Zero);
kernel = program.CreateKernel("ArrayCopy");
}
private void InitClooApi()
{
try
{
CvInvoke.UseOpenCL = true;
// pick first platform
ComputePlatform platform = ComputePlatform.Platforms[1];
// create context with all gpu devices
clooCtx = new ComputeContext(ComputeDeviceTypes.Gpu,
new ComputeContextPropertyList(platform), null, IntPtr.Zero);
// load opencl source
StreamReader streamReader = new StreamReader(LASER_CL_PATH);
string clSource = streamReader.ReadToEnd();
streamReader.Close();
// build program.
ctxLaserCL = new ComputeProgram(clooCtx, clSource);
// compile opencl source
ctxLaserCL.Build(null, null, null, IntPtr.Zero);
// load chosen kernel from program
ctxMinMaxKernel = ctxLaserCL.CreateKernel("minMaxValues");
ctxMaskImageKernel = ctxLaserCL.CreateKernel("maskImage");
ctxCenterMassKernel = ctxLaserCL.CreateKernel("centerMass");
ctxTransform3DKernel = ctxLaserCL.CreateKernel("transformPixelsTo3D");
// create a command queue with first gpu found
queue = new ComputeCommandQueue(clooCtx,
clooCtx.Devices[0], ComputeCommandQueueFlags.None);
// execute kernel
events = new ComputeEventList();
} catch (Exception ex)
{
MessageBox.Show(ex.Message);
MessageBox.Show(ctxLaserCL.GetBuildLog(clooCtx.Devices[0]));
}
}
static void Main(string[] args)
{
int w = 11, h = 11, sx = 5, sy = 5, iters = 2;
var properties = new ComputeContextPropertyList(ComputePlatform.Platforms[0]);
var context = new ComputeContext(ComputeDeviceTypes.All, properties, null, IntPtr.Zero);
var quene = new ComputeCommandQueue(context, ComputePlatform.Platforms[0].Devices[0], ComputeCommandQueueFlags.None);
//Компиляция программы
var prog = new ComputeProgram(context, Source);
try
{
prog.Build(context.Devices, "", null, IntPtr.Zero);
}
catch
{
Console.WriteLine(prog.GetBuildLog(context.Devices[0]));
}
//Создание ядра
var kernel = prog.CreateKernel("Test");
var mat = new float[w * h];
for (int i = 0; i < w * h; i++)
{
mat[i] = (i == w * sy + sx) ? 1f : 0f;
}
var mat1 = new ComputeBuffer <float>(context, ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.UseHostPointer, mat);
var mat2 = new ComputeBuffer <float>(context, ComputeMemoryFlags.ReadWrite, w * h);
kernel.SetMemoryArgument(0, mat1);
kernel.SetMemoryArgument(1, mat2);
kernel.SetValueArgument(2, iters);
kernel.SetValueArgument(3, w);
kernel.SetValueArgument(4, h);
quene.Execute(kernel, null, new long[] { (long)h, (long)w }, null, null);
quene.ReadFromBuffer(mat1, ref mat, true, null);
for (int i = 0; i < h; i++)
{
for (int j = 0; j < w; j++)
{
Console.Write($"{mat[i*w+j]:.00} ");
}
Console.WriteLine();
}
Console.ReadKey();
}
private void InitializeCellGrid()
{
ComputeProgram prog = new ComputeProgram(Ctx, InitilizeGridKernelString);
prog.Build(Ctx.Devices, "", null, IntPtr.Zero);
InitilizeGridKernel = prog.CreateKernel("InitializeGrid");
InitilizeGridKernel.SetMemoryArgument(0, CurrImage);
InitilizeGridKernel.SetMemoryArgument(1, NextImage);
CQ = new ComputeCommandQueue(Ctx, Ctx.Devices[0], ComputeCommandQueueFlags.None);
CQ.Execute(InitilizeGridKernel, null, new long[] { Width, Height }, null, null);
}
static void Main(string[] args)
{
int[] r1 = new int[]
{ 8, 2, 3, 4 };
int[] r2 = new int[]
{ 4, 3, 2, 5 };
int[] r3 = new int[4];
int rowSize = r1.Length;
// pick first platform
ComputePlatform platform = ComputePlatform.Platforms[0];
// create context with all gpu devices
ComputeContext context = new ComputeContext(ComputeDeviceTypes.Gpu,
new ComputeContextPropertyList(platform), null, IntPtr.Zero);
// create a command queue with first gpu found
ComputeCommandQueue queue = new ComputeCommandQueue(context,
context.Devices[0], ComputeCommandQueueFlags.None);
// load opencl source and
// create program with opencl source
ComputeProgram program = new ComputeProgram(context, CalculateKernel);
// compile opencl source
program.Build(null, null, null, IntPtr.Zero);
// load chosen kernel from program
ComputeKernel kernel = program.CreateKernel("Calc");
// allocate a memory buffer with the message (the int array)
ComputeBuffer <int> row1Buffer = new ComputeBuffer <int>(context,
ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.UseHostPointer, r1);
// allocate a memory buffer with the message (the int array)
ComputeBuffer <int> row2Buffer = new ComputeBuffer <int>(context,
ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.UseHostPointer, r2);
// allocate a memory buffer with the message (the int array)
ComputeBuffer <int> resultBuffer = new ComputeBuffer <int>(context,
ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.UseHostPointer, new int[4]);
kernel.SetMemoryArgument(0, row1Buffer); // set the integer array
kernel.SetMemoryArgument(1, row2Buffer); // set the integer array
kernel.SetValueArgument(2, rowSize); // set the array size
kernel.SetMemoryArgument(3, resultBuffer); // set the integer array
// execute kernel
queue.ExecuteTask(kernel, null);
// wait for completion
queue.Finish();
GCHandle arrCHandle = GCHandle.Alloc(r3, GCHandleType.Pinned);
queue.Read <int>(resultBuffer, true, 0, r3.Length, arrCHandle.AddrOfPinnedObject(), null);
Console.WriteLine("display result from gpu buffer:");
for (int i = 0; i < r3.Length; i++)
{
Console.WriteLine(r3[i]);
}
arrCHandle.Free();
Console.WriteLine("Finished");
Console.ReadKey();
}
public void Can_copy()
{
var platform = ComputePlatform.Platforms.First();
var device = platform.Devices.First();
var context = new ComputeContext(new[] { device }, new ComputeContextPropertyList(platform), null, IntPtr.Zero);
using var program = new ComputeProgram(context, Copy_kernel);
try
{
program.Build(new[] { device }, "-cl-std=CL1.2", null, IntPtr.Zero);
}
catch (Exception ex)
{
OnProgramBuilt(program, device);
return;
}
using var queue = new ComputeCommandQueue(context, device, ComputeCommandQueueFlags.None);
using var kernel = program.CreateKernel("copy");
var sourcePath = Path.GetFullPath(Path.Combine(Environment.CurrentDirectory, "sample00.png"));
using var sourceImage = (Bitmap)Image.FromFile(sourcePath);
var w = sourceImage.Width;
var h = sourceImage.Height;
var source = sourceImage.ToBytes();
source.SaveFormatted("source.txt", w, h, channels: 4);
using var sourceBuffer = context.CreateImage2D(source, w, h);
var target = new byte[h * w * 4];
using var targetBuffer = context.CreateBuffer(target);
kernel.SetMemoryArgument(0, sourceBuffer);
kernel.SetMemoryArgument(1, targetBuffer);
queue.Execute(kernel, null, new long[] { w, h }, null, null);
queue.Finish();
target.SaveFormatted("target.txt", w, h, channels: 4);
var result = target.ToBitmap(w, h, 4);
var resultPath = Path.GetFullPath(Path.Combine(Environment.CurrentDirectory, "copy.png"));
result.Save(resultPath);
Run("source.txt");
Run("target.txt");
Run(resultPath);
}
public int ArrayTest(int[,] a1, int[,] b)
{
Point f;
int[,] a = new int[2, 2];
MyPoint[] plist = new MyPoint[100000];
MyPoint p1 = new MyPoint()
{
X = 1, Y = 1
};
MyPoint p2 = new MyPoint()
{
X = 2, Y = 2
};
plist[0] = p1;
plist[1] = p2;
IntPtr dataaddr_a = Marshal.UnsafeAddrOfPinnedArrayElement(plist, 0);
ComputeBuffer <MyPoint> abuffer = new ComputeBuffer <MyPoint>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, plist);
string source = Encoding.ASCII.GetString(FZYK.Nest.Properties.Resources.nest);
ComputeProgram program = new ComputeProgram(context, source);
try
{
program.Build(null, null, null, IntPtr.Zero);
}
catch (Exception)
{
var log = program.GetBuildLog(context.Devices[0]);
Debugger.Break();
}
ComputeKernel kernel = program.CreateKernel("Sum");
ComputeBuffer <MyPoint> outdata = new ComputeBuffer <MyPoint>(context, ComputeMemoryFlags.ReadWrite, plist.Length);
kernel.SetMemoryArgument(0, abuffer);
kernel.SetMemoryArgument(1, outdata);
commands.Execute(kernel, null, new long[] { 2 }, null, events);
MyPoint[] myresult = new MyPoint[2];
GCHandle arrCHandle = GCHandle.Alloc(myresult, GCHandleType.Pinned);
commands.Read(outdata, true, 0, 2, arrCHandle.AddrOfPinnedObject(), events);
MyPoint[] test = myresult;
arrCHandle.Free();
return(0);
}
////////////////////////////////////////////////////////////////////////////////////////////////////
/// <summary> Creates the kernel. </summary>
///
/// <seealso cref="M:KelpNet.Common.Functions.IParallelizable.CreateKernel()"/>
////////////////////////////////////////////////////////////////////////////////////////////////////
public void CreateKernel()
{
if (GpuEnable)
{
string kernelSource = KernelString;
if (Activator != null)
{
kernelSource = Activator.ActivateFunctionString + KernelString;
}
if (Verbose)
{
RILogManager.Default?.SendDebug("CreateKernel -> KernelSource is " + kernelSource);
}
ComputeProgram program = Weaver.CreateProgram(kernelSource);
ForwardKernel = program?.CreateKernel(ForwardKernelName);
BackwardgWKernel = program?.CreateKernel(BackwardgWKernelName);
BackwardgXKernel = program?.CreateKernel(BackwardgXKernelName);
}
}
private static ComputeKernel GetKernel(ComputeProgram program)
{
try
{
return(program.CreateKernel("place"));
}
catch
{
string log = program.GetBuildLog(program.Context.Platform.Devices[0]);
Console.WriteLine(log);
throw;
}
}
protected T[] InternalExecuteOpencl <T>(
String source,
String function,
int bufferSize,
ParallelTaskParams loaderParams,
params Object[] kernelParams)
where T : struct
{
TriggerCheckpoint(ParallelExecutionCheckpointType.CheckpointStart);
ComputeCommandQueue queue = QueueWithDevice(loaderParams.OpenCLDevice);
TriggerCheckpoint(ParallelExecutionCheckpointType.CheckpointPlatformInit);
String updatedSource = "#define OpenCL\r\n" + source;
ComputeProgram program = new ComputeProgram(queue.Context, updatedSource);
program.Build(new ComputeDevice[] { queue.Device }, null, null, IntPtr.Zero);
TriggerCheckpoint(ParallelExecutionCheckpointType.CheckpointKernelBuild);
T[] resultBuffer = new T[bufferSize];
ComputeBuffer <T> resultBufferVar = new ComputeBuffer <T>(queue.Context, ComputeMemoryFlags.WriteOnly, bufferSize);
List <ComputeMemory> vars = new List <ComputeMemory>();
vars.Add(resultBufferVar);
vars.AddRange(WrapDeviceVariables(kernelParams, queue.Context));
ComputeKernel kernel = program.CreateKernel(function);
for (int i = 0; i < vars.Count; i++)
{
kernel.SetMemoryArgument(i, vars[i]);
}
TriggerCheckpoint(ParallelExecutionCheckpointType.CheckpointDeviceWrite);
long[] workersGlobal = new long[2] {
loaderParams.GlobalWorkers.Width, loaderParams.GlobalWorkers.Height
};
queue.Execute(kernel, null, workersGlobal, null, null);
TriggerCheckpoint(ParallelExecutionCheckpointType.CheckpointKernelExecute);
queue.ReadFromBuffer <T>(resultBufferVar, ref resultBuffer, false, null);
TriggerCheckpoint(ParallelExecutionCheckpointType.CheckpointDeviceRead);
queue.Finish();
TriggerCheckpoint(ParallelExecutionCheckpointType.CheckpointPlatformDeinit);
return(resultBuffer);
}
