public override void Proccess()
{
// execute kernel
queue.Execute(kernel, null, new long[] { DataGenerator.InputCount }, null, null);
queue.Finish();
/*
* short[] results2 = new short[this.results.Length];
* GCHandle arrCHandle = GCHandle.Alloc(results2, GCHandleType.Pinned);
* queue.Read(result_dev, true, 0, DataFeeder.GetInputCount(), arrCHandle.AddrOfPinnedObject(), events);
*/
//bool[] results2 = new bool[DataFeeder.GetInputCount()];
queue.ReadFromBuffer(result_dev, ref resultsBytes, true, null);
queue.ReadFromBuffer(resultCalc_dev, ref calculatables, true, null);
//queue.ReadFromBuffer()
/*
* bool[] arrC = new bool[5];
* GCHandle arrCHandle = GCHandle.Alloc(arrC, GCHandleType.Pinned);
* queue.Read<bool>(result_dev, true, 0, 5, arrCHandle.AddrOfPinnedObject(), null);
*/
// wait for completion
//queue.Finish();
//kernel.Dispose();
//queue.Dispose();
//context.Dispose();
}
public int[] ReadIntBuffer(string key, int length)
{
int[] rawBuffer = _intBuffers[key];
if (HardwareAccelerationEnabled)
{
_commands.ReadFromBuffer(_intComputeBuffers[key], ref rawBuffer, true, 0, 0, length, null);
_commands.Finish();
}
return(rawBuffer);
}
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);
}
/// <summary>
/// Renders the provided potential force field or gradient map.
/// </summary>
/// <param name="queue">The queue.</param>
/// <param name="pointsToRender">The points to render.</param>
/// <param name="computeBuffer">The compute buffer to use.</param>
/// <returns>A bitmap representing <paramref name="pointsToRender"/></returns>
private Bitmap RenderPoints(ComputeCommandQueue queue, float[] pointsToRender, ComputeBuffer <float> computeBuffer)
{
// Calculate the max and min values here, as it's easier than on the GPU
// and there's no real time cost
var max = pointsToRender.AsParallel().Max();
var min = pointsToRender.AsParallel().Min();
// Set up the colourKernel's arguments
colourKernel.SetValueArgument(0, max);
colourKernel.SetValueArgument(1, min);
colourKernel.SetMemoryArgument(2, computeBuffer);
colourKernel.SetMemoryArgument(3, outPix);
// Run the colourKernl
queue.Execute(colourKernel, new long[] { 0 }, new long[] { gridWidth, gridHeight },
new long[] { WorkGroupSize, WorkGroupSize }, null);
// and read out the data produced
queue.ReadFromBuffer(outPix, ref pixels, true, null);
// Copy the data into a new Bitmap
var bitmap = new Bitmap(gridWidth, gridHeight, PixelFormat.Format24bppRgb);
var info = bitmap.LockBits(new Rectangle(0, 0, gridWidth, gridHeight), ImageLockMode.ReadWrite,
PixelFormat.Format24bppRgb);
Marshal.Copy(pixels, 0, info.Scan0, pixels.Length);
bitmap.UnlockBits(info);
// Flip the bitmap to make the coordinate axis match those in the simulator
bitmap.RotateFlip(RotateFlipType.RotateNoneFlipY);
return(bitmap);
}
public T[] ExecuteReturn <T>(long Worksize, long WorkOffset, int OutSize) where T : struct
{
T[] Returned = new T[OutSize];
SetArgs();
if (WorkOffset != 0)
{
queue.Execute(kernel, new long[] { WorkOffset }, new long[] { Worksize }, null, null);
}
else
{
queue.Execute(kernel, null, new long[] { Worksize }, null, null);
}
for (int i = 0; i < MethodInfo.Arguments.Length; i++)
{
if (MethodInfo.Arguments[i].CopyBack)
{
queue.ReadFromBuffer((ComputeBuffer <T>)MethodInfo.Arguments[i].ComputeMemory, ref Returned, false, null);
}
}
queue.Finish();
return(Returned);
}
public uint FindProofOfWork(byte[] header, byte[] bits, uint nonceStart, uint iterations, out long elapsedMilliseconds)
{
this.stopwatch.Restart();
using var headerBuffer = new ComputeBuffer <byte>(this.computeContext, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, header);
using var bitsBuffer = new ComputeBuffer <byte>(this.computeContext, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, bits);
using var powBuffer = new ComputeBuffer <uint>(this.computeContext, ComputeMemoryFlags.WriteOnly, 1);
this.computeKernel.SetMemoryArgument(0, headerBuffer);
this.computeKernel.SetMemoryArgument(1, bitsBuffer);
this.computeKernel.SetValueArgument(2, nonceStart);
this.computeKernel.SetMemoryArgument(3, powBuffer);
using var commands = new ComputeCommandQueue(this.computeContext, this.computeDevice, ComputeCommandQueueFlags.None);
commands.Execute(this.computeKernel, null, new long[] { iterations }, null, null);
var nonceOut = new uint[1];
commands.ReadFromBuffer(powBuffer, ref nonceOut, true, null);
commands.Finish();
elapsedMilliseconds = this.stopwatch.ElapsedMilliseconds;
return(nonceOut[0]);
}
/// <summary>
/// Computes the field used when the robot posesses the ball
/// </summary>
/// <param name="queue">The queue to execute the kernel on.</param>
/// <param name="field">The calculated field points</param>
private void ComputePosessionField(ComputeCommandQueue queue, float[] field)
{
var ball =
new Vector2((float)currentEnvironment.CurrentBall.Position.X - currentEnvironment.FieldBounds.Left,
(float)currentEnvironment.CurrentBall.Position.Y - currentEnvironment.FieldBounds.Bottom
);
var goalTarget =
new Vector2((97.3632f) - currentEnvironment.FieldBounds.Left,
(33.932f + 49.6801f) / 2.0f - currentEnvironment.FieldBounds.Bottom);
// Collect together all the points that will repel the robot
var repulsers =
currentEnvironment.Opponents.Select(o => o.Position).Concat(
currentEnvironment.Home.Select(h => h.Position)).Select(
p =>
new Vector2((float)p.X - currentEnvironment.FieldBounds.Left,
(float)p.Y - currentEnvironment.FieldBounds.Bottom)).ToArray();
using (var inRepulsers = new ComputeBuffer <Vector2>(context,
ComputeMemoryFlags.ReadOnly |
ComputeMemoryFlags.CopyHostPointer, repulsers))
{
possessionKernel.SetValueArgument(0, ball);
possessionKernel.SetValueArgument(1, goalTarget);
possessionKernel.SetValueArgument(2, GridResolution);
possessionKernel.SetMemoryArgument(3, inRepulsers);
possessionKernel.SetMemoryArgument(4, outCl);
queue.Execute(possessionKernel, new long[] { 0 }, new long[] { gridWidth, gridHeight },
new long[] { WorkGroupSize, WorkGroupSize }, null);
queue.ReadFromBuffer(outCl, ref field, true, null);
}
}
public unsafe static void MatrixMulti_OpenCL(double[,] result, double[,] a, double[,] b)
{
InitCloo();
var ncols = result.GetUpperBound(0) + 1;
var nrows = result.GetUpperBound(1) + 1;
fixed(double *rp = result, ap = a, bp = b)
{
ComputeBuffer <double> aBuffer = new ComputeBuffer <double>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, a.Length, (IntPtr)ap);
ComputeBuffer <double> bBuffer = new ComputeBuffer <double>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, b.Length, (IntPtr)bp);
ComputeBuffer <double> rBuffer = new ComputeBuffer <double>(context, ComputeMemoryFlags.WriteOnly, result.Length);
kernel.SetMemoryArgument(0, aBuffer);
kernel.SetMemoryArgument(1, bBuffer);
kernel.SetValueArgument(2, ncols);
kernel.SetMemoryArgument(3, rBuffer);
ComputeEventList events = new ComputeEventList();
ComputeCommandQueue commands = new ComputeCommandQueue(context, context.Devices[0], ComputeCommandQueueFlags.None);
commands.Execute(kernel, null, new long[] { result.Length }, null, events);
commands.ReadFromBuffer(rBuffer, ref result, false, new SysIntX2(), new SysIntX2(), new SysIntX2(ncols, nrows), events);
commands.Finish();
}
}
/// <summary>
/// Finds the nonce for a block header hash that meets the given target.
/// </summary>
/// <param name="header">serialized block header</param>
/// <param name="bits">the target</param>
/// <param name="nonceStart">the first nonce value to try</param>
/// <param name="iterations">the number of iterations</param>
/// <returns></returns>
public uint FindPow(byte[] header, byte[] bits, uint nonceStart, uint iterations)
{
if (this.computeDevice == null)
{
throw new InvalidOperationException("GPU not found");
}
this.ConstructOpenCLResources();
using var headerBuffer = new ComputeBuffer <byte>(this.computeContext, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, header);
using var bitsBuffer = new ComputeBuffer <byte>(this.computeContext, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, bits);
using var powBuffer = new ComputeBuffer <uint>(this.computeContext, ComputeMemoryFlags.WriteOnly, 1);
this.computeKernel.SetMemoryArgument(0, headerBuffer);
this.computeKernel.SetMemoryArgument(1, bitsBuffer);
this.computeKernel.SetValueArgument(2, nonceStart);
this.computeKernel.SetMemoryArgument(3, powBuffer);
using var commands = new ComputeCommandQueue(this.computeContext, this.computeDevice, ComputeCommandQueueFlags.None);
commands.Execute(this.computeKernel, null, new long[] { iterations }, null, null);
var nonceOut = new uint[1];
commands.ReadFromBuffer(powBuffer, ref nonceOut, true, null);
commands.Finish();
this.DisposeOpenCLResources();
return(nonceOut[0]);
}
private double[] MatrixMultiply(Matrix <double> L, Matrix <double> R)
{
var L_array = L.To1D();
var R_array = R.To1D();
var O_array = new double[L.M * R.N];
ComputeBuffer <double> a = new ComputeBuffer <double>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, L_array);
ComputeBuffer <double> b = new ComputeBuffer <double>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, R_array);
ComputeBuffer <double> c = new ComputeBuffer <double>(context, ComputeMemoryFlags.WriteOnly, O_array.Length);
kernel.SetMemoryArgument(0, a);
kernel.SetMemoryArgument(1, b);
kernel.SetMemoryArgument(2, c);
kernel.SetValueArgument(3, L.N);
kernel.SetValueArgument(4, L.M);
kernel.SetValueArgument(5, R.N);
kernel.SetValueArgument(6, R.M);
commands.Execute(kernel, null, new long[] { R.N, L.M }, null, null);
commands.ReadFromBuffer(c, ref O_array, true, null);
commands.Finish();
// cleanup buffers
a.Dispose();
b.Dispose();
c.Dispose();
return(O_array);
}
/// <summary>
/// Computes the gradient of the field at all points.
/// </summary>
/// <param name="queue">The queue to execute the kernel on.</param>
/// <param name="gradPoints">The calculated gradient points.</param>
private void ComputeGradient(ComputeCommandQueue queue, float[] gradPoints)
{
gradientKernel.SetMemoryArgument(0, outCl);
gradientKernel.SetMemoryArgument(1, outGradient);
queue.Execute(gradientKernel, new long[] { 1, 1 }, new long[] { gridWidth - 2, gridHeight - 2 },
null, null);
queue.ReadFromBuffer(outGradient, ref gradPoints, true, null);
}
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();
}
public List <Vector2> CalculateSquareDistance(List <Vector2> positions)
{
// Create data
var xResultData = new float[positions.Count];
var yResultData = new float[positions.Count];
var xData = positions.Select(x => x.X).ToArray();
var yData = positions.Select(x => x.Y).ToArray();
// Put data on buffers
var xResultBuffer = new ComputeBuffer <float>(context, ComputeMemoryFlags.WriteOnly | ComputeMemoryFlags.CopyHostPointer, xResultData);
var yResultBuffer = new ComputeBuffer <float>(context, ComputeMemoryFlags.WriteOnly | ComputeMemoryFlags.CopyHostPointer, yResultData);
var xBuffer = new ComputeBuffer <float>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, xData);
var yBuffer = new ComputeBuffer <float>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, yData);
// Set memory arguments to kernel
kernel.SetMemoryArgument(0, xResultBuffer);
kernel.SetMemoryArgument(1, yResultBuffer);
kernel.SetMemoryArgument(2, xBuffer);
kernel.SetMemoryArgument(3, yBuffer);
// Create queue
var queue = new ComputeCommandQueue(context, device, ComputeCommandQueueFlags.None);
queue.Execute(kernel, null, new long[] { positions.Count }, null, null);
// Read data
queue.ReadFromBuffer(xResultBuffer, ref xResultData, true, null);
queue.ReadFromBuffer(yResultBuffer, ref yResultData, true, null);
var result = new List <Vector2>();
for (int i = 0; i < xResultData.Length; i++)
{
result.Add(new Vector2(xResultData[i], xResultData[i]));
}
return(result);
}
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);
}
public PrOpenClCalculation()
{
m_Platform = ComputePlatform.Platforms.First();
m_Device = m_Platform.Devices.First();
m_ComputeContext = new ComputeContext(new [] { m_Device }, new ComputeContextPropertyList(m_Platform), null, IntPtr.Zero);
m_CommandQueue = new ComputeCommandQueue(m_ComputeContext, m_Device, ComputeCommandQueueFlags.None);
m_Program = new ComputeProgram(m_ComputeContext, ProgramSource);
m_Program.Build(new [] { m_Device }, "", null, IntPtr.Zero);
m_Kernel = m_Program.CreateKernel("Test");
long count = 100;
var result = new int[count];
var a = new int[count];
for (int i = 0; i < count; i++)
{
a[i] = i;
}
var resultDev = new ComputeBuffer <int>(m_ComputeContext,
ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.UseHostPointer,
result);
var aDev = new ComputeBuffer <int>(m_ComputeContext,
ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.UseHostPointer,
a);
var bDev = new ComputeBuffer <int>(m_ComputeContext,
ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.UseHostPointer,
a);
//Задаем их для нашего ядра
m_Kernel.SetMemoryArgument(0, resultDev);
m_Kernel.SetMemoryArgument(1, aDev);
m_Kernel.SetMemoryArgument(2, bDev);
//Вызываем ядро количество потоков равно count
m_CommandQueue.Execute(m_Kernel, null, new[] { count }, null, null);
//Читаем результат из переменной
m_CommandQueue.ReadFromBuffer(resultDev, ref result, true, null);
//Выводим результат
foreach (var i in result)
{
Console.WriteLine(i);
}
}
public Bitmap GetScreenshot(CameraConfig camera, int screenshotHeight, int slowRender)
{
var screenshotWidth = (int)(screenshotHeight * ScreenshotAspectRatio);
var computeBuffer = new ComputeBuffer <Vector4>(_program.Context, ComputeMemoryFlags.ReadWrite, screenshotWidth * screenshotHeight);
var queue = new ComputeCommandQueue(_program.Context, _program.Context.Devices[0], ComputeCommandQueueFlags.None);
var globalSize = GlobalLaunchsizeFor(screenshotWidth, screenshotHeight);
for (var i = 0; i < slowRender; i++)
{
CoreRender(computeBuffer, queue, _kernels, new Vector4((Vector3)camera.Position), new Vector4((Vector3)camera.Lookat), new Vector4((Vector3)camera.Up), i, camera.Fov, slowRender, camera.FocalDistance, screenshotWidth, screenshotHeight, globalSize, _localSize);
}
for (var i = 0; i < camera.Frame * slowRender; i++)
{
CoreRender(computeBuffer, queue, _kernels, new Vector4((Vector3)camera.Position), new Vector4((Vector3)camera.Lookat), new Vector4((Vector3)camera.Up), i, camera.Fov, slowRender, camera.FocalDistance, screenshotWidth, screenshotHeight, globalSize, _localSize);
}
var pixels = new Vector4[screenshotWidth * screenshotHeight];
queue.ReadFromBuffer(computeBuffer, ref pixels, true, null);
queue.Finish();
computeBuffer.Dispose();
queue.Dispose();
var bmp = new Bitmap(screenshotWidth, screenshotHeight);
var destBuffer = new int[screenshotWidth * screenshotHeight];
for (var y = 0; y < screenshotHeight; y++)
{
for (var x = 0; x < screenshotWidth; x++)
{
var pixel = pixels[x + y * screenshotWidth];
if (float.IsNaN(pixel.X) || float.IsNaN(pixel.Y) || float.IsNaN(pixel.Z))
{
Console.WriteLine("Warning! Caught NAN pixel while taking screenshot!");
continue;
}
destBuffer[y * screenshotWidth + x] = (byte)(pixel.X * 255) << 16 | (byte)(pixel.Y * 255) << 8 | (byte)(pixel.Z * 255);
}
}
var bmpData = bmp.LockBits(new Rectangle(0, 0, screenshotWidth, screenshotHeight), ImageLockMode.ReadWrite, PixelFormat.Format32bppRgb);
Marshal.Copy(destBuffer, 0, bmpData.Scan0, destBuffer.Length);
bmp.UnlockBits(bmpData);
return(bmp);
}
private static void ConductSearch(ComputeContext context, ComputeKernel kernel)
{
var todos = GetQueenTaskPartition(NumQueens, 4);
var done = new List <QueenTask>();
ComputeEventList eventList = new ComputeEventList();
var commands = new ComputeCommandQueue(context, context.Devices[1], ComputeCommandQueueFlags.None);
Console.WriteLine("Starting {0} tasks, and working {1} at a time.", todos.Count, Spread);
QueenTask[] inProgress = GetNextAssignment(new QueenTask[] {}, todos, done);
var sw = new Stopwatch();
sw.Start();
while (inProgress.Any())
{
var taskBuffer =
new ComputeBuffer <QueenTask>(context,
ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.CopyHostPointer,
inProgress);
kernel.SetMemoryArgument(0, taskBuffer);
commands.WriteToBuffer(inProgress, taskBuffer, false, null);
for (int i = 0; i < 12; i++)
{
commands.Execute(kernel, null, new long[] { inProgress.Length }, null, eventList);
}
commands.ReadFromBuffer(taskBuffer, ref inProgress, false, eventList);
commands.Finish();
inProgress = GetNextAssignment(inProgress, todos, done);
}
sw.Stop();
Console.WriteLine(sw.ElapsedMilliseconds / 1000.0);
ulong sum = done.Select(state => state.solutions)
.Aggregate((total, next) => total + next);
Console.WriteLine("Q({0})={1}", NumQueens, sum);
}
public void ExecuteReturn <T>(int Worksize, T[] Returned) where T : struct
{
SetArgs();
queue.Execute(kernel, null, new long[] { Worksize }, null, null);
for (int i = 0; i < MethodInfo.Arguments.Length; i++)
{
var arg = MethodInfo.Arguments[i];
if (arg.CopyBack)
{
queue.ReadFromBuffer <T>((Cloo.ComputeBuffer <T>)MethodInfo.Arguments[i].ComputeMemory, ref Returned, false, null);
}
}
queue.Finish();
}
private static void ConductSearch(ComputeContext context, ComputeKernel kernel)
{
var todos = GetQueenTaskPartition(NumQueens, 4);
var done = new List<QueenTask>();
ComputeEventList eventList = new ComputeEventList();
var commands = new ComputeCommandQueue(context, context.Devices[1], ComputeCommandQueueFlags.None);
Console.WriteLine("Starting {0} tasks, and working {1} at a time.", todos.Count, Spread);
QueenTask[] inProgress = GetNextAssignment(new QueenTask[] {}, todos, done);
var sw = new Stopwatch();
sw.Start();
while (inProgress.Any())
{
var taskBuffer =
new ComputeBuffer<QueenTask>(context,
ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.CopyHostPointer,
inProgress);
kernel.SetMemoryArgument(0, taskBuffer);
commands.WriteToBuffer(inProgress, taskBuffer, false, null);
for (int i = 0; i < 12; i++)
commands.Execute(kernel, null, new long[] { inProgress.Length }, null, eventList);
commands.ReadFromBuffer(taskBuffer, ref inProgress, false, eventList);
commands.Finish();
inProgress = GetNextAssignment(inProgress, todos, done);
}
sw.Stop();
Console.WriteLine(sw.ElapsedMilliseconds / 1000.0);
ulong sum = done.Select(state => state.solutions)
.Aggregate((total, next) => total + next);
Console.WriteLine("Q({0})={1}", NumQueens, sum);
}
public override void RunKernel(ComputeContext context, ComputeKernel kernel, ComputeCommandQueue commands, long[] dimensions)
{
var tradeprofitsBuffer = new ComputeBuffer <short>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, TradeProfits);
var tradearbitrageBuffer = new ComputeBuffer <byte>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, TradeArbitrage);
var fit_functionBuffer = new ComputeBuffer <float>(context, ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.CopyHostPointer, FitFunction);
kernel.SetMemoryArgument(0, allvarsBuffer);
kernel.SetMemoryArgument(1, loboundsBuffer);
kernel.SetMemoryArgument(2, upboundsBuffer);
kernel.SetMemoryArgument(3, tradeprofitsBuffer);
kernel.SetMemoryArgument(4, tradearbitrageBuffer);
kernel.SetMemoryArgument(5, fit_functionBuffer);
kernel.SetValueArgument <int>(6, VariablesCount);
var eventList = new ComputeEventList();
commands.Execute(kernel, null, dimensions, null, eventList);
commands.ReadFromBuffer(fit_functionBuffer, ref FitFunction, true, null);
commands.Finish();
}
public T[] ExecuteReturn <T>(long Worksize, long?LocalWorksize, long WorkOffset, int OutSize) where T : struct
{
SetArgs();
if (LocalWorksize.HasValue)
{
queue.Execute(kernel, new long[] { WorkOffset }, new long[] { Worksize }, new long[] { LocalWorksize.Value }, null);
}
else
{
queue.Execute(kernel, new long[] { WorkOffset }, new long[] { Worksize }, null, null);
}
T[] Returned = new T[OutSize];
var copyBack = MethodInfo.Arguments.First(x => x.CopyBack);
queue.ReadFromBuffer((ComputeBuffer <T>)copyBack.ComputeMemory, ref Returned, true, null);
queue.Finish();
return(Returned);
}
public static void RunKernels(int nofKernels)
{
try
{
// Create the event wait list. An event list is not really needed for this example but it is important to see how it works.
// Note that events (like everything else) consume OpenCL resources and creating a lot of them may slow down execution.
// For this reason their use should be avoided if possible.
ComputeEventList eventList = new ComputeEventList();
commandQueue1.WriteToBuffer(input, CB_input, false, eventList);
commandQueue1.WriteToBuffer(weightIDs, CB_networkIndex, false, eventList);
// Execute the kernel "count" times. After this call returns, "eventList" will contain an event associated with this command.
// If eventList == null or typeof(eventList) == ReadOnlyCollection<ComputeEventBase>, a new event will not be created.
commandQueue1.Execute(kernel, null, new long[] { nofKernels }, null, eventList);
// Read back the results. If the command-queue has out-of-order execution enabled (default is off), ReadFromBuffer
// will not execute until any previous events in eventList (in our case only eventList[0]) are marked as complete
// by OpenCL. By default the command-queue will execute the commands in the same order as they are issued from the host.
// eventList will contain two events after this method returns.
commandQueue1.ReadFromBuffer(CB_output, ref output, false, eventList); // , eventList
// A blocking "ReadFromBuffer" (if 3rd argument is true) will wait for itself and any previous commands
// in the command queue or eventList to finish execution. Otherwise an explicit wait for all the opencl commands
// to finish has to be issued before "arrC" can be used.
// This explicit synchronization can be achieved in two ways:
// 1) Wait for the events in the list to finish,
eventList.Wait();
// 2) Or simply use
commandQueue1.Finish();
}
catch (Exception e)
{
Console.WriteLine(e.ToString());
}
}
private void Run(OpenCLInfo[] info, OpenCLPointInfo[] points, OpenCLNote[] envs, float[] result)
{
//var s = sw.ElapsedMilliseconds;
ComputeBuffer <OpenCLInfo> infoBuffer = new ComputeBuffer <OpenCLInfo>(FContext, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, info);
ComputeBuffer <OpenCLPointInfo> pointsBuffer = new ComputeBuffer <OpenCLPointInfo>(FContext, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, points);
ComputeBuffer <OpenCLNote> envsBuffer = new ComputeBuffer <OpenCLNote>(FContext, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, envs);
ComputeBuffer <float> resultBuffer = new ComputeBuffer <float>(FContext, ComputeMemoryFlags.WriteOnly, result.Length);
kernel.SetMemoryArgument(0, infoBuffer);
kernel.SetMemoryArgument(1, pointsBuffer);
// 2 - FWaveformBuffer
kernel.SetMemoryArgument(3, envsBuffer);
kernel.SetMemoryArgument(4, resultBuffer);
//var f = sw.ElapsedMilliseconds;
commands.Execute(kernel, null, new long[] { result.Length / 2 }, null, null);
//var e = sw.ElapsedMilliseconds;
commands.ReadFromBuffer(resultBuffer, ref result, true, null);
//var r = sw.ElapsedMilliseconds;
commands.Finish();
infoBuffer.Dispose();
pointsBuffer.Dispose();
envsBuffer.Dispose();
resultBuffer.Dispose();
//var d = sw.ElapsedMilliseconds;
//if (App.DebugMode)
// Debug.WriteLine($"OpenCLWaveProvider.Run: tot: {d - s}");
}
/// <summary>
/// Run kernel against all elements.
/// </summary>
/// <typeparam name="TSource">Struct type that corresponds to kernel function type</typeparam>
/// <param name="array">Array of elements to process</param>
/// <param name="kernelCode">The code of kernel function</param>
/// <param name="kernelSelector">Method that selects kernel by function name; if null uses first</param>
/// <param name="deviceSelector">Method that selects device by index, description, OpenCL version; if null uses first</param>
public static void ClooForEach <TSource>(this TSource[] array, string kernelCode, Func <string, bool> kernelSelector = null, Func <int, string, Version, bool> deviceSelector = null) where TSource : struct
{
kernelSelector = kernelSelector ?? ((k) => true);
deviceSelector = deviceSelector ?? ((i, d, v) => true);
var device = ComputePlatform.Platforms.SelectMany(p => p.Devices).Where((d, i) => deviceSelector(i, $"{d.Name} {d.DriverVersion}", d.Version)).First();
var properties = new ComputeContextPropertyList(device.Platform);
using (var context = new ComputeContext(new[] { device }, properties, null, IntPtr.Zero))
using (var program = new ComputeProgram(context, kernelCode))
{
program.Build(new[] { device }, null, null, IntPtr.Zero);
var kernels = program.CreateAllKernels().ToList();
try
{
var kernel = kernels.First((k) => kernelSelector(k.FunctionName));
using (var primesBuffer = new ComputeBuffer <TSource>(context, ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.UseHostPointer, array))
{
kernel.SetMemoryArgument(0, primesBuffer);
using (var queue = new ComputeCommandQueue(context, context.Devices[0], 0))
{
queue.Execute(kernel, null, new long[] { primesBuffer.Count }, null, null);
queue.Finish();
queue.ReadFromBuffer(primesBuffer, ref array, true, null);
}
}
}
finally
{
kernels.ForEach(k => k.Dispose());
}
}
}
/// <summary>
/// Executes the specified kernel function name.
/// </summary>
/// <typeparam name="TSource">The type of the source.</typeparam>
/// <param name="functionName">Name of the function.</param>
/// <param name="inputs">The inputs.</param>
/// <param name="returnInputVariable">The return result.</param>
/// <returns></returns>
/// <exception cref="ExecutionException">
/// </exception>
public override void Execute <TSource>(string functionName, params object[] args)
{
ComputeKernel kernel = _compiledKernels.FirstOrDefault(x => (x.FunctionName == functionName));
ComputeCommandQueue commands = new ComputeCommandQueue(_context, _defaultDevice, ComputeCommandQueueFlags.None);
if (kernel == null)
{
throw new ExecutionException(string.Format("Kernal function {0} not found", functionName));
}
try
{
var ndobject = (TSource[])args.FirstOrDefault(x => (x.GetType() == typeof(TSource[])));
long length = ndobject != null ? ndobject.Length : 1;
var buffers = BuildKernelArguments <TSource>(args, kernel, length);
commands.Execute(kernel, null, new long[] { length }, null, null);
foreach (var item in buffers)
{
TSource[] r = (TSource[])args[item.Key];
commands.ReadFromBuffer(item.Value, ref r, true, null);
//args[item.Key] = r;
item.Value.Dispose();
}
commands.Finish();
}
catch (Exception ex)
{
throw new ExecutionException(ex.Message);
}
finally
{
commands.Dispose();
}
}
public void CopyFromDevice()
{
_queue.ReadFromBuffer(_gpubuffer, ref _cpubuffer, true, null);
}
static void Main(string[] args)
{
#region
const string programName = "Prime Number";
Stopwatch stopWatch = new Stopwatch();
string clProgramSource = KernelProgram();
Console.WriteLine("Environment OS:");
Console.WriteLine("-----------------------------------------");
Console.WriteLine(Environment.OSVersion);
#endregion
if (ComputePlatform.Platforms.Count == 0)
{
Console.WriteLine("No OpenCL Platforms are availble!");
}
else
{
#region 1
// step 1 choose the first available platform
ComputePlatform platform = ComputePlatform.Platforms[0];
// output the basic info
BasicInfo(platform);
Console.WriteLine("Program: " + programName);
Console.WriteLine("-----------------------------------------");
#endregion
//Cpu 10 seconds Gpu 28 seconds
int count = 64;
int[] output_Z = new int[count * count * count];
int[] input_X = new int[count * count * count];
for (int x = 0; x < count * count * count; x++)
{
input_X[x] = x;
}
#region 2
// step 2 create context for that platform and all devices
ComputeContextPropertyList properties = new ComputeContextPropertyList(platform);
ComputeContext context = new ComputeContext(platform.Devices, properties, null, IntPtr.Zero);
// step 3 create and build program
ComputeProgram program = new ComputeProgram(context, clProgramSource);
program.Build(platform.Devices, null, null, IntPtr.Zero);
#endregion
// step 4 create memory objects
ComputeBuffer<int> a = new ComputeBuffer<int>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, input_X);
ComputeBuffer<int> z = new ComputeBuffer<int>(context, ComputeMemoryFlags.WriteOnly, output_Z.Length);
// step 5 create kernel object with same kernel programe name VectorAdd
ComputeKernel kernel = program.CreateKernel("PrimeNumber");
// step 6 set kernel arguments
//kernel.SetMemoryArgument(0, a);
kernel.SetMemoryArgument(0, a);
kernel.SetMemoryArgument(1, z);
ComputeEventList eventList = new ComputeEventList();
//for (int j = 0; j < context.Devices.Count; j++)
// query available devices n,...,1,0. cpu first then gpu
for (int j = context.Devices.Count-1; j > -1; j--)
{
#region 3
stopWatch.Start();
// step 7 create command queue on that context on that device
ComputeCommandQueue commands = new ComputeCommandQueue(context, context.Devices[j], ComputeCommandQueueFlags.None);
// step 8 run the kernel program
commands.Execute(kernel, null, new long[] { count, count, count }, null, eventList);
//Application.DoEvents();
#endregion
// step 9 read results
commands.ReadFromBuffer(z, ref output_Z, false, eventList);
#region 4
commands.Finish();
string fileName = "C:\\primenumber\\PrimeNumberGPU.txt";
StreamWriter file = new StreamWriter(fileName, true);
FileInfo info = new FileInfo(fileName);
long fs = info.Length;
// 1 MegaByte = 1.049e+6 Byte
int index = 1;
if (fs == 1.049e+6)
{
fileName = "C:\\primenumber\\PrimeNumberGPU" + index.ToString() + ".txt";
file = new System.IO.StreamWriter(fileName, true);
index++;
}
#endregion
for (uint xx = 0; xx < count * count * count; xx++)
{
if (output_Z[xx] != 0 && output_Z[xx] != 1)
{
Console.WriteLine(output_Z[xx]);
file.Write(output_Z[xx]);
file.Write("x");
}
}
#region 5
file.Close();
stopWatch.Stop();
ComputeCommandProfilingInfo start = ComputeCommandProfilingInfo.Started;
ComputeCommandProfilingInfo end = ComputeCommandProfilingInfo.Ended;
double time = 10e-9 * (end - start);
//Console.WriteLine("Nanosecond: " + time);
TimeSpan ts = stopWatch.Elapsed;
string elapsedTime = String.Format("{0:00}:{1:00}:{2:00}.{3:00}", ts.Hours, ts.Minutes, ts.Seconds, ts.Milliseconds);
Console.WriteLine(context.Devices[j].Name.Trim() + " Elapsed Time " + elapsedTime);
Console.WriteLine("-----------------------------------------");
#endregion
}
Console.ReadLine();
}
}
public void Run(ComputeContext context, TextWriter log)
{
try
{
// Create the arrays and fill them with random data.
int count = 10;
float[] arrA = new float[count];
float[] arrB = new float[count];
float[] arrC = new float[count];
Random rand = new Random();
for (int i = 0; i < count; i++)
{
arrA[i] = (float)(rand.NextDouble() * 100);
arrB[i] = (float)(rand.NextDouble() * 100);
}
// Create the input buffers and fill them with data from the arrays.
// Access modifiers should match those in a kernel.
// CopyHostPointer means the buffer should be filled with the data provided in the last argument.
ComputeBuffer<float> a = new ComputeBuffer<float>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, arrA);
ComputeBuffer<float> b = new ComputeBuffer<float>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, arrB);
// The output buffer doesn't need any data from the host. Only its size is specified (arrC.Length).
ComputeBuffer<float> c = new ComputeBuffer<float>(context, ComputeMemoryFlags.WriteOnly, arrC.Length);
// Create and build the opencl program.
program = new ComputeProgram(context, clProgramSource);
program.Build(null, null, null, IntPtr.Zero);
// Create the kernel function and set its arguments.
ComputeKernel kernel = program.CreateKernel("VectorAdd");
kernel.SetMemoryArgument(0, a);
kernel.SetMemoryArgument(1, b);
kernel.SetMemoryArgument(2, c);
// Create the event wait list. An event list is not really needed for this example but it is important to see how it works.
// Note that events (like everything else) consume OpenCL resources and creating a lot of them may slow down execution.
// For this reason their use should be avoided if possible.
ComputeEventList eventList = new ComputeEventList();
// Create the command queue. This is used to control kernel execution and manage read/write/copy operations.
ComputeCommandQueue commands = new ComputeCommandQueue(context, context.Devices[0], ComputeCommandQueueFlags.None);
// Execute the kernel "count" times. After this call returns, "eventList" will contain an event associated with this command.
// If eventList == null or typeof(eventList) == ReadOnlyCollection<ComputeEventBase>, a new event will not be created.
commands.Execute(kernel, null, new long[] { count }, null, eventList);
// Read back the results. If the command-queue has out-of-order execution enabled (default is off), ReadFromBuffer
// will not execute until any previous events in eventList (in our case only eventList[0]) are marked as complete
// by OpenCL. By default the command-queue will execute the commands in the same order as they are issued from the host.
// eventList will contain two events after this method returns.
commands.ReadFromBuffer(c, ref arrC, false, eventList);
// A blocking "ReadFromBuffer" (if 3rd argument is true) will wait for itself and any previous commands
// in the command queue or eventList to finish execution. Otherwise an explicit wait for all the opencl commands
// to finish has to be issued before "arrC" can be used.
// This explicit synchronization can be achieved in two ways:
// 1) Wait for the events in the list to finish,
//eventList.Wait();
// 2) Or simply use
commands.Finish();
// Print the results to a log/console.
for (int i = 0; i < count; i++)
log.WriteLine("{0} + {1} = {2}", arrA[i], arrB[i], arrC[i]);
// cleanup commands
commands.Dispose();
// cleanup events
foreach (ComputeEventBase eventBase in eventList)
{
eventBase.Dispose();
}
eventList.Clear();
// cleanup kernel
kernel.Dispose();
// cleanup program
program.Dispose();
// cleanup buffers
a.Dispose();
b.Dispose();
c.Dispose();
}
catch (Exception e)
{
log.WriteLine(e.ToString());
}
}
public unsafe void CreateCollisionCellArray()
{
time.Reset();
int num_of_bodies = Program.window.number_of_bodies;
uint num_of_elements = (uint)num_of_bodies * 8;
float ge = Program.window.grid_edge;
InitializeQueueAndContext();
InitializeComponents();
time.Start();
uint[] sortedCellIDArray = new uint[num_of_elements];
uint[] indexArrayIn = new uint[num_of_elements];
for (int j = 0; j < num_of_elements; j++)
{
indexArrayIn[j] = (uint)j;
}
uint[] indexArrayOut = new uint[num_of_elements];
GCHandle gch_iai = GCHandle.Alloc(indexArrayIn, GCHandleType.Pinned);
IntPtr ptr_i = gch_iai.AddrOfPinnedObject();
GCHandle gch_ge = GCHandle.Alloc(ge, GCHandleType.Pinned);
IntPtr ptr_ge = gch_ge.AddrOfPinnedObject();
array = new BodyData[num_of_bodies];
for (int i = 0; i < num_of_bodies; i++)
{
Body body = Program.window.bodies.ElementAt(i);
array[i].pos = new float[3];
array[i].cellIDs = new uint[8];
//array[i].ctrl_bits = 0;
array[i].pos[0] = body.getPos().X;
array[i].pos[1] = body.getPos().Y;
array[i].pos[2] = body.getPos().Z;
array[i].ID = (uint)i;
array[i].radius = body.getBSphere().radius;
}
structSize = Marshal.SizeOf(array[0]);
#region INITIALIZATION AND POPULATION OF DEVICE ARRAYS
IntPtr ptr = Marshal.AllocHGlobal(structSize * num_of_bodies);
for (int i = 0; i < num_of_bodies; i++)
{
Marshal.StructureToPtr(array[i], ptr + i * structSize, false);
}
byte[] input = new byte[structSize * num_of_bodies];
Marshal.Copy(ptr, input, 0, structSize * num_of_bodies);
b_start = time.ElapsedMilliseconds;
b_objectData = new ComputeBuffer <byte>
(context, ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.CopyHostPointer, input);
b_objectIDArray = new ComputeBuffer <ulong>
(context, ComputeMemoryFlags.ReadWrite, num_of_elements);
b_cellIDArray = new ComputeBuffer <uint>
(context, ComputeMemoryFlags.ReadWrite, num_of_elements);
b_gridEdge = new ComputeBuffer <float>
(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, 1, ptr_ge);
b_count += time.ElapsedMilliseconds - b_start;
k_dataInitialization.SetMemoryArgument(0, b_objectData);
k_dataInitialization.SetMemoryArgument(1, b_gridEdge);
k_dataInitialization.SetMemoryArgument(2, b_cellIDArray);
k_dataInitialization.SetMemoryArgument(3, b_objectIDArray);
dataInit_start = time.ElapsedMilliseconds;
queue.Execute(k_dataInitialization, null, new long[] { num_of_bodies }, null, null);
queue.Finish();
Console.WriteLine("TIME SPENT EXECUTIND DATA INITIALIZATION KERNEL: " + (time.ElapsedMilliseconds - dataInit_start) + "ms");
#endregion
#if PRINT
uint[] rs_array = new uint[num_of_bodies * 8];
queue.ReadFromBuffer <uint>(b_cellIDArray, ref rs_array, true, null);
queue.Finish();
#region READ DATA INITIALIZED IN DEVICE
array = new BodyData[num_of_bodies];
byte[] result = new byte[structSize * num_of_bodies];
IntPtr intPtr = Marshal.AllocHGlobal(structSize * num_of_bodies);
queue.ReadFromBuffer <byte>(b_objectData, ref result, true, null);
queue.Finish();
#endregion
#region CHECK CORRECTNESS
Marshal.Copy(result, 0, intPtr, structSize * num_of_bodies);
for (int i = 0; i < num_of_bodies; i++)
{
array[i] = (BodyData)Marshal.PtrToStructure(intPtr + i * structSize, typeof(BodyData));
}
checkCorrectness();
Marshal.FreeHGlobal(intPtr);
#endregion
#region CHECK OBJECT ID AND CELL ID ARRAYS
uint[] unsortedCellIDArray = new uint[num_of_elements];
ulong[] unsortedObjectIDArray = new ulong[num_of_elements];
queue.ReadFromBuffer <uint>(b_cellIDArray, ref unsortedCellIDArray, true, null);
queue.Finish();
string cellIDArrayLog = "";
for (int i = 0; i < num_of_elements; i++)
{
cellIDArrayLog += "[" + i + "]" + unsortedCellIDArray[i] + "\n";
}
File.WriteAllText(Application.StartupPath + @"\unsortedCellIDArrayLog.txt", cellIDArrayLog);
string objectIDArrayLog = "";
queue.ReadFromBuffer <ulong>(b_objectIDArray, ref unsortedObjectIDArray, true, null);
queue.Finish();
for (int i = 0; i < num_of_elements; i++)
{
objectIDArrayLog += "[" + i + "]" + (uint)unsortedObjectIDArray[i] + "\n";
}
File.WriteAllText(Application.StartupPath + @"\unsortedObjectIDArrayLog.txt", objectIDArrayLog);
#endregion
#endif
#if ALL
#region RADIX SORT
#region INITIALIZING RADIX SORT MEMBERS
uint block_count = (uint)Math.Ceiling((float)num_of_elements / BLOCK_SIZE);
uint lScanCount = block_count * (1 << DIGITS_PER_ROUND) / 4;
uint lScanSize = (uint)Math.Ceiling(((float)lScanCount / BLOCK_SIZE)) * BLOCK_SIZE;
uint globalSize = block_count * BLOCK_SIZE;
#endregion
#region POPULATING RADIX SORT BUFFERS
b_start = time.ElapsedMilliseconds;
b_blockScan = new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite, block_count * (1 << DIGITS_PER_ROUND));
b_blockOffset = new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite, block_count * (1 << DIGITS_PER_ROUND));
b_blockSum = new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite, BLOCK_SIZE);
b_temp = new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite, num_of_elements);
b_iArrayIn = new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite
| ComputeMemoryFlags.CopyHostPointer, num_of_elements, ptr_i);
b_iArrayOut = new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite, num_of_elements);
b_count += time.ElapsedMilliseconds - b_start;
#endregion
GCHandle gch_sc = GCHandle.Alloc(lScanCount, GCHandleType.Pinned);
IntPtr ptr_sc = gch_sc.AddrOfPinnedObject();
GCHandle gch_ec = GCHandle.Alloc(num_of_elements, GCHandleType.Pinned);
IntPtr ptr_ec = gch_ec.AddrOfPinnedObject();
b_start = time.ElapsedMilliseconds;
b_scanCount =
new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.CopyHostPointer, 1, ptr_sc);
b_numberOfElems =
new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.CopyHostPointer, 1, ptr_ec);
b_count += time.ElapsedMilliseconds - b_start;
#region libCL RADIX SORT ITERATION
for (uint j = 0; j < 32; j += DIGITS_PER_ROUND)
{
GCHandle gch_iter = GCHandle.Alloc(j, GCHandleType.Pinned);
IntPtr ptr_j = gch_iter.AddrOfPinnedObject();
b_start = time.ElapsedMilliseconds;
ComputeBuffer <uint> b_iteration =
new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, 1, ptr_j);
b_count += time.ElapsedMilliseconds - b_start;
#region BLOCK SORT
kernel_block_sort.SetMemoryArgument(0, b_cellIDArray);
kernel_block_sort.SetMemoryArgument(1, b_temp);
kernel_block_sort.SetMemoryArgument(2, b_iArrayIn);
kernel_block_sort.SetMemoryArgument(3, b_iArrayOut);
kernel_block_sort.SetMemoryArgument(4, b_iteration);
kernel_block_sort.SetMemoryArgument(5, b_blockScan);
kernel_block_sort.SetMemoryArgument(6, b_blockOffset);
kernel_block_sort.SetMemoryArgument(7, b_numberOfElems);
radixSort_start = time.ElapsedMilliseconds;
queue.Execute(kernel_block_sort, null, new long[] { globalSize }, new long[] { BLOCK_SIZE }, null);
queue.Finish();
radixSort_count += time.ElapsedMilliseconds - radixSort_start;
#endregion
#region BLOCK SCAN
kernel_block_scan.SetMemoryArgument(0, b_blockScan);
kernel_block_scan.SetMemoryArgument(1, b_blockSum);
kernel_block_scan.SetMemoryArgument(2, b_scanCount);
radixSort_start = time.ElapsedMilliseconds;
queue.Execute(kernel_block_scan, null, new long[] { lScanSize }, new long[] { BLOCK_SIZE }, null);
queue.Finish();
radixSort_count += time.ElapsedMilliseconds - radixSort_start;
#endregion
#region BLOCK PREFIX
kernel_block_prefix.SetMemoryArgument(0, b_blockScan);
kernel_block_prefix.SetMemoryArgument(1, b_blockSum);
kernel_block_prefix.SetMemoryArgument(2, b_scanCount);
radixSort_start = time.ElapsedMilliseconds;
queue.Execute(kernel_block_prefix, null, new long[] { lScanSize }, new long[] { BLOCK_SIZE }, null);
queue.Finish();
radixSort_count += time.ElapsedMilliseconds - radixSort_start;
#endregion
#region REORDER
kernel_reorder.SetMemoryArgument(0, b_temp);
kernel_reorder.SetMemoryArgument(1, b_cellIDArray);
kernel_reorder.SetMemoryArgument(2, b_iArrayOut);
kernel_reorder.SetMemoryArgument(3, b_iArrayIn);
kernel_reorder.SetMemoryArgument(4, b_blockScan);
kernel_reorder.SetMemoryArgument(5, b_blockOffset);
kernel_reorder.SetMemoryArgument(6, b_iteration);
kernel_reorder.SetMemoryArgument(7, b_numberOfElems);
radixSort_start = time.ElapsedMilliseconds;
queue.Execute(kernel_reorder, null, new long[] { globalSize }, new long[] { BLOCK_SIZE }, null);
queue.Finish();
radixSort_count += time.ElapsedMilliseconds - radixSort_start;
#endregion
b_iteration.Dispose();
gch_iter.Free();
}
#endregion
Console.WriteLine("TIME SPENT EXECUTING RADIX SORT: " + radixSort_count + " ms");
#endregion
#if PRINT
#region CHECK CELL ID ARRAY ORDER
queue.ReadFromBuffer <uint>(b_cellIDArray, ref sortedCellIDArray, true, null);
queue.Finish();
string orderedCellIDArrayLog = "";
for (int i = 0; i < num_of_elements; i++)
{
orderedCellIDArrayLog += "[" + i + "]" + sortedCellIDArray[i] + "\n";
}
File.WriteAllText(Application.StartupPath + @"\radixSortLog.txt", orderedCellIDArrayLog);
#endregion
#endif
#region REORDER OBJECT ID ARRAY
ulong[] o_array = new ulong[num_of_elements];
b_start = time.ElapsedMilliseconds;
b_reorder = new ComputeBuffer <ulong>(context, ComputeMemoryFlags.ReadWrite, num_of_elements);
b_count += time.ElapsedMilliseconds - b_start;
k_reorder.SetMemoryArgument(0, b_iArrayIn);
k_reorder.SetMemoryArgument(1, b_objectIDArray);
k_reorder.SetMemoryArgument(2, b_reorder);
reorder_start = time.ElapsedMilliseconds;
queue.Execute(k_reorder, null, new long[] { num_of_elements }, null, null);
queue.Finish();
Console.WriteLine("TIME SPENT REORDERING OBJECT ID ARRAY: " + (time.ElapsedMilliseconds - reorder_start) + " ms");
#endregion
#if PRINT
#region CHECK OBJECT ID ORDER
indexArrayOut = new uint[num_of_elements];
queue.ReadFromBuffer <uint>(b_iArrayIn, ref indexArrayOut, true, null);
queue.Finish();
queue.ReadFromBuffer <ulong>(b_reorder, ref o_array, true, null);
queue.Finish();
string orderedObjIDArray = "";
for (int i = 0; i < num_of_elements; i++)
{
orderedObjIDArray += "[" + i + "]" + (uint)o_array[i] + "\n";
}
string indexOut = "";
for (int i = 0; i < num_of_elements; i++)
{
indexOut += "[" + i + "]" + indexArrayOut[i] + "\n";
}
File.WriteAllText(Application.StartupPath + @"\reorderLog.txt", orderedObjIDArray);
File.WriteAllText(Application.StartupPath + @"\indexLog.txt", indexOut);
#endregion
#endif
#region ELEMENT COUNT
//511 -> max value a cell hash can be
//uint[] occurrences = new uint[512];
uint[] n_occurrences = new uint[512];
uint[] temp_array = new uint[num_of_elements];
uint[] temp_array2 = new uint[num_of_elements];
uint[] nocc = new uint[1];
b_start = time.ElapsedMilliseconds;
b_occPerRad =
new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite, 512);
b_temp2 =
new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite, num_of_elements);
b_flags =
new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite, num_of_elements);
b_numOfCC =
new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite, 1);
b_count = time.ElapsedMilliseconds - b_start;
k_elementCount.SetMemoryArgument(0, b_reorder);
k_elementCount.SetMemoryArgument(1, b_temp2);
k_elementCount.SetMemoryArgument(2, b_numOfCC);
k_elementCount.SetMemoryArgument(3, b_occPerRad);
k_elementCount.SetMemoryArgument(4, b_flags);
elemCount_start = time.ElapsedMilliseconds;
try{
queue.Execute(k_elementCount, null, new long[] { num_of_elements }, null, null);
}
catch (Exception e)
{
File.WriteAllText(Application.StartupPath + @"\exeLog.txt", e.Message);
}
queue.Finish();
Console.WriteLine("TIME SPENT EXECUTING ELEMENT COUNT: " + (time.ElapsedMilliseconds - elemCount_start) + " ms");
queue.ReadFromBuffer <uint>(b_temp2, ref temp_array, true, null);
queue.ReadFromBuffer <uint>(b_numOfCC, ref nocc, true, null);
queue.Finish();
#endregion
#if PRINT
#region CHECK ELEMENT COUNT
queue.ReadFromBuffer <uint>(b_occPerRad, ref n_occurrences, true, null);
queue.Finish();
string hs = "";
for (int h = 0; h < 512; h++)
{
hs += "[" + h + "] " +
//temp_array[h] + "\n\t" +
n_occurrences[h] + "\n";
}
File.WriteAllText(Application.StartupPath + @"\elementCountLog_nOcc.txt", hs);
Array.Sort <uint>(rs_array); //sort by .NET
string s = "";
for (int h = 0; h < num_of_elements; h++)
{
s += "[" + h + "] " +
//s += sortedCellIDArray[h] + "\n"; //OCL
//s += rs_array[h] + "\n"; //CPU
//s += unchecked((uint)o_array[h]) + "\n"; //OCL reordered array cell ID
//s += ((o_array[h] & ((ulong)281470681743360)) >> 32) + "\n"; //OCL reordered array obj ID
//s += (indexArrayOut[h] / 8) + "\n"; // cell index divided by 8 --> should be equal to the line above;
temp_array[h] + "\n";
}
File.WriteAllText(Application.StartupPath + @"\elementCountLog_temp.txt", s);
#endregion
#endif
#region PREFIX SUM
int maxIter = (int)Math.Log(num_of_elements, 2);
for (uint d = 0; d < maxIter; d++)
{
GCHandle gch_iteration = GCHandle.Alloc(d, GCHandleType.Pinned);
IntPtr ptr_iteration = gch_iteration.AddrOfPinnedObject();
b_start = time.ElapsedMilliseconds;
ComputeBuffer <uint> b_iteration2 =
new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.CopyHostPointer, 1, ptr_iteration);
b_count += time.ElapsedMilliseconds - b_start;
k_prefixSum.SetMemoryArgument(0, b_temp2);
k_prefixSum.SetMemoryArgument(1, b_iteration2);
prefixSum_start = time.ElapsedMilliseconds;
try
{
queue.Execute(k_prefixSum, null, new long[] { num_of_elements }, new long[] { num_of_elements }, null);
}
catch (Exception e)
{
File.WriteAllText(Application.StartupPath + @"\exeLog.txt", e.Message);
}
queue.Finish();
prefixSum_count += time.ElapsedMilliseconds - prefixSum_start;
b_iteration2.Dispose();
gch_iteration.Free();
}
Console.WriteLine("TIME SPENT EXECUTING PREFIX SUM: " + prefixSum_count + " ms");
queue.ReadFromBuffer <uint>(b_temp2, ref temp_array2, true, null);
#endregion
#if PRINT
#region CHECK SUM
string sscan = "";
for (int h = 0; h < num_of_elements; h++)
{
//TO BE FIXED! -- fixed?!
sscan += temp_array2[h] + "\n";
}
File.WriteAllText(Application.StartupPath + @"\scanLog.txt", sscan);
uint sum = 0;
for (int p = 0; p < num_of_elements; p++) //TODO CHANGE TO nocc[0] -- but why??
{
sum += temp_array[p];
}
Console.WriteLine(".NET sum: " + sum);
Console.WriteLine("OCL sum: " + nocc[0]);
#endregion
#endif
#region COLLISION CELL ARRAY CREATION
//uint[] outArray = new uint[nocc[0]];
ulong[] out_array = new ulong[nocc[0]];
uint[] indexes = new uint[nocc[0]];
b_ccArray = new ComputeBuffer <ulong>(context, ComputeMemoryFlags.ReadWrite, nocc[0]);//TODO !!!!! CHANGE TO nocc[0]
GCHandle gch_ta = GCHandle.Alloc(temp_array, GCHandleType.Pinned);
IntPtr ptr_ta = gch_ta.AddrOfPinnedObject();
b_start = time.ElapsedMilliseconds;
b_temp3 =
new ComputeBuffer <ulong>(context, ComputeMemoryFlags.ReadWrite | ComputeMemoryFlags.CopyHostPointer, num_of_elements, ptr_ta);
b_ccIndexes =
new ComputeBuffer <uint>(context, ComputeMemoryFlags.ReadWrite, nocc[0]);//TODO!!! CHANGE TO nocc[0]
b_count += time.ElapsedMilliseconds - b_start;
k_ccArrayCreation.SetMemoryArgument(0, b_reorder);
k_ccArrayCreation.SetMemoryArgument(1, b_ccArray);
k_ccArrayCreation.SetMemoryArgument(2, b_temp2);
k_ccArrayCreation.SetMemoryArgument(3, b_occPerRad);
k_ccArrayCreation.SetMemoryArgument(4, b_temp3);
k_ccArrayCreation.SetMemoryArgument(5, b_ccIndexes);
k_ccArrayCreation.SetMemoryArgument(6, b_flags);
ccArrayC_start = time.ElapsedMilliseconds;
try
{
queue.Execute(k_ccArrayCreation, null, new long[] { num_of_elements }, null, null);
}
catch (Exception e)
{
File.WriteAllText(Application.StartupPath + @"\exeLog.txt", e.Message);
}
queue.Finish();
Console.WriteLine("TIME SPENT POPULATING COLLISION CELL ARRAY: " + (time.ElapsedMilliseconds - ccArrayC_start) + " ms");
time.Stop();
#endregion
Console.WriteLine("TIME SPENT EXECUTING BROAD-PHASE COLLISION DETECTION: " + time.ElapsedMilliseconds + " ms");
Console.WriteLine("TIME SPENT CREATING DEVICE BUFFERS: " + b_count + " ms");
#if PRINT
#region CHECK RESULT
queue.ReadFromBuffer(b_ccArray, ref out_array, true, null);
queue.ReadFromBuffer(b_ccIndexes, ref indexes, true, null);
queue.Finish();
string output = "";
for (int t = 0; t < nocc[0]; t++)//CHANGE TO nocc[0]
{
output += "---INDEX--- " + t + "\n\t";
if ((out_array[t] & ((ulong)1 << 63)) != (ulong)0)
{
output += "[H] ";
}
output += (uint)out_array[t] + "\n\t";
output += indexes[t] + "\n";
}
File.WriteAllText(Application.StartupPath + @"\outputLog.txt", output);
#endregion
#endif
#region POINTERS DISPOSAL
Marshal.FreeHGlobal(ptr);
gch_ta.Free();
gch_ec.Free();
gch_sc.Free();
gch_ge.Free();
#endregion
try
{
DisposeBuffers();
DisposeComponents();
DisposeQueueAndContext();
}
catch (Exception e) {
Console.WriteLine("Error encountered while releasing buffers - " + e.Message);
File.WriteAllText(Application.StartupPath + @"\exeLog.txt", e.Message);
}
#else
Marshal.FreeHGlobal(ptr);
gch_iai.Free();
gch_ge.Free();
b_objectData.Dispose();
b_objectIDArray.Dispose();
b_cellIDArray.Dispose();
b_gridEdge.Dispose();
#endif
}
public static void Calculate(List<Calculation> calculations)
{
Stopwatch s = new Stopwatch();
s.Start();
int count = calculations.Count;
IntVec2[] p_p = new IntVec2[count];
IntVec2[] p_a = new IntVec2[count];
IntVec2[] p_b = new IntVec2[count];
IntVec2[] p_c = new IntVec2[count];
FloatVec3[] c = new FloatVec3[count];
int[] c_valid = new int[count];
Parallel.For(0, count, i =>
{
var calc = calculations[i];
p_p[i] = new IntVec2(calc.P);
p_a[i] = new IntVec2(calc.A);
p_b[i] = new IntVec2(calc.B);
p_c[i] = new IntVec2(calc.C);
});
mark(s, "memory init");
ComputeBuffer<IntVec2> _p_p = new ComputeBuffer<IntVec2>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, p_p);
ComputeBuffer<IntVec2> _p_a = new ComputeBuffer<IntVec2>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, p_a);
ComputeBuffer<IntVec2> _p_b = new ComputeBuffer<IntVec2>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, p_b);
ComputeBuffer<IntVec2> _p_c = new ComputeBuffer<IntVec2>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, p_c);
ComputeBuffer<FloatVec3> _c = new ComputeBuffer<FloatVec3>(context, ComputeMemoryFlags.WriteOnly, c.Length);
ComputeBuffer<int> _c_valid = new ComputeBuffer<int>(context, ComputeMemoryFlags.WriteOnly, c_valid.Length);
mark(s, "memory buffer init");
ComputeKernel kernel = program.CreateKernel("Barycentric");
kernel.SetMemoryArgument(0, _p_p);
kernel.SetMemoryArgument(1, _p_a);
kernel.SetMemoryArgument(2, _p_b);
kernel.SetMemoryArgument(3, _p_c);
kernel.SetMemoryArgument(4, _c);
kernel.SetMemoryArgument(5, _c_valid);
mark(s, "memory init 2");
ComputeEventList eventList = new ComputeEventList();
ComputeCommandQueue commands = new ComputeCommandQueue(context, context.Devices[0], ComputeCommandQueueFlags.None);
commands.Execute(kernel, null, new long[] { count }, null, eventList);
mark(s, "execute");
commands.ReadFromBuffer(_c, ref c, false, eventList);
commands.ReadFromBuffer(_c_valid, ref c_valid, false, eventList);
commands.Finish();
mark(s, "read 1");
Parallel.For(0, count, i =>
{
var calc = calculations[i];
calc.Coords = new BarycentricCoordinates(c[i].U,c[i].V,c[i].W);
if (c_valid[i] == 1)
{
lock (calc.Tri)
calc.Tri.Points.Add(new DrawPoint(calc.Coords, calc.P));
}
});
mark(s, "read 2");
// cleanup commands
commands.Dispose();
// cleanup events
foreach (ComputeEventBase eventBase in eventList)
{
eventBase.Dispose();
}
eventList.Clear();
// cleanup kernel
kernel.Dispose();
_p_p.Dispose();
_p_a.Dispose();
_p_b.Dispose();
_p_c.Dispose();
_c.Dispose();
_c_valid.Dispose();
mark(s, "dispose");
}
private void CalculateConvolution(ComputeContext computeContext)
{
Stopwatch stopwatch = new Stopwatch();
stopwatch.Start();
float dx;
bool shiftXParse = float.TryParse(textBoxShiftX.Text, NumberStyles.Float, CultureInfo.InvariantCulture.NumberFormat, out dx);
if (!shiftXParse)
throw new SyntaxErrorException(", needs to be .");
float dy;
bool shiftYParse = float.TryParse(textBoxShiftX.Text, NumberStyles.Float, CultureInfo.InvariantCulture.NumberFormat, out dy);
if (!shiftYParse)
throw new SyntaxErrorException(", needs to be .");
float dz;
bool shiftZParse = float.TryParse(textBoxShiftX.Text, NumberStyles.Float, CultureInfo.InvariantCulture.NumberFormat, out dz);
if (!shiftZParse)
throw new SyntaxErrorException(", needs to be .");
int pixelCount = _imageDimensionX*_imageDimensionY*_imageDimensionZ;
Console.WriteLine("Computing...");
Console.WriteLine("Reading kernel...");
String kernelPath = Directory.GetParent(Directory.GetCurrentDirectory()).Parent.Parent.FullName;
String kernelString;
using (var sr = new StreamReader(kernelPath + "\\convolution.cl"))
kernelString = sr.ReadToEnd();
Console.WriteLine("Reading kernel... done");
float[] selectedTransformation = Transformations.GetTransformation((TransformationType)comboBoxTransform.SelectedItem, 1.0f / float.Parse(textBoxPixelSize.Text), 1.0f / float.Parse(textBoxPixelSize.Text), 1.0f / float.Parse(textBoxPixelSize.Text), dx, dy, dz);
//create openCL program
ComputeProgram computeProgram = new ComputeProgram(computeContext, kernelString);
computeProgram.Build(computeContext.Devices, null, null, IntPtr.Zero);
ComputeProgramBuildStatus computeProgramBuildStatus = computeProgram.GetBuildStatus(_selectedComputeDevice);
Console.WriteLine("computeProgramBuildStatus\n\t"+computeProgramBuildStatus);
String buildLog = computeProgram.GetBuildLog(_selectedComputeDevice);
Console.WriteLine("buildLog");
if (buildLog.Equals("\n"))
Console.WriteLine("\tbuildLog is empty...");
else
Console.WriteLine("\t" + buildLog);
float[] fluorophores = CsvData.ReadFluorophores(_sourceFilename);
/////////////////////////////////////////////
// Create a Command Queue & Event List
/////////////////////////////////////////////
ComputeCommandQueue computeCommandQueue = new ComputeCommandQueue(computeContext, _selectedComputeDevice, ComputeCommandQueueFlags.None);
////////////////////////////////////////////////////////////////
// Create Buffers Transform
////////////////////////////////////////////////////////////////
ComputeBuffer<float> fluorophoresCoords = new ComputeBuffer<float>(computeContext, ComputeMemoryFlags.ReadWrite, fluorophores.LongLength);
ComputeBuffer<float> transformationMatrix = new ComputeBuffer<float>(computeContext, ComputeMemoryFlags.ReadOnly, selectedTransformation.LongLength);
/////////////////////////////////////////////
// Create the transformFluorophoresKernel
///////////////////////////////////////////////////////////
ComputeKernel transformFluorophoresKernel = computeProgram.CreateKernel("transform_fluorophores");
/////////////////////////////////////////////
// Set the transformFluorophoresKernel arguments
/////////////////////////////////////////////
transformFluorophoresKernel.SetMemoryArgument(0, fluorophoresCoords);
transformFluorophoresKernel.SetMemoryArgument(1, transformationMatrix);
/////////////////////////////////////////////
// Configure the work-item structure
/////////////////////////////////////////////
long[] globalWorkOffsetTransformFluorophoresKernel = null;
long[] globalWorkSizeTransformFluorophoresKernel = new long[] { fluorophores.Length / 4 };
long[] localWorkSizeTransformFluorophoresKernel = null;
////////////////////////////////////////////////////////
// Enqueue the transformFluorophoresKernel for execution
////////////////////////////////////////////////////////
computeCommandQueue.WriteToBuffer(fluorophores, fluorophoresCoords, true, null);
computeCommandQueue.WriteToBuffer(selectedTransformation, transformationMatrix, true, null);
computeCommandQueue.Execute(transformFluorophoresKernel, globalWorkOffsetTransformFluorophoresKernel, globalWorkSizeTransformFluorophoresKernel, localWorkSizeTransformFluorophoresKernel, null);
// computeCommandQueue.ExecuteTask(transformFluorophoresKernel, transformFluorophoresEvents);
float[] transformedFluorophores = new float[fluorophores.Length];
computeCommandQueue.ReadFromBuffer(fluorophoresCoords, ref transformedFluorophores, true, null);
computeCommandQueue.Finish();
//TODO remove, only for testing
// for (int i = 0; i < transformedFluorophores.Length; i++)
// {
// Console.WriteLine(transformedFluorophores[i]);
// }
// /TODO remove, only for testing
stopwatch.Stop();
Console.WriteLine("Transform fluophores duration:\n\t" + stopwatch.Elapsed);
stopwatch.Reset();
stopwatch.Start();
// fluorophoresCoords are now transformed (done in place)
////////////////////////////////////////////////////////////////
// Create Buffers Convolve Fluorophores
////////////////////////////////////////////////////////////////
const int convolve_kernel_lwgs = 16;
int totalBuffer = (int) Math.Ceiling(pixelCount / (float)convolve_kernel_lwgs) * convolve_kernel_lwgs;
ComputeBuffer<float> resultImage = new ComputeBuffer<float>(computeContext, ComputeMemoryFlags.WriteOnly, totalBuffer);
/////////////////////////////////////////////
// Create the transformFluorophoresKernel
/////////////////////////////////////////////
ComputeKernel convolveFluorophoresKernel = computeProgram.CreateKernel("convolve_fluorophores");
/////////////////////////////////////////////
// Set the convolveFluorophoresKernel arguments
/////////////////////////////////////////////
convolveFluorophoresKernel.SetMemoryArgument(0, resultImage);
convolveFluorophoresKernel.SetValueArgument(1, _imageDimensionX);
convolveFluorophoresKernel.SetValueArgument(2, _imageDimensionY);
convolveFluorophoresKernel.SetMemoryArgument(3, fluorophoresCoords);
convolveFluorophoresKernel.SetLocalArgument(4, convolve_kernel_lwgs);
convolveFluorophoresKernel.SetValueArgument(5, fluorophores.Length / 4);
/////////////////////////////////////////////
// Configure the work-item structure
/////////////////////////////////////////////
long[] globalWorkOffsetTransformConvolveFluorophoresKernel = null;
long[] globalWorkSizeTransformConvolveFluorophoresKernel = new long[] { pixelCount };
long[] localWorkSizeTransformConvolveFluorophoresKernel = new long[] {convolve_kernel_lwgs};
////////////////////////////////////////////////////////
// Enqueue the convolveFluorophoresKernel for execution
////////////////////////////////////////////////////////
computeCommandQueue.Execute(convolveFluorophoresKernel, globalWorkOffsetTransformConvolveFluorophoresKernel, globalWorkSizeTransformConvolveFluorophoresKernel, localWorkSizeTransformConvolveFluorophoresKernel, null);
float[] resultImageData = new float[totalBuffer];
computeCommandQueue.ReadFromBuffer(resultImage, ref resultImageData, true, null);
computeCommandQueue.Finish();
for (int i = 0; i < pixelCount; i++)
{
Console.WriteLine(resultImageData[i]);
}
Console.WriteLine("Writing data to file...");
// CsvData.WriteToDisk("..\\..\\..\\output.csv", resultImageData);
TiffData.WriteToDisk(resultImageData, _saveFilename, _imageDimensionX, _imageDimensionY);
Bitmap bitmap = new Bitmap(_imageDimensionX, _imageDimensionY);
float max = resultImageData.Max();
float scale = 255/(float)max;
// for (int r = 0; r < _imageDimensionY; r++)
// {
// for (int c = 0; c < _imageDimensionX; c++)
// {
// float value = resultImageData[c*(r + 1)];
// Color newColor = Color.FromArgb((int)(value * scale), (int)(value * scale), (int)(value * scale));
// bitmap.SetPixel(c,r, newColor);
// }
// }
ushort[] ushortdata = new ushort[resultImageData.Length];
for (int i = 0; i < resultImageData.Length; i++)
{
ushortdata[i] = (ushort)resultImageData[i];
}
uint[] convertGray16ToRgb = ConvertGray16ToRGB(ushortdata, 16);
byte[] bytes = new byte[convertGray16ToRgb.Length * 4];
//
// int[] resultImageData2 = new int[resultImageData.Length];
//
for (int index = 0; index < convertGray16ToRgb.Length; index++)
{
// resultImageData2[index] = (int)(scale*resultImageData[index]);
byte[] bytes1 = BitConverter.GetBytes(convertGray16ToRgb[index]);
bytes[index] = bytes1[0];
bytes[4 * index + 1] = bytes1[1];
bytes[4 * index + 2] = bytes1[2];
bytes[4 * index + 3] = bytes1[3];
}
//
// for (int r = 0; r < _imageDimensionY; r++)
// {
// for (int c = 0; c < _imageDimensionX; c++)
// {
// float value = resultImageData2[c*(r + 1)];
// Color newColor = Color.FromArgb((int)(value), (int)(value), (int)(value));
// bitmap.SetPixel(c,r, newColor);
// }
// }
// bitmap.Save("c:\\temp.bmp");
using (MemoryStream ms = new MemoryStream(bytes))
{
Image image = Bitmap.FromStream(ms);
image.Save("c:\\temp.bmp");
}
Console.WriteLine("Writing data to file... done");
stopwatch.Stop();
Console.WriteLine("Convolve fluophores duration:\n\t" + stopwatch.Elapsed);
Console.WriteLine("Computing... done");
}
public void Run(IComputeContext context, TextWriter log)
{
try
{
log.Write("Creating command queue... ");
var commands = new ComputeCommandQueue(context, context.Devices[0], ComputeCommandQueueFlags.None);
log.WriteLine("done.");
log.Write("Generating data... ");
int linearSize = 24;
SysIntX2 rectSize = new SysIntX2(4, 6);
SysIntX3 cubicSize = new SysIntX3(2, 3, 4);
float[] linearIn = new float[linearSize];
float[] linearOut = new float[linearSize];
float[,] rectIn = new float[(int)rectSize.Y, (int)rectSize.X];
float[,] rectOut = new float[(int)rectSize.Y, (int)rectSize.X];
float[,,] cubicIn = new float[(int)cubicSize.Z, (int)cubicSize.Y, (int)cubicSize.X];
float[,,] cubicOut = new float[(int)cubicSize.Z, (int)cubicSize.Y, (int)cubicSize.X];
for (int i = 0; i < linearSize; i++)
{
linearIn[i] = i;
}
for (int i = 0; i < (int)rectSize.X; i++)
{
for (int j = 0; j < (int)rectSize.Y; j++)
{
rectIn[j, i] = (float)(rectSize.X.ToInt32() * j + i);
}
}
for (int i = 0; i < (int)cubicSize.X; i++)
{
for (int j = 0; j < (int)cubicSize.Y; j++)
{
for (int k = 0; k < (int)cubicSize.Z; k++)
{
cubicIn[k, j, i] = (float)(k * cubicSize.Y.ToInt32() * cubicSize.X.ToInt32() + cubicSize.X.ToInt32() * j + i);
}
}
}
log.WriteLine("done.");
log.Write("Creating buffer... ");
var buffer = new ComputeBuffer <float>(context, ComputeMemoryFlags.ReadWrite, linearSize);
log.WriteLine("done.");
GC.Collect();
log.Write("Writing to buffer (linear)... ");
commands.WriteToBuffer(linearIn, buffer, false, null);
log.WriteLine("done.");
log.Write("Reading from buffer (linear)... ");
commands.ReadFromBuffer(buffer, ref linearOut, false, null);
log.WriteLine("done.");
GC.Collect();
commands.Finish();
log.Write("Comparing data... ");
Compare(linearIn, linearOut);
log.WriteLine("passed.");
GC.Collect();
log.Write("Writing to buffer (rectangular)... ");
commands.WriteToBuffer(rectIn, buffer, false, new SysIntX2(), new SysIntX2(), rectSize, null);
log.WriteLine("done.");
GC.Collect();
log.Write("Reading from buffer (rectangular)... ");
commands.ReadFromBuffer(buffer, ref rectOut, false, new SysIntX2(), new SysIntX2(), rectSize, null);
log.WriteLine("done.");
GC.Collect();
commands.Finish();
log.Write("Comparing data... ");
Compare(rectIn, rectOut);
log.WriteLine("passed.");
GC.Collect();
log.Write("Writing to buffer (cubic)... ");
commands.WriteToBuffer(cubicIn, buffer, false, new SysIntX3(), new SysIntX3(), cubicSize, null);
log.WriteLine("done.");
GC.Collect();
log.Write("Reading from buffer (cubic)... ");
commands.ReadFromBuffer(buffer, ref cubicOut, false, new SysIntX3(), new SysIntX3(), cubicSize, null);
log.WriteLine("done.");
GC.Collect();
commands.Finish();
log.Write("Comparing data... ");
Compare(cubicIn, cubicOut);
log.WriteLine("passed.");
}
catch (Exception e)
{
log.WriteLine(e.ToString());
}
}
// tick: renders one frame
public void Tick()
{
// initialize timer
if (firstFrame)
{
timer.Reset();
timer.Start();
firstFrame = false;
}
// handle keys, only when running time set to -1 (infinite)
if (runningTime == -1)
{
if (camera.HandleInput())
{
// camera moved; restart
ClearAccumulator();
}
}
// render
if (false) // if (useGPU)
{
// add your CPU + OpenCL path here
// mind the gpuPlatform parameter! This allows us to specify the platform on our
// test system.
// note: it is possible that the automated test tool provides you with a different
// platform number than required by your hardware. In that case, you can hardcode
// the platform during testing (ignoring gpuPlatform); do not forget to put back
// gpuPlatform before submitting!
long[] workSize = { screen.width, screen.height };
long[] localSize = { 16, 2 };
queue.Execute(kernel, null, workSize, null, null);
queue.Finish();
queue.ReadFromBuffer(outputBuffer, ref screen.pixels, true, null);
Console.WriteLine(screen.pixels[0]);
Random r = RTTools.GetRNG();
for (int i = 0; i < 1000; i++)
{
randoms[i] = (float)r.NextDouble();
}
rndBuffer = new ComputeBuffer <float>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.UseHostPointer, randoms);
cameraBuffer = new ComputeBuffer <Vector3>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.UseHostPointer, camera.toCL());
kernel.SetMemoryArgument(6, cameraBuffer);
kernel.SetMemoryArgument(7, rndBuffer);
}
else
{
// this is your CPU only path
float scale = 1.0f / (float)++spp;
Parallel.For(0, screen.height, y => {
for (int x = 0; x < screen.width; x++)
{
// generate primary ray
Ray ray = camera.Generate(RTTools.GetRNG(), x, y);
// trace path
int pixelIdx = x + y * screen.width;
accumulator[pixelIdx] += Sample(ray, 0, x, y);
// plot final color
screen.pixels[pixelIdx] = RTTools.Vector3ToIntegerRGB(scale * accumulator[pixelIdx]);
}
});
}
// stop and report when max render time elapsed
int elapsedSeconds = (int)(timer.ElapsedMilliseconds / 1000);
if (runningTime != -1)
{
if (elapsedSeconds >= runningTime)
{
OpenTKApp.Report((int)timer.ElapsedMilliseconds, spp, screen);
}
}
}
public String SearchPassword (byte[] hash, HashType type, int maxLength, String[] keySpace)
{
if (type != HashType.MD5) {
throw new NotImplementedException ("sums other than MD5 not supported");
}
if (maxLength > 6) {
throw new NotImplementedException ("doesn't support longer passwords than 7");
}
var joinedKeySpace = new List<byte> ();
foreach (var k in keySpace) {
if (k.Length > 1) {
throw new NotImplementedException ("doesn't support longer keyspaces than 1");
}
joinedKeySpace.AddRange (Encoding.ASCII.GetBytes (k));
}
byte[] resultData = new byte[20];
byte[] keyspaceJoined = joinedKeySpace.ToArray ();
var resultBuffer = new ComputeBuffer<byte> (Context, ComputeMemoryFlags.WriteOnly | ComputeMemoryFlags.CopyHostPointer, resultData);
var hashBuffer = new ComputeBuffer<byte> (Context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, hash);
var keyspaceBuffer = new ComputeBuffer<byte> (Context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, keyspaceJoined);
var passLenBuffer = new ComputeBuffer<byte> (Context, ComputeMemoryFlags.WriteOnly, 1);
var flagBuffer = new ComputeBuffer<int> (Context, ComputeMemoryFlags.None, 1);
Kernel.SetMemoryArgument (0, hashBuffer);
Kernel.SetMemoryArgument (1, keyspaceBuffer);
Kernel.SetMemoryArgument (2, resultBuffer);
Kernel.SetMemoryArgument (3, passLenBuffer);
Kernel.SetMemoryArgument (4, flagBuffer);
// execute kernel
var queue = new ComputeCommandQueue (Context, Device, ComputeCommandQueueFlags.None);
long firstDim = joinedKeySpace.Count;
var globalWorksize = new long[] { firstDim, 57 * 57, 57 * 57 };
queue.Execute (Kernel, new long[] { 0, 0, 0 }, globalWorksize, null, null);
byte[] passLen = new byte[1];
queue.ReadFromBuffer (resultBuffer, ref resultData, true, null);
queue.ReadFromBuffer (passLenBuffer, ref passLen, true, null);
String password = null;
if (passLen [0] > 0) {
logger.Info ("pass len {0}", passLen [0]);
password = Encoding.ASCII.GetString (resultData, 0, passLen [0]);
logger.Info ("Found password: \"{0}\"", password);
} else {
logger.Info ("Password not found.");
}
queue.Finish ();
return password;
}
public unsafe void EndSend()
{
for (int i = 0; i < points.Count; i++)
{
inx[i].x = (float)points[i].Item3.Real;
inx[i].y = (float)points[i].Item3.Imaginary;
inc[i].x = (float)points[i].Item4.Real;
inc[i].y = (float)points[i].Item4.Imaginary;
}
_krnl.SetMemoryArgument(0, x);
_krnl.SetMemoryArgument(1, c);
for (int i = 0; i < _ld.Count; i++)
{
_krnl.SetMemoryArgument(2 + i, outp[i]);
}
ComputeCommandQueue command = new ComputeCommandQueue(_context, _context.Devices[0], ComputeCommandQueueFlags.None);
command.WriteToBuffer(inx, x, false, null);
command.WriteToBuffer(inc, c, false, null);
command.Execute(_krnl, null, new long[] { points.Count }, null, null);
for (int i = 0; i < _ld.Count; i++)
command.ReadFromBuffer(outp[i], ref opl[i], false, null);
command.Finish();
output = new Queue<Tuple<int, int, List<ProcessLayer>>>();
for (int i = 0; i < points.Count; i++)
{
List<ProcessLayer> pl = new List<ProcessLayer>();
for (int ii = 0; ii < _ld.Count; ii++)
{
ProcessLayer p = _ld[ii].Clone();
p.c_active = opl[ii][i].c_active != 0;
p.c_calc = opl[ii][i].c_calc;
p.c_cmean = opl[ii][i].c_cmean;
p.c_cvariance = opl[ii][i].c_cvariance;
p.c_cvarsx = opl[ii][i].c_cvarsx;
p.c_isin = opl[ii][i].c_isin != 0;
p.c_n = opl[ii][i].c_n;
p.c_old2x = new Complex(opl[ii][i].c_old2x.x,opl[ii][i].c_old2x.y);
p.c_oldx = new Complex(opl[ii][i].c_oldx.x,opl[ii][i].c_oldx.y);
p.c_resn = opl[ii][i].c_resn;
p.c_resx = new Complex(opl[ii][i].c_resx.x,opl[ii][i].c_resx.y);
p.c_x = new Complex(opl[ii][i].c_x.x,opl[ii][i].c_x.y);
pl.Add(p);
}
output.Enqueue(Tuple.Create(points[i].Item1, points[i].Item2, pl));
}
}
private Bitmap Download(ComputeCommandQueue queue, ComputeBuffer<Vector4> buffer, int width, int height)
{
var pixels = new Vector4[width * height];
queue.ReadFromBuffer(buffer, ref pixels, true, 0, 0, width * height, null);
queue.Finish();
var intPixels = Array.ConvertAll(pixels, pixel =>
{
pixel = Vector4.Clamp(pixel, new Vector4(0), new Vector4(1));
return (byte)(pixel.X * 255) << 16 | (byte)(pixel.Y * 255) << 8 | (byte)(pixel.Z * 255);
});
var bmp = new Bitmap(width, height, PixelFormat.Format32bppRgb);
var bmpData = bmp.LockBits(new Rectangle(0, 0, bmp.Width, bmp.Height), ImageLockMode.ReadWrite, PixelFormat.Format32bppRgb);
Marshal.Copy(intPixels, 0, bmpData.Scan0, intPixels.Length);
bmp.UnlockBits(bmpData);
return bmp;
}
public void CopyFromDevice()
{
queue.ReadFromBuffer(gpuBuffer, ref cpuBuffer, true, null);
}
public static void CallOpenCL(int[,] libertyGroups,
int[,] groupNumbers, int x, int y,
int[] surroundingLibs, ref int emptySurroundings,
int ourSgn, ref int duplicateGroups)
{
//Create arguments
//Does not split yet
//int[,] libertyGroups,
//int[,] groupNumbers,
//int x,
//int y,
//int[] surroundingLibs,
//ref int emptySurroundings,
//ref int duplicateGroups,
//int ourSgn,
//We have to map 2 dimension to 1 dimension
//Set arguments
ComputeBuffer<int> libertyGroupsIn = new ComputeBuffer<int>(openCLContext, ComputeMemoryFlags.UseHostPointer, twoDtoOneD(libertyGroups));
openCLKernel.SetMemoryArgument(0, libertyGroupsIn);
ComputeBuffer<int> groupNumbersIn = new ComputeBuffer<int>(openCLContext, ComputeMemoryFlags.UseHostPointer, twoDtoOneD(groupNumbers));
openCLKernel.SetMemoryArgument(1, groupNumbersIn);
openCLKernel.SetValueArgument<int>(2, x);
openCLKernel.SetValueArgument<int>(3, y);
ComputeBuffer<int> surroundingLibsIn = new ComputeBuffer<int>(openCLContext, ComputeMemoryFlags.UseHostPointer, surroundingLibs);
openCLKernel.SetMemoryArgument(4, surroundingLibsIn);
int[] emptySurroundRef = new int[1];
ComputeBuffer<int> emptySurroundRefIn = new ComputeBuffer<int>(openCLContext, ComputeMemoryFlags.UseHostPointer, emptySurroundRef);
openCLKernel.SetMemoryArgument(5, emptySurroundRefIn);
int[] duplicateGroupsRef = new int[1];
ComputeBuffer<int> duplicateGroupsRefIn = new ComputeBuffer<int>(openCLContext, ComputeMemoryFlags.UseHostPointer, duplicateGroupsRef);
openCLKernel.SetMemoryArgument(6, duplicateGroupsRefIn);
openCLKernel.SetValueArgument<int>(7, ourSgn);
//long localWorkSize = Math.Min(openCLDevice.MaxComputeUnits, sideSize);
//Display input data
//Runs commands
ComputeCommandQueue commands = new ComputeCommandQueue(openCLContext,
openCLContext.Devices[0], ComputeCommandQueueFlags.None);
long executionTime = DateTime.Now.Ticks;
//Execute kernel
//globalWorkSize in increments of localWorkSize (max of device.MaxComputeUnits or kernel.GetWorkGroupSize())
commands.Execute(openCLKernel, null,
new long[] { 1 },
new long[] { 1 }, null);
//Also, you should probably use this
//kernel.GetPreferredWorkGroupSizeMultiple(device);
commands.Finish();
//int[] surroundingLibs,
//ref int emptySurroundings,
//ref int duplicateGroups,
//Read output data
commands.ReadFromBuffer(surroundingLibsIn, ref surroundingLibs, true, null);
commands.ReadFromBuffer(emptySurroundRefIn, ref emptySurroundRef, true, null); emptySurroundings = emptySurroundRef[0];
commands.ReadFromBuffer(duplicateGroupsRefIn, ref duplicateGroupsRef, true, null); duplicateGroups = duplicateGroupsRef[0];
//We could set blocking to false on reads and then read them all back in then, we could (possiblity) gain some performance
//by telling it that commands can be executed out of order and then by queuing them up and calling Finish
commands.Finish();
executionTime = DateTime.Now.Ticks - executionTime;
GC.Collect();
// openCLProgram.Dispose();
//display output data
//Test are done by our caller now
Console.WriteLine(executionTime / 10000.0);
}
public void Run(ComputeContext context, TextWriter log)
{
try
{
// Create the arrays and fill them with random data.
int count = 640*480; //
float[] arrA = new float[count];
float[] arrB = new float[count];
float[] arrC = new float[count];
Random rand = new Random();
for (int i = 0; i < count; i++)
{
arrA[i] = (float)(rand.NextDouble() * 100);
arrB[i] = (float)(rand.NextDouble() * 100);
}
// Create the input buffers and fill them with data from the arrays.
// Access modifiers should match those in a kernel.
// CopyHostPointer means the buffer should be filled with the data provided in the last argument.
program = new ComputeProgram(context, clProgramSource);
program.Build(null, null, null, IntPtr.Zero);
ComputeBuffer<float> a = new ComputeBuffer<float>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, arrA);
//ComputeBuffer<float> b = new ComputeBuffer<float>(context, ComputeMemoryFlags.ReadOnly | ComputeMemoryFlags.CopyHostPointer, arrB);
// The output buffer doesn't need any data from the host. Only its size is specified (arrC.Length).
ComputeBuffer<float> c = new ComputeBuffer<float>(context, ComputeMemoryFlags.WriteOnly, arrC.Length);
// Create and build the opencl program.
// Create the kernel function and set its arguments.
ComputeKernel kernel = program.CreateKernel("CompareGPUCPU");
DateTime ExecutionStartTime; //Var will hold Execution Starting Time
DateTime ExecutionStopTime;//Var will hold Execution Stopped Time
TimeSpan ExecutionTime;//Var will count Total Execution Time-Our Main Hero
ComputeCommandQueue commands = new ComputeCommandQueue(context, context.Devices[0], ComputeCommandQueueFlags.None);
ExecutionStartTime = DateTime.Now; //Gets the system Current date time expressed as local time
int repeatTimes = 100;
for (int repeatCounter = 0; repeatCounter < repeatTimes; repeatCounter++)
{
kernel.SetMemoryArgument(0, a);
//kernel.SetMemoryArgument(1, b);
//kernel.SetMemoryArgument(2, c);
kernel.SetMemoryArgument(1, c);
// Create the event wait list. An event list is not really needed for this example but it is important to see how it works.
// Note that events (like everything else) consume OpenCL resources and creating a lot of them may slow down execution.
// For this reason their use should be avoided if possible.
//ComputeEventList eventList = new ComputeEventList();
// Create the command queue. This is used to control kernel execution and manage read/write/copy operations.
// Execute the kernel "count" times. After this call returns, "eventList" will contain an event associated with this command.
// If eventList == null or typeof(eventList) == ReadOnlyCollection<ComputeEventBase>, a new event will not be created.
//commands.Execute(kernel, null, new long[] { count }, null, eventList);
commands.Execute(kernel, null, new long[] { count }, null, null);
// Read back the results. If the command-queue has out-of-order execution enabled (default is off), ReadFromBuffer
// will not execute until any previous events in eventList (in our case only eventList[0]) are marked as complete
// by OpenCL. By default the command-queue will execute the commands in the same order as they are issued from the host.
// eventList will contain two events after this method returns.
//commands.ReadFromBuffer(c, ref arrC, false, eventList);
commands.ReadFromBuffer(c, ref arrC, false, null);
// A blocking "ReadFromBuffer" (if 3rd argument is true) will wait for itself and any previous commands
// in the command queue or eventList to finish execution. Otherwise an explicit wait for all the opencl commands
// to finish has to be issued before "arrC" can be used.
// This explicit synchronization can be achieved in two ways:
// 1) Wait for the events in the list to finish,
//eventList.Wait();
// 2) Or simply use
commands.Finish();
}
ExecutionStopTime = DateTime.Now;
ExecutionTime = ExecutionStopTime - ExecutionStartTime;
double perTaskTime = ExecutionTime.TotalMilliseconds / repeatTimes;
log.WriteLine("Use {0} ms using GPU", perTaskTime);
// Do that using CPU
/*
ExecutionStartTime = DateTime.Now; //Gets the system Current date time expressed as local time
for (int repeatCounter = 0; repeatCounter < repeatTimes; repeatCounter++)
{
for (int i = 0; i < count; i++)
{
//arrC[i] = arrA[i] + arrB[i];
int j;
for (j = 0; j < 330 * 10; j++)
arrC[i] = arrA[i] + j;
}
}
ExecutionStopTime = DateTime.Now;
ExecutionTime = ExecutionStopTime - ExecutionStartTime;
perTaskTime = ExecutionTime.TotalMilliseconds / repeatTimes;
log.WriteLine("Use {0} ms using CPU", ExecutionTime.TotalMilliseconds.ToString());
*/
log.WriteLine("arrA[0]:{0}, arrC[0]:{1}", arrA[0], arrC[0]);
}
catch (Exception e)
{
log.WriteLine(e.ToString());
}
}
public Bitmap GetScreenshot(CameraConfig camera, int screenshotHeight, int slowRender)
{
var screenshotWidth = (int)(screenshotHeight * ScreenshotAspectRatio);
var computeBuffer = new ComputeBuffer<Vector4>(_program.Context, ComputeMemoryFlags.ReadWrite, screenshotWidth * screenshotHeight);
var queue = new ComputeCommandQueue(_program.Context, _program.Context.Devices[0], ComputeCommandQueueFlags.None);
var globalSize = GlobalLaunchsizeFor(screenshotWidth, screenshotHeight);
for (var i = 0; i < slowRender; i++)
CoreRender(computeBuffer, queue, _kernels, new Vector4((Vector3)camera.Position), new Vector4((Vector3)camera.Lookat), new Vector4((Vector3)camera.Up), i, camera.Fov, slowRender, camera.FocalDistance, screenshotWidth, screenshotHeight, globalSize, _localSize);
for (var i = 0; i < camera.Frame * slowRender; i++)
CoreRender(computeBuffer, queue, _kernels, new Vector4((Vector3)camera.Position), new Vector4((Vector3)camera.Lookat), new Vector4((Vector3)camera.Up), i, camera.Fov, slowRender, camera.FocalDistance, screenshotWidth, screenshotHeight, globalSize, _localSize);
var pixels = new Vector4[screenshotWidth * screenshotHeight];
queue.ReadFromBuffer(computeBuffer, ref pixels, true, null);
queue.Finish();
computeBuffer.Dispose();
queue.Dispose();
var bmp = new Bitmap(screenshotWidth, screenshotHeight);
var destBuffer = new int[screenshotWidth * screenshotHeight];
for (var y = 0; y < screenshotHeight; y++)
{
for (var x = 0; x < screenshotWidth; x++)
{
var pixel = pixels[x + y * screenshotWidth];
if (float.IsNaN(pixel.X) || float.IsNaN(pixel.Y) || float.IsNaN(pixel.Z))
{
Console.WriteLine("Warning! Caught NAN pixel while taking screenshot!");
continue;
}
destBuffer[y * screenshotWidth + x] = (byte)(pixel.X * 255) << 16 | (byte)(pixel.Y * 255) << 8 | (byte)(pixel.Z * 255);
}
}
var bmpData = bmp.LockBits(new Rectangle(0, 0, screenshotWidth, screenshotHeight), ImageLockMode.ReadWrite, PixelFormat.Format32bppRgb);
Marshal.Copy(destBuffer, 0, bmpData.Scan0, destBuffer.Length);
bmp.UnlockBits(bmpData);
return bmp;
}
static void Main(string[] args)
{
//Test2();
Test1();
ComputePlatform plat = ComputePlatform.Platforms[0];
Console.WriteLine("Plat:" + plat.Name);
ComputeContext context = new ComputeContext(ComputeDeviceTypes.Gpu, new ComputeContextPropertyList(plat), null, IntPtr.Zero);
ComputeCommandQueue queue = new ComputeCommandQueue(context, context.Devices[0], ComputeCommandQueueFlags.None);
StreamReader rs = new StreamReader("Foom/CL/testProg.txt");
string clSrc = rs.ReadToEnd();
rs.Close();
ComputeProgram prog = new ComputeProgram(context, clSrc);
prog.Build(null, null, null, IntPtr.Zero);
Console.WriteLine("BS:" + prog.GetBuildStatus(context.Devices[0]).ToString());
Console.WriteLine("Info:" + prog.GetBuildLog(context.Devices[0]));
ComputeKernel kern = prog.CreateKernel("vector_add");
int[] data = new int[1024];
for (int i = 0; i < 1024; i++)
{
data[i] = 100;
}
ComputeBuffer <int> b1 = new ComputeBuffer <int>(context, ComputeMemoryFlags.CopyHostPointer, data);
ComputeBuffer <int> b2 = new ComputeBuffer <int>(context, ComputeMemoryFlags.WriteOnly, 1024);
// queue.WriteToBuffer<int>(data, b1, true, null);
kern.SetMemoryArgument(0, b1);
kern.SetMemoryArgument(1, b2);
long[] wo = new long[1];
wo[0] = 0;
long[] ws = new long[1];
ws[0] = 1024;
long[] tc = new long[1];
tc[0] = 16;
queue.Execute(kern, wo, ws, tc, null);
int c = Environment.TickCount;
queue.Finish();
c = Environment.TickCount - c;
queue.ReadFromBuffer <int>(b2, ref data, true, null);
for (int i = 0; i < 10; i++)
{
Console.WriteLine("C:" + (int)data[i]);
}
Console.WriteLine("Done:" + c);
while (true)
{
}
}
public Bitmap Screenshot(int screenshotHeight, int slowRenderPower, Action<string> displayInformation)
{
displayInformation("Rendering screenshot");
var ccontext = _kernel.ComputeContext;
_kernelInUse++;
var screenshotWidth = (int)(screenshotHeight * ScreenshotAspectRatio);
Bitmap bmp;
try
{
bmp = new Bitmap(screenshotWidth, screenshotHeight, PixelFormat.Format24bppRgb);
}
catch (ArgumentException)
{
MessageBox.Show("Image size too big", "Error");
return null;
}
var nancount = 0;
var bmpData = bmp.LockBits(new Rectangle(0, 0, screenshotWidth, screenshotHeight), ImageLockMode.ReadWrite, PixelFormat.Format24bppRgb);
var scan0 = bmpData.Scan0.ToInt64();
var queue = new ComputeCommandQueue(ccontext, ccontext.Devices[0], ComputeCommandQueueFlags.None);
var localSize = _kernel.Threadsize(queue);
for (var i = 0; i < localSize.Length; i++)
localSize[i] *= slowRenderPower;
var computeBuffer = new ComputeBuffer<Vector4>(ccontext, ComputeMemoryFlags.ReadWrite, localSize[0] * localSize[1]);
const int numFrames = 200;
var frameDependantControls = _parameters as IFrameDependantControl;
var framesToRender = frameDependantControls == null ? 1 : numFrames;
var totalYs = (screenshotHeight + localSize[1] - 1) / localSize[1];
var totalXs = (screenshotWidth + localSize[0] - 1) / localSize[0];
var stopwatch = new Stopwatch();
for (var y = 0; y < totalYs; y++)
{
for (var x = 0; x < totalXs; x++)
{
stopwatch.Restart();
for (var frame = 0; frame < framesToRender; frame++)
{
if (frameDependantControls != null)
frameDependantControls.Frame = frame;
displayInformation(string.Format("Screenshot {0}% done", 100 * (y * totalXs * framesToRender + x * framesToRender + frame) / (totalXs * totalYs * framesToRender)));
_kernel.Render(computeBuffer, queue, _parameters, new Size(screenshotWidth, screenshotHeight), slowRenderPower, new Size(x, y), (int)localSize[0]);
}
var pixels = new Vector4[localSize[0] * localSize[1]];
queue.ReadFromBuffer(computeBuffer, ref pixels, true, 0, 0, localSize[0] * localSize[1], null);
queue.Finish();
stopwatch.Stop();
var elapsed = stopwatch.Elapsed.TotalMilliseconds / framesToRender;
_kernel.AverageKernelTime = (elapsed + _kernel.AverageKernelTime * 4) / 5;
var blockWidth = Math.Min(localSize[0], screenshotWidth - x * localSize[0]);
var blockHeight = Math.Min(localSize[1], screenshotHeight - y * localSize[1]);
var intPixels = new byte[blockWidth * blockHeight * 3];
for (var py = 0; py < blockHeight; py++)
{
for (var px = 0; px < blockWidth; px++)
{
var pixel = pixels[py * localSize[1] + px];
if (float.IsNaN(pixel.X) || float.IsNaN(pixel.Y) || float.IsNaN(pixel.Z))
nancount++;
// BGR
if (float.IsNaN(pixel.Z) == false)
intPixels[(py * blockWidth + px) * 3 + 0] = (byte)(pixel.Z * 255);
if (float.IsNaN(pixel.Y) == false)
intPixels[(py * blockWidth + px) * 3 + 1] = (byte)(pixel.Y * 255);
if (float.IsNaN(pixel.X) == false)
intPixels[(py * blockWidth + px) * 3 + 2] = (byte)(pixel.X * 255);
}
}
for (var line = 0; line < blockHeight; line++)
Marshal.Copy(intPixels, line * (int)blockWidth * 3, new IntPtr(scan0 + ((y * localSize[1] + line) * bmpData.Stride) + x * localSize[0] * 3), (int)blockWidth * 3);
}
}
bmp.UnlockBits(bmpData);
if (nancount != 0)
MessageBox.Show(string.Format("Caught {0} NAN pixels while taking screenshot", nancount), "Warning");
_kernelInUse--;
return bmp;
}
public static void Test()
{
string source = File.ReadAllText("MonteCarloSimulate.cl");
//Choose Device
ComputePlatform platform = ComputePlatform.Platforms[0];
ComputeDevice device = platform.QueryDevices()[0];
ComputeContextPropertyList properties =
new ComputeContextPropertyList(platform);
//Setup of stuff on our side
ComputeContext context = new ComputeContext(ComputeDeviceTypes.All,
properties, null, IntPtr.Zero);
//Build the program, which gets us the kernel
ComputeProgram program = new ComputeProgram(context, source);
program.Build(null, null, null, IntPtr.Zero);
//can use notify as the 3rd command... if you want this to be non-blocking
ComputeKernel kernel = program.CreateKernel("MonteCarloSimulate");
//Create arguments
int sideSize = 4096;
int[] inMatrixA = new int[sideSize * sideSize];
int[] inMatrixB = new int[sideSize * sideSize];
int[] outMatrixC = new int[sideSize * sideSize];
Random random = new Random((int)DateTime.Now.Ticks);
if (sideSize <= 32)
for (int y = 0; y < sideSize; y++)
for (int x = 0; x < sideSize; x++)
{
inMatrixA[y * sideSize + x] = random.Next(3);
inMatrixB[y * sideSize + x] = random.Next(3);
outMatrixC[y * sideSize + x] = 0;
}
ComputeBuffer<int> bufferMatrixA = new ComputeBuffer<int>(context,
ComputeMemoryFlags.UseHostPointer, inMatrixA);
ComputeBuffer<int> bufferMatrixB = new ComputeBuffer<int>(context,
ComputeMemoryFlags.UseHostPointer, inMatrixB);
ComputeBuffer<int> bufferMatrixC = new ComputeBuffer<int>(context,
ComputeMemoryFlags.UseHostPointer, outMatrixC);
long localWorkSize = Math.Min(device.MaxComputeUnits, sideSize);
//Sets arguments
kernel.SetMemoryArgument(0, bufferMatrixA);
kernel.SetMemoryArgument(1, bufferMatrixB);
kernel.SetMemoryArgument(2, bufferMatrixC);
kernel.SetLocalArgument(3, sideSize * 2);
kernel.SetValueArgument<int>(4, sideSize);
//kernel.SetLocalArgument(1, localWorkSize);
string offset = " ";
for (int x = 0; x < sideSize; x++)
offset += " ";
if (sideSize <= 32)
for (int y = 0; y < sideSize; y++)
{
Console.Write(offset);
for (int x = 0; x < sideSize; x++)
Console.Write(inMatrixA[y * sideSize + x] + " ");
Console.WriteLine();
}
//Runs commands
ComputeCommandQueue commands = new ComputeCommandQueue(context,
context.Devices[0], ComputeCommandQueueFlags.None);
long executionTime = DateTime.Now.Ticks;
//Execute kernel
//globalWorkSize in increments of localWorkSize (max of device.MaxComputeUnits or kernel.GetWorkGroupSize())
commands.Execute(kernel, null,
new long[] { Math.Min(sideSize, 16), Math.Min(sideSize, 16) },
new long[] { localWorkSize, 1 }, null);
//globalWorkSize can be any size
//localWorkSize product much not be greater than device.MaxComputeUnits
//and it must not be greater than kernel.GetWorkGroupSize()
//ESSENTIALLY, the program iterates through globalWorkSize
//in increments of localWorkSize. Both are multidimensional,
//but this just saves us the time of doing that
//(1 dimension can be put to multiple if the max dimension lengths
//are known very easily with remainder).
//Also, you should probably use this
//kernel.GetPreferredWorkGroupSizeMultiple(device);
commands.Finish();
commands.ReadFromBuffer(bufferMatrixC, ref outMatrixC, true, null);
commands.Finish();
executionTime = DateTime.Now.Ticks - executionTime;
GC.Collect();
program.Dispose();
Console.WriteLine();
if (sideSize <= 32)
for (int y = 0; y < sideSize; y++)
{
for (int x = 0; x < sideSize; x++)
Console.Write(inMatrixB[y * sideSize + x] + " ");
Console.Write(" ");
for (int x = 0; x < sideSize; x++)
Console.Write(outMatrixC[y * sideSize + x] + " ");
Console.WriteLine();
}
int testY = random.Next(sideSize);
int testX = random.Next(sideSize);
int sum = 0;
for (int q = 0; q < sideSize; q++)
sum += inMatrixA[q * sideSize + testX] *
inMatrixB[testY * sideSize + q];
Console.WriteLine(sum == outMatrixC[testY * sideSize + testX]);
Console.WriteLine(executionTime / 10000.0);
}
