internal OpenCLEvent(CLEventHandle handle, OpenCLCommandQueue queue)
{
Handle = handle;
SetID(Handle.Value);
CommandQueue = queue;
Type = (OpenCLCommandType)GetInfo<CLEventHandle, OpenCLEventInfo, int>(Handle, OpenCLEventInfo.CommandType, CL10.GetEventInfo);
Context = queue.Context;
if (OpenCLTools.ParseVersionString(CommandQueue.Device.Platform.Version, 1) > new Version(1, 0))
HookNotifier();
//Trace.WriteLine("Create " + this + " in Thread(" + Thread.CurrentThread.ManagedThreadId + ").", "Information");
}
internal OpenCLEvent(CLEventHandle handle, OpenCLCommandQueue queue)
{
Handle = handle;
SetID(Handle.Value);
CommandQueue = queue;
Type = (OpenCLCommandType)GetInfo <CLEventHandle, OpenCLEventInfo, int>(Handle, OpenCLEventInfo.CommandType, CL10.GetEventInfo);
Context = queue.Context;
if (OpenCLTools.ParseVersionString(CommandQueue.Device.Platform.Version, 1) > new Version(1, 0))
{
HookNotifier();
}
//Trace.WriteLine("Create " + this + " in Thread(" + Thread.CurrentThread.ManagedThreadId + ").", "Information");
}
private static void RunKernel(OpenCLPlatform platform, OpenCLDevice device)
{
var context = new OpenCLContext(new List<OpenCLDevice> {device}, new OpenCLContextPropertyList(platform), null, IntPtr.Zero);
var program = LoadProgram(context, device, "ReductionUsingFSCLOpenCLManagedWrapper.reduction.cl");
var kernel1 = program.CreateKernel("reductionVector");
var kernel2 = program.CreateKernel("reductionComplete");
const int numValues = 1024 * 1024;
const int numValuesPerWorkItem = 4;
var globalWorkSize = numValues / numValuesPerWorkItem;
const int localWorkSize = 32;
var initialNumWorkGroups = globalWorkSize/localWorkSize;
const int value = 42;
var data = Enumerable.Repeat(value, numValues).Select(n => (float)n).ToArray();
var commandQueue = new OpenCLCommandQueue(context, device, OpenCLCommandQueueProperties.None);
var floatType = typeof (float);
var floatSize = sizeof (float);
var dataBuffer1 = new OpenCLBuffer(context, OpenCLMemoryFlags.ReadWrite | OpenCLMemoryFlags.AllocateHostPointer, floatType, new long[] {numValues});
var dataBuffer2 = new OpenCLBuffer(context, OpenCLMemoryFlags.ReadWrite | OpenCLMemoryFlags.AllocateHostPointer, floatType, new long[] {initialNumWorkGroups*numValuesPerWorkItem});
var sumBuffer = new OpenCLBuffer(context, OpenCLMemoryFlags.WriteOnly | OpenCLMemoryFlags.AllocateHostPointer, floatType, new long[] { 1 });
var resultDataBuffer = dataBuffer2;
using (var pinnedData = new PinnedObject(data))
{
commandQueue.WriteToBuffer(pinnedData, dataBuffer1, true, 0L, numValues);
}
foreach (var index in Enumerable.Range(0, int.MaxValue))
{
var dataBufferIn = index%2 == 0 ? dataBuffer1 : dataBuffer2;
var dataBufferOut = index%2 == 0 ? dataBuffer2 : dataBuffer1;
resultDataBuffer = dataBufferOut;
kernel1.SetMemoryArgument(0, dataBufferIn);
kernel1.SetMemoryArgument(1, dataBufferOut);
kernel1.SetLocalArgument(2, localWorkSize*numValuesPerWorkItem*floatSize);
Console.WriteLine($"Calling commandQueue.Execute(kernel1) with globalWorkSize: {globalWorkSize}; localWorkSize: {localWorkSize}; num work groups: {globalWorkSize / localWorkSize}");
commandQueue.Execute(kernel1, null, new long[] {globalWorkSize}, new long[] {localWorkSize});
globalWorkSize /= localWorkSize;
if (globalWorkSize <= localWorkSize) break;
}
kernel2.SetMemoryArgument(0, resultDataBuffer);
kernel2.SetLocalArgument(1, globalWorkSize*numValuesPerWorkItem*floatSize);
kernel2.SetMemoryArgument(2, sumBuffer);
Console.WriteLine($"Calling commandQueue.Execute(kernel2) with globalWorkSize: {globalWorkSize}; localWorkSize: {globalWorkSize}");
commandQueue.Execute(kernel2, null, new long[] { globalWorkSize }, new long[] { globalWorkSize });
commandQueue.Finish();
var sum = new float[1];
using (var pinnedSum = new PinnedObject(sum))
{
commandQueue.ReadFromBuffer(sumBuffer, pinnedSum, true, 0L, 1L);
}
const int correctAnswer = numValues * value;
Console.WriteLine($"OpenCL final answer: {Math.Truncate(sum[0]):N0}; Correct answer: {correctAnswer:N0}");
}
