/// <summary>
/// Creates a new <see cref="ComputeBuffer{T}"/>.
/// </summary>
/// <param name="context"> A <see cref="ComputeContext"/> used to create the <see cref="ComputeBuffer{T}"/>. </param>
/// <param name="flags"> A bit-field that is used to specify allocation and usage information about the <see cref="ComputeBuffer{T}"/>. </param>
/// <param name="count"> The number of elements of the <see cref="ComputeBuffer{T}"/>. </param>
/// <param name="dataPtr"> A pointer to the data for the <see cref="ComputeBuffer{T}"/>. </param>
public ComputeBuffer(ComputeContext context, ComputeMemoryFlags flags, long count, IntPtr dataPtr)
: base(context, flags)
{
ComputeErrorCode error = ComputeErrorCode.Success;
Handle = CL10.CreateBuffer(context.Handle, flags, new IntPtr(Marshal.SizeOf(typeof(T)) * count), dataPtr, out error);
ComputeException.ThrowOnError(error);
Init();
}
/// <summary>
/// Creates a new <see cref="ComputeBuffer{T}"/>.
/// </summary>
/// <param name="context"> A <see cref="ComputeContext"/> used to create the <see cref="ComputeBuffer{T}"/>. </param>
/// <param name="flags"> A bit-field that is used to specify allocation and usage information about the <see cref="ComputeBuffer{T}"/>. </param>
/// <param name="data"> The data for the <see cref="ComputeBuffer{T}"/>. </param>
/// <remarks> Note, that <paramref name="data"/> cannot be an "immediate" parameter, i.e.: <c>new T[100]</c>, because it could be quickly collected by the GC causing Cloo to send and invalid reference to OpenCL. </remarks>
public ComputeBuffer(ComputeContext context, ComputeMemoryFlags flags, T[] data)
: base(context, flags)
{
GCHandle dataPtr = GCHandle.Alloc(data, GCHandleType.Pinned);
try
{
ComputeErrorCode error = ComputeErrorCode.Success;
Handle = CL10.CreateBuffer(context.Handle, flags, new IntPtr(Marshal.SizeOf(typeof(T)) * data.Length), dataPtr.AddrOfPinnedObject(), out error);
ComputeException.ThrowOnError(error);
}
finally
{
dataPtr.Free();
}
Init();
}
public void sortKeysValue(CLMemoryHandle key, CLMemoryHandle value,
int numElements)
{
debugRead = new int[Math.Max(numElements, numCounters)];
ComputeErrorCode error;
ComputeEvent eve;
/*
* error = CL10.EnqueueReadBuffer(cqCommandQueue, input, Bool.True, IntPtr.Zero, (IntPtr)(numElements * 4),
* debugRead, 0, null, out eve);
* CheckErr(error, "CL10.EnqueueReadBuffer");
*/
mCounters = CL10.CreateBuffer(cxGPUContext, ComputeMemoryFlags.ReadWrite, gpuConstants.numGroupsPerBlock * gpuConstants.numRadices * gpuConstants.numBlocks * sizeof(int),
out error);
CheckErr(error, "CL10.CreateBuffer");
mRadixPrefixes = CL10.CreateBuffer(cxGPUContext, ComputeMemoryFlags.ReadWrite, gpuConstants.numRadices * sizeof(int), out error);
CheckErr(error, "CL10.CreateBuffer");
CLMemoryHandle outputValue = CL10.CreateBuffer(cxGPUContext, ComputeMemoryFlags.ReadWrite, (IntPtr)(8 * numElements),
out error);
CheckErr(error, "CL10.CreateBuffer");
CLMemoryHandle outputKey = CL10.CreateBuffer(cxGPUContext, ComputeMemoryFlags.ReadWrite, (IntPtr)(4 * numElements),
out error);
CheckErr(error, "CL10.CreateBuffer");
gpuConstants.numElementsPerGroup = (numElements / (gpuConstants.numBlocks * gpuConstants.numGroupsPerBlock)) + 1;
gpuConstants.numTotalElements = numElements;
int i;
for (i = 0; i < 8; i++)
{
error = CL10.EnqueueWriteBuffer(cqCommandQueue, mCounters, Bool.True, IntPtr.Zero, (IntPtr)(numCounters * 4),
counters, 0, null, out eve);
CheckErr(error, "CL10.EnqueueWriteBuffer Counter initialize");
if (i % 2 == 0)
{
SetupAndCount(key, 4 * i);
SumIt(key, 4 * i);
ReorderingKeyValue(key, outputKey, value, outputValue, 4 * i);
}
else
{
SetupAndCount(outputKey, 4 * i);
SumIt(outputKey, 4 * i);
ReorderingKeyValue(outputKey, key, outputValue, value, 4 * i);
}
}
if (i % 2 == 0)
{
error = CL10.EnqueueCopyBuffer(cqCommandQueue, outputKey, key, IntPtr.Zero, IntPtr.Zero, (IntPtr)(numElements * 4), 0, null, out eve);
CheckErr(error, "CL10.EnqueueCopyBuffer");
error = CL10.Finish(cqCommandQueue);
CheckErr(error, "CL10.Finish Copybuffer");
error = CL10.EnqueueCopyBuffer(cqCommandQueue, outputValue, value, IntPtr.Zero, IntPtr.Zero, (IntPtr)(numElements * 8), 0, null, out eve);
CheckErr(error, "CL10.EnqueueCopyBuffer");
error = CL10.Finish(cqCommandQueue);
CheckErr(error, "CL10.Finish Copybuffer");
}
error = CL10.ReleaseMemObject(outputKey);
CheckErr(error, "CL10.ReleaseMemObj");
error = CL10.ReleaseMemObject(outputValue);
CheckErr(error, "CL10.ReleaseMemObj");
error = CL10.ReleaseMemObject(mRadixPrefixes);
CheckErr(error, "CL10.ReleaseMemObj");
error = CL10.ReleaseMemObject(mCounters);
CheckErr(error, "CL10.ReleaseMemObj");
Log_Idx++;
}
private void InternalCreateBuffer(IComputeContext context, ComputeMemoryFlags flags, long count, IntPtr dataPtr)
{
Handle = CL10.CreateBuffer(context.Handle, flags, new IntPtr(Marshal.SizeOf(typeof(T)) * count), dataPtr, out ComputeErrorCode error);
ComputeException.ThrowOnError(error);
Init();
}
public void sortKeysOnly(CLMemoryHandle input, CLMemoryHandle output,
int numElements)
{
debugRead = new int[Math.Max(numElements, numCounters)];
ComputeErrorCode error;
Compute ComputeEvent eve;
mCounters = CL10.CreateBuffer(cxGPUContext, ComputeMemoryFlags.ReadWrite, gpuConstants.numGroupsPerBlock * gpuConstants.numRadices * gpuConstants.numBlocks * sizeof(int),
out error);
CheckErr(error, "CL10.CreateBuffer");
mRadixPrefixes = CL10.CreateBuffer(cxGPUContext, ComputeMemoryFlags.ReadWrite, gpuConstants.numRadices * sizeof(int), out error);
CheckErr(error, "CL10.CreateBuffer");
gpuConstants.numElementsPerGroup = (numElements / (gpuConstants.numBlocks * gpuConstants.numGroupsPerBlock)) + 1;
gpuConstants.numTotalElements = numElements;
if (DEBUG)
{
CL10.EnqueueReadBuffer(cqCommandQueue, input, Bool.True, IntPtr.Zero, (IntPtr)(gpuConstants.numTotalElements * 4), debugRead, 0,
null, out eve);
CheckErr(error, "CL10.EnqueueReadBuffer");
PrintAsArray(debugRead, gpuConstants.numTotalElements);
}
int i;
for (i = 0; i < 8; i++)
{
error = CL10.EnqueueWriteBuffer(cqCommandQueue, mCounters, true, IntPtr.Zero, (IntPtr)(numCounters * 4),
counters, 0, null, out eve);
CheckErr(error, "CL10.EnqueueWriteBuffer Counter initialize");
if (i % 2 == 0)
{
DateTime before = DateTime.Now;
SetupAndCount(input, 4 * i);
if (DEBUG_CONSOLE_OUTPUT)
{
Console.WriteLine("Setup and Count =" + (DateTime.Now - before).TotalMilliseconds);
}
before = DateTime.Now;
SumIt(input, 4 * i);
if (DEBUG_CONSOLE_OUTPUT)
{
Console.WriteLine("SumIt =" + (DateTime.Now - before).TotalMilliseconds);
}
before = DateTime.Now;
ReorderingKeysOnly(input, output, 4 * i);
if (DEBUG_CONSOLE_OUTPUT)
{
Console.WriteLine("Reorder =" + (DateTime.Now - before).TotalMilliseconds);
}
}
else
{
SetupAndCount(output, 4 * i);
SumIt(output, 4 * i);
ReorderingKeysOnly(output, input, 4 * i);
}
}
if (i % 2 != 0)
{
error = CL10.EnqueueCopyBuffer(cqCommandQueue, input, output, IntPtr.Zero, IntPtr.Zero, (IntPtr)(numElements * 4), 0, null, out eve);
CheckErr(error, "CL10.EnqueueCopyBuffer");
error = CL10.Finish(cqCommandQueue);
CheckErr(error, "CL10.Finish Copybuffer");
}
error = CL10.ReleaseMemObject(mRadixPrefixes);
CheckErr(error, "CL10.ReleaseMemObj");
error = CL10.ReleaseMemObject(mCounters);
CheckErr(error, "CL10.ReleaseMemObj");
Log_Idx++;
}