public MyParallelKernel(MyNode owner, int nGPU, ParallelKernelDescriptor descriptor, int bufferSize)
{
m_owner = owner;
m_nGPU = nGPU;
m_threadCount = ThreadCount.T32;
m_outTypeSize = descriptor.outTypeSize;
m_inTypeSize = descriptor.inTypeSize;
m_TSize = Marshal.SizeOf(typeof(T));
if (m_TSize > m_inTypeSize || m_TSize > m_outTypeSize || m_inTypeSize % m_TSize != 0 ||
m_outTypeSize % m_TSize != 0 || m_inTypeSize == 0 || m_outTypeSize == 0)
{
MyLog.Writer.WriteLine(MyLogLevel.WARNING, "MyReduction.cs: MemoryBlock type can be incompatible with reduction in/out types.");
}
Array threadCountValues = Enum.GetValues(typeof(ThreadCount));
m_kernels = new Dictionary <ThreadCount, MyCudaKernel>();
foreach (ThreadCount threadCount in Enum.GetValues(typeof(ThreadCount)))
{
string kernelName = descriptor.GetKernelName(threadCount);
m_kernels.Add(threadCount, MyKernelFactory.Instance.Kernel(owner.GPU, @"Common\Reduction\Reduction", kernelName));
}
m_buffer = MyMemoryManager.Instance.CreateMemoryBlock <float>(owner);
m_buffer.Name = "Buffer(" + descriptor.modeName + ")";
m_buffer.Count = bufferSize;
m_blockDims = new dim3((int)m_threadCount, 1, 1);
m_gridDims = new dim3(1, 1, 1);
}
public static ParallelKernelDescriptor GetDescriptor(Type type, string name)
{
MemberInfo[] memInfo = type.GetMember(name);
object[] attributes = memInfo[0].GetCustomAttributes(typeof(ParallelKernelDescriptor), false);
if (attributes.Length == 0)
{
throw new Exception("Kernel descriptor is missing for " + name + ".");
}
ParallelKernelDescriptor descriptor = attributes[0] as ParallelKernelDescriptor;
descriptor.modeName = name;
return(descriptor);
}
public MyProductKernel(MyNode owner, int nGPU, ProductMode mode, int bufferSize = BUFFER_SIZE)
: base(owner, nGPU, ParallelKernelDescriptor.GetDescriptor(typeof(ProductMode), mode.ToString()), bufferSize)
{
ResetParameters();
}
