public virtual void CopyToMemoryBlockAsync(MyMemoryBlock <T> destination, int srcOffset, int destOffset, int count,
CudaStream stream = null)
{
destination.GetDevice(Owner.GPU).AsyncCopyToDevice(
Device[Owner.GPU], srcOffset * ESize, destOffset * ESize, count * ESize,
MyKernelFactory.GetCuStreamOrDefault(stream));
}
// TODO(Premek): Add more copy functions to be consistent with Async variants. (And cleanup the API in general.)
#region Async copy device to device
public virtual void CopyFromMemoryBlockAsync(MyMemoryBlock <T> source, int srcOffset, int destOffset, int count,
CudaStream stream = null)
{
Device[Owner.GPU].AsyncCopyToDevice(
source.GetDevice(Owner.GPU), srcOffset * ESize, destOffset * ESize, count * ESize,
MyKernelFactory.GetCuStreamOrDefault(stream));
}
public virtual void FillAsync(uint value, CudaStream stream = null)
{
Device[Owner.GPU].MemsetAsync(value, MyKernelFactory.GetCuStreamOrDefault(stream));
}
/// <summary>
/// Asynchonously copies the whole block or part of it when the destination is not large enough.
/// </summary>
public virtual void CopyToMemoryBlockAsync(MyMemoryBlock <T> destination, CudaStream stream = null)
{
destination.GetDevice(Owner.GPU).AsyncCopyToDevice(Device[Owner.GPU], MyKernelFactory.GetCuStreamOrDefault(stream));
}
/// <summary>
/// Asynchonously copies the whole source block or part of it when the destination is not large enough.
/// </summary>
public virtual void CopyFromMemoryBlockAsync(MyMemoryBlock <T> source, CudaStream stream = null)
{
Device[Owner.GPU].AsyncCopyToDevice(source.GetDevice(Owner.GPU), MyKernelFactory.GetCuStreamOrDefault(stream));
}