/// <inheritdoc />
protected override Util.ISigmaDiffDataBuffer <T> _InternalDeepCopy()
{
T[] copyData = SigmaDiffSharpBackendProvider.Instance.GetBackend <T>(BackendTag).BackendHandle.CreateUninitialisedArray((int)Length);
CudaSigmaDiffDataBuffer <T> copy = new CudaSigmaDiffDataBuffer <T>(copyData, 0L, Length, BackendTag, CudaContext, Type);
if (_initialisedInContext && !_flagHostModified)
{
copy.InitialiseCudaBuffer(copyHostToDevice: false);
copy._cudaBuffer.AsyncCopyToDevice(_cudaBuffer, CudaFloat32Handler.GetStreamForContext(CudaContext));
copy._flagHostModified = false;
copy._flagDeviceModified = true;
}
else
{
OnReadAccess();
Buffer.BlockCopy(_data, (int)(Offset * Type.SizeBytes), copyData, 0, (int)(Length * Type.SizeBytes));
copy._flagHostModified = true;
}
return(copy);
}
/// <inheritdoc />
public override Util.ISigmaDiffDataBuffer <T> GetStackedValues(int totalRows, int totalCols, int rowStart, int rowFinish, int colStart, int colFinish)
{
int colLength = colFinish - colStart + 1;
int newSize = (rowFinish - rowStart + 1) * colLength;
Backend <T> backendHandle = SigmaDiffSharpBackendProvider.Instance.GetBackend <T>(BackendTag).BackendHandle;
CudaSigmaDiffDataBuffer <T> values = (CudaSigmaDiffDataBuffer <T>)backendHandle.CreateDataBuffer(backendHandle.CreateUninitialisedArray(newSize));
if (_initialisedInContext)
{
SynchroniseFromDeviceToHost();
}
for (int m = rowStart; m <= rowFinish; m++)
{
long sourceIndex = Offset + m * totalCols + colStart;
long destinationIndex = (m - rowStart) * colLength;
Buffer.BlockCopy(_data, (int)(sourceIndex * Type.SizeBytes), values.Data, (int)(destinationIndex * Type.SizeBytes), colLength * Type.SizeBytes);
}
if (_initialisedInContext)
{
values.SynchroniseFromHostToDevice();
}
return(values);
}
/// <inheritdoc />
public override void Fill(INDArray filler, INDArray arrayToFill, long[] sourceBeginIndices, long[] sourceEndIndices, long[] destinationBeginIndices, long[] destinationEndIndices)
{
CudaSigmaDiffDataBuffer <float> fillerData = (CudaSigmaDiffDataBuffer <float>)InternaliseArray(filler).Data;
CudaSigmaDiffDataBuffer <float> arrayToFillData = (CudaSigmaDiffDataBuffer <float>)InternaliseArray(arrayToFill).Data;
int sourceOffset = (int)NDArrayUtils.GetFlatIndex(filler.Shape, filler.Strides, sourceBeginIndices);
int sourceLength = (int)NDArrayUtils.GetFlatIndex(filler.Shape, filler.Strides, sourceEndIndices) - sourceOffset + 1; // +1 because end is inclusive
int destinationOffset = (int)NDArrayUtils.GetFlatIndex(arrayToFill.Shape, arrayToFill.Strides, destinationBeginIndices);
int destinationLength = (int)NDArrayUtils.GetFlatIndex(arrayToFill.Shape, arrayToFill.Strides, destinationEndIndices) - destinationOffset + 1; // same here
if (sourceLength < 0)
{
throw new ArgumentOutOfRangeException($"Source begin indices must be smaller than source end indices, but source length was {sourceLength}.");
}
if (destinationLength < 0)
{
throw new ArgumentOutOfRangeException($"Destination begin indices must be smaller than destination end indices, but destination length was {destinationLength}.");
}
if (sourceLength != destinationLength)
{
throw new ArgumentException($"Source and destination indices length must batch, but source length was {sourceLength} and destination legnth was {destinationLength}.");
}
Array.Copy(fillerData.Data, sourceOffset, arrayToFillData.Data, destinationOffset, sourceLength);
arrayToFillData.CopyFromHostToDevice();
}
/// <inheritdoc />
public override void Fill(INDArray filler, INDArray arrayToFill)
{
CudaSigmaDiffDataBuffer <float> fillerData = (CudaSigmaDiffDataBuffer <float>)InternaliseArray(filler).Data;
CudaSigmaDiffDataBuffer <float> arrayToFillData = (CudaSigmaDiffDataBuffer <float>)InternaliseArray(arrayToFill).Data;
arrayToFillData.SetValues(fillerData.Data, fillerData.Offset, 0, Math.Min(arrayToFill.Length, filler.Length));
arrayToFillData.CopyFromHostToDevice();
}
private void PrepareCudaBuffer(CudaContext cudaContext, T[] data, long offset, long length)
{
CudaContext = cudaContext;
_cudaZero = new SizeT(0);
_cudaOffsetBytes = new SizeT(offset * Type.SizeBytes);
_cudaLengthBytes = new SizeT(length * Type.SizeBytes);
_cudaContextDeviceId = cudaContext.DeviceId;
_underlyingCudaBuffer = GetUnderlyingBuffer() as CudaSigmaDiffDataBuffer <T>;
}
/// <summary>
/// Shallow copy constructor with the same system- and device memory buffers.
/// </summary>
/// <param name="other"></param>
internal CudaSigmaDiffDataBuffer(CudaSigmaDiffDataBuffer <T> other) : base(other, other.BackendTag)
{
_cudaBuffer = other._cudaBuffer;
CudaContext = other.CudaContext;
_cudaOffsetBytes = other._cudaOffsetBytes;
_cudaLengthBytes = other._cudaLengthBytes;
_flagDeviceModified = other._flagDeviceModified;
_flagHostModified = other._flagHostModified;
}
/// <inheritdoc />
protected override INDArray _RowWiseDirect(INDArray array, CudaFloat32NDArray internalArray, Func <INDArray, INDArray> function)
{
INDArray result = base._RowWiseDirect(array, internalArray, function);
// TODO optimise ofRows operation for CUDA on device
// flag host array modified because for some reason the host modification logic isn't triggered from F# SigmaDiff when appling ofRows (using System.Array.Copy)
CudaSigmaDiffDataBuffer <float> resultBuffer = (CudaSigmaDiffDataBuffer <float>)InternaliseArray(result).Data;
resultBuffer.FlagHostModified();
return(result);
}
/// <inheritdoc />
public override void Fill <TOther>(TOther value, INDArray arrayToFill)
{
CudaSigmaDiffDataBuffer <float> arrayToFillData = (CudaSigmaDiffDataBuffer <float>)InternaliseArray(arrayToFill).Data;
float floatValue = (float)System.Convert.ChangeType(value, typeof(float));
for (int i = 0; i < arrayToFillData.Length; i++)
{
arrayToFillData.Data.SetValue(floatValue, i);
}
arrayToFillData.CopyFromHostToDevice();
}
/// <inheritdoc />
public override void Fill <T>(T[] filler, INDArray arrayToFill, long[] destinationBeginIndices, long[] destinationEndIndices)
{
CudaSigmaDiffDataBuffer <float> arrayToFillData = (CudaSigmaDiffDataBuffer <float>)InternaliseArray(arrayToFill).Data;
int destinationOffset = (int)NDArrayUtils.GetFlatIndex(arrayToFill.Shape, arrayToFill.Strides, destinationBeginIndices);
int destinationLength = (int)NDArrayUtils.GetFlatIndex(arrayToFill.Shape, arrayToFill.Strides, destinationEndIndices) - destinationOffset + 1; // +1 because end is inclusive
if (destinationLength < 0)
{
throw new ArgumentOutOfRangeException($"Destination begin indices must be smaller than destination end indices, but destination length was {destinationLength}.");
}
Array.Copy(filler, 0, arrayToFillData.Data, destinationOffset, destinationLength);
arrayToFillData.CopyFromHostToDevice();
}
/// <inheritdoc />
public override IDataBuffer <T> GetValues(long startIndex, long length)
{
if (_initialisedInContext && !_flagHostModified)
{
CudaSigmaDiffDataBuffer <T> vData = new CudaSigmaDiffDataBuffer <T>(_data, startIndex, length, BackendTag, CudaContext);
vData.InitialiseCudaBuffer(copyHostToDevice: false);
vData._cudaBuffer.AsyncCopyToDevice(_cudaBuffer.DevicePointer, new SizeT(startIndex * sizeof(float)), vData._cudaZero, vData._cudaLengthBytes,
CudaFloat32Handler.GetStreamForContext(CudaContext));
vData._flagHostModified = false;
vData._flagDeviceModified = false;
return(vData);
}
else
{
OnReadAccess();
return(new CudaSigmaDiffDataBuffer <T>(this, startIndex, length, BackendTag, CudaContext));
}
}
