public void Im2Col(float alpha,
TensorDescriptor srcDesc,
CudaDeviceVariable <float> srcData,
FilterDescriptor filterDesc,
ConvolutionDescriptor convDesc,
CudaDeviceVariable <byte> colBuffer
)
{
res = CudaDNNNativeMethods.cudnnIm2Col(_handle, ref alpha, srcDesc.Desc, srcData.DevicePointer, filterDesc.Desc, convDesc.Desc, colBuffer.DevicePointer);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnIm2Col", res));
if (res != cudnnStatus.Success)
{
throw new CudaDNNException(res);
}
}
public void GetConvolutionForwardAlgorithm(TensorDescriptor srcDesc,
FilterDescriptor filterDesc,
ConvolutionDescriptor convDesc,
TensorDescriptor destDesc,
cudnnConvolutionFwdPreference preference,
SizeT memoryLimitInbytes,
ref cudnnConvolutionFwdAlgo algo
)
{
res = CudaDNNNativeMethods.cudnnGetConvolutionForwardAlgorithm(_handle, srcDesc.Desc, filterDesc.Desc, convDesc.Desc, destDesc.Desc, preference, memoryLimitInbytes, ref algo);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnGetConvolutionForwardAlgorithm", res));
if (res != cudnnStatus.Success)
{
throw new CudaDNNException(res);
}
}
public SizeT GetConvolutionForwardWorkspaceSize(TensorDescriptor srcDesc,
FilterDescriptor filterDesc,
ConvolutionDescriptor convDesc,
TensorDescriptor destDesc,
cudnnConvolutionFwdAlgo algo
)
{
SizeT sizeInBytes = 0;
res = CudaDNNNativeMethods.cudnnGetConvolutionForwardWorkspaceSize(_handle, srcDesc.Desc, filterDesc.Desc, convDesc.Desc, destDesc.Desc, algo, ref sizeInBytes);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnGetConvolutionForwardWorkspaceSize", res));
if (res != cudnnStatus.Success)
{
throw new CudaDNNException(res);
}
return(sizeInBytes);
}
public void ConvolutionBackwardData(double alpha,
FilterDescriptor filterDesc,
CudaDeviceVariable <double> filterData,
TensorDescriptor diffDesc,
CudaDeviceVariable <double> diffData,
ConvolutionDescriptor convDesc,
double beta,
TensorDescriptor gradDesc,
CudaDeviceVariable <double> gradData
)
{
res = CudaDNNNativeMethods.cudnnConvolutionBackwardData(_handle, ref alpha, filterDesc.Desc, filterData.DevicePointer, diffDesc.Desc, diffData.DevicePointer, convDesc.Desc, ref beta, gradDesc.Desc, gradData.DevicePointer);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnConvolutionBackwardData", res));
if (res != cudnnStatus.Success)
{
throw new CudaDNNException(res);
}
}
/* Convolution functions: All of the form "output = alpha * Op(inputs) + beta * output" */
/* Function to perform the forward multiconvolution */
public void ConvolutionForward(double alpha,
TensorDescriptor srcDesc,
CudaDeviceVariable <double> srcData,
FilterDescriptor filterDesc,
CudaDeviceVariable <double> filterData,
ConvolutionDescriptor convDesc,
cudnnConvolutionFwdAlgo algo,
CudaDeviceVariable <byte> workSpace,
SizeT workSpaceSizeInBytes,
double beta,
TensorDescriptor destDesc,
CudaDeviceVariable <double> destData
)
{
res = CudaDNNNativeMethods.cudnnConvolutionForward(_handle, ref alpha, srcDesc.Desc, srcData.DevicePointer, filterDesc.Desc, filterData.DevicePointer, convDesc.Desc, algo, workSpace.DevicePointer, workSpaceSizeInBytes, ref beta, destDesc.Desc, destData.DevicePointer);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnConvolutionForward", res));
if (res != cudnnStatus.Success)
{
throw new CudaDNNException(res);
}
}
public void Im2Col(double alpha,
TensorDescriptor srcDesc,
CudaDeviceVariable<double> srcData,
FilterDescriptor filterDesc,
ConvolutionDescriptor convDesc,
CudaDeviceVariable<byte> colBuffer
)
{
res = CudaDNNNativeMethods.cudnnIm2Col(_handle, ref alpha, srcDesc.Desc, srcData.DevicePointer, filterDesc.Desc, convDesc.Desc, colBuffer.DevicePointer);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnIm2Col", res));
if (res != cudnnStatus.Success) throw new CudaDNNException(res);
}
public SizeT GetConvolutionForwardWorkspaceSize(TensorDescriptor srcDesc,
FilterDescriptor filterDesc,
ConvolutionDescriptor convDesc,
TensorDescriptor destDesc,
cudnnConvolutionFwdAlgo algo
)
{
SizeT sizeInBytes = 0;
res = CudaDNNNativeMethods.cudnnGetConvolutionForwardWorkspaceSize(_handle, srcDesc.Desc, filterDesc.Desc, convDesc.Desc, destDesc.Desc, algo, ref sizeInBytes);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnGetConvolutionForwardWorkspaceSize", res));
if (res != cudnnStatus.Success) throw new CudaDNNException(res);
return sizeInBytes;
}
public void GetConvolutionForwardAlgorithm(TensorDescriptor srcDesc,
FilterDescriptor filterDesc,
ConvolutionDescriptor convDesc,
TensorDescriptor destDesc,
cudnnConvolutionFwdPreference preference,
SizeT memoryLimitInbytes,
ref cudnnConvolutionFwdAlgo algo
)
{
res = CudaDNNNativeMethods.cudnnGetConvolutionForwardAlgorithm(_handle, srcDesc.Desc, filterDesc.Desc, convDesc.Desc, destDesc.Desc, preference, memoryLimitInbytes, ref algo);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnGetConvolutionForwardAlgorithm", res));
if (res != cudnnStatus.Success) throw new CudaDNNException(res);
}
/* Convolution functions: All of the form "output = alpha * Op(inputs) + beta * output" */
/* Function to perform the forward multiconvolution */
public void ConvolutionForward(double alpha,
TensorDescriptor srcDesc,
CudaDeviceVariable<double> srcData,
FilterDescriptor filterDesc,
CudaDeviceVariable<double> filterData,
ConvolutionDescriptor convDesc,
cudnnConvolutionFwdAlgo algo,
CudaDeviceVariable<byte> workSpace,
SizeT workSpaceSizeInBytes,
double beta,
TensorDescriptor destDesc,
CudaDeviceVariable<double> destData
)
{
res = CudaDNNNativeMethods.cudnnConvolutionForward(_handle, ref alpha, srcDesc.Desc, srcData.DevicePointer, filterDesc.Desc, filterData.DevicePointer, convDesc.Desc, algo, workSpace.DevicePointer, workSpaceSizeInBytes, ref beta, destDesc.Desc, destData.DevicePointer);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnConvolutionForward", res));
if (res != cudnnStatus.Success) throw new CudaDNNException(res);
}
public void ConvolutionBackwardFilter(double alpha,
TensorDescriptor srcDesc,
CudaDeviceVariable<double> srcData,
TensorDescriptor diffDesc,
CudaDeviceVariable<double> diffData,
ConvolutionDescriptor convDesc,
double beta,
FilterDescriptor gradDesc,
CudaDeviceVariable<double> gradData
)
{
res = CudaDNNNativeMethods.cudnnConvolutionBackwardFilter(_handle, ref alpha, srcDesc.Desc, srcData.DevicePointer, diffDesc.Desc, diffData.DevicePointer, convDesc.Desc, ref beta, gradDesc.Desc, gradData.DevicePointer);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnConvolutionBackwardFilter", res));
if (res != cudnnStatus.Success) throw new CudaDNNException(res);
}
/// <summary>
/// This function computes the convolution gradient with respect to the output tensor using
/// the specified algo, returning results in gradDesc. Scaling factors alpha and beta can
/// be used to scale the input tensor and the output tensor respectively.
/// </summary>
/// <param name="alpha">Pointer to scaling factors (in host memory) used to blend the computation
/// result with prior value in the output layer as follows: dstValue =
/// alpha[0]*result + beta[0]*priorDstValue. Please refer to this section for
/// additional details.</param>
/// <param name="filterDesc">Handle to a previously initialized filter descriptor.</param>
/// <param name="filterData">Data pointer to GPU memory associated with the filter descriptor filterDesc.</param>
/// <param name="diffDesc">Handle to the previously initialized input differential tensor descriptor.</param>
/// <param name="diffData">Data pointer to GPU memory associated with the input differential tensor descriptor diffDesc.</param>
/// <param name="convDesc">Previously initialized convolution descriptor.</param>
/// <param name="algo">Enumerant that specifies which backward data convolution algorithm shoud be used to compute the results</param>
/// <param name="workSpace">Data pointer to GPU memory to a workspace needed to able to execute
/// the specified algorithm. If no workspace is needed for a particular
/// algorithm, that pointer can be nil</param>
/// <param name="beta">Pointer to scaling factors (in host memory) used to blend the computation
/// result with prior value in the output layer as follows: dstValue =
/// alpha[0]*result + beta[0]*priorDstValue. Please refer to this section for
/// additional details.</param>
/// <param name="gradDesc">Handle to the previously initialized output tensor descriptor.</param>
/// <param name="gradData">Data pointer to GPU memory associated with the output tensor descriptor
/// gradDesc that carries the result.</param>
public void ConvolutionBackwardData(double alpha,
FilterDescriptor filterDesc,
CudaDeviceVariable<double> filterData,
TensorDescriptor diffDesc,
CudaDeviceVariable<double> diffData,
ConvolutionDescriptor convDesc,
cudnnConvolutionBwdDataAlgo algo,
CudaDeviceVariable<byte> workSpace,
double beta,
TensorDescriptor gradDesc,
CudaDeviceVariable<double> gradData
)
{
res = CudaDNNNativeMethods.cudnnConvolutionBackwardData(_handle, ref alpha, filterDesc.Desc, filterData.DevicePointer, diffDesc.Desc, diffData.DevicePointer, convDesc.Desc, algo, workSpace.DevicePointer, workSpace.SizeInBytes, ref beta, gradDesc.Desc, gradData.DevicePointer);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnConvolutionBackwardData", res));
if (res != cudnnStatus.Success) throw new CudaDNNException(res);
}
/// <summary>
/// This function returns the amount of GPU memory workspace the user needs
/// to allocate to be able to call cudnnConvolutionBackwardData_v3 with the
/// specified algorithm. The workspace allocated will then be passed to the routine
/// cudnnConvolutionBackwardData_v3. The specified algorithm can be the result of the
/// call to cudnnGetConvolutionBackwardDataAlgorithm or can be chosen arbitrarily
/// by the user. Note that not every algorithm is available for every configuration of the
/// input tensor and/or every configuration of the convolution descriptor.
/// </summary>
/// <param name="filterDesc">Handle to a previously initialized filter descriptor.</param>
/// <param name="diffDesc">Handle to the previously initialized input differential tensor descriptor.</param>
/// <param name="convDesc">Previously initialized convolution descriptor.</param>
/// <param name="gradDesc">Handle to the previously initialized output tensor descriptor.</param>
/// <param name="algo">Enumerant that specifies the chosen convolution algorithm</param>
/// <returns>Amount of GPU memory needed as workspace to be able to execute a forward convolution with the specified algo</returns>
public SizeT GetConvolutionBackwardDataWorkspaceSize(FilterDescriptor filterDesc,
TensorDescriptor diffDesc,
ConvolutionDescriptor convDesc,
TensorDescriptor gradDesc,
cudnnConvolutionBwdDataAlgo algo
)
{
SizeT sizeInBytes = new SizeT();
res = CudaDNNNativeMethods.cudnnGetConvolutionBackwardDataWorkspaceSize(_handle, filterDesc.Desc, diffDesc.Desc, convDesc.Desc, gradDesc.Desc, algo, ref sizeInBytes);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnGetConvolutionBackwardDataWorkspaceSize", res));
if (res != cudnnStatus.Success) throw new CudaDNNException(res);
return sizeInBytes;
}
/// <summary>
/// This function serves as a heuristic for obtaining the best suited algorithm for
/// cudnnConvolutionBackwardData_v3 for the given layer specifications. Based
/// on the input preference, this function will either return the fastest algorithm or the
/// fastest algorithm within a given memory limit. For an exhaustive search for the fastest
/// algorithm, please use cudnnFindConvolutionBackwardDataAlgorithm.
/// </summary>
/// <param name="filterDesc">Handle to a previously initialized filter descriptor.</param>
/// <param name="diffDesc">Handle to the previously initialized input differential tensor descriptor.</param>
/// <param name="convDesc">Previously initialized convolution descriptor.</param>
/// <param name="gradDesc">Handle to the previously initialized output tensor descriptor.</param>
/// <param name="preference">Enumerant to express the preference criteria in terms of memory
/// requirement and speed.</param>
/// <param name="memoryLimitInbytes">It is to specify the maximum amount of GPU memory the user is willing to
/// use as a workspace. This is currently a placeholder and is not used.</param>
/// <returns>Enumerant that specifies which convolution algorithm should be used to
/// compute the results according to the specified preference</returns>
public cudnnConvolutionBwdDataAlgo GetConvolutionBackwardDataAlgorithm(FilterDescriptor filterDesc,
TensorDescriptor diffDesc,
ConvolutionDescriptor convDesc,
TensorDescriptor gradDesc,
cudnnConvolutionBwdDataPreference preference,
SizeT memoryLimitInbytes
)
{
cudnnConvolutionBwdDataAlgo algo = new cudnnConvolutionBwdDataAlgo();
res = CudaDNNNativeMethods.cudnnGetConvolutionBackwardDataAlgorithm(_handle, filterDesc.Desc, diffDesc.Desc, convDesc.Desc, gradDesc.Desc, preference, memoryLimitInbytes, ref algo);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnGetConvolutionBackwardDataAlgorithm", res));
if (res != cudnnStatus.Success) throw new CudaDNNException(res);
return algo;
}
/// <summary>
/// This function attempts all cuDNN algorithms for
/// cudnnConvolutionBackwardData_v3 and outputs performance metrics to a user-
/// allocated array of cudnnConvolutionBwdDataAlgoPerf_t. These metrics are written
/// in sorted fashion where the first element has the lowest compute time.
/// </summary>
/// <param name="filterDesc">Handle to a previously initialized filter descriptor.</param>
/// <param name="diffDesc">Handle to the previously initialized input differential tensor descriptor.</param>
/// <param name="convDesc">Previously initialized convolution descriptor.</param>
/// <param name="gradDesc">Handle to the previously initialized output tensor descriptor.</param>
/// <param name="requestedAlgoCount">The maximum number of elements to be stored in perfResults.</param>
/// <returns>An array to store performance metrics sorted ascending by compute time.</returns>
public cudnnConvolutionBwdDataAlgoPerf[] FindConvolutionBackwardDataAlgorithm(FilterDescriptor filterDesc,
TensorDescriptor diffDesc,
ConvolutionDescriptor convDesc,
TensorDescriptor gradDesc,
int requestedAlgoCount
)
{
cudnnConvolutionBwdDataAlgoPerf[] temp = new cudnnConvolutionBwdDataAlgoPerf[requestedAlgoCount];
int returnedAlgoCount = 0;
res = CudaDNNNativeMethods.cudnnFindConvolutionBackwardDataAlgorithm(_handle, filterDesc.Desc, diffDesc.Desc, convDesc.Desc, gradDesc.Desc, requestedAlgoCount, ref returnedAlgoCount, temp);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnFindConvolutionBackwardDataAlgorithm", res));
if (res != cudnnStatus.Success) throw new CudaDNNException(res);
if (returnedAlgoCount <= 0) return null;
cudnnConvolutionBwdDataAlgoPerf[] perfResults = new cudnnConvolutionBwdDataAlgoPerf[returnedAlgoCount];
Array.Copy(temp, perfResults, returnedAlgoCount);
return perfResults;
}
