|
private cudnnDivisiveNormalizationBackward ( |
||
| handle | ||
| normDesc | ||
| mode | cudnnDivNormMode | |
| alpha | double | |
| srcDesc | ||
| srcData | ManagedCuda.BasicTypes.CUdeviceptr | |
| srcMeansData | ManagedCuda.BasicTypes.CUdeviceptr | |
| srcDiffData | ManagedCuda.BasicTypes.CUdeviceptr | |
| tempData | ManagedCuda.BasicTypes.CUdeviceptr | |
| tempData2 | ManagedCuda.BasicTypes.CUdeviceptr | |
| beta | double | |
| destDataDesc | ||
| destDataDiff | ManagedCuda.BasicTypes.CUdeviceptr | |
| destMeansDiff | ManagedCuda.BasicTypes.CUdeviceptr | |
| Résultat | cudnnStatus | |
public void cudnnDivisiveNormalizationBackward(
cudnnDivNormMode mode,
double alpha,
cudnnTensorDescriptor srcDesc, // same desc for diff, means, temp, temp2
CUdeviceptr srcData,
CUdeviceptr srcMeansData, // if NULL, means are assumed to be zero
CUdeviceptr srcDiffData,
CUdeviceptr tempData,
CUdeviceptr tempData2,
double betaData,
cudnnTensorDescriptor destDataDesc, // same desc for dest, means, meansDiff
CUdeviceptr destDataDiff, // output data differential
CUdeviceptr destMeansDiff // output means differential, can be NULL
)
{
res = CudaDNNNativeMethods.cudnnDivisiveNormalizationBackward(_handle, _desc, mode, ref alpha, srcDesc, srcData, srcMeansData, srcDiffData, tempData, tempData2, ref betaData, destDataDesc, destDataDiff, destMeansDiff);
Debug.WriteLine(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnDivisiveNormalizationBackward", res));
if (res != cudnnStatus.Success)
{
throw new CudaDNNException(res);
}
}
/// <summary>
/// This function performs the backward DivisiveNormalization layer computation.
/// </summary>
/// <param name="mode">DivisiveNormalization layer mode of operation. Currently only
/// CUDNN_DIVNORM_PRECOMPUTED_MEANS is implemented. Normalization
/// is performed using the means input tensor that is expected to be
/// precomputed by the user.</param>
/// <param name="alpha">Pointer to scaling factors (in host memory) used to blend the layer output
/// value with prior value in the destination tensor as follows: dstValue =
/// alpha[0]*resultValue + beta[0]*priorDstValue. Please refer to this section
/// for additional details.</param>
/// <param name="xDesc">Tensor descriptor and pointers in device memory for the bottom layer's
/// data and means. (Bottom layer is the earlier layer in the computation
/// graph during inference). Note: the means tensor is expected to be
/// precomputed by the user. It can also contain any valid values (not required
/// to be actual means, and can be for instance a result of a convolution with
/// a Gaussian kernel).</param>
/// <param name="x">Tensor descriptor and pointers in device memory for the bottom layer's
/// data and means. (Bottom layer is the earlier layer in the computation
/// graph during inference). Note: the means tensor is expected to be
/// precomputed by the user. It can also contain any valid values (not required
/// to be actual means, and can be for instance a result of a convolution with
/// a Gaussian kernel).</param>
/// <param name="means">Tensor descriptor and pointers in device memory for the bottom layer's
/// data and means. (Bottom layer is the earlier layer in the computation
/// graph during inference). Note: the means tensor is expected to be
/// precomputed by the user. It can also contain any valid values (not required
/// to be actual means, and can be for instance a result of a convolution with
/// a Gaussian kernel).</param>
/// <param name="dy">Tensor pointer in device memory for the top layer's cumulative loss
/// differential data (error backpropagation). (Top layer is the later layer in
/// the computation graph during inference).</param>
/// <param name="temp">Temporary tensors in device memory. These are used for computing
/// intermediate values during the backward pass. These tensors do not have
/// to be preserved from forward to backward pass. Both use srcDesc as a
/// descriptor.</param>
/// <param name="temp2">Temporary tensors in device memory. These are used for computing
/// intermediate values during the backward pass. These tensors do not have
/// to be preserved from forward to backward pass. Both use srcDesc as a
/// descriptor.</param>
/// <param name="beta">Pointer to scaling factors (in host memory) used to blend the layer output
/// value with prior value in the destination tensor as follows: dstValue =
/// alpha[0]*resultValue + beta[0]*priorDstValue. Please refer to this section
/// for additional details.</param>
/// <param name="dXdMeansDesc">Tensor descriptor for destDataDiff and destMeansDiff.</param>
/// <param name="dx">Tensor pointers (in device memory) for the bottom layer's resulting
/// differentials (data and means). Both share the same descriptor.</param>
/// <param name="dMeans">Tensor pointers (in device memory) for the bottom layer's resulting
/// differentials (data and means). Both share the same descriptor.</param>
public void cudnnDivisiveNormalizationBackward(
cudnnDivNormMode mode,
float alpha,
cudnnTensorDescriptor xDesc, // same desc for diff, means, temp, temp2
CUdeviceptr x,
CUdeviceptr means, // if NULL, means are assumed to be zero
CUdeviceptr dy,
CUdeviceptr temp,
CUdeviceptr temp2,
float beta,
cudnnTensorDescriptor dXdMeansDesc, // same desc for dest, means, meansDiff
CUdeviceptr dx, // output data differential
CUdeviceptr dMeans // output means differential, can be NULL
)
{
res = CudaDNNNativeMethods.cudnnDivisiveNormalizationBackward(_handle, _desc, mode, ref alpha, xDesc, x, means, dy, temp, temp2, ref beta, dXdMeansDesc, dx, dMeans);
Debug.Write("");//Line(String.Format("{0:G}, {1}: {2}", DateTime.Now, "cudnnDivisiveNormalizationBackward", res));
if (res != cudnnStatus.Success)
{
throw new CudaDNNException(res);
}
}
