mkldnn

Type Members

1. class AvgPooling extends AbstractModule[Activity, Activity, Float] with MklDnnLayer

   

2. class CAddTable extends AbstractModule[Activity, Activity, Float] with MklDnnLayer with MklInt8Convertible

   

3. class ConcatTable extends DynamicContainer[Activity, Activity, Float] with MklDnnContainer with MklInt8Convertible

   

4. class DnnGraph extends Graph[Float] with MklDnnLayer with MklInt8Convertible

   

5. class Dropout extends AbstractModule[Activity, Activity, Float] with MklDnnLayer

   

6. case class HeapData(_shape: Array[Int], _layout: Int, _dataType: Int = DataType.F32) extends MemoryData with Product with Serializable

   

7. class Identity extends AbstractModule[Activity, Activity, Float] with MklDnnLayer

   

   Identity just return the input to output.

   Identity just return the input to output. It's useful in same parallel container to get an origin input.

8. class Input extends ReorderMemory

   

9. class JoinTable extends AbstractModule[Activity, Activity, Float] with MklDnnLayer

   

10. class LRN extends AbstractModule[Activity, Activity, Float] with MklDnnLayer

    

11. class Linear extends AbstractModule[Activity, Activity, Float] with MklDnnLayer with Initializable with MklInt8Convertible

    

12. class MaxPooling extends AbstractModule[Activity, Activity, Float] with MklDnnLayer

    

13. sealed trait MemoryData extends Serializable

    

14. trait MklDnnContainer extends DynamicContainer[Activity, Activity, Float] with MklDnnModule

    

    Helper utilities when integrating containers with MKL-DNN

15. trait MklDnnLayer extends AbstractModule[Activity, Activity, Float] with MklDnnModule

    

16. trait MklDnnModule extends MklDnnModuleHelper

    

    Helper utilities when integrating Module with MKL-DNN

17. trait MklDnnModuleHelper extends MemoryOwner

    

18. abstract class MklDnnNativeMemory extends Releasable

    

19. class MklDnnRuntime extends AnyRef

    

20. class MklMemoryAttr extends MklDnnNativeMemory

    

21. class MklMemoryDescInit extends MklDnnNativeMemory

    

22. class MklMemoryPostOps extends MklDnnNativeMemory

    

23. class MklMemoryPrimitive extends MklDnnNativeMemory

    

24. class MklMemoryPrimitiveDesc extends MklDnnNativeMemory

    

25. case class NativeData(_shape: Array[Int], _layout: Int, _dataType: Int = DataType.F32) extends MemoryData with Product with Serializable

    

26. class Output extends AbstractModule[Activity, Activity, Float] with MklDnnLayer

    

    Convert output to user defined layout and appoint gradOutput layout

27. sealed class Phase extends AnyRef

    

28. class RNN extends AbstractModule[Activity, Activity, Float] with MklDnnLayer with Initializable

    

29. class ReLU extends AbstractModule[Activity, Activity, Float] with MklDnnLayer with MklInt8Convertible

    

30. class ReorderMemory extends AbstractModule[Activity, Activity, Float] with MklDnnLayer with Releasable

    

31. case class ResNet50PerfParams(batchSize: Int = 16, iteration: Int = 50, training: Boolean = true, model: String = "vgg16") extends Product with Serializable

    

32. class SelectTable extends AbstractModule[Activity, Activity, Float] with MklDnnLayer

    

    Creates a module that takes a table as input and outputs the element at index index (positive or negative).

    Creates a module that takes a table as input and outputs the element at index index (positive or negative). This can be either a table or a Tensor. The gradients of the non-index elements are zeroed Tensors of the same size. This is true regardless of the depth of the encapsulated Tensor as the function used internally to do so is recursive.

    Annotations

    @SerialVersionUID()

33. class Sequential extends DynamicContainer[Activity, Activity, Float] with MklDnnContainer with MklInt8Convertible

    

34. class SoftMax extends AbstractModule[Activity, Activity, Float] with MklDnnLayer

    

35. class SpatialBatchNormalization extends AbstractModule[Activity, Activity, Float] with MklDnnLayer with Initializable with MklInt8Convertible

    

36. class SpatialConvolution extends AbstractModule[Activity, Activity, Float] with MklDnnLayer with Initializable with Serializable with MklInt8Convertible

    

    Applies a 2D convolution over an input image composed of several input planes.

    Applies a 2D convolution over an input image composed of several input planes. The input tensor in forward(input) is expected to be a 3D tensor (nInputPlane x height x width).

    nInputPlane The number of expected input planes in the image given into forward() nOutputPlane: The number of output planes the convolution layer will produce. kernelW: the kernel width of the convolution kernelH: The kernel height of the convolution strideW: Int = 1, The step of the convolution in the width dimension. strideH: Int = 1, The step of the convolution in the height dimension padW: Int = 0, The additional zeros added per width to the input planes. padH: Int = 0, The additional zeros added per height to the input planes. nGroup: Int = 1, Kernel group number propagateBack: Boolean = true, propagate gradient back wRegularizer: Regularizer[Float] = null, bRegularizer: Regularizer[Float] = null, initWeight: Tensor[Float] = null, initBias: Tensor[Float] = null, initGradWeight: Tensor[Float] = null, initGradBias: Tensor[Float] = null, withBias: Boolean = true, format: DataFormat = DataFormat.NCHW, dilationW: Int = 1, dilationH: Int = 1

    When padW and padH are both -1, we use a padding algorithm similar to the "SAME" padding of tensorflow. That is

    outHeight = Math.ceil(inHeight.toFloat/strideH.toFloat) outWidth = Math.ceil(inWidth.toFloat/strideW.toFloat)

    padAlongHeight = Math.max(0, (outHeight - 1) * strideH + kernelH - inHeight) padAlongWidth = Math.max(0, (outWidth - 1) * strideW + kernelW - inWidth)

    padTop = padAlongHeight / 2 padLeft = padAlongWidth / 2