Instructions

Interface Definition

Initialize SoftmaxLayerFWD. During the construction, the tensor information of the input matrix, weight matrix, and output matrix needs to be passed. The bias matrix is optional. If it is not passed, the second constructor will be called for initialization.

To perform operator computation, the memory addresses for storing the input and output data must be passed.

Run(const void *src, void *dst)->void

ValidateInput verifies the input parameters of SoftMax. It is automatically triggered during operator construction.

ValidateInput(const TensorInfo &srcInfo, const TensorInfo &dstInfo, SizeType axis, SoftmaxAlgorithmKind algorithm)->KuDNN::Status

Supported Data Types

• SoftMax supports the FP32 data type. (The Shape, Type, and Layout parameters must be passed during TensorInfo object initialization. The following lists the data types supported by Type.)

  Table 4 Type supported during TensorInfo object initialization

  srcInfo	dstInfo
  KuDNN::Element::TypeT::F32(fp32)	KuDNN::Element::TypeT::F32(fp32)

• A maximum of 5 dimensions are supported. The supported sequential data layouts include a, ab, abc, abcd, and abcde.

  They correspond to KuDNN::Layout::A, KuDNN::Layout::AB, KuDNN::Layout::ABC, KuDNN::Layout::ABCD, and KuDNN::Layout::ABCDE, respectively.

  Table 5 Layout supported during TensorInfo object initialization

  Dimension	srcInfo	dstInfo
  1D	a	a
  2D	ab	ab
  3D	abc	abc
  4D	abcd	abcd
  5D	abcde	abcde

Examples

SoftMax operation with two-dimensional data of the F32 type:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17

// Example
using SizeType = KuDNN::SizeType;
using Shape = KuDNN::Shape;
using Type KuDNN::Element::TypeT
// Tensor initialization
const KuDNN::TensorInfo srcTensor= {{3, 2}, KuDNN::Element::TypeT::F32, KuDNN::Layout::AB};
const KuDNN::TensorInfo dstTensor= {{3, 2}, KuDNN::Element::TypeT::F32, KuDNN::Layout::AB};
SizeType axis = 1;KuDNN::SoftmaxAlgorithmKind algKind = KuDNN::SoftmaxAlgorithmKind::SOFTMAX
SizeType srcSize = srcATensor.GetTotalTensorSize();
SizeType dstSize = weiBTensor.GetTotalTensorSize();
// Allocate memory for storing the input parameters and result.
float * src = malloc(srcSize* sizeof(float));
float * dst = malloc(dstSize* sizeof(float));
// Construct the operator.
KuDNN::SoftmaxLayerFWD softMaxLayerFwd(srcTensor, dstTensor, axis, algKind);
// Run the operator.
softMaxLayerFwd.Run(src, dst);