ListTensorDesc

Applicability

Product	Supported
Atlas 350 Accelerator Card	x
Atlas A3 training product/Atlas A3 inference product	√
Atlas A2 training product/Atlas A2 inference product	√
Atlas 200I/500 A2 inference product	x
Atlas inference product AI Core	√
Atlas inference product Vector Core	x
Atlas training product	x

Function Usage

Parses the data that meets the following memory layouts and obtains the address and shape information of the corresponding data based on the index on the kernel.

Header File to Be Included

#include "kernel_operator_list_tensor_intf.h"

Prototype

class ListTensorDesc {
    ListTensorDesc();
    ListTensorDesc(__gm__ void* data, uint32_t length = 0xffffffff, uint32_t shapeSize = 0xffffffff);
    void Init(__gm__ void* data, uint32_t length = 0xffffffff, uint32_t shapeSize = 0xffffffff);
    template<class T> void GetDesc(TensorDesc<T>& desc, uint32_t index);
    template<class T> T* GetDataPtr(uint32_t index);
    uint32_t GetSize();
}

Functions

**Table 1** Template parameters
Parameter	Description
T	Data type of tensor elements.

**Table 2** Function and parameter description
Function Name	Input Parameter	Description
ListTensorDesc	-	Default constructor, which must be used together with the Init function.
ListTensorDesc	data: initial address of the data to be parsed length: length of the memory to be parsed shapeSize: number of data pointers length and shapeSize are used only for verification. If their values are not set, verification is not performed.	Constructor of the ListTensorDesc class, which is used to parse the corresponding memory layout.
Init	data: initial address of the data to be parsed length: length of the memory to be parsed shapeSize: number of data pointers length and shapeSize are used only for verification. If their values are not set, verification is not performed.	Initialization function, which is used to parse the corresponding memory layout.
GetDesc	desc: output parameter, description of the parsed tensor index: index value	Obtains the corresponding TensorDesc information in the function usage diagram based on index. Before using GetDesc, call TensorDesc.SetShapeAddr to specify the address for storing the shape information for desc. After GetDesc is called, the shape information is written to this address. For the Atlas inference product AI Core, this function is supported. For the Atlas training product, this function is not supported. For the Atlas A2 training product/Atlas A2 inference product, this function is supported. For the Atlas A3 training product/Atlas A3 inference product, this function is supported. For the Atlas 200I/500 A2 inference product, this function is not supported.
GetDataPtr	index: index value	Obtains the address for storing the corresponding data.
GetSize	-	Obtains the number of data pointers contained in ListTensor.

Examples

The following figure shows the memory layout of srcGm to be parsed.

AscendC::ListTensorDesc listTensorDesc(reinterpret_cast<__gm__ void *>(srcGm)); // srcGm indicates the GM address to be parsed.
uint32_t size = listTensorDesc.GetSize();                                       // size = 2
auto dataPtr0 = listTensorDesc.GetDataPtr<int32_t>(0);                          // Obtains ptr0.
auto dataPtr1 = listTensorDesc.GetDataPtr<int32_t>(1);                          // Obtains ptr1.

uint64_t buf[100] = {0}; // In the example, dim of the tensor is 3, and the value 100 indicates that sufficient space is reserved.
AscendC::TensorDesc<int32_t> desc;
desc.SetShapeAddr(buf);          // Specifies the address for storing shape information for desc.
listTensorDesc.GetDesc(desc, 0); // Obtains the shape information of index 0.

uint64_t dim = desc.GetDim();   // dim = 3
uint64_t idx = desc.GetIndex(); // idx = 0
uint64_t shape[3] = {0};
for (uint32_t i = 0; i < desc.GetDim(); i++)
{
    shape[i] = desc.GetShape(i); // GetShape(0) = 1, GetShape(1) = 2, GetShape(2) = 3
}
auto ptr = desc.GetDataPtr();

Parent topic: Other Data Types