aclGetDataType

Function Usage

Obtains the DataType of the aclTensor created by calling aclCreateTensor.

Prototype

aclnnStatus aclGetDataType(const aclTensor *tensor, aclDataType *dataType)

Parameters

aclDataType is an enumeration class of data types defined by the framework. For details, see "aclDataType".

Parameter

Input/Output

Description

tensor

Input

Input aclTensor.

dataType

Output

Returned data type of the aclTensor.

Returns

0 on success; else, failure. For details about the return codes, see Common APIs and Return Codes.

Possible causes:

  • If error code 161001 is returned, tensor or dataType is a null pointer.

Constraints

None

Examples

Assume that there is an aclTensor object (xTensor). Obtain its attributes such as the data type, data layout format, dimension, stride, and offset, and create an aclTensor object (yTensor) based on these attributes.

The following code examples are for reference only and are not intended for direct copying and execution:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
// 1. Create an xTensor.
int64_t xViewDims = {2, 4};       
int64_t xStridesValue = {4, 1};  // The stride of dimension 1 is 4, and the stride of dimension 2 is 1.
int64_t xStorageDims = {2, 4};    
xTensor = aclCreateTensor(xViewDims, 2, ACL_FLOAT16, xStridesValue, 0, ACL_FORMAT_ND, xStorageDims, 2, nullptr);

// 2. Obtain the attribute values of xTensor.
// Obtain the logical shape of xTensor. viewDims is {2, 4}, and viewDimsNum is 2.
int64_t *viewDims = nullptr;
uint64_t viewDimsNum = 0;
auto ret = aclGetViewShape(xTensor, &viewDims, &viewDimsNum);
// Obtain the data type (ACL_FLOAT16) of xTensor.
aclDataType dataType = aclDataType::ACL_DT_UNDEFINED;
ret = aclGetDataType(xTensor, &dataType);
//Obtain the stride information about xTensor. stridesValue is {4, 1}, and stridesNum is 2.
int64_t *stridesValue = nullptr;
uint64_t stridesNum = 0;
ret = aclGetViewStrides(xTensor, &stridesValue, &stridesNum);
//Obtain the offset of the first element of xTensor relative to storage. The offset is 0.
int64_t offset = 0;
ret = aclGetViewOffset(xTensor, &offset);
// Obtain the data layout format (ACL_FORMAT_ND) of xTensor.
aclFormat format = aclFormat::ACL_FORMAT_UNDEFINED;
ret = aclGetFormat(xTensor, &format);
// Obtain the actual physical shape of xTensor. storageDims is {2, 4}, and storageDimsNum is 2.
int64_t *storageDims = nullptr;
uint64_t storageDimsNum = 0;
ret = aclGetStorageShape(xTensor, &storageDims, &storageDimsNum);
// Device-side address
void *deviceAddr;

// 3. Create a tensor based on the xTensor attributes.
aclTensor *yTensor = aclCreateTensor(viewDims, viewDimsNum, dataType, stridesValue, offset, format, storageDims, storageDimsNum, deviceAddr);

// 4. Manually free the memory.
delete[] viewDims;
delete[] stridesValue;
delete[] storageDims;