aclGetStorageShape

Function Usage

Obtains the StorageShape of the aclTensor created by calling aclCreateTensor.

StorageShape indicates the actual physical layout shape of the aclTensor, which is the actual size of the tensor in the memory. If StorageShape is [10, 20], the aclTensor is arranged in the memory based on [10, 20].

Prototype

aclnnStatus aclGetStorageShape(const aclTensor *tensor, int64_t **storageDims, uint64_t *storageDimsNum)

Parameters

Parameter	Input/Output	Description
tensor	Input	Address of the input aclTensor. The aclTensor must be created by calling aclCreateTensor in advance.
storageDims	Output	Dimension value of StorageShape.
storageDimsNum	Output	Number of dimensions of StorageShape.

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, storageDims, or storageDimsNum is a null pointer.

Constraints

The memory for the storageDims parameter is internally allocated by this API. After use, you need to delete the memory to manually release it.

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. 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;

Parent topic: Common APIs