aclSetRawTensorAddr

Description

Updates the device memory address recorded in the aclTensor created by calling aclCreateTensor.

Typically, if the network needs to frequently reuse aclTensor (that is, the attributes such as shape and format remain the same), you can call this API to update the original device memory address of aclTensor.

Prototype

aclnnStatus aclSetRawTensorAddr(aclTensor *tensor, void *addr)

Parameters

Parameter	Input/Output	Description
tensor	Input	Pointer to the aclTensor to be updated.
addr	Input	Device storage address to be updated to the specified aclTensor. The address must be 32-byte aligned. Otherwise, an undefined error may occur.

Returns

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

Possible causes:

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

Constraints

This API must be called before the first-phase API aclxxXxxGetWorkspaceSize is called or after the second-phase API aclxxXxx is called. It cannot be called between the first-phase and second-phase API calls.
This API can be used together with aclGetRawTensorAddr to check whether the updated result meets the expectation.

Examples

The following sample code is for reference only. Do not copy and run it.

// Create input and output tensors inputTensor and outputTensor.
std::vector<int64_t> shape = {1, 2, 3};
aclTensor inputTensor = aclCreateTensor(shape.data(), shape.size(), aclDataType::ACL_FLOAT, nullptr, 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), nullptr);
aclTensor outputTensor = aclCreateTensor(shape.data(), shape.size(), aclDataType::ACL_FLOAT, nullptr, 0, aclFormat::ACL_FORMAT_ND, shape.data(), shape.size(), nullptr);
// Call the first- and second-phase APIs of the Xxx operator.
auto ret = aclxxXxxGetWorkspaceSize(inputTensor, outputTensor, &workspaceSize, &executor);
ret = aclxxXxx(workspace, workspaceSize, executor, stream);
...

void *addr1;
void *addr2;
... // Allocate device memory addresses addr1 and addr2.
ret = aclSetRawTensorAddr(inputTensor, addr1); // Update the device address of inputTensor.
ret = aclSetRawTensorAddr(outputTensor, addr2); // Update the device address of outputTensor.
...
// After inputTensor and outputTensor are reused, call the first- and second-phase APIs of the Yyy operator in aclnn.
auto ret = aclnnYyyGetWorkspaceSize(inputTensor, outputTensor, &workspaceSize, &executor);
ret = aclnnYyy(workspace, workspaceSize, executor, stream);
...

Parent topic: Basic APIs